Transportation Research Part B: Zhi-Chun Li, Qian-Wen Guo, William H.K. Lam, S.C. Wong

Transportation Research Part B 78 (2015) 318–340
Contents lists available at ScienceDirect
Transportation Research Part B

journal homepage: www.elsevier.com/locate/trb
Transit technology investment and selection under urban

population volatility: A real option perspective
Zhi-Chun Li a,⇑, Qian-Wen Guo a, William H.K. Lam b,c, S.C. Wong d
a
School of Management, Huazhong University of Science and Technology, Wuhan 430074, China
b
Department of Civil and Environmental Engineering, The Hong Kong Polytechnic University, Kowloon, Hong Kong, China
c
School of Traffic & Transportation, Beijing Jiaotong University, Beijing 100044, China
d
Department of Civil Engineering, The University of Hong Kong, Pokfulam Road, Hong Kong, China
a r t i c l e i n f o a b s t r a c t
Article history: This paper addresses transit technology investment issues under urban population volatil-
Received 23 December 2013 ity using a real option approach. Two important problems are investigated: which transit
Received in revised form 7 May 2015 technology should be selected and when should it be introduced. A real option model is
Accepted 8 May 2015
proposed to incorporate explicitly the effects of transit technology investment on urban
Available online 26 May 2015
spatial structure in terms of households’ residential location choices and housing market.
The trigger population thresholds for investing in a transit technology project and for shift-
Keywords:
ing from a transit technology to another are explored analytically. Comparative static anal-
Transit technology investment and selection
Population volatility
yses of the urban system and transit technology investment are also carried out. It was
Real option approach found that (i) transit technology investment can induce urban sprawl; (ii) ignoring the
Net present value approach effects of transit technology investment on urban spatial equilibrium can lead to a late
Comparative static analysis investment; and (iii) there is a significant difference in the trigger population thresholds
Urban spatial equilibrium for transit technology shift estimated by the net present value approach and the real option
approach.
Ó 2015 Elsevier Ltd. All rights reserved.
1. Introduction
To respond to growing traffic congestion problems, the Chinese government has launched a number of public transporta-
tion infrastructure investment projects in recent years. These projects include constructing metro, light rapid transit (LRT),
and bus rapid transit (BRT) lines. For instance, a recent statistical report by Liu et al. (2015) showed that by the end of 2014,
83 metro lines with a total length of over 2500 km had been constructed in 22 metropolitan cities in mainland China, and
seven cities had built LRT systems with a total length of over 300 km. The China BRT database (ChinaBRT, 2014) showed that
so far, 19 cities in mainland China had already introduced BRT systems with a total length of about 1500 km.
These mass public transportation modes can accommodate a large number of passengers, but require huge investment
costs. For example, the capital cost of the Shanghai Metro Line 2 was about RMB600 million per kilometer (RMB is the
Chinese currency ‘‘Renminbi’’. US$1 approximates RMB6.2 as of January 1, 2015) and the Guangzhou BRT Line located on
Zhongshan Road cost about RMB50 million per kilometer. Each type of transit technologies has its advantages and disadvan-
tages. For instance, a metro system can carry more passengers at a faster operating speed, but needs higher capital and
⇑ Corresponding author. Tel.: +86 27 8754 6073; fax: +86 27 8755 6437.
E-mail addresses: [email protected] (Z.-C. Li), [email protected] (Q.-W. Guo), [email protected] (W.H.K. Lam), [email protected]
(S.C. Wong).
http://dx.doi.org/10.1016/j.trb.2015.05.006
0191-2615/Ó 2015 Elsevier Ltd. All rights reserved.
Z.-C. Li et al. / Transportation Research Part B 78 (2015) 318–340 319
operating costs than a bus service. BRT is regarded as a high-quality, cost-effective bus-based transit mode. BRT systems use
buses to provide faster and more efficient services than an ordinary bus line. These systems combine the speed, reliability,
and amenities of rail transit with the flexibility of bus transit. The investment cost of a BRT system is much lower than that of
a metro system with the same length. However, the speed and capacity of a BRT system are low, making it difficult to satisfy
a high level of passenger demand, particularly in densely populated urban areas. LRT systems are above-ground rail transit
systems. They have lower speed and capacity than metros but higher than BRT systems. Their capital cost is between that of
a metro and of a BRT system. The trade-off between the operating speed and capacity and the investment cost poses an
important question. Which transit technology should an authority introduce so as to improve the efficiency of urban trans-
portation system?
In addition, the feasibility of a transit investment project depends heavily on the urban population size, which directly
governs the level of passenger demand for transit services. However, the future population size for a city is uncertain. In par-
ticular, with the rapid urbanization, a large amount of rural inhabitants in China migrate to urban areas, leading to a
large-scale migrant population. It has been reported that China is currently experiencing the largest population migration
in the human history, causing a big variation in the urban population size. An investment made too early may cause a
low level of travel demand due to a small population size and thus revenue shortfalls and low efficiency, whereas an invest-
ment made too late may result in large social costs (such as increasing congestion and pollution) due to leaving demand
unmet for too long. This raises another intriguing and important question. When is the most appropriate time for an author-
ity to introduce a new transit technology in a fast-growing city with future population volatility?
In the literature, there are some studies on the transit technology selection and investment timing problems. For instance,
several studies have compared the costs or operating efficiencies of selecting different transit technologies, including metro,
LRT, and BRT systems (e.g., see Allport, 1981; Stutsman, 2002; Bruun, 2005; Tirachini et al., 2010a). Parajuli and Wirasinghe
(2001) developed a multi-attribute decision model to address the transit technology selection problem based on a field sur-
vey and an empirical statistical approach. Tirachini et al. (2010b) presented a social welfare and a private profit maximiza-
tion model to compare the investment benefits of different transit technologies on a linear transportation corridor. Chen
et al. (2015) proposed a static (i.e., steady-state), deterministic model to address the transit technology selection problem.
In their model, travel demand was assumed to be time-independent and deterministic. Szymanski (1991) compared the tim-
ing of infrastructure investment made by a welfare-maximizing public agency and a profit-maximizing private firm. Chu and
Polzin (2000) studied the timing rules for major transportation investment using a cost-benefit analysis model. Sivakumaran
et al. (2014) explored how the access mode (e.g., walking, bicycle, or feeder bus) affected the choice of transit technology
(e.g., heavy rail, BRT, or ordinary bus) for the trunk-line portion of a transit network.
The above-mentioned studies on transit technology investment issues mainly focused on static and deterministic prob-
lems. However, the future urban population size and thus the travel demand will stochastically and dynamically fluctuate
over time (Saez et al., 2012). This is particularly true for some of the fast-growing cities in China, which have uncertain future
population sizes due to large-scale population migration. The return or benefit of a transit technology investment project
will therefore also change stochastically and dynamically over time. It is therefore necessary to incorporate the dynamics
and uncertainty of urban population size over time into transit technology investment models.
These previous related studies have usually used the standard benefit-cost analysis method of net present value (NPV)
(Snell, 2011). It has been shown that the traditional NPV approach cannot properly capture management’s flexibility to post-
pone, abandon, or expand an investment opportunity, particularly in an irreversible and uncertain investment environment
(Dixit and Pindyck, 1994; Zhao and Tseng, 2003). The real option (RO) valuation approach provides a useful way to capture
the value of the flexibility that goes unrecognized in an NPV analysis (McDonald and Siegel, 1986; Trigeorgis, 1996; de
Neufville and Scholtes, 2011).
In the past decade, RO theory has attracted considerable attention from transportation researchers. For the convenience of
readers, we have summarized in Table 1 some principal contributions of the RO theory to transportation infrastructure
investment issues, in terms of the type of transportation infrastructure, decision variables, objective function, investment
effects on land use or spatial structure, model solution, and source of uncertainty. This table shows that the previous related
studies have mainly focused on the timing issue of transit technology investment, and little attention has been paid to the
selection issue of transit technologies. In addition, they usually ignored the effects of transportation infrastructure invest-
ment on the urban spatial structure. However, McDonald and Osuji (1995), Bowes and Ihlanfeldt (2001), and Li et al.
(2012a, 2013) showed that introducing a new transit line could induce a change in the urban land-use pattern, property val-
ues, and housing market in terms of housing rental prices and space, due to improvements in trip accessibility. The change in
the urban spatial structure in turn affected the value of the investment project through a change in the travel demand pat-
tern. The effects of transit technology investment projects on households’ residential location choices and housing market
should therefore be explicitly considered.
In view of the above, this paper addresses transit technology investment and selection issues under urban population
volatility based on the RO approach. These issues include which transit technology should be invested in and when. The main
contributions of this paper are as follows. First, an RO model is proposed to investigate the transit technology selection and
investment timing problems. In the proposed model, the effects of population uncertainty and transit technology investment
on the urban spatial structure are considered in terms of households’ residential location choices and housing market. A
comparison between the solutions of the models with and without considering urban spatial/land use equilibrium is made.
Second, the trigger population thresholds for investing in a transit technology project and for shifting from a transit
320 Z.-C. Li et al. / Transportation Research Part B 78 (2015) 318–340
Table 1
Applications of real option theory to transportation infrastructure investment issues.
Citation Type of Decision variables Objective function Considering Model Source of

