Transportation Research Part C 164 (2024) 104666

Contents lists available at ScienceDirect

Transportation Research Part C

journal homepage: www.elsevier.com/locate/trc

Revenue sharing and resource allocation for cooperative

multimodal transport systems☆
Xiaoshu Ding , Sisi Jian *
Department of Civil and Environmental Engineering, The Hong Kong University of Science and Technology, Hong Kong Special Administrative Regions
of China

A R T I C L E I N F O A B S T R A C T

Keywords: The past decade has witnessed a significant growth in the diversity of transport services. The
Cooperative game cooperation among different transport service providers (TSPs) helps to reduce operational costs
Non-cooperative game and improve resource utilization rates by pooling the mobility resources together and facilitating
Revenue sharing
the centralized optimal resource allocation. However, TSPs are selfish entities who seek to
Multimodal transportation
maximize their own profits. They will form a coalition only when cooperation brings them more
benefits. Hence, to ensure the stability of such a coalition, the key challenge lies in designing
revenue-sharing schemes that make TSPs better off and the coalition reaches Pareto efficiency,
while accounting for individual TSPs’ decentralized resource allocation decisions. In this study,
we propose a two-stage hierarchical game theoretical model to design an efficient revenue-
sharing rule that can stabilize the coalition. The first stage solves the revenue-sharing problem
in a cooperative game, while the second stage solves TSPs’ resource allocation strategies in a
decentralized non-cooperative game. We analytically prove the stability and efficiency of the
revenue-sharing scheme and derive managerial insights through numerical experiments.

1. Introduction

Well-coordinated, efficient multimodal transport systems bring tremendous benefits to cities and the people who live in them.
Multimodal transportation provides seamless and end-to-end mobility solutions, enabling travelers to use a combination of trans­
portation modes to reach their destinations. For example, individuals can take a bus or train from the city center to a stop near their
destinations and then use a flexible transport service as the first-and-last-mile connection. Aligning with the global urbanization trend,
multimodal transportation is particularly advantageous for individuals living in dense, but spatially large cities such as London and
Beijing. It is also prevalent in long-distance intercity trips, such as railway and local metro connections in Europe and the Guangdong-
Hong Kong-Macao Greater Bay Area (GBA) in China. Some business models such as Mobility-as-a-Service (MaaS) and cooperative
platforms between ride-sourcing and public transit (e.g., Uber Transit) also provide hassle-free combined payment options to travelers
and offer discounts for using multimodal services. Because of its affordability and efficiency, multimodal transportation has been
advocated as a promising alternative to private vehicles, which serves to alleviate traffic congestion and reduce carbon emissions.
These benefits have prompted cities worldwide to develop multimodal networks that combine conventional forms of mass transit with

☆
This article belongs to the Virtual Special Issue on “special issue full title”.
* Corresponding author at: Department of Civil and Environmental Engineering, The Hong Kong University of Science and Technology, Hong
Kong Special Administrative Regions of China.
E-mail address: cesjian@ust.hk (S. Jian).

https://doi.org/10.1016/j.trc.2024.104666
Received 22 November 2023; Received in revised form 15 March 2024; Accepted 15 May 2024
Available online 2 June 2024
0968-090X/© 2024 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

emerging mobility services such as ride-sourcing, on-demand transit, and bike-sharing. However, several challenges impede the
development of multimodal transport systems.
First, developing an efficient multimodal transport system requires cooperation between different transport service providers
(TSPs). However, TSPs are primarily self-interested entities aiming to maximize their profits. They will cooperate with each other only
when they can mutually benefit from such collaboration. Intuitively, if two TSPs’ service links do not overlap, and combining their
service links into a path can serve travelers from more origin–destination (OD) pairs, this cooperation can benefit them, which is
known as vertical cooperation. Fig. 1 (a) illustrates this type of cooperation. Suppose TSP A and TSP B serve travelers from the origin to
the destination sequentially. Through collaboration, both TSPs enhance their customer base and service coverage by leveraging the
complementary services offered by their partners. Nevertheless, it is common for TSPs to also operate parallel services, providing
substitutes and thus competing with each other for the same group of travelers, as shown in Fig. 1 (b). In this situation, cooperation can
be advantageous under certain circumstances, e.g., when both TSPs face supply–demand imbalances. More specifically, if TSP A
experiences an oversupply while TSP B has a shortage of mobility resources, pooling their resources will be beneficial. In the real-world
transport network, operations are considerably more complex, as TSPs concurrently manage sequential and parallel services. The
decision of whether to cooperate depends on the potential benefits it will bring, which are contingent on various factors, including
supply–demand balance, travel demand uncertainty, competition intensity, and the method of dividing the aggregate revenue.
Revenue sharing is an essential aspect of cooperation between TSPs. A partition rule can either be negotiated by the TSPs them­
selves or determined by a third-party platform. Irrespective of the specific business model employed, the crucial issue is to establish a
rational basis for a revenue-sharing rule that ensures all participating TSPs are better off when forming a coalition. The second
challenge lies in designing such a revenue-sharing rule. Cooperative game theory has traditionally been employed as the primary
method for investigating coalition revenue-sharing problems (Peleg & Sudhölter, 2007; Niyato et al., 2011). However, within the scope
of our research, the problem of sharing aggregate revenue among TSPs cannot be solely studied as a cooperative game, as TSPs’ non-
cooperative behavior cannot be ignored. In fact, even in cases where TSPs form a coalition, they continue to operate their mobility
services independently. For instance, Uber and a local transit agency may offer a joint Uber Transit service to multimodal users, while
still operating independent ride-sourcing-only and transit-only services for their respective customers. This entails the need to develop
a model that captures how TSPs determine the amount of resources to allocate to the coalition. Furthermore, it should also be noted
that the strategies of TSPs are influenced by the uncertain nature of travel demand. Hence, demand uncertainty cannot be ignored
when establishing revenue-sharing rules. Conventionally, researchers solve revenue-sharing rules using the Shapley value (Shapley,
1953; Bahinipati et al., 2009) or Nucleolus (Schmeidler, 1969; Kimms and Çetiner, 2012). However, these methods require exponential
computation time. The consideration of TSPs’ non-cooperative resource allocation and demand uncertainty further increases the
computational complexity. Consequently, it is necessary to design an effective revenue-sharing method that can be solved in poly­
nomial time.
To address these challenges, we propose a two-stage hierarchical game theoretical model in this study. The first stage tackles the
revenue-sharing problem within a cooperative game framework, while the second stage, given the designed revenue-sharing rule,
solves TSPs’ resource allocation strategies in a decentralized non-cooperative game, taking into account travel demand uncertainty.
The two stages interact with each other through the revenue-sharing rule, such that the rule affects TSPs’ resource sharing decisions in
the second stage, while TSPs’ responses also influence the determination of the revenue-sharing rule in the first stage. More specif­
ically, our contributions can be summarized as follows:

• Model development: Unlike previous research on the interactions of multimodal TSPs, which either employed cooperative games
to model cooperation or non-cooperative games to depict competition, we develop a two-stage model that combines both coop­
erative and non-cooperative games to account for the coexistence of TSPs’ cooperation and competition in practice. The two stages
are connected by a revenue-sharing rule and TSPs’ resource allocation strategies, which are the key challenges to maintaining
sustainable cooperation.
• Solution approach: We design a revenue-sharing rule that stabilizes the coalition (i.e., no TSPs will be better off by leaving the
coalition) and achieves the maximum revenue for the grand coalition by efficiently allocating mobility resources. Unlike con­
ventional sharing rules solved by Shapley value or Nucleolus given deterministic aggregate revenue, our rule proactively considers

Fig. 1. Two types of cooperation between TSPs.

2
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

TSPs’ responses to the rule in advance, thus guaranteeing the maximization of aggregate revenue. The proposed rule can be
calculated in polynomial time, enhancing its applicability compared to the Shapley value method. Furthermore, in contrast to the
conventional Shapley value or Nucleolus methods, which do not provide a guarantee for the existence of a core, our proposed
revenue sharing rule has been demonstrated to always ensure the presence of a core. This, in turn, ensures the stability of the
coalition.
• Managerial insights: We conduct a numerical experiment with a realistic intercity transport network to validate the efficiency and
stability of the proposed revenue-sharing and resource allocation scheme. By comparing various market structures and demand
scenarios, we observe that: (1) small-sized TSPs with limited mobility resources benefit more from joining the coalition; (2)
aggregate revenue growth is most substantial when supply and demand are approximately equal in magnitude; and (3) external
subsidization might be required when the demand distribution undergoes a dramatic shift. These managerial discoveries can offer
policymakers valuable insights into the viability and necessary financial support of establishing a cooperative multimodal transport
platform.

The rest of the paper is organized as follows. Section 2 reviews related works. The revenue-sharing rule and resource allocation
model is presented in Section 3. Section 4 analyzes the properties of the proposed scheme. In Section 5, we extend the model to fit a
more realistic setting and further discuss subsidies for the participating TSPs. Two numerical studies are conducted in Section 6,
followed by the conclusions and future research directions summarized in Section 7.

2. Literature review

In this section, we briefly review related works on the cooperative and non-cooperative games in transportation engineering and
summarize research gaps.

2.1. Cooperative games in transportation

Cooperative games have been widely applied to addressing revenue and cost sharing problems when multiple operators pool re­
sources and form coalitions to facilitate an integrated service (Nash, 1953), including airline code sharing (Hu et al., 2013), supply
chain cooperation (Özener and Ergun, 2008; Lozano et al., 2013), and wireless communications (Singh et al., 2011; Wu et al., 2014).
However, the application of cooperative game theory in urban transport systems is scarce. Engevall et al. (2004) solved cooperative
vehicle routing games to determine the allocation of transportation costs between customers. Rosenthal (2017) applied the cooperative
game approach to solving the cost allocation problem in a rapid transit network. Pantelidis et al. (2023) designed risk-pooling con­
tracts for TSPs to facilitate horizontal cooperation in the context of network disruptions. Atasoy et al. (2020) studied the vehicle
routing problem in the carrier-shipper market and motivated carriers to collaborate with a third-party platform through individual
rationality constraints. They applied dynamic pricing to minimize the price that the platform pays to the carriers. Most of these studies
disregard the interactions among TSPs, including both vertical cooperation and horizontal competition, which are prevalent in urban
transport networks.
In terms of multimodal transportation, a stream of studies quantify the impacts of cooperation on travel demand and social welfare
by comparing the equilibrium outcomes, e.g., in high-speed rail and airline cooperation (Jiang and Zhang, 2014; Álvarez-SanJaime
et al., 2021). They demonstrated the potential advantages of such cooperation under specific conditions. On the operational level,
Algaba et al. (2019) studied the profit allocation problem in a multimodal public transport system and designed two profit allocation
rules that can stabilize the coalition. Pantelidis et al. (2020) studied the matching problem between travelers and TSPs in a MaaS
platform. They utilized cooperative game theory to derive stable matchings between travelers and TSPs. Schlicher and Lurkin (2022)
investigated collaborative pricing for TSPs to maximize joint profit and proposed a market share exchange rule that always generates
core allocations. These applications of cooperative games in transport systems laid a solid foundation for investigating a cooperative
multimodal transport system.