transportation effects of solution uncertainty
infrastructure investment on
land use or urban
spatial structure
Pichayapan et al. Expressway Immediate Max. expected net No Simulation Traffic volume
(2003) investment or project benefit with
defer to invest binomial method
Zhao et al. (2004) A highway Number of lanes, Max. expected profit No Simulation Traffic demand,
corridor right-of-way land price, and
width, highway
rehabilitation or deterioration
not level
Saphores and Boarnet A highway Investment timing Max. expected utility No Closed- Urban
(2006) corridor or a changes of residents form population
transit line before and after
investment
Friesz et al. (2008) Transportation Flow distribution Max. expected net trip No Simulation Trip (or asset)
network over time and value cost
space
Galera and Solino A highway Highway Max. expected cash No Closed- Traffic volume
(2010) corridor concession flows or revenue form
Chow and Regan Transportation Link improvement Max. option value No Simulation Travel demand
(2011a) network timing (including options to
defer and re-design
network)
Chow and Regan Transportation Link improvement Max. option value No Simulation Travel demand
(2011b) network timing and (including options to
capacity defer and re-order
expansions project)
Gao and Driouchi A rail transit line Investment timing Alpha-maxmin. No Closed- Urban
(2013) expected utility form population and
changes of residents decision-making
before and after uncertainty
investment
This paper A transit line Transit technology Max. expected social Yes Closed- Urban
selection and welfare form population
investment timing
technology to another are analytically explored and compared. Third, comparative static analyses of the urban system and
the transit technology investment decision are conducted. Sensitivity analyses are also carried out to assess the effects of key
model parameters, such as the household income level, population volatility, project construction duration and discount
rate, on timely investment decisions about urban transit technology projects. In addition, the loss in project value due to
adoption of NPV valuation method but not RO valuation method is also estimated. The results show that transit technology
investment can change urban spatial structure, and ignoring the effects of transit technology investment on urban spatial
equilibrium can cause a late investment. In addition, there is a big difference in the trigger population thresholds for the
transit technology selection estimated by the NPV and RO valuation approaches.
The remainder of the paper is organized as follows. In the next section, housing market equilibrium with transit technol-
ogy investment is described. Section 3 presents the RO model for transit technology selection and investment timing deci-
sions. In Section 4, two examples are used to illustrate the applications of the proposed model. Finally, the conclusions are
given in Section 5, together with recommendations for further studies.
2. Housing market equilibrium
In this paper, we assume that a transit investment project consists of three stages (represented by i): before project
investment (i.e., before the project exists), during project construction, and after project operations. For presentation pur-
pose, these three stages are indexed by the subscripts 0, 1, and 2, respectively. The public transit modes adopted at stages
0 and 1 are conventional/ordinary public transit modes (e.g., regular bus or minibus), and those at stage 2 are mass public
transit modes (e.g., metro, LRT or BRT). In this section, we formulate the housing market equilibrium problem in stage i. To
facilitate the presentation of the essential ideas without loss of generality, basic model assumptions made in this paper are
listed in Table 2.
Table 2
Basic model assumptions.
No. of Description
assumption
A1 We assume that the authority or government is the investor of transit project, and makes transit investment decisions in a
sequential way. Specifically, it first decides which transit technology to choose, and then when to invest. The time horizon
considered in this paper is infinite. The objective of transit technology investment is to maximize the expected social welfare of the
urban system over the infinite time horizon. We also assume that the interest rate is riskless and is a known constant over the time
horizon. In China, the current interest rate is around 6%. However, in other countries, it may range from 9% to 15% for infrastructure
investment projects
A2 The city studied is assumed to be linear, closed, and monocentric, implying that all job opportunities are located in the urban central
business district (CBD). The value of land at or beyond the city boundary is equal to the agricultural rent or opportunity cost of the
land. These assumptions have been widely adopted in the field of urban economics (see, for example, Alonso, 1964; Mills, 1972;
Fujita, 1989; O’Sullivan, 2000; Kraus, 2006; Li et al., 2013)
A3 The housing market contains the authority, property developer, and household agents. The authority aims to choose a transit
technology and the investment timing to maximize the (expected) investment return or value of the transit project in terms of the
expected social welfare. The property developers determine the intensity of their capital investment in the land market to maximize
the net profit generated by the supply of housing. Each property developer is assumed to adopt a Cobb-Douglas housing production
function (see, for example, Beckmann, 1974; Quigley, 1984; Li et al., 2013)
A4 All households in the city are assumed to be homogenous, implying that their income levels and utility functions are identical. Each
household has a Cobb-Douglas utility function. A household’s income is spent on transportation, housing, and composite goods. The
objective of each household is to maximize its utility by choosing a residential location, housing space size, and amount of other
goods within its budget constraints (see, for example, Solow, 1972, 1973; Beckmann, 1969, 1974; Anas, 1982; Fujita, 1989). It is
further assumed that the urban population size (represented by N > 0) stochastically fluctuates over time and follows a geometric
Brownian motion (see also Saphores and Boarnet, 2006; Gao and Driouchi, 2013)
A5 We assume that the average daily number of trips to the CBD by transit modes per household is d. For example, d ¼ 1 indicates that
each household makes an average of one trip to the CBD by transit modes per day
2.1. Households’ residential location choices
We first define the household utility function. Let x be the distance from a household’s residential location to the CBD, and
Bi be the distance from the urban corridor boundary to the CBD (i.e., the city size) in stage i. A Cobb-Douglas utility function
(e.g., see Beckmann, 1969, 1974; Solow, 1972, 1973; Li et al., 2012a, 2013) is used:
U i ðxÞ ¼ a ln zi ðxÞ þ b ln g i ðxÞ; a; b > 0; a þ b ¼ 1; x 2 ½0; Bi ; i ¼ 0; 1; 2; ð1Þ
where U i ðxÞ is the utility of the households at location x in stage i; zi ðxÞ is the consumption of non-housing goods at location x
in stage i, of which the price is normalized to 1; g i ðxÞ is the consumption of housing at location x in stage i, measured in
square meters of floor space; and a and b are positive constants.
According to A4, each household within the city chooses a residential location that maximizes its utility subject to budget
constraints. The household utility maximization problem can be expressed as
max U i ðxÞ ¼ a ln zi ðxÞ þ b ln g i ðxÞ; ð2Þ
zi ;g i
s:t: zi ðxÞ þ pi ðxÞg i ðxÞ ¼ Y i ui ðxÞ; 8x 2 ½0; Bi ; i ¼ 0; 1; 2; ð3Þ

where pi ðxÞ is the average annual rental price per unit of housing area at location x in stage i and Y i is the average annual
household income in stage i. ui ðxÞ is the average annual travel cost from location x to the CBD in stage i and can be estimated
by
ui ðxÞ ¼ 2qC i ðxÞ; 8x 2 ½0; Bi ; i ¼ 0; 1; 2; ð4Þ
where ‘‘2’’ denotes a round trip between the CBD and location x, and q is the average annual number of trips to the CBD by
transit modes per household, based on the average number of daily trips, d, by transit modes per household (see A5). As an
approximation and for illustration purpose, we use q ¼ 365d in this paper. C i ðxÞ is the (one-way) average travel cost from
location x to the CBD. For simplicity, C i ðxÞ is assumed to be a function of travel distance:
x
C i ðxÞ ¼ s þ f i; 8i ¼ 0; 1; 2; ð5Þ
Vi
where V i is the average speed when traveling from location x to the CBD in stage i; s is the value of travel time, and f i is the
transit fare (where f 0 ¼ f 1 ¼ 0). In this paper, we assume that conventional public transit modes (e.g., regular bus, minibus)
are adopted at stages 0 and 1, whereas mass public transit modes (e.g., metro, LRT or BRT) are adopted at stage 2. In addition,
the transit project construction may affect the vehicle travel speed due to a decrease in the capacity of transit services.
Therefore, V 1 < V 0 < V 2 holds. The values of V 2 and f 2 change with the type of transit vehicles. For example, among the
metro, LRT and BRT, the metro has the fastest average speed, but the highest fare, whereas the BRT has the slowest average
speed, but the lowest fare. The LRT has the middle values for both.
From the first-order optimality conditions of the maximization problem (2), (3), we obtain
1
u ðxÞ b
pi ðxÞ ¼ pi ð0Þ 1 i ; 8i ¼ 0; 1; 2; ð6Þ
Yi
a
bY i u ðxÞ b
g i ðxÞ ¼ 1 i ; 8i ¼ 0; 1; 2; and ð7Þ
pi ð0Þ Yi
zi ðxÞ ¼ aðY i ui ðxÞÞ; 8i ¼ 0; 1; 2; ð8Þ

where pi ð0Þ represents the average rental price of housing in the CBD area, which is determined below. Eqs. (6)–(8) define
the equilibrium rental price per unit of housing area, the equilibrium amount of housing floor space per household, and the
equilibrium consumption of non-housing goods at location x in stage i, respectively.
Substituting Eqs. (6)–(8) into U i ðxÞ in Eq. (1), we obtain the equilibrium household utility:

bY i
U i ¼ a ln ðaY i Þ þ b ln ; 8i ¼ 0; 1; 2: ð9Þ
pi ð0Þ
Once pi ð0Þ is given, the terms on the right-hand side of Eq. (9) become constants. The households’ residential location choice
equilibrium state is then reached: all of the households in the linear city have the same utility regardless of their residential
locations and housing spaces.
2.2. Housing production
We now examine the supply side of the housing market. Let Si ðxÞ be the capital investment per unit of land area at loca-
tion x in stage i, also called the capital investment intensity. It will serve as an indicator to describe the intensity of land
development permitted in a particular area. According to A3, the property developers have the following Cobb–Douglas
housing production function:
hðSi ðxÞÞ ¼ l ðSi ðxÞÞh ; 0 < h < 1; ð10Þ

where hðSi ðxÞÞ is the housing supply per unit of land area at location x in stage i, and l and h are positive parameters.
Let Ki ðxÞ be the net profit per unit of land area due to the housing supply at location x in stage i, expressed as
Ki ðxÞ ¼ pi ðxÞhðSi ðxÞÞ ðri ðxÞ þ kSi ðxÞÞ; 8i ¼ 0; 1; 2; ð11Þ
where ri ðxÞ is the rent or value per unit of land area at location x in stage i and k is the price of the capital (i.e., the discount or
interest rate). The rental price pi ðxÞ per unit of housing floor area can be determined by Eq. (6). The first term on the
right-hand side of Eq. (11) is the total revenue from housing rent. The final two terms are the land rent cost and capital cost,
respectively.
According to A3, each property developer in the housing market aims to maximize its net profit by determining the cap-
ital investment intensity, expressed as
maxKi ðxÞ ¼ lpi ðxÞðSi ðxÞÞh ðr i ðxÞ þ kSi ðxÞÞ: ð12Þ

Si ðxÞ
The first-order optimality condition of maximization problem (12) yields

@ Ki ðxÞ
¼ hlpi ðxÞðSi ðxÞÞh1 k ¼ 0: ð13Þ
@Si ðxÞ
Substituting Eq. (6) into (13) yields
1=ð1hÞ !1=ð1hÞ
1 1 ui ðxÞ 1=b
Si ðxÞ ¼ pi ðxÞlhk ¼ k lhpi ð0Þ 1 ; 8i ¼ 0; 1; 2: ð14Þ
Yi
Let ni ðxÞ be the household residential density at location x in stage i. It is given by
hðSi ðxÞÞ 1 h=ð1hÞ

u ðxÞ ðaþbhÞ=ðbbhÞ
1
ni ðxÞ ¼ ¼ ðpi ð0ÞlÞ1=ð1hÞ hk 1 i ; 8i ¼ 0; 1; 2: ð15Þ
g i ðxÞ bY i Yi
Under perfect competition, the property developers earn zero profit. Thus,
ri ðxÞ ¼ lpi ðxÞðSi ðxÞÞh kSi ðxÞ; 8i ¼ 0; 1; 2: ð16Þ

Substituting Eqs. (6) and (14) into Eq. (16) yields
!1=ð1hÞ
1 ui ðxÞ 1=b h
ri ðxÞ ¼ k 1 Si ðxÞ ¼ ð1 hÞhh=ð1hÞ lpi ð0Þ 1 k ; 8i ¼ 0; 1; 2: ð17Þ
h Yi
Eqs. (14), (15), and (17) define the equilibrium capital investment intensity, household residential density, and land value at
any location along a linear urban corridor in stage i, respectively. It is evident that given the values of pi ð0Þ; Y i ; l, and h, the
capital investment intensity, household residential density, and land value decrease with an increase in either the travel cost
or interest rate, and vice versa.
2.3. Housing demand–supply equilibrium
The equilibrium of the housing market must satisfy two conditions. First, all of the households fit exactly inside the urban
corridor boundary, so that
Z Bi
ni ðxÞdx ¼ N; ð18Þ
0
where N is the total number of households in the city.

Second, the equilibrium rent per unit of land area devoted to housing at the city’s edge equals the exogenous agricultural
rent or the opportunity cost of the land according to A2:
ri ðBi Þ ¼ RA ; ð19Þ
where RA is a constant agricultural rent.
Note that there are two unknowns in Eqs. (18) and (19), pi ð0Þ and Bi . We can determine their values by solving the two
equations jointly.
Proposition 1. Given the agricultural rent RA , the rental price of housing in the CBD area pi ð0Þ and the city boundary Bi in stage i
are given by Eqs. (20) and (21), respectively:
h 1h 1=b ð1hÞ
k RA 2qf i 2Nqs
pi ð0Þ ¼ l1 1 þ1 ; 8i ¼ 0; 1; 2; and ð20Þ
h 1h Yi RA V i
bðh1Þ !
Vi Yi 2Nqs
Bi ¼ fi 1 þ1 ; 8i ¼ 0; 1; 2: ð21Þ
s 2q RA V i
The proof of Proposition 1 is provided in Appendix A. According to Proposition 1, once the housing rental price in the CBD
area pi ð0Þ and the city boundary Bi are obtained by Eqs. (20) and (21), the equilibrium rent per unit of housing space pi ðxÞ, the
equilibrium amount of housing space g i ðxÞ, the equilibrium household utility U i , the equilibrium capital investment intensity
Si ðxÞ, the equilibrium household residential density ni ðxÞ, and the equilibrium land value ri ðxÞ can be determined by Eqs. (6),
(7), (9), (14), (15), and (17), respectively.
Proposition 2 provides the comparative static results of the housing rental price in the CBD area pi ð0Þ and the city bound-
ary Bi .
Proposition 2. Under the assumption RA V 2 2N qs, the housing rental price in the city center pi ð0Þ increases with urban
population size N and transit fare f i , but decreases with household income Y i and transit operating speed V i . The city boundary Bi
increases with household income Y i , urban population size N, and transit operating speed V i , but decreases with transit fare f i . The
effects of Y i ; N; V i , and f i on pi ð0Þ and Bi are summarized in Table 3.
The proof of Proposition 2 is given in Appendix B. The assumption RA V 2 2N qs generally holds because population size N is
usually a large number, particularly in large Chinese cities. Proposition 2 can be explained as follows. (1) If the number of
urban population increases, the demand for housing and thus the housing rental price in the city center increases. As a result,
some inhabitants would like to move to the city suburbs to enjoy a larger housing space, and the city thus expands. (2) A rise
in the transit fare means an increase in the annual household transportation expenses. Some households would thus like to
live in the central area of the city to reduce their transportation expenses. As a result, the city shrinks and the housing rental
Table 3
The effects of the model parameters on the housing rental price in the CBD area and the city boundary.
Yi N Vi fi
pi ð0Þ + +
Bi + + +
Note: ‘‘+’’ represents a positive correlation and ‘‘’’ represents a negative correlation.
price in the city center increases. (3) If the household income level increases, households then have more budget for trans-
portation. Some households would like to reside in the city suburbs and enjoy a larger housing space. Thus, the city expands
and the housing rental price in the city center falls. (4) An increase in the transit operating speed means a decrease in travel
time, leading some households to move to the city suburbs. Accordingly, the size of the city enlarges and the housing rental
price in the city central area decreases.
3. Model formulation
3.1. Cost of transit project
The annual project cost is incurred by the annual transit line facility cost and annual transit vehicle operating cost. The
(annual) transit line facility cost is the sum of the fixed costs (e.g., the line overhead cost) and variable costs (e.g., the land
acquisition, line construction, maintenance, and labor costs), proportional to the transit line length, i.e.,
C L ¼ k0 þ k1 L; ð22Þ
where k0 is the fixed cost, k1 is the variable cost per kilometer, and L is the transit line length. For simplicity, the transit line
length L is assumed to equal the length of the city under the transit project operations, i.e.,
L ¼ B2 ; ð23Þ
where B2 is determined by Eq. (21).
The annual transit operating cost includes the fixed and variable operating costs, which are represented by
H
C O ¼ x0 þ x1 ; ð24Þ
H
where x0 is the fixed operating cost, x1 is the variable operating cost per vehicle, H is the transit headway, H is the average
vehicle round journey time, and H=H is the fleet size on the transit line. H can be calculated by
2L
H¼ ; ð25Þ
V2
where V 2 is the average vehicle operating speed after the transit line is put into operation and L=V 2 is the (one-way) average
vehicle journey time.
It should be pointed out that the parameters k0 ; k1 ; x0 , and x1 in Eqs. (22) and (24) depend on the type of transit vehicles.
In general, the values of these parameters for a metro are larger than those for an LRT, which are in turn larger than those for
a BRT.
3.2. Transit technology investment timing problem
The return on or benefit of a transit investment project is the improved accessibility of travel along the urban corridor to
the CBD. The derived value from the transit investment project is uncertain due to the uncertainty in the urban population
and thus travel demand. To describe the change in the urban population over time, we denote NðtÞ as the urban population
size at time t. Suppose that the urban population size follows the geometric Brownian motion in terms of A4. NðtÞ then
satisfies
dNðtÞ ¼ gNðtÞdt þ rNðtÞdwðtÞ; ð26Þ

where g is the growth rate of the urban population, r is the volatility rate of the urban population, dt is an infinitesimal time
increment, and dwðtÞ is an increment of a standard Wiener process. For any given period t; dwðtÞ satisfies the equation
pffiffi
dwðtÞ ¼ et t, where et is a random variable that follows the standard normal distribution with a mean of 0 and a standard
deviation of 1.
Transit infrastructure investment is irreversible due to the huge sunk cost. Studies showed that the traditional NPV
method failed to accurately capture the economic value of investment in an uncertain and irreversible environment, whereas
the RO approach provided a proper avenue for quantifying the value of management flexibility (McDonald and Siegel, 1986;
Dixit and Pindyck, 1994). An RO refers to the right, but not the obligation, to undertake an action (e.g., deferring, expanding,
contracting, or abandoning) at a predetermined cost (the exercise price) for a predetermined period (the life of the option).
We use an RO approach to address the investment timing problem for a given transit technology when the urban population
size fluctuates over time.
Determination of the optimal time for introducing a new transit technology is a standard stopping problem. In a low pop-
ulation density city, the passenger demand level is low. Waiting is thus a good strategy when the transit investment cost is
high, implying that a ‘‘waiting region’’ is determined. However, in a high population density city, the passenger demand level
is high and immediate investment may be a good decision. Accordingly, a ‘‘stopping region’’ is determined. The transit tech-
nology investment timing issue aims to determine the population threshold N that separates the waiting region from the
stopping region. When the urban population size NðtÞ at time t exceeds the population threshold N (i.e., NðtÞ > N ), the tran-
sit investment project should be introduced, otherwise it should be deferred.
By the definition of the population threshold N , the authority has no preference between the two assets (i.e., no invest-
ment or immediate investment) at N . That is, at N , the values of continuing to wait and immediate investment are equal.
We assume that the transit investment project is based on the authority’s perspective of expected social welfare maximiza-
tion (see A1 and A3). The value of continuing to wait is the sum of the expected discounted social welfare without the transit
project and the option value of waiting to invest in the transit project. The value of making the investment immediately is
the total expected discounted social welfare with the transit investment project, which is the sum of the expected dis-
counted social welfare during the project construction and during the project operation. Let FðNÞ denote the value of the
option of investing in the transit project at population size N. Let SW 0 ðNÞ be the expected discounted social welfare without
the transit investment project. Let SW 1 ðNÞ and SW 2 ðNÞ be the expected discounted social welfare during the project con-
struction and operation, respectively. At the population threshold N , we have
FðN Þ þ SW 0 ðN Þ ¼ SW 1 ðN Þ þ SW 2 ðN Þ: ð27Þ

Eq. (27) is the value-matching condition. At N , the expected discounted social welfare without the transit project plus the
value of the investment opportunity equals the expected discounted social welfare with the transit project.
We now define the expected discounted social welfare at each investment stage, SW 0 ðNÞ; SW 1 ðNÞ, and SW 2 ðNÞ, as follows.
The expected discounted social welfare is the sum of the expected discounted consumer surplus and the expected dis-
counted producer surplus (i.e., the expected net profit of the transit investment). When there is no new transit technology
investment project, the expected discounted social welfare equals the expected discounted consumer surplus, which
depends on the resulting household utility U 0 ðNðtÞÞ. SW 0 ðNÞ can thus be expressed as
Z þ1
SW 0 ðNÞ ¼ EN n U 0 ðNðtÞÞNðtÞekt dt ; ð28Þ
0
where EN ½ is the expectation operator with regard to the population size N; n is a parameter used to convert the household
utility into equivalent monetary units, k is the riskless interest rate and is assumed to be a known constant across time, and
U 0 ðNðtÞÞ is the household utility or consumer surplus without the transit investment project, which can be calculated by Eq.
(9).
Under the immediate investment strategy, the transit technology investment project includes construction and operation
stages. When the transit investment project is under construction, the resultant expected discounted social welfare SW 1 ðNÞ
equals the corresponding expected discounted consumer surplus minus the expected discounted transit line facility cost,
represented by
Z D Z D
SW 1 ðNÞ ¼ EN n U 1 ðNðtÞÞNðtÞekt dt C L ekt dt ; ð29Þ
0 0
where D is the construction duration of the transit project, which is assumed to be a constant in this paper, and the transit
line facility cost C L can be determined by Eq. (22). The transit operating cost is not included in Eq. (29) because the transit
facilities are still under construction and thus are not in operation.
When the transit technology investment project is in operation (i.e., interval ½D; þ1Þ), the expected discounted social
welfare SW 2 ðNÞ is the difference between the corresponding expected discounted consumer surplus minus the expected dis-
counted transit operating cost:
Z þ1 Z þ1
SW 2 ðNÞ ¼ EN n U 2 ðNðtÞÞNðtÞekt dt C O ekt dt ; ð30Þ
D D
where the transit operating cost C O can be determined by Eq. (24).

Substituting Eqs. (28)–(30) into Eq. (27) and carrying out algebraic operations, we obtain
FðN Þ ¼ UðN Þ; ð31Þ
where UðNÞ is given by
Z þ1 Z þ1 Z D
UðNÞ ¼ EN nðU 1 ðÞ U 0 ðÞÞNðtÞekt dt þ ðnðU 2 ðÞ U 1 ðÞÞNðtÞ C O Þekt dt C L ekt dt : ð32Þ
0 D 0
It should be noted that UðNÞ represents the change in the expected discounted social welfare due to the introduction of the
transit investment project, also called the investment return/benefit or project value in this paper. Under the mild assump-
tion that RA V 2 2N qs, we can derive the analytical expression for UðNÞ as below. The proof is provided in Appendix C.
Proposition 3. Suppose that RA V 2 2N qs holds, the value of the transit investment project can then be determined by

N V1 V2 2qf 2 NeðgkÞD C L ð1 ekD Þ C O ekD
UðNÞ ¼ nbð1 hÞ ln þ n bð1 hÞ ln þ ln 1 : ð33Þ
kg V0 V1 Y2 kg k k
According to the RO theory (e.g., see Dixit and Pindyck, 1994), the expected return (or value) of the investment opportunity
at equilibrium equals its expected rate of capital appreciation over the short time dt. This equilibrium condition can be
expressed as
kF ðNðtÞÞdt ¼ Et ½dF ðNðtÞÞ: ð34Þ

Eq. (34) actually represents the Bellman equation for the option value F ðNðtÞÞ. Applying Ito’s lemma (e.g., see Dixit and
Pindyck, 1994), we can determine the population threshold, N , for introducing a new transit project and the option value
function F ðNðtÞÞ as follows.
Proposition 4. Under the assumption RA V 2 2N qs, the trigger population threshold N and the option value function F ðNðtÞÞ
are, respectively, given by

b1 ðk gÞ C L 1 ekD þ C O ekD
N ¼ ; ð35Þ
nkðb1 1Þ bð1 hÞ ln VV 10 þ bð1 hÞ ln VV 21 þ ln 1 2Yq2f 2 eðgkÞD
F ðNðtÞÞ ¼ a1 ðNðtÞÞb1 ; ð36Þ

where
8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
g 1
2 2k
>
> 1 g
< b1 ¼ 2 r2 þ r2 2 þ r2 ;
b1 b1 1 ð37Þ
>
> n V1 V2 2qf 2 ðgkÞD kðb1 1Þ
: a1 ¼ b1 ðk gÞ bð1 hÞ ln þ bð1 hÞ ln þ ln 1 e :
L ð1e Þ O
V0 V1 Y2 C kD þC ekD
The proof of Proposition 4 is provided in Appendix D. We carry out a comparative static analysis of the population threshold
N and obtain the following results.
Proposition 5. N is positively correlated with the urban population volatility rate r, the marginal capital cost of the transit line