2.2. Non-cooperative games in multimodal transport systems

Non-cooperative game theory has been mainly applied to investigating the competition among different agents (Nash, 1951). Clark
et al. (2014) modeled the competition between two TSPs that operate complementary transport services in a natural linear transport
chain. In their model, travelers need to make transfers at the interconnected node, so the decisions of two TSPs and the network
structure have an impact on travel demand and ultimately influence the equilibrium outcome. The emergence of shared mobility and
the concern of it competing with incumbent TSPs have motivated more researchers to study its substitute and complement relationship
with public transit and taxis. Heilker and Sieg (2018) applied game theory to analyze the asymmetric competition between a ride­
sharing company and a taxi company and concluded that ridesourcing will harm the social welfare of a transport system in a strictly
regulated taxi market. Cohen and Zhang (2018) studied the competition and cooperation of two ridesourcing companies that operate
independent services, meanwhile pooling their resources in an integrated ridesourcing platform. They characterized such a scenario as
co-opetition and discussed the resulting impacts of co-opetition on ridesourcing companies’ competition intensity, profits, and drivers’
payoffs. Ni et al. (2021) proposed an origin–destination competitive network equilibrium model for the ride-sourcing market. They
modeled the competition among drivers, passengers, and transport network companies (TNCs) in a non-cooperative game, where
passengers seek utility maximization and drivers and TNCs aim at profit maximization.

3
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

Several studies placed the focus on the competition and cooperation between ridesourcing and public transit. Zhu et al. (2021)
examined the cooperative and competitive relationship between ridesourcing and government-owned public transit using a bi-level
model, in which the upper-level solves the cooperative subsidy decision, while the lower-level solves their own competitive strate­
gies. In solving multi-class multimodal multi-provider market equilibrium, Najmi et al. (2023) factored in the behavior of various
market participants, such as travelers’ route selection and operators’ service pricing. They applied a Nash-Cournot structure and
sought an equilibrium structure that simultaneously considered a set of interrelated optimization problems. These studies examine the
effect of competition in multimodal transport systems on various stakeholders, with the majority deriving managerial insights from the
comparison of market equilibria. They provide useful guidelines for modeling the strategic behavior of TSPs in this study.

2.3. Research gaps

Most multimodal transport studies assumed TSPs to be fully coordinated under centralized control but ignored the strategic be­
haviors of TSPs when participating in the coalition. Studies on the interactions among TSPs, on the contrary, assumed TSPs to compete
with each other without coordination. However, in reality, cooperation and competition coexist among TSPs in the multimodal
transport systems. TSPs can form a coalition to offer bundled mobility services to travelers, meanwhile providing independent services
to compete with each other simultaneously. Cooperative game theory is a useful tool to investigate cooperation strategies while taking
individual entity’s incentive into account, however, its application in multimodal transportation is very limited. The two studies by
Algaba et al. (2019) and Pantelidis et al. (2020) are the most relevant to our research. However, their works either considered TSPs to
contribute all resources to the coalition or modeled the cooperation between TSPs and travelers. The revenue-sharing problem for the
cooperative multimodal transport system, allowing TSPs to partially pool resources and maintain independent services, has not yet
been studied.
This study attempts to investigate the revenue sharing and resource allocation problems in a cooperative multimodal transport
system. A distinguishing feature of our research is the two-stage model combining both cooperative and non-cooperative game theory.
We present a revenue-sharing method that is both stable and efficient; it proactively considers the responses of TSPs in advance and can
be solved in polynomial time.

3. The model

3.1. Problem statement

We consider two types of players in a multimodal transport system: the cooperative multimodal platform (CMP) operator and
individual transport service providers (TSPs). The CMP serves to integrate mobility resources from various TSPs, offering compre­
hensive multimodal mobility services to travelers. It enables collaboration among TSPs in both vertical and horizontal manners, as
depicted in Fig. 1. Notably, several successful prototypes of this business model have emerged, such as MaaS platforms that provide
multimodal packages and trip booking applications that offer airline-railway bundled options. The multimodal transport platform can
be implemented on both urban and intercity levels, catering to short-haul trips within cities or regional transport. In the urban context,
individual TSPs encompass services such as metros and taxis, with MaaS platforms like Whim, a pioneering European operator,
functioning as the CMP. Conversely, in the intercity scenario, TSPs comprise airlines, trains and coaches, while trip booking appli­
cations like Ctrip, a leading all-in-one travel platform in China, serve as the CMP.
TSPs contribute a portion of their mobility resources to the CMP, expecting to receive a share of the joint revenues generated. The
nature of these mobility resources depends on the specific TSP; for instance, taxi companies contribute vehicles, whereas train
companies provide seats. Upon obtaining these resources, the CMP operator supplies multimodal mobility services to users and de­
termines the distribution of aggregate revenues among participating TSPs. It should be noted that we assume that the CMP does not
share the aggregate revenues for profit-seeking purposes. Instead, the CMP functions more as a coordinator aiming to maximize
aggregate revenues. In this context, the CMP can be perceived as a government-led corporation dedicated to offering convenient and
affordable transport services to the public, with the overarching objective of constructing an efficient transport system.
Fig. 2 delineates the interactions between the CMP and TSPs. To describe these interactions, we propose a two-stage model. In the
first cooperative stage, the CMP establishes a revenue-sharing rule that ensures the coalition’s stability, taking into account the re­
sources contributed by each TSP. In the subsequent non-cooperative stage, each TSP, upon receiving the revenue-sharing rule, stra­
tegically allocates its mobility resources between cooperative and independent operations to maximize its own profit. The TSPs’
resource allocation strategies are influenced by the announced revenue-sharing rule, which, in turn, depends on the resources that
TSPs share with the CMP. To maintain the stability of the coalition formed by multiple TSPs under the CMP’s leadership, it is essential
to design a reasonable revenue-sharing rule and efficient resource allocation schemes that guarantee all participants benefit from the
collaboration.