@N @N
k1 , and the marginal operating cost of the transit line x1 , i.e., @N
@ r > 0; @k1 > 0, and @ x1 > 0. When the marginal capital cost of the

transit line is larger than its marginal operating cost (i.e., C L P C O Þ; N is positively correlated with the transit project construction

duration D, i.e., @N
@ D > 0.
The proof of Proposition 5 is given in Appendix E. It should be pointed out that the signs of N with regard to the house-

hold income level and the discount rate (i.e., @N
@Y i
and @N
@k
) are difficult to determine analytically. A simulation method is there-
fore used to ascertain their relationships in the later numerical example.
For comparison purpose, we derive the population threshold of the traditional NPV model as follows. In the NPV
approach, a transit project is introduced if and only if the expected discounted social welfare with the transit project just
exceeds that without the project, i.e.,
SW 1 ðNNPV Þ þ SW 2 ðNNPV Þ P SW 0 ðNNPV Þ: ð38Þ
From Eq. (38), the population threshold in the NPV approach is obtained as below:

ðk gÞ C L 1 ekD þ C O ekD
NNPV ¼ : ð39Þ
nk bð1 hÞ ln VV 10 þ bð1 hÞ ln VV 21 þ ln 1 2Yq2f 2 eðgkÞD
Given the model parameters, from Eqs. (35) and (39), we can obtain the relationship between the population thresholds
under the RO and NPV approaches as follows:
b1
N ¼ N : ð40Þ
b1 1 NPV
Eq. (40) implies that the population threshold under the RO approach is always larger than that under the NPV approach, as
the RO approach incorporates the value of flexibility through the option to wait and defer investment.
3.3. Transit technology selection problem
As assumed in A1, the authority or government makes transit investment decisions in a sequential way, i.e., first deciding
which transit technology to choose, and then when to invest. Accordingly, once a transit technology is chosen, the optimal
timing (i.e., the trigger population threshold) for a transit technology investment can then be determined by using the
method developed in Section 3.2. In this section, we address the transit technology selection problem: given multiple transit
technology options, which transit technology is the most appropriate to invest in.
Let us consider multiple alternative transit technologies, such as the metro, LRT, and BRT, which are undergoing rapid
developments and applications in some large Chinese cities. For a given urban population size, different transit technologies
can bring different investment returns. Following A3, the authority will choose to introduce the transit technology that
yields the highest return on the investment. Let M be the set of alternative transit technologies and m1 and m2 be two transit
technology options, where m1 ; m2 2 M. Let Wm1 ðN Þ and Wm2 ðN Þ be the investment benefits of transit technologies m1 and m2
under population size N, respectively. The definition of WðÞ is applicable to different investment valuation methods.
Specifically, ‘‘WðÞ ¼ FðÞ’’ (see Eq. (36)) implies that an RO valuation method is used, and ‘‘WðÞ ¼ UðÞ’’ (see Eq. (33)) implies
that the NPV method is used. For presentation purpose, we introduce two definitions.
Definition 1 (A dominant transit technology). Transit technology m1 is dominant over technology m2 in the region ½N L ; N U if
the transit investment benefit function satisfies
Wm1 ðNÞ P Wm2 ðNÞ; N 2 ½NL ; NU ; ð41Þ
where N L and N U are the lower and upper bounds of the dominant region, respectively. When ½N L ; N U is the whole real num-
ber field, it is called wholly dominant.
Definition 2 (Trigger population threshold for transit technology shift). The population size N is at the critical (or trigger) level
for shifting from transit technology m2 to m1 if and only if Inequality (41) becomes binding at N, i.e.,

Wm1 N ¼ Wm2 N : ð42Þ

The following proposition further reveals the difference of the NPV and RO valuation approaches from the perspective of the
transit technology shift.
Proposition 6. (i) Under the NPV approach, there is a critical population size N NPV for transit technology shift from m2 to m1 ,
expressed as
1

m2

ðk gÞ C m 1 ekDm1 þ C m 1 kDm1
O e C L 1 ekDm2 þ C m 2 kDm2
O e
NNPV ¼ m L m
m
m
: ð43Þ
V 1 2qf 1 V 2 2qf 2
nk bð1 hÞ ln V21 þ ln 1 Y 22 eðgkÞDm1 bð1 hÞ ln V21 þ ln 1 Y 22 eðgkÞDm2
(ii) Under the RO approach, a transit technology is wholly dominant over the other transit technology in terms of the option value.
Specifically, transit technology m1 (e.g., metro) wholly dominates technology m2 (e.g., LRT) when Inequality (44) holds, and vice
versa.
m m

V 1 2qf 1 !b1 1
bð1 hÞ ln V1
þ bð1 hÞ ln V21 þ ln 1 Y 22 eðgkÞDm1
V0 Cm
L
1
1 ekDm1 þ C m 1 kDm1
O e
b1
m
m
P m2 kDm2

m2 kDm : ð44Þ
V 2 2qf 2 CL 1 e þ CO e 2
bð1 hÞ ln VV 10 þ bð1 hÞ ln V21 þ ln 1 Y 22 eðgkÞDm2
Proof. Property (i) with the NPV approach can directly be derived by substituting Eq. (33) into Eq. (42). For Property (ii),
when Inequality (44) holds, we can obtain am 1 P a1
1 m2
by Eq. (37), and thus F m1 ðN Þ P F m2 ðN Þ for any N by Eq. (36).
According to Definition 1, m1 wholly dominates m2 . This completes the proof of this proposition.
Proposition 6 shows that given transit technologies m1 and m2 , for the NPV model, when the population size reaches the
critical level of N NPV , then the transit technology to be adopted should shift from m2 to m1 . However, for the RO model, as
long as Inequality (44) is satisfied (this is usually the case), transit technology m1 is wholly dominant over transit technology
m2 .
Remark. It can be seen in Eq. (43) that the critical population size N NPV relies on such parameters as transit vehicle operating
speed V 2 , transit fare f 2 , construction duration of transit line D, transit line facility cost C L and transit operating cost C O . These
parameters describe the intrinsic characteristics of a transit technology and their values may change with the type of transit
technologies. Among these parameters, V 2 ; f 2 , and C O are relatively stable. However, C L and D may be significantly affected
by the urban environment and site condition in terms of hydro-geological characteristics, and are thus sensitive and varied
considerably, particularly for the metro technology. It can easily be shown by Eq. (43) that when C L P C O holds for transit
technology m, the critical population size N NPV with the NPV approach satisfies
@NNPV @NNPV @N NPV @NNPV

> 0; > 0; < 0; and < 0: ð45Þ
@C m
L
1
@ Dm 1 @C m
L
2
@ Dm 2
Eq. (45) means that N NPV increases with the increase in the values of C m
L and Dm1 of transit technology m1 , but decreases with
1
m2
the increase in the values of C L and Dm2 of transit technology m2 .
Note that N in Eq. (35) or (39) presents the population threshold for introducing a transit technology. N in Eq. (43)
defines the trigger population threshold for shifting from one transit technology to another under the NPV approach. The
value of N may differ from that of N . We thus have the following definition:
Definition 3. The trigger population threshold N for selecting a transit technology from multiple options is defined as the
maximum of the population threshold of that transit technology itself N and the critical population threshold for transit
technology shift N, i.e.,
N ¼ max N ; N : ð46Þ

Once the population size of a city reaches the trigger population size N , the new transit technology can then be launched. It
should be pointed out that according to Proposition 6(ii) and Eq. (46), under the RO approach, the value of N is equal to that
of N . However, both may not be equal under the NPV approach.
4. Model applications
In this section, two test examples are used to illustrate the applications of the proposed model and the contributions of
this study. The first example is intended to reveal the difference in the transit technology investment decisions with and
without urban spatial or land use equilibrium consideration. The case without considering the land use equilibrium implies
that the urban spatial structure in terms of urban residential distribution, housing rental price and space, and urban size is
the same as that before transit technology investment. The effects of a transit technology investment on the urban spatial
structure and the effects of the key model parameters (e.g., the project construction duration, discount rate, household
income level, and population volatility) on the trigger population threshold for a transit technology investment are also
investigated. The second example illustrates the real application of the proposed model in two candidate Chinese cities.
In the following analysis, unless specifically stated otherwise, the input parameters and baseline values used in the model
are the same as those shown in Table 4. Three types of transit technologies, the BRT, LRT, and metro, are considered. The
input parameters for these technologies are given in Table 5.
4.1. Example 1: Illustration of model properties
4.1.1. Comparison of investment decisions with and without land use integration
Fig. 1a and b depict the option value curves (in bold line) and the NPV curves (in dotted line) under different transit tech-
nology investments with and without incorporating the land use equilibrium, which can be calculated by Eqs. (36) and (33),
respectively. In Fig. 1a and b, the intersection point (D or D0 ) between the option value curve and the NPV curve for a given
transit technology represents the population threshold N of the corresponding transit technology under the RO approach in
terms of Eq. (31). The intersection point (G or G0 ) between the NPV curve and the horizontal line represents the population
threshold N NPV under the NPV approach according to Eq. (39). The intersection point (X or X0 ) between the NPV curves of
different transit technologies represents the critical population size N NPV for transit technology shift by Eq. (43). The trigger
Table 4
Input parameters for model application.
Parameter Definition Baseline value

Y Average annual household income (RMB/year) 80,000
2
RA Agricultural rent at the city boundary (RMB/km /year) 100
a; b Parameters in the household utility function 0.75, 0.25
h; l Parameters in the housing production function 0.7, 0.05
V0; V1 Average journey speed before and during project construction (km/h) 20, 18
f 0; f 1 Fare before and during project construction (RMB) 0, 0
H Transit service headway (hour/vehicle) 0.1
n Parameter in the social welfare function 6350
g Annual population growth rate 1.1%
r Population volatility rate 8.0%
k Riskless interest rate 6.0%
D Average daily number of trips to the CBD area per household 1.0
Source: please refer to Li et al. (2012a), Saphores and Boarnet (2006), and Chen et al. (2015).
Table 5
Input parameters for the metro, LRT, and BRT technologies.
Parameter Definition Metro LRT BRT

V2 Average vehicle operating speed (km/h) 40 35 30
f2 Transit fare (RMB) 4.0 3.0 2.0
k0 Fixed component of the annual transit facility cost (million RMB/year) 1.2 1.0 0.7
k1 Variable component of the annual transit facility cost (million RMB/km/year) 50.0 7.0 5.0
x0 Fixed component of the annual transit operating cost (million RMB/year) 1.2 0.5 0.35
x1 Variable component of the annual transit operating cost (million RMB/vehicle/year) 6.5 1.8 1.5
D Duration of the construction of the transit line (years) 5 3 1
Source: CPSITS (2011), BAOGAO (2012), and Li et al. (2012a).
With land use equilibrium Without land use equilibrium

20 Metro, Option value 20 Metro, Option value D′3(9.50, 19.77)
Option value and NPV (billion RMB)

18 LRT, Option value D3(8.20, 16.60) 18 LRT, Option value

16 BRT, Option value 16 BRT, Option value
Metro, NPV Metro, NPV
14 14
LRT, NPV LRT, NPV
12 12
BRT, NPV BRT, NPV
10 10
8 8
6 6
X2(4.85, 1.20) X′2(4.95, 0.83) X′3(7.35, 5.33)
4 X3(6.38, 4.38) 4
D1(4.39, 0.92) D′1(4.70, 0.71)
2 D2(6.20, 4.02) 2 D′2(6.50, 3.78)
0 0
G1(3.05,0) G2(4.31,0) G3(5.70,0) G′1(3.20,0) G′2(4.50,0) G′3(6.60,0)
3 4 5 6 7 8 9 3 4 5 6 7 8 9
Population (million people) Population (million people)
(a) (b)
Fig. 1. Option value curves (in bold line) and NPV curves (in dotted line) under different transit technology investments with and without land use
equilibrium consideration.
Table 6
Trigger population thresholds (million people) for different transit technology investments for the RO and NPV models with and without land use equilibrium.
Transit technology Approach used With land use equilibrium Without land use
equilibrium
N N N N
Metro RO 8.20 8.20 9.50 9.50
NPV 5.70 6.38 6.60 7.35
LRT RO 6.20 6.20 6.50 6.50
NPV 4.31 4.85 4.50 4.95
BRT RO 4.39 4.39 4.70 4.70
NPV 3.05 3.05 3.20 3.20
Note: N denotes the trigger population threshold for a single transit technology, and is given by Eq. (35).
N denotes the trigger population threshold for selecting among multiple transit technology options, and is determined by Eq. (46).
population threshold N for transit technology selection from multiple options is determined by Eq. (46). The population
threshold solutions with and without the land use equilibrium consideration under the RO and NPV approaches are shown
in Table 6.
Some main insights in Fig. 1a and b and Table 6 are summarized as follows. First, the option value curve of the metro is
above that of the LRT, which is above that of the BRT. This means that a costlier technology has a higher option value.
Specifically, the option value of the metro is the highest, and that of the BRT is the lowest, and that of the LRT is in between.
That is, the metro wholly dominates the LRT and the BRT, and the LRT wholly dominates the BRT in terms of the option value,
which is consistent with that in Proposition 6(ii). This result favors prior introduction of a costlier technology from the per-
spective of the option value. As a result, a long wait for investment is needed such that the urban population size grows to
the level of the trigger population threshold. In reality, the public cannot see such an option value that will be realized at a
future time, and want their tax money to benefit them now as opposed to the future. Therefore, from the public perspective,
an ‘‘effective’’ discount rate, which may be much higher than the real discount rate, can be used in a real application. Second,
ignoring the effects of the transit technology investment on the land use or urban spatial equilibrium will lead to a late
investment due to an underestimation of the investment benefits or returns. We take the RO solution as an example.
With the land use equilibrium consideration, the trigger population thresholds N for the metro, LRT, and BRT are 8.20,
6.20, and 4.39 million people, associated with points D3 ; D2 , and D1 in Fig. 1a, respectively. The corresponding values of
N without the land use equilibrium are 9.50, 6.50, 4.70 million people, which are points D03 ; D02 , and D01 in Fig. 1b, respec-
tively. Similar results can be observed for the NPV solution. Third, under the NPV approach, the population threshold solu-
tions for a single transit technology (N ) and for selecting between multiple transit technology options (N ) are different. For
example, for the case with land use equilibrium consideration, the NPV solutions N for the metro and LRT are 5.70 and 4.31
million people (i.e., points G3 and G2 in Fig. 1a), respectively. However, the corresponding values of N are 6.38 and 4.85
million people (i.e., points X 3 and X 2 in Fig. 1a), respectively. This result also holds for the case without the land use equi-
librium consideration (see Fig. 1b). Fourth, for a given transit technology, the RO approach leads to a higher population
threshold N than the NPV approach, which satisfies Eq. (40). For example, with the land use equilibrium consideration,
the population thresholds N for the metro under the RO and NPV approaches are 8.20 and 5.70 million, for the LRT are
6.20 and 4.31 million, and for the BRT are 4.39 and 3.05 million people, respectively. The NPV approach therefore leads
to a premature transit investment decision, which further illustrates the result underlying Eq. (40).
4.1.2. Effects of transit technology investment project on the urban structure

Fig. 2a–d display the profiles of the equilibrium household residential density, equilibrium housing rental price, equilib-
rium housing space per household, and equilibrium capital investment intensity, respectively, for an urban population size of
8.2 million people, which is the RO population threshold for introducing the metro technology (see Fig. 1a or Table 6). Fig. 2a
shows that in contrast with the ‘‘doing nothing’’ scenario (no new transit technology is introduced), introducing a transit
technology can cause urban expansion or growth due to the increased accessibility and convenience to the city center.
Introducing a metro, LRT, or BRT causes the city to grow by 15.6 km, 10.3 km, or 6.3 km, respectively, as shown in
Table 7. The residential density in the urban central area then decreases, whereas that in the city’s outskirts increases, com-
pared with the ‘‘doing nothing’’ scenario. The housing rental prices and capital investment intensity in the suburbs then
increase, whereas those in the urban central area decrease, as shown in Fig. 2b and d. As a result of the residential density
2000 9000
Residential density (people/km)
Housing rental price (RMB/m 2)
1800 Metro Metro

7500
1600 LRT LRT
1400 BRT BRT
6000
1200 Doing nothing Doing nothing
1000 4500
800
3000
600
400 1500
200
0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Distance from CBD (km) Distance from CBD (km)
(a) (b)
Capital investment density (RMB/m 2)
Housing space per household (m2)
90 300
Metro Metro
75 250
LRT LRT
BRT BRT
60 200
Doing nothing Doing nothing
45 150
30 100
15 50
0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Distance from CBD (km) Distance from CBD (km)
(c) (d)
Fig. 2. (a)–(d) represent household residential density, housing rental price, housing space per household, and capital investment intensity for a given
population size of 8.2 million people, respectively.
Table 7
Equilibrium solutions of the urban system with different transit technology investments when the urban population size is fixed as 8.2 million people.
Performance index Metro LRT BRT Doing nothing

City length (km) 55.3 50.0 46.0 39.7
2
Average urban residential density (households/km ) 682 776 852 920
Average housing rental price (RMB/m2 ) 4550 5100 5522 5642
Average housing space per household (m2 /household) 39 37 34 29
Average land value (RMB/m2 ) 776 840 912 941
2
Average capital investment intensity (million RMB/km ) 166.9 174.1 186.2 196.2
decrease in the urban central area and the capital investment intensity increase in the suburbs, the housing spaces per
household in the urban central area and in the suburbs increase, as shown in Fig. 2c.
Table 7 further summarizes the effects of introducing a transit technology on the equilibrium solution of the urban sys-
tem when the urban population size is fixed as 8.2 million people. It shows that introducing a transit technology can lead to a
decrease in the average urban residential density, average housing rental price, average land value, and average capital
investment intensity, but an increase in the average housing space. Of the three technologies, the effects of the metro on
the urban system are the greatest, those of the LRT are the second greatest, and those of the BRT are the smallest.
Specifically, the introduction of the metro, LRT, or BRT leads to a decrease in the average urban residential density of 238,
144, or 68 households per square meter, in the average housing rental price of RMB1092, RMB542, or RMB120 per square
meter, in the average land value of RMB165, RMB101, or RMB29 per square meter, and in the average capital investment
intensity of RMB29.3, RMB22.1 or RMB10.0 million per square kilometer, but an increase in the average housing space
per household of 10, 8, or 5 square meter per household, respectively.
4.1.3. Effects of the metro project construction duration and discount rate on the population threshold
In reality, a metro project construction duration is closely related to the urban geological conditions, whereas the dura-
tion of an LRT or BRT project construction is relatively stable. Fig. 3 shows the effects of the metro project construction dura-
tion and the discount (or interest) rate on the trigger population threshold. For a given discount rate, the longer the metro
project construction duration, the higher the population threshold, and vice versa. A long construction duration implies a
high investment opportunity cost, which reduces the attraction of the investment project. Accordingly, it would be better
to postpone the investment and wait for the project value to increase with the population growth. On the contrary, for a
given metro project construction duration, as the discount rate increases, the population threshold decreases, and vice versa,
because a high discount rate means a high project investment cost. Therefore, a wise decision is to invest early at a low dis-
count rate to reduce the investment cost.
4.1.4. Effects of the household income and population volatility on the population threshold
Fig. 4 plots the change of the population threshold as the household income level and urban population volatility change.
It shows that for a given population volatility, the trigger population threshold decreases with an increase in the household
income level. A growth in the household income level implies an increase in the budget for transportation and housing con-
sumption, according to Eq. (3). Some inhabitants would therefore like to live in the city suburbs and enjoy a larger housing
space. A mass rapid transit mode should thus be introduced earlier to satisfy the inhabitants’ needs, leading to a decreasing
population threshold. In contrast, for a fixed household income level, as the population volatility increases, the trigger pop-
ulation threshold also increases. This is because an increase in the urban population volatility means an increase in the
17
Population threshold (million people)
15
= 15
13
11 = 10
9
=5
7
5
0.03 0.05 0.07 0.09 0.11 0.13 0.15
Discount rate
Fig. 3. Change of the population threshold with the discount rate and metro project construction duration.
16
Population threshold (million people)

14
12
10
8
= 0.08
= 0.10
6
= 0.06
4
2
70 80 90 100 110 120 130 140
Household income level (thousand RMB)
Fig. 4. Change of the population threshold with the household income and population volatility.
flexibility (or option) value of the transit investment project (in fact, it is easy to show that @FðN Þ=@ r > 0 holds in terms of
Eq. (36)). Deferring the investment is therefore a better strategy than immediate investment.
4.2. Example 2: Case studies of two cities in China
For further illustration purpose, we apply the proposed transit technology investment model to two candidate Chinese
cities: Dalian and Changzhou. Dalian is a medium-density city located in Liaoning province, northeastern China.
Changzhou is a low-density city located in Jiangsu province, eastern China. Table 8 shows the numbers of residents (which
is the sum of the registered permanent residents and the migrant residents) and the per capita income levels of the two can-
didate cities between 1994 and 2014, according to the China Statistical Yearbooks (NBSC, 1994–2014). The associated annual
growth rates of the population sizes and the per capita income levels for these two cities are also shown in Table 8, together
with the volatility rates of the population sizes. In order to apply the households’ residential location choice model that is
proposed in Section 2, the population size and the per capita income level in Table 8 need to be converted to a household
level from a personal level. To do so, we assume in this example that each household has three persons (i.e., parents and
one child) in terms of the China’s one-child policy, and each household makes an average of 3.0 trips to the CBD by transit
Table 8
Population sizes and per capita income levels of two candidate cities between 1994 and 2014.
Year Population size (million people) Per capita income level (RMB/year/
person)
Dalian Changzhou Dalian Changzhou
1994 5.039 3.310 3152 4123
1995 5.295 3.390 3574 4532
1996 5.350 3.510 4001 4847
1997 5.566 3.550 4544 5322
1998 5.690 3.704 5077 5633
1999 5.770 3.770 5434 6192
2000 5.890 3.795 6098 6392
2001 5.966 3.855 6234 7205
2002 5.984 3.902 6765 7833
2003 5.992 3.923 7255 8705
2004 6.015 4.023 8183 9851
2005 6.020 4.110 11,994 11,379
2006 6.060 4.257 13,456 14,356
2007 6.080 4.352 15,456 18,765
2008 6.130 4.407 17,234 24,092
2009 6.170 4.452 19,734 26,723
2010 6.690 4.593 21,923 27,455
2011 6.740 4.650 24,276 29,559
2012 6.850 4.687 27,480 33,587
2013 6.954 4.692 29,434 35,232
2014 6.966 4.723 30,238 39,483
Average annual growth rate (%) 1.65 1.80 12.27 12.20
Population volatility rate (%) 2.05 1.16 – –
Source: China Statistical Yearbook (NBSC, 1994–2014).