4
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

Fig. 2. Illustration of the two-stage model.

Fig. 3. An illustrative example of multimodal transport systems.

To elucidate the model, we utilize an illustrative example to explicate the decisions made by the two parties. Fig. 3 depicts a simple
transport network operated by TSP A and TSP B, connecting three locations: X, Y, and Z. The network consists of three OD pairs: X-Y, Y-
Z, and X-Z. TSP A and TSP B independently operate service link1 1 and service link 2 to accommodate the travel demand from X to Y
and Y to Z, respectively. TSPs A and B allocate a portion of their mobility resources on each link to the CMP, enabling it to operate the
multimodal path from X to Z.
To provide a concrete example, assume TSP A is a bus company with a fixed route from X to Y, and TSP B is an on-demand mobility
company with available vehicles for transporting passengers from Y to Z. Through cooperation, the CMP can offer a convenient
bundled option by coordinating the mobility resources from TSPs; for instance, travelers who book the CMP’s service will be picked up
by a pre-arranged vehicle at Y immediately after disembarking the bus, eliminating additional waiting time. Such seamless services can
attract travelers who previously relied on private vehicles, thereby increasing demand within the system. The demand modeling will
be elaborated later in the paper. TSPs A and B must determine the allocation of mobility resources to the service links operated solely
by themselves to serve their customers (e.g., service link 1 for TSP A and service link 2 for TSP B) and those shared with the CMP, with
the objective of maximizing their own profit. The CMP, on the other hand, must decide on the distribution of revenues collected from
the CMP service path among TSPs A and B based on the amount of resources they contributed. The CMP aims to establish a stable
coalition with TSP A and B, in which both TSPs cannot benefit from leaving the coalition.
We then introduce the notations of the model. Let I denote the set of TSPs. Each TSP i ∈ I operates a distinct set of service links Li
(Note: we use service link and link interchangeably throughout the paper). Let L denote the set of service links operated by all TSPs, i.e.,
L = ∪i∈I {Li }. A link l ∈ L is associated with fare pl and capacity cl . These links construct K paths in the network and we use δlk to indicate
whether a link l is on path k, i.e., δlk = 1 if link l is on path k, otherwise it equals 0. Let Ik denote the set of TSPs that jointly operate on
path k. Note that some paths only consist of one service link, e.g., service link 1 in Fig. 3 is a path itself, indicating it is solely operated
by TSP A. A path k ∈ K is associated with travel fare Pk , which equals the summation of the fare of all links on path k, i.e., Pk =
∑ S M
l∈L δlk pl . Let Ki denote the set of paths solely operated by TSP i and Ki denote the set of multimodal paths in which TSP i participates.
Then, Ki is the subset of paths in which TSP i at least operates one link, i.e., Ki = KSi ∪ KM M
i . We use K to represent all the multimodal
M
{ M}
paths in the network, i.e., K = ∪i∈I Ki . The CMP only operates the multimodal paths. Table 1 summarizes the notations used in this
study for reference.

1
A service link is distinct from a physical transport link in that it signifies the end-to-end transport service between an OD pair provided by a
single TSP, encapsulating the continuity of certain travel modes. For instance, if the TSP is a taxi company and mobility resources pertain to vehicles,
when a traveler takes a taxi from the origin to the destination, they will be served by only one vehicle for the entire trip. Consequently, the mobility
resource should be defined on a link that directly connects the origin and the destination.

5
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

Table 1
Summary of Notations.
Notations Descriptions

Set
I Set of TSPs
L Set of all service links
Li Set of service links operated by TSP i
K Set of all paths
KM Set of multimodal paths
KSi Set of paths solely operated by TSP i
KMi
Set of multimodal paths in which TSP i participates
Ki Set of paths in which TSP i at least operates one link
Ik Set of TSPs that jointly operate path k
Θ Set of feasible revenue sharing rules
ΘS Set of stable revenue sharing rules
ΘE Set of efficient revenue sharing rules
X (θ)/W(θ) Set of all equilibrium solutions under revenue-sharing rule θ

Parameters
δlk Indicator of whether service link l is on path k
cl Capacity of service link l
pl Fare of service link l
Pk Fare of path k
dk Travel demand for path k which follows a certain distribution

Variables
xi : xik Resource allocation strategy of TSP i: the amount of mobility resources that TSP i allocates to path k
xMi
Resource allocation strategy of TSP i for multimodal paths
xSi Resource allocation strategy of paths solely operated by TSP i
θ : θik Revenue-sharing rule: the proportion of the revenues that TSP i shares on path k
w : wk Auxiliary variable: public signal of amount of mobility resources wk on path multimodal k

Functions
πi ( • ) TSP i’s revenue
κi ( • ) TSP i’s total operational cost
fik ( • ) TSP i’s operational cost for operating path k.
Dk ( • ) Cumulative distribution function of travel demand for path k
Dk ( • ) Survival function of travel demand for path k

We use vector xi = (xik )k∈Ki to represent the resource allocation strategy of TSP i, where xik denotes the amount of mobility re­
sources that TSP i allocates to path k. To distinguish between multimodal paths operated by the CMP and the paths solely operated by
∑
TSP i, we further introduce xM S
i = (xik )k∈KM and xi = (xik )k∈KS . A resource allocation strategy is feasible for TSP i if k∈Ki δlk xik ≤ cl ,
i i