12
Dalian, Trajectory 1
11
Population (million people)

10 Actual Simulated
9 Changzhou, Trajectory 1
Changzhou, Trajectory 2
8
Changzhou, Trajectory 3
7
3 1994
1996
1998
2000
2002
2004
2006
2008
2010
2012
2014
2016
2018
2020
2022
2024
2026
2028
2030
Year
Fig. 5. Actual and simulated population sizes of two candidate cities between 1994 and 2030. The population size after 2014 is simulated according to
geometric Brownian motion.
Annual per capita income (RMB/person)
160000
Dalian
140000
Changzhou
120000 Actual Forecasted
100000
80000
60000
40000
20000
0
1994
1996
1998
2000
2002
2004
2006
2008
2010
2012
2014
2016
2018
2020
2022
2024
2026
2028
2030
Year
Fig. 6. Actual and forecasted annual per capita income levels of two candidate cities between 1994 and 2030. The per capita income level after 2014 grows
by a constant rate.
modes per day, i.e., d ¼ 3:0. We also assume that since 2014, the change in the population size over time follows a geometric
Brownian motion and the per capita income level grows at the current average annual growth rate (see Table 8). Without
loss of generality, we generate three trajectories (or paths) of the urban population size over time for each of the two cities
according to the geometric Brownian motion, as indicated on the right-hand side of Fig. 5. The change trajectory of the per
capita income level over time (based on the current average annual growth rate) for each city is shown on the right-hand
side of Fig. 6.
Fig. 7a and b show the change curves of the annual option value and annual NPV when different transit technologies are,
respectively, introduced into Dalian and Changzhou and the changes in the future population sizes of these two cities follow
the simulated population trajectory 1 (see Fig. 5). Fig. 7a shows that the option value and NPV curves for the LRT intersect at
point Q1, which is associated with an investment in 2007 and yields a project value of RMB7.31 billion. The option value and
NPV curves for the metro intersect at point Q2 with an investment in 2015, yielding a project value of RMB17.32 billion.
Points Q1 and Q2 are the results estimated with the RO approach in terms of Eq. (35). Fig. 7a also shows that the NPV curves
for the LRT and metro intersect with the horizontal axis at points Q3 and Q4, respectively, which are the results estimated
with the NPV approach. Their corresponding investment timings are 1997 and 2001, respectively. Similarly, Fig. 7b shows
that the optimal investment decisions generated with the RO approach for the BRT, LRT, and metro occur at points W1,
W2, and W3, with investments in 2003, 2009, and 2020 and project values of RMB2.98, RMB5.67, and RMB23.81 billion,
respectively. The optimal investment timings generated with the NPV approach are 1997, 2001, and 2005, which are asso-
ciated with points W4, W5, and W6, respectively.
Table 9 summarizes the estimated transit technology investment timings with the RO and NPV approaches under the
simulated population trajectory 1 for Dalian and Changzhou. It can be seen that the NPV approach induces earlier invest-
ments than the RO approach of 14 and 10 years for the metro and LRT for Dalian, and of 15, 8, and 6 years for the metro,
LRT, and BRT for Changzhou, respectively.
18
Metro, Option value
Q2 (2015, 17.32)

LRT, Option value
15 Metro, NPV
LRT, NPV
12
9
Q′4 Q1 (2007, 7.31)
6
Q′3
3
Q3 Q4 Q′2
0
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
Year
Fig. 7a. Option value and NPV of introducing a metro or LRT in Dalian under simulated population trajectory 1.
24
Metro, Option value W3 (2020, 23.81)
LRT, Option value

20
BRT, Option value
Metro, NPV
16
LRT, NPV
BRT, NPV
12
8
W′6 W2 (2009, 5.67)
4 W′5
W1 (2003, 2.98)
W′4
W4 W5 W6
0
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
Year
Fig. 7b. Option value and NPV of introducing a metro, LRT, or BRT in Changzhou under simulated population trajectory 1.
Table 9
Estimated investment timings for two candidate cities under population trajectory 1.
Transit technology Approach used Estimated investment timing

(Year)
Dalian Changzhou
Metro RO 2015 2020
NPV 2001 2005
LRT RO 2007 2009
NPV 1997 2001
BRT RO – 2003
NPV – 1997
Note: The population size of Dalian city has exceeded 5.0 million since 1994 (see Table 8). A BRT cannot
satisfy the needs of the urban development of Dalian city, and thus a BRT project is not considered.
In addition, from Fig. 7a and b, one can calculate the total project values accrued from the investment time estimated with
the NPV approach to that with the RO approach for the two cities under the simulated population trajectory 1, as well as the
corresponding loss in the project values if the investment occurs at the time estimated with the NPV approach but not with
the RO approach. According to Fig. 7a, for the metro investment project in Dalian, the total option value and the total NPV
accrued from 2001 (i.e., the estimated investment time with the NPV approach) to 2015 (i.e., the estimated investment time
with the RO approach) are, respectively, the areas of Q 2 Q 02 Q 4 Q 04 and Q 2 Q 02 Q 4 , causing a loss in the project value equal to the
area of Q 2 Q 4 Q 04 . If Dalian invests in an LRT project in 1997 but not 2007, a loss in the project value equal to the area of
Q 1 Q 3 Q 03 is then incurred. Similarly, the investment timing decisions about metro, LRT and BRT projects in Changzhou under
Table 10
Average project value over the simulated population trajectories for each candidate city.
Transit technology Approach used Average project values (billion

RMB)
Dalian Changzhou
Metro RO 151.76 192.87
NPV 107.13 118.67
Difference 44.63 74.20
LRT RO 54.93 37.43
NPV 35.82 25.56
Difference 19.11 11.87
BRT RO – 15.11
NPV – 10.26
Difference – 4.85
the NPV approach can, respectively, lead to losses in the project values equal to the areas of W 3 W 6 W 06 ; W 2 W 5 W 05 , and
W 1 W 4 W 04 , compared with those under the RO approach.
In the foregoing discussions, we have presented the solution of the transit technology investment decision problem under
a specific population trajectory (i.e., trajectory 1). Note that different population trajectories could yield different investment
value curves and thus different investment decisions for a given valuation approach (RO or NPV approach). Therefore, it is
meaningful to calculate the average project value over the simulated population trajectories and the associated average loss
for each city concerned, as shown in Table 10. It can be seen in this table that the average option value and the average NPV
for the metro investment project in Dalian are, respectively, RMB151.76 billion and RMB107.13 billion, causing an average
loss of RMB44.63 billion. For the LRT investment project in Dalian, the NPV approach will incur an average loss of RMB19.11
billion compared to the RO approach. In addition, the investment decisions given by the NPV approach for metro, LRT and
BRT projects in Changzhou can, respectively, yield average losses of RMB74.20 billion, RMB11.87 billion, and RMB4.85 bil-
lion, in contrast to those given by the RO approach. These results further illustrate the bias of the NPV approach in determin-
ing timely investment decisions for transit technology.
5. Conclusion and further studies
In this paper, an RO model was proposed to address transit technology investment and selection problems with consid-
eration of uncertainty in urban population size. The effects of a transit technology investment on the urban system in terms
of households’ residential location choices and housing market were endogenously incorporated in the proposed model. The
properties of the proposed model were explored analytically, together with the trigger population thresholds for a transit
technology investment and for shifting from a transit technology to another. The comparative static analyses of the urban
system and the trigger threshold for a transit technology investment were also carried out. The transit project investment
decisions with and without urban spatial equilibrium consideration were compared. The loss in the project value caused
by bias in the NPV approach was also estimated.
The proposed model offers some new insights and important findings. First, transit technology investment can change
urban residential distribution and housing market and induce urban sprawl, thus leading to a decentralized city. Ignoring
the effects of transit technology investment on land use or urban spatial structure could underestimate the investment
return, thus resulting in a late investment of the transit technology. Second, there is a significant difference in the relation-
ships among the investment benefit curves of different transit technologies estimated by the NPV and RO valuation
approaches. For the NPV model, there is a trigger population threshold for shifting from a transit technology to another.
However, for the RO model, one transit technology is always dominant over the other transit technology. Compared with
the RO approach, the NPV approach underestimates the value of a transit technology investment project, causing a prema-
ture investment and thus a loss in the project value. Third, the optimal investment timing for a transit technology project is
dependent of the type of transit technologies (e.g., BRT, LRT, or metro), project cost and construction duration, discount rate,
urban population size, household income level, and population volatility. The investment decisions for single and multiple
transit technologies may be different. The proposed modeling methodology can serve as a useful tool for making decisions
about public transportation infrastructure investments and for evaluating the effects of transit project investment decisions
and urban development policies on urban systems at a strategic level. We believe that the closed-form solution of the pro-
posed model would be helpful for giving insightful findings in making decision for transit technology investment under dif-
ferent boundary conditions.
Although the proposed model provides useful insights for transit infrastructure investment decisions and policy evalua-
tion, some important extensions below could be made for further studies.
(1) The correlation effect between alternative transit technologies was not explicitly considered in the transit technology
selection problem. A feasible approach to analyze such effect is to assume that the values of underlying transit tech-
nology investment projects are stochastic and evolve over time according to a stochastic process (see e.g., Geltner
et al., 1996). For example, we may consider two alternative transit technology investment projects m1 and m2 , with
values that follow geometric Brownian motion:
dUj ðtÞ ¼ gj Uj ðtÞdt þ rj Uj ðtÞdwj ðtÞ; j ¼ m1 ; m2 ;
where Uj ðtÞ is the value of investing in transit technology project j at time t. gj and rj are the expected drift rate and
the volatility rate of the value of transit investment project j, respectively. wj ðtÞ is a standardized Wiener process. The
correlation coefficient c of transit investment projects m1 and m2 satisfies E½dwm1 dwm2 ¼ cdt and j c j6 1. In line of this
vein, we can explore the effect of correlation coefficient on transit technology investment decision-making for further
study.
(2) In this paper, the variation of population size is considered as the major source of uncertainty to affect the transit tech-
nology investment decisions. Other sources of uncertainty should also be studied, such as various random factors from
the supply side (e.g., fluctuation in investment cost and/or interest rate) and demand side (e.g., attitudes of
decision-makers, see Gao and Driouchi, 2013). These factors would have certain impacts on the returns or values of
public transportation investment projects. It is thus of great importance to incorporate the investment risk due to
other sources of uncertainty in the transit technology investment model.
(3) The transit investment problems in this paper are only investigated for a single transit line in a transportation corridor.
However, adding a transit line can usually cause re-distribution of traffic flows in a multi-modal transport network. It
is, therefore, necessary to develop a network-based real option model to assess the effects of transit line investment
with multiple technology options on the interrelated flows in a realistic network. In this regard, the studies of Chow
and Regan (2011a,b), and Chow et al. (2011) provided an important basis for developing such a model for investigating
the transit investment problems.
(4) It was assumed in the proposed model that the government was the investor of transit projects with an objective of
maximizing the expected social welfare. Recently, as investment markets have been progressively deregulated in
China, private sectors have been allowed to be involved in public infrastructure investment projects in various ways,
such as Build-Operate-Transfer and Public–Private-Partnership. The objective of private investors is however to max-
imize their own net profit (Li et al., 2012b). It is thus worthwhile to extend the proposed model to consider the inter-
ests of the private and public investors under different transit market regulatory regimes.
Acknowledgments
The authors would like to thank three anonymous referees for their helpful comments and suggestions on an earlier draft
of the paper. The work described in this paper was jointly supported by the National Natural Science Foundation of China
(71222107), the National Basic Research Program of China (973 Program) (2012CB725400, 2012CB725406), and the
Research Grants Council of the Hong Kong Special Administrative Region, China (Project Nos. PolyU 5181/13E and
17208614).
Appendix A. Proof of Proposition 1
Substituting Eq. (15) into Eq. (18), we obtain

Z Bi
1 h=ð1hÞ Z Bi
u ðxÞ ðaþbhÞ=ðbbhÞ
1
N¼ ni ðxÞdx ¼ ðpi ð0ÞlÞ1=ð1hÞ hk 1 i dx
0 bY i 0 Yi
h=ð1hÞ V 1=ðbbhÞ 1=ðbbhÞ !
1=ð1hÞ 1 i 2q sBi 2qf i
¼ ðpi ð0ÞlÞ hk ð1 hÞ 1 þ fi 1 : ðA:1Þ
2qs Yi Vi Yi
Combining Eqs. (17) and (19) yields

1=b !1=ð1hÞ
ui ðBi Þ h
RA ¼ ð1 hÞhh=ð1hÞ lpi ð0Þ 1 k : ðA:2Þ
Yi
From Eqs. (4) and (5), we have

sBi
ui ðBi Þ ¼ 2q þ fi ; 8i ¼ 0; 1; 2: ðA:3Þ
Vi
Substituting Eq. (A.3) into (A.2), we obtain
h 1h 1=b
k RA 2q sBi
pi ð0Þ ¼ l1 1 þ fi ; 8i ¼ 0; 1; 2: ðA:4Þ
h 1h Yi Vi
Substituting Eq. (A.4) into (A.1), we obtain
bðh1Þ !
vi Yi 2Nqs
Bi ¼ fi 1 þ1 ; 8i ¼ 0; 1; 2: ðA:5Þ
s 2q RA V i
Substituting Eq. (A.5) into (A.4), we have

h 1h 1=b ð1hÞ
k RA 2qf i 2Nqs
pi ð0Þ ¼ l1 1 þ1 ; 8i ¼ 0; 1; 2: ðA:6Þ
h 1h Yi RA V i
This completes the proof of Proposition 1.
Appendix B. Proof of Proposition 2
The partial derivatives of pi ð0Þ and Bi with regard to Y i ; N; V i , and f i are, respectively, as follows:
h 1h ð1hÞ ð1=bþ1Þ
@pi ð0Þ 2qf i k RA 2Nqs 2qf i
¼ 2
þ1 1 ; ðB:1Þ
@Y i blY i h 1h RA V i Yi
h 1h 1=b h

@pi ð0Þ 2qs k RA 2qf i 2N qs
¼ ð1 hÞ 1 þ1 ; ðB:2Þ
@N l RA V i h 1h Yi RA V i
h 1h 1=b h
@pi ð0Þ 2Nqs k RA 2qf i 2Nqs
¼ ð 1 hÞ 1 þ 1 ; ðB:3Þ
@V i lRA V 2i h 1h Yi RA V i
h 1h ð1=bþ1Þ ð1hÞ

@pi ð0Þ 2q k RA 2qf i 2Nqs
¼ 1 þ1 ; ðB:4Þ
@f i blY i h 1h Yi RA V i
bðh1Þ !
@Bi 1 Vi 2Nqs
¼ 1 þ1 ; ðB:5Þ
@Y i 2q s RA V i
bðh1Þ1
@Bi 2q s V i Y i 2Nqs
¼ bð1 hÞ fi þ1 ; ðB:6Þ
@N RA V i s 2q RA V i
bðh1Þ ! bðh1Þ1
@Bi 1 Y i 2Nqs 2Nqs V i Y i 2Nqs
¼ fi 1 þ1 þ bðh 1Þ f i þ 1 ; ðB:7Þ
@V i s 2q RA V i RA V 2i s 2q RA V i
bðh1Þ !
@Bi Vi 2N qs
¼ 1 þ1 : ðB:8Þ
@f i s RA V i
As 0 < h < 1 and 0 < b < 1 hold, we immediately obtain

@pi ð0Þ @pi ð0Þ @pi ð0Þ @pi ð0Þ
< 0; > 0; < 0; > 0: ðB:9Þ
@Y i @N @V i @f i
In addition, under the assumption of RA V 2 2N qs, we have
@Bi @Bi @Bi @Bi
> 0; > 0; > 0; < 0: ðB:10Þ
@Y i @N @V i @f i
Appendix C. Proof of Proposition 3
We first derive the expressions for U 1 ðÞ U 0 ðÞ and U 2 ðÞ U 1 ðÞ. According to Eqs. (9) and (20), we have

bY 1 bY 1
U 1 ðÞ U 0 ðÞ ¼ a ln ðaY 1 Þ þ b ln a ln ðaY 0 Þ þ b ln
p1 ð0Þ p0 ð0Þ
0 1=b 1h 1
2qf 0 2NðtÞqs
Y1 BY 1 1 Y 0 RA V 0
þ1 C
¼ a ln þ b ln @ 1=b 1h A: ðC:1Þ
Y0 2qf 1 2NðtÞqs
Y 0 1 Y1 RA V 1
þ1
Suppose that the household’s income level is unchanged over time, i.e., Y 0 ¼ Y 1 ¼ Y 2 . Note that f 0 ¼ 0 and f 1 ¼ 0 hold
because the transit project is not yet in operation. Substituting them into Eq. (C.1), we obtain
ð2NðtÞqs þ RA V o ÞV 1
U 1 ðÞ U 0 ðÞ ¼ bð1 hÞ ln : ðC:2Þ
ð2NðtÞqs þ RA V 1 ÞV 0
Similarly, U 2 ðÞ U 1 ðÞ can be given by
1=b 1h !
p1 ð0Þ 2q f 2 ð2NðtÞqs þ RA V 1 ÞV 2
U 2 ðÞ U 1 ðÞ ¼ b ln ¼ b ln 1
p2 ð0Þ Y2 ð2NðtÞqs þ RA V 2 ÞV 1

ð2NðtÞqs þ RA V 1 ÞV 2 2q f 2
¼ bð1 hÞ ln þ ln 1 : ðC:3Þ
ð2NðtÞqs þ RA V 2 ÞV 1 Y2
When RA V 2 2N qs holds, Eqs. (C.2) and (C.3) can, respectively, be expressed as

V1
U 1 ðÞ U 0 ðÞ bð1 hÞ ln ; and ðC:4Þ
V0

V2 2qf 2
U 2 ðÞ U 1 ðÞ bð1 hÞ ln þ ln 1 : ðC:5Þ
V1 Y2
Substituting Eqs. (C.4) and (C.5) into Eq. (32) yields
Z þ1
V1
UðÞ ¼ E NðtÞekt dt
nbð1 hÞ ln
0 V0
Z þ1 Z D
V2 2qf 2
þE n bð1 hÞ ln þ ln 1 NðtÞ C O ekt dt E C L ekt dt : ðC:6Þ
D V1 Y2 0
As NðtÞ follows geometric Brownian motion (Dixit and Pindyck, 1994), we have
E½NðtÞjN0 ¼ N ¼ Negt : ðC:7Þ

Substituting Eq. (C.7) into (C.6), we obtain

N V1 V2 2qf 2 NeðgkÞD C L ð1 ekD Þ C O ekD
UðNÞ ¼ nbð1 hÞ ln þ n bð1 hÞ ln þ ln 1 : ðC:8Þ
kg V0 V1 Y2 kg k k
Appendix D. Proof of Proposition 4
By Ito’s lemma, we can obtain the solution of Eq. (34) as below
F ðNðtÞÞ ¼ a1 ðNðtÞÞb1 : ðD:1Þ

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
g 1
2 2k
where b1 ¼ 12 rg2 þ r2 2 þ r2 > 1 and a1 is a positive constant to be determined.
From the value-matching condition in Eq. (31) as well as Eqs. (D.1) and (33), we obtain
ðgkÞD
b1 N V1 V2 2qf 2 N e C L ð1 ekD Þ C O ekD
a1 ðN Þ ¼ nbð1 hÞ ln þ n bð1 hÞ ln þ ln 1 : ðD:2Þ
kg V0 V1 Y2 kg k k
By the smooth-pasting condition (see Dixit and Pindyck, 1994), we have

dFðNÞ dUðNÞ
¼ : ðD:3Þ
dN N¼N dN N¼N
Eq. (D.3) can be written as

ðgkÞD
b1 1 1 V1 V2 2qf 2 e ð1 ekD Þ dC L ekD dC O
a1 b1 ðN Þ ¼ nbð1 hÞ ln þ n bð1 hÞ ln þ ln 1 ;
kg V0 V1 Y2 kg k dN N¼N k dN N¼N
ðD:4Þ
where
bðh1Þ1
dC L 2q Y 2 2N qs
¼ bð1 hÞk1 f2 þ1 ; and ðD:5Þ
dN N¼N RA 2q RA V 2
bðh1Þ1
dC O 4qx1 Y 2 2N qs
¼ bð 1 h Þ f þ 1 : ðD:6Þ
dN N¼N V 2 HRA 2q 2
RA V 2
When RA V 2 2N qs holds, the right-hand sides of Eqs. (D.5) and (D.6) approximate to zero. The final two terms of Eq. (D.4)
then disappear. Solving the system of Eqs. (D.2) and (D.4), one obtains expression (35) for the population threshold N . Then,
one can further determine the parameter a1 by Eq. (D.4), and thus expression (36) for the option value F ðNðtÞÞ according to
Eq. (D.1). This completes the proof of this proposition.
Appendix E. Proof of Proposition 5
For ease of presentation, we denote

V1 V2 2qf 2
f1 ¼ bð1 hÞ ln ; and f2 ¼ bð1 hÞ ln þ ln 1 : ðE:1Þ
V0 V1 Y2
The partial derivatives of N with regard to r; k1 ; x1 , and D are then given as follows:

@N 2rb1 ðk gÞ C L 1 ekD þ C O ekD
¼ ; ðE:2Þ
@r nkðb1 1Þðr2 ð2b1 1Þ þ 2gÞðf1 þ f2 eðgkÞD Þ

@N b1 V 2 ðk gÞðY 2 2qf 2 Þ 1 ekD

¼ ; ðE:3Þ
@k1 2nkqsðb1 1Þðf1 þ f2 eðgkÞD Þ
@N b1 ðk gÞðY 2 2qf 2 ÞekD

¼ ; ðE:4Þ
@ x1 nkqsHðb1 1Þðf1 þ f2 eðgkÞD Þ
@N b1 ðk gÞðC L C O ÞekD þ N nðb1 1Þðk gÞf2 eðgkÞD

¼ : ðE:5Þ
@D nðb1 1Þðf1 þ f2 eðgkÞD Þ
Note that 0 < h < 1; b1 > 1; Y 2 > 2qf 2 , and k > g hold. We thus obtain
@N @N @N
> 0; > 0; and > 0: ðE:6Þ
@r @k1 @ x1
@N
The sign of @D
depends on the sign of C L C O . As C L P C O , we have