∀l ∈ Li , which guarantees that the total amount of mobility resources on link l does not exceed its capacity.
The CMP needs to determine a fixed proportion revenue-sharing rule. A feasible revenue-sharing rule ensures that the revenue
collected on path k is only distributed to the TSPs that share resources on path k. Hence, the set of feasible proportion rates Θ is given
by:
{ }
∑
Θ := θ ≥ 0, s.t. θik = 1, ∀k ∈ K, θik = 0, ∀i ∕
∈ Ik (1)
i∈Ik

where θik is the proportion of the revenues that TSP i shares on path k.
Given a revenue-sharing rule θ ∈ Θ, TSP i’s revenue πi consists of the revenues earned from independent operation πSi and the
revenues collected from the CMP πM i , subtracting the operational cost κ i , which can be calculated by:

πi (xi , x− i , θ) = πSi (xi ) + π M

i (xi ; x− i , θ) − κi (xi )

∑ ∑ [ ( )] ∑
= Pk E[min(dk , xik ) ] + θik Pk E min dk , min{xîk } − fik ( • ) (2)
î∈Ik k∈Ki
k∈KSi k∈KM
i

where x− i represents the resource allocation strategies of other TSPs except TSP i. dk denotes the demand of path k and fik ( • ) denotes
the cost of TSP i for operating path k.
We consider travelers choosing each path as stochastic processes, resulting in the demand of each path following a certain dis­
tribution. This is a common practice in modeling the uncertainty of travel demand (see Wang and Yang (2019) for a comprehensive

6
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

review of demand-related research for ridesourcing systems), especially in data-driven optimization (Rasouli and Timmermans, 2012;
Hu et al., 2013; Xu et al., 2022). We assume that the demand for each path is independent, and its distribution can be estimated from
historical data. The demand distribution may fluctuate over time; for instance, travel demand is typically greater during peak hours
than off-peak hours. With adequate data, it is viable to characterize demand distribution and solve the revenue sharing and resource
allocation problem for different time periods independently. Without loss of generality, the cumulative distribution function of travel
demand of path k is denoted as Dk ( • ). Let Dk ( • ) denote the survival function of this distribution, i.e., Dk ( • ) = 1 − Dk ( • ).
For paths exclusively operated by TSP i, the revenue collected on path k equals the path fare Pk multiplied by the expected number
of served travelers, which is the expectation of the minimum between the demand dk and the mobility resources allocated to this path
xik . For multimodal paths in which TSP i participates, the revenue collected on a specific path k is further multiplied by the revenue-
sharing proportion θik . It should be noted that the number of served travelers on multimodal path k is contingent upon the resource
allocation decisions made by all the TSPs that jointly operate the path. For instance, in Fig. 3, if TSPs A and B decide to share xA and xB
with the CMP, then the effective amount of mobility resources on the CMP service path is min{xA , xB }.
∑
The total cost of TSP i equals the summation of the costs for operating all paths, i.e., κi (xi ) = k∈Ki fik (xik ), where fik ( • ) is the
operational cost of TSP i for operating path k. We assume that the cost function f( • ) is a convex increasing function with respect to the
amount of resources TSP i allocates to path k, i.e., xik . Such convex cost function forms have been widely adopted in prior studies, e.g.,
Jian et el. (2020), Ding et al., (2023).
The revenue of TSP i is influenced not only by the revenue-sharing rule θ, but also by the resource allocation strategies of other TSPs
through the value of min{xîk }, which raises the issue that the profit maximization problems of TSPs are coupled together, making it
î∈Ik
challenging to solve the Nash equilibrium. We will discuss how to tackle this issue in Section 4.

3.2. The two-stage model

In this subsection, we present a formal mathematical formulation of the two-stage model. The objective of the CMP is to attain a
centralized optimal outcome for the coalition, by maximizing the collective revenue generated by the cooperating TSPs. The imple­
mentation of a centralized optimum necessitates that TSPs allocate their mobility resources in a centralized optimal manner.
Nevertheless, TSPs are only inclined to do so if this maximizes their own profits. Therefore, the CMP designs a revenue-sharing rule
with the dual purpose of fostering cooperation among all TSPs and establishing a stable grand coalition, while also providing incentives
for adherence to the coalition’s centralized optimal resource allocation. This indicates that the CMP’s decision is contingent upon the
attainment of the centralized optimum.
With this in mind, in the first stage, we assume that all the TSPs pool their mobility resources together to solve the centralized
optimum, based on which the CMP establishes a stable revenue-sharing rule to ensure every cooperating TSP will be better off.
Subsequently, in the second stage, notified with this rule, TSPs will strategically partition their resources to maximize their own profits.
It is anticipated that with the proposed revenue-sharing rule, the resource allocation schemes implemented by TSPs will be in
accordance with the centralized optimal solution, which aims to maximize the aggregate revenue. In brief, the proposed revenue-
sharing rule should guarantee the following properties:

• Stability: all TSPs are ensured to obtain higher profits by cooperating with the CMP such that they will not leave the coalition.
• Efficiency: the aggregate revenue of the whole multimodal system is maximized through TSPs’ strategic resource allocation
decisions.

The formal definitions of stability and efficiency will be presented in the following subsections.

3.2.1. The first stage: CMP’s decision in the cooperative game

In this subsection, we formulate the first-stage cooperative game. First, we define the characteristic function of the game, i.e., the
aggregate revenue of the coalition. Suppose a subset of TSPs N ⊆ I join the coalition coordinated by the CMP and contribute all of their
resources. The CMP can decide the amount of mobility resources allocated to each path (including paths solely operated by TSPs in N
since we assume they share all the resources in this stage). Then, the maximum revenue of the CMP with cooperating TSPs in set N is
solved from the following problem:
∑ [ ( ) ] ∑∑ ( N)
P1 max Pk E min dk , xNk − fik xk (3)
xN ≥0
k∈KN k∈KN i∈Ik

Subject to
∑
δlk xNk ≤ cl , ∀l ∈ LN (4)
k∈KN

( )
The decision variable xN = xNk k∈KN represents the amount of mobility resources that CMP would allocate to each path. Generally, all
the symbols with a subscript or superscript N can be interpreted as the parameters or variables associated with the TSPs in set N. For
example, KN and LN denote the sets of paths and links that the CMP can operate after TSPs in set N sharing all their resources. In P1, we

7
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

do not identify the decisions of TSP i because all TSPs are assumed to follow the central decision made by the CMP in the cooperative
stage. Constraints (4) ensure that the resource allocation solution by the CMP is feasible considering link capacity constraints.
Let V(N) represent the optimal objective value of P1 with respect to set N. V( • ) is also known as the characteristic function of the
cooperative game. Let xI be the optimal solution to P1 for the grand coalition, which is formed when all TSPs in set I agree to cooperate
with the CMP. As discussed before, a stable revenue-sharing rule guarantees that no subset of TSPs can be better off by suspending the
grand coalition and forming a smaller coalition. In other words, the maximal aggregate revenue of any smaller coalition (the left-hand
side of condition (5)) is no greater than the total revenues of these TSPs that can obtain from the grand coalition, less the operational
cost (the right-hand side of condition (5)). The formal definition is given as follows.
Definition 1 (Stable revenue-sharing rules):. A revenue-sharing rule θ ∈ Θ is stable if for any coalition N ⊆ I, we have
∑∑{ [ ( )] ( )}
V(N) ≤ θik Pk E min dk , xIk − fik xIk (5)
k∈K i∈N

We define ΘS ∈ Θ to be the set of stable revenue-sharing rules.

3.2.2. The second stage: TSPs’ non-cooperative decisions

In this subsection, we model the second-stage non-cooperative game of TSPs, in which each TSP aims to maximize its own profit
through strategic mobility resource allocation, given the CMP’s revenue-sharing rule in the first stage.
We start by charactering the Nash equilibrium with respect to the given revenue-sharing rule θ. Each TSP determines the resource
allocation strategy xi to maximize its revenue πi by solving the following problem P2:
P2 maxπi (xi ; x− i , θ) (6)
xi ≥0

Subject to
∑
δlk xik ≤ cl , ∀l ∈ Li (7)
k∈Ki

where πi (xi ; x− i , θ) is calculated by Eq. (2).

( )
Let x* = x*1 , x*2 , ⋯, x*i i∈I denote the Nash equilibrium associated with θ where resource allocation strategy x*i for TSP i is solved
from P2. We use X (θ) to represent the set of all equilibrium solutions under revenue-sharing rule θ.
An efficient revenue-sharing rule ensures that the optimal resource allocation solution xI for the CMP in the first stage can be
implemented by one of the Nash equilibria associated with this rule in the second stage. In other words, it realizes a situation in which
the shared resources on each multimodal path determined by selfish TSPs in the second stage can satisfy the optimal resource allo­
cation solution in the first stage. The formal definition is given as follows.
Definition 2 (Efficient revenue-sharing rules):. A revenue-sharing rule θ ∈ Θ is efficient if

x ∈ X (θ)
I
(8)

We define ΘE ∈ Θ to be the set of efficient revenue-sharing rules. The goal of this study is to derive the revenue-sharing rules in the set
ΘS ∩ ΘE , which are both stable and efficient. We develop the above two-stage model to solve the problem and propose an eligible
revenue-sharing rule in Section 4.

4. Theoretical analysis

In this section, we propose a stable and efficient revenue-sharing rule and present corresponding proofs. Our proposed revenue-
sharing rule θ* is given as follows:
∑ ( I)
l∈Li δlk μl + fik xk
I ’
θ*ik = ∑ ∑ ( I ), ∀i ∈ I (9)
î∈Ik fîk xk
I ’
l∈L δlk μl +

where μIl is the Lagrange multiplier (shadow price) of link l solved from P1 for the grand coalition.
This rule is calculated based on the shadow price and derivative of the operational cost function of each link. TSP i’s share on path k
is proportional to the sum of the shadow price of its links on path k and the derivative of its cost function of path k, where the shadow
price exhibits a strong correlation with the capacity of the link. It should be noted that there may be multiple stable and efficient
revenue-sharing rules in existence, of which our proposed rule is an easily computable one, i.e., θ* ∈ ΘS ∩ ΘE . This rule reflects the
marginal contribution of TSP i on path k, which is consistent with the fundamental principle of the classic distribution approach, such
as the Shapley value. However, in contrast to conventional methods that require lengthy computation time, these proportions in Eq. (9)
can be obtained immediately once the grand coalition problem P1 is solved. We present Lemma 1 to claim the existence of the pro­
posed optimal revenue-sharing rule.

8
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

Fig. 4. Illustration of the proving process.

Lemma 1. The revenue-sharing rule θ* defined by Eq. (9) always exists.