@N
> 0: ðE:7Þ
@D
@N
However, when C L < C O , the sign of @D
is ambiguous. This completes the proof of Proposition 5.
References
Allport, R.J., 1981. The costing of bus, light rail transit and metro public transport systems. Traffic Engineering & Control 22, 633–639.
Alonso, W., 1964. Location and Land Use: Toward a General Theory of Land Rent. Harvard University Press, Cambridge, MA.
Anas, A., 1982. Residential Location Markets and Urban Transportation. Academic Press, New York.
BAOGAO, 2012. Report on Urban Light Rail Transit Project Feasibility Analyses. <http://www.BaoGaoBaoGao.com/DiaoYan/2012-10/
chengshiqingguijianshexiangmukexingx.html>.
Beckmann, M.J., 1969. On the distribution of urban rent and residential density. Journal of Economic Theory 1, 60–67.
Beckmann, M.J., 1974. Spatial equilibrium in housing market. Journal of Urban Economics 1, 99–107.
Bowes, D.R., Ihlanfeldt, K.R., 2001. Identifying the impacts of rail transit stations on residential property values. Journal of Urban Economics 50, 1–25.
Bruun, E., 2005. Bus rapid transit and light rail: comparing operating costs with a parametric cost model. Transportation Research Record 1927, 11–21.
Chen, Y.J., Li, Z.C., Lam, W.H.K., 2015. Modeling transit technology selection in a linear transportation corridor. Journal of Advanced Transportation 49, 48–
72.
ChinaBRT, 2014. Information on Bus Rapid Transit in Chinese and Asian Cities. <http://www.chinabrt.org>.
Chow, J.Y.J., Regan, A.C., 2011a. Real option pricing of network design investments. Transportation Science 45, 50–63.
Chow, J.Y.J., Regan, A.C., 2011b. Network-based real option models. Transportation Research Part B 45, 67–74.
Chow, J.Y.J., Regan, A.C., Ranaiefar, F., Arkhipov, D.I., 2011. A network option portfolio management framework for adaptive transportation planning.
Transportation Research Part A 45 (8), 765–778.
Chu, X., Polzin, S.E., 2000. Timing rules for major transportation investments. Transportation 27, 201–219.
CPSITS, 2011. Analysis of BRT Systems in Some Typical Chinese Cities. <http://www.cpsits.com/bencandy.php?fid=157&id=4054>.
de Neufville, R., Scholtes, S., 2011. Flexibility in Engineering Design. MIT Press, Cambridge, MA.
Dixit, A., Pindyck, R., 1994. Investment under Uncertainty. Princeton University Press, Princeton.
Friesz, T.L., Mookherjee, R., Yao, T., 2008. Securitizing congestion: the congestion call option. Transportation Research Part B 42, 407–437.
Fujita, M., 1989. Urban Economic Theory. Cambridge University Press, Cambridge.
Galera, A.L.L., Solino, A.S., 2010. A real options approach for the valuation of highway concessions. Transportation Science 44, 416–427.
Gao, Y., Driouchi, T., 2013. Incorporating Knightian uncertainty into real options analysis: using multiple-priors in the case of rail transit investment.
Transportation Research Part B 55, 23–40.
Geltner, D., Riddiough, T., Stojanovic, S., 1996. Insights on the effect of land use choice: the perpetual option on the best of two underlying assets. Journal of
Urban Economics 39, 20–50.
Kraus, M., 2006. Monocentric cities. In: Arnott, R.J., McMillan, D.P. (Eds.), A Companion to Urban Economics. Blackwell Publishing, Oxford, pp. 96–108.
Li, Z.C., Chen, Y.J., Wang, Y.D., Lam, W.H.K., Wong, S.C., 2013. Optimal density of radial major roads in a two-dimensional monocentric city with endogenous
residential distribution and housing prices. Regional Science and Urban Economics 43, 927–937.
Li, Z.C., Lam, W.H.K., Wong, S.C., Choi, K., 2012a. Modeling the effects of integrated rail and property development on the design of rail line services in a
linear monocentric city. Transportation Research Part B 46, 710–728.
Li, Z.C., Lam, W.H.K., Wong, S.C., Sumalee, A., 2012b. Design of a rail transit line for profit maximization in a linear transportation corridor. Transportation
Research Part E 48, 50–70.
Liu, Z., Gu, B.N., Sun, S.C., Deng, C.Y., 2015. Statistics and analysis of urban rail transit lines in 2014, China. Urban Rail Transit Research 1, 1–6.
McDonald, J.F., Osuji, C.I., 1995. The effect of anticipated transportation improvement on residential land values. Regional Science and Urban Economics 25,
261–278.
McDonald, R., Siegel, D., 1986. The value of waiting to invest. Quarterly Journal of Economics 101, 707–727.
Mills, E.S., 1972. Urban Economics. Scott Foresman, Glenview, IL.
NBSC (National Bureau of Statistics of China), 1994–2014. China Statistical Yearbook (1994–2014). China Statistics Press, Beijing, China.
O’Sullivan, A., 2000. Urban Economics. Irwin/McGraw-Hill, Boston.
Parajuli, P.M., Wirasinghe, S.C., 2001. A line haul transit technology selection model. Transportation Planning and Technology 24, 271–308.
Pichayapan, P., Hino, S., Kishi, K., Satoh, K., 2003. Real option analysis in evaluation of expressway projects under uncertainties. Journal of the Eastern Asia
Society for Transportation Studies 5, 3015–3030.
Quigley, J.M., 1984. The production of housing services and the derived demand for residential energy. RAND Journal of Economics 15, 555–567.
Saez, D., Cortes, C.E., Milla, F., Nunez, A., Tirachini, A., Riquelme, M., 2012. Hybrid predictive control strategy for a public transport system with uncertain
demand. Transportmetrica 8, 61–86.
Saphores, J.D.M., Boarnet, M.G., 2006. Uncertainty and the timing of an urban congestion relief investment: the no-land case. Journal of Urban Economics 59,
189–208.
Sivakumaran, K., Li, Y., Cassidy, M., Madanat, S., 2014. Access and the choice of transit technology. Transportation Research Part A 59, 204–221.
Snell, M., 2011. Cost-benefit Analysis: A Practical Guide. Thomas Telford Limited, London.
Solow, R.M., 1972. Congestion, density and the use of land in transportation. Swedish Journal of Economics 74, 161–173.
Solow, R.M., 1973. Congestion cost and the use of land for streets. Bell Journal of Economics Management Science 4, 602–618.
Stutsman, J.M., 2002. Bus rapid transit or light, rail transit: how to decide? Los Angeles case study. Transportation Research Record 1793, 55–61.
Szymanski, S., 1991. The optimal timing of infrastructure investment. Journal of Transport Economics and Policy 12, 247–258.
Tirachini, A., Hensher, D.A., Jara-Diaz, S.R., 2010a. Comparing operator and users costs of light rail, heavy rail and bus rapid transit over a radial public
transport network. Research in Transportation Economics 29, 231–242.
Tirachini, A., Hensher, D.A., Jara-Diaz, S.R., 2010b. Restating modal investment priority with an improved model for public transport analysis. Transportation
Research Part E 46, 1148–1168.
Trigeorgis, L., 1996. Real Options: Managerial Flexibility and Strategy in Resource Allocation. The MIT Press, Cambridge, MA.
Zhao, T., Sundararajan, S.K., Tseng, C.L., 2004. Highway development decision-making under uncertainty: a real options approach. Journal of Infrastructure
Systems 10, 23–32.
Zhao, T., Tseng, C.L., 2003. Valuing flexibility in infrastructure expansion. Journal of Infrastructure Systems 9, 89–97.

Transportation Research Part B: Zhi-Chun Li, Qian-Wen Guo, William H.K. Lam, S.C. Wong

Uploaded by

Copyright:

Available Formats

Transportation Research Part B: Zhi-Chun Li, Qian-Wen Guo, William H.K. Lam, S.C. Wong

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Transportation Research Part B: Zhi-Chun Li, Qian-Wen Guo, William H.K. Lam, S.C. Wong

Uploaded by

Copyright:

Available Formats

Transportation Research Part B 78 (2015) 318–340

Contents lists available at ScienceDirect

Transportation Research Part B

Transit technology investment and selection under urban

Citation Type of Decision variables Objective function Considering Model Source of

2. Housing market equilibrium

2.1. Households’ residential location choices

s:t: zi ðxÞ þ pi ðxÞg i ðxÞ ¼ Y i ui ðxÞ; 8x 2 ½0; Bi ; i ¼ 0; 1; 2; ð3Þ

zi ðxÞ ¼ aðY i ui ðxÞÞ; 8i ¼ 0; 1; 2; ð8Þ

2.2. Housing production

hðSi ðxÞÞ ¼ l ðSi ðxÞÞh ; 0 < h < 1; ð10Þ

maxKi ðxÞ ¼ lpi ðxÞðSi ðxÞÞh ðr i ðxÞ þ kSi ðxÞÞ: ð12Þ

The ﬁrst-order optimality condition of maximization problem (12) yields

Let ni ðxÞ be the household residential density at location x in stage i. It is given by

hðSi ðxÞÞ 1 h=ð1hÞ

ri ðxÞ ¼ lpi ðxÞðSi ðxÞÞh kSi ðxÞ; 8i ¼ 0; 1; 2: ð16Þ

2.3. Housing demand–supply equilibrium

where N is the total number of households in the city.

3.1. Cost of transit project

3.2. Transit technology investment timing problem

dNðtÞ ¼ gNðtÞdt þ rNðtÞdwðtÞ; ð26Þ

where the transit operating cost C O can be determined by Eq. (24).

kF ðNðtÞÞdt ¼ Et ½dF ðNðtÞÞ: ð34Þ

F ðNðtÞÞ ¼ a1 ðNðtÞÞb1 ; ð36Þ

3.3. Transit technology selection problem

Wm1 ðNÞ P Wm2 ðNÞ; N 2 ½NL ; NU ; ð41Þ

Wm1 N ¼ Wm2 N : ð42Þ

@NNPV @NNPV @N NPV @NNPV

N ¼ max N ; N : ð46Þ

4.1. Example 1: Illustration of model properties

Parameter Deﬁnition Baseline value

Parameter Deﬁnition Metro LRT BRT

Source: CPSITS (2011), BAOGAO (2012), and Li et al. (2012a).

With land use equilibrium Without land use equilibrium

Option value and NPV (billion RMB)

18 LRT, Option value D3(8.20, 16.60) 18 LRT, Option value

4.1.2. Effects of transit technology investment project on the urban structure

Housing rental price (RMB/m 2)

1800 Metro Metro

Performance index Metro LRT BRT Doing nothing

Population threshold (million people)

4.2. Example 2: Case studies of two cities in China

Source: China Statistical Yearbook (NBSC, 1994–2014).

Population (million people)

Option value and NPV (billion RMB)

LRT, Option value

Transit technology Approach used Estimated investment timing

Transit technology Approach used Average project values (billion

5. Conclusion and further studies

Appendix A. Proof of Proposition 1

Substituting Eq. (15) into Eq. (18), we obtain

Combining Eqs. (17) and (19) yields

From Eqs. (4) and (5), we have

Substituting Eq. (A.5) into (A.4), we have

Appendix B. Proof of Proposition 2

h 1h 1=b h

h 1h ð1=bþ1Þ ð1hÞ

As 0 < h < 1 and 0 < b < 1 hold, we immediately obtain

hðSi ðxÞÞ 1 h=ð1hÞ

N ¼ max N ; N : ð46Þ

@N b1 V 2 ðk gÞðY 2 2qf 2 Þ 1 ekD

@N b1 ðk gÞðY 2 2qf 2 ÞekD

@N b1 ðk gÞðC L C O ÞekD þ N nðb1 1Þðk gÞf2 eðgkÞD