The detailed proof is provided in the Appendix A.
Fig. 4 provides a summary of the methodology employed to establish the stability and efficiency of the proposed revenue-sharing
rule. To begin with, we decouple TSPs’ profit maximization problems P2 in the second stage into multiple independent problems P3 by
introducing an auxiliary vector w. This auxiliary vector can be interpreted as a public signal announced by the CMP (Hu et al., 2013).
Note that in our context, a public signal does not refer to a physical traffic signal, but rather to a quota of the maximal amount of
mobility resources that TSPs share on multimodal paths. More detailed explanations of the public signal will be presented later. A
specific revenue-sharing rule can trigger multiple Nash equilibria, each associated with a unique public signal (Lemma 2). According to
Definition 2, we were supposed to show xI implements a Nash equilibrium under θ* . Nevertheless, this is hard to prove. Instead, we
formulate the condition to identify which public signal is capable of generating a Nash equilibrium by examining the Kar­
ush–Kuhn–Tucker (KKT) conditions (Proposition 1). Next, we demonstrate that when the grand coalition optimal solution xI acts as a
public signal for multimodal paths, the derived condition is satisfied under our proposed revenue-sharing rule θ* . This indicates that xI
belongs to the set of Nash equilibria that are induced by θ* , which implies efficiency (Proposition 2). Finally, we prove the stability of
the proposed revenue-sharing rule with cooperative game theory (Proposition 3).
We start by analyzing the Nash equilibria in the second stage, then back to the cooperative game in the first stage. Recall that the
profit of TSP i given by Eq. (2) is not only dependent on its own strategy xi , but also influenced by the resource allocation strategies of
other TSPs through the minimum amount of mobility resources allocated to those cooperative multimodal paths, i.e., min{xîk }. The
î∈Ik
coupled problem complicates the equilibrium solution.
We develop a decomposition method to tackle this difficulty. Specifically, we decompose P2 into a set of independent subproblems
for each TSP by introducing an auxiliary vector w = (wk )k∈KM . This auxiliary vector, also known as the public signal, indicates the
maximal amount of mobility resources that TSPs can share on multimodal paths. The CMP shall determine an optimal public signal that
maximizes the grand coalition’s aggregate revenues. TSPs will adhere to this public signal as long as they are members of the CMP-
coordinated coalition. Considering the majority of multimodal transport systems globally are currently managed by governments,
this assumption applies to our context. Upon agreeing to participate in the collaboration, TSPs will essentially comply with the
∑
centralized decisions made by the CMP. A public signal is feasible if k∈KM δlk wk ≤ cl for all link l ∈ L.
Given the public signal w, the profit maximization problem for TSP i can be reformulated as follows:
∑ ∑ ∑
P3 max Pk E[min(dk , xik ) ] + θik Pk E[min(dk , xik ) ] − fik (xik ) (10)
xi ≥0
k∈KSi k∈KM k∈Ki
i

Subject to
∑
δlk xik ≤ cl , ∀l ∈ Li (11)
k∈Ki

xik ≤ wk , ∀k ∈ KM
i (12)

The objective function in P3 is still to maximize the profit for each TSP, except that now the profit πi of TSP i is only dependent on its
own resource allocation strategy because the impact of other TSPs has been eliminated by the public signal. Constraints (12) indicate
that no TSP will share more resources on a multimodal path than the public signal.
Next, we discuss how to find a public signal that achieves efficient resource allocation. Let xi (w, θ) denote the solution of P3 for TSP
i given public signal w and revenue-sharing rule θ. Now we can introduce another way to characterize the Nash equilibria by defining
the following set:
{ }
W(θ) := w ≥ 0 : xik (w, θ) = wk , ∀k ∈ KMi ,i ∈ I (13)

9
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

This set contains the public signals that all TSPs would follow, i.e., every TSP allocates exactly wk for multimodal path k. The following
lemma demonstrates that W(θ) also characterizes the set of all Nash equilibria and can be regarded as an alternative to X (θ).
Lemma 2. For any w ∈ W(θ), there exists a unique Nash equilibrium x(w) ∈ X (θ) such that xik (w) = wk for all k ∈ KM
i , i ∈ I.
Conversely, for any x* ∈ X (θ), there exists a unique w(x* ) ∈ W(θ) such that wk = x*ik for all k ∈ KM
i , i ∈ I.
The detailed proof is provided in the Appendix B.
Lemma 2. indicates that we can find the Nash equilibrium by finding those public signals w such that Constraints (12) hold with
equality at optimality. Based on Lemma 2, P3 can be reformulated as follows:
∑ ∑
P4 max Pk E[min(dk , xik ) ] − fik (xik ) (14)
xSi ≥0
k∈KSi k∈KSi

Subject to
∑ ∑
δlk xik ≤ cl − δlk wk , ∀l ∈ Li (15)
k∈KSi k∈KM
i

The decision variables of TSP i in P4 are the resource allocation among paths solely operated by TSP i. P4 is a concave problem, so we
can characterize its optimal solution (primal and dual) using KKT conditions.
∑
Proposition 1:. Given a feasible w, i.e., w ≥ 0, k∈KM δlk wk ≤ cl ,∀l ∈ L, and wk ≤ dk , we shall have w ∈ W(θ) if and only if there exists
S
a vector of Lagrange multipliers μi for problem P4 such that
∑
θik Pk Dk (wk ) − fik’ (wk ) ≥ δlk μSil , ∀i ∈ I, k ∈ KM
i (16)
l∈Li

This inequality ensures that the marginal profit that TSP i can obtain by allocating one additional unit of mobility resource on the
multimodal path is greater than or equal to the opportunity cost of allocating the same amount of resource on the path solely operated
by TSP i. Hence, in a Nash equilibrium with respect to public signal w, no TSP can obtain a higher payoff by deviating from the resource
allocation strategy that the public signal suggests. With Proposition 1, to prove the proposed revenue-sharing rule is efficient, we only
need to show condition (16) is satisfied by the rule θ* .
Proposition 2:. The revenue-sharing rule θ* defined by Eq. (9) is efficient.
Proposition 3:. The revenue-sharing rule θ* defined by Eq. (9) is stable.
Proposition 1 characterizes the condition of a public signal that is capable of inducing a Nash equilibrium. Then by deriving the KKT
conditions of the grand coalition maximization problem P1, we demonstrate that when xI functions as a public signal for multimodal
paths, condition (16) is satisfied under θ* . Hence, xI is one of the Nash equilibria led by θ* , which indicates that θ* is efficient.
By comparing the maximal attainable revenue of any smaller coalition to the revenue collected when remaining in the grand
coalition, we demonstrate that the proposed revenue-sharing rule ensures stability in Proposition 3. The detailed proofs of Propo­
sitions 1–3 are presented in the Appendix C, D, and E, respectively.
Next, we use the example presented in Fig. 3 to elucidate that the proposed revenue-sharing rule guarantees the stability and
efficiency of the coalition. The properties for links and paths are presented in Table 2. In this example, we assume that the cost function
adheres to a quadratic structure, resulting in a positive relationship between the assigned mobility resources and the corresponding
marginal cost. There exist three OD pairs, each characterized by a certain distribution of travel demand. TSPs A and B need to
determine how much mobility resources to allocate to the OD pairs they independently operate, i.e., X-Y for TSP A and Y-Z for TSP B,
and how much to allocate to the multimodal path, i.e., X-Y-Z. The CMP needs to decide how to share the aggregate revenue of the
multimodal path between TSP A and TSP B.
Without cooperation with the CMP, TSP A and TSP B independently operate their service links to serve OD pairs X-Y and Y-Z,
respectively. Then the optimal resource allocation scheme is xA1 = 22.06, xB2 = 21.13 by solving P2, and the corresponding maximal
revenues of two TSPs are πA = $55.15, πB = $105.63, respectively. The total revenue equals $160.78. With the CMP, we can calculate
the optimal revenue-sharing rule by Eq. (9), which is θ*A = 0.534, θ*B = 0.466. Under this rule, the optimal resource allocation of the

Table 2
Attributes of service links and paths in Fig. 3.
Attributes Operator Capacity Fare ($) Length (km) Cost function

Service link 1 A 30 5 4 0.03x2

2 B 40 10 6 0.07x2

Attributes Operator Fare ($) Demand distribution

Path 1: X-Y A 5 Unif[0, 30]
2: Y-Z B 10 Unif[0, 30]
3: X-Y-Z CMP 15 Unif[0, 20]

10
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

grand coalition is achieved: xA1 = 15.72, xB2 = 21.13 and xA3 = xB3 = 14.28. The total revenue equals $273.57 with π*A = $118.06 and
π*B = $155.51, respectively. In this example, both TSPs obtain higher payoffs and serve more travel demand through cooperation,
which demonstrates the benefits that CMP brings to the transport system. This also guarantees the stability of the cooperation.
To show the efficiency of the proposed revenue-sharing rule, we consider an alternative way to partition the revenue from the CMP.
Suppose we announce that TSP A obtains $5 from the CMP while TSP B obtains $10, i.e., θʹA = 13, θʹB = 23. Then, there exist multiple
market equilibria. The corresponding optimal revenues for two TSPs at equilibrium are in a range, i.e., πʹA ∈ [88.74, 91.08], πʹB ∈
[180.97, 183.76]. Under this revenue-sharing rule, both TSPs also obtain greater revenues compared with the situation without the
CMP. However, the maximal total revenue of all equilibria is lower than that under the optimal revenue-sharing rule. Consequently,
our proposed revenue-sharing rule not only stabilizes the coalition but also achieves efficiency by maximizing the total revenue of the
grand coalition.

5. Extension – Variation in demand distribution

In this section, we extend the model to account for demand elasticity and discuss subsidization scheme when necessary.

5.1. Demand variation

In the aforementioned model, we assume the demand distribution remains the same before and after introducing the CMP. There
are several reasons that justify this assumption. First, in the short run, travel demand would not fluctuate significantly with the newly
implemented CMP. For regular trips, e.g., commute trips, most travelers tend to choose a fixed travel mode and route as part of their
daily routine. Such travel patterns are formed based on their past experience regarding the travel time and cost of each route, which
will not change with the introduction of the CMP at once. There exists a transition period for travelers diverting from their original
choice to the emerging seamless services offered by the CMP. Secondly, we model travel demand as a stochastic variable and char­
acterize it with a certain distribution. The variation and uncertainty of the demand, to some extent, have already been captured in the
distribution. Hence, we keep the demand distribution unchanged for theoretical analysis to derive the optimal revenue-sharing rule in
Section 4.
In this section, we consider the problem in the long-run, which means the operations of the CMP will gradually impact the travel
pattern and the demand distribution accordingly. In practice, in order to attract travelers to use its mobility services, the CMP usually
offers fare discounts to travelers. Such promotions will influence travelers’ choices, further changing the distribution of travel demand.
In the example illustrated in Fig. 3, without the CMP, some travelers depart from X and arrive at Z by transferring at Y. They are
counted into the demand of OD pairs X-Y and Y-Z separately. After introducing the CMP, they have a more convenient and economic
option: X-Z service by the CMP. Then, the travel demand of X-Y and Y-Z may decrease accordingly as some travelers will be attracted to
use the CMP path. Considering demand variation, efficiency and stability of our proposed revenue-sharing rule in Section 4 will be
affected. Hence, the subsequent discussion is to identify whether the two properties remain valid and to devise countermeasures if they
are not guaranteed in the new setting.
Remark 1. The proposed revenue-sharing rule defined by Eq. (9) is still efficient considering demand distribution variation.
The reason is that when demand distribution is changed, we can solve P1 under the new distribution for the grand coalition and
obtain the optimal resource allocation solution. With this solution and corresponding Lagrange multipliers, we can calculate the
revenue-sharing rule, which still satisfies condition (16) of Proposition 1 under the new demand distribution. Therefore, the efficiency
of our proposed revenue-sharing rule still holds. The new demand distribution can be characterized by collecting customer travel data
when the CMP’s operation reaches a state of stability.
Remark 2. The proposed revenue-sharing rule defined by Eq. (9) is not necessarily stable considering demand changing.
Compared with not cooperating with the CMP, TSPs’ total payoffs are not necessarily higher. This is due to the potential drop in the
number of travelers using the paths that TSPs independently operate, as well as the possibility that the revenues shared from the CMP
may not compensate for the loss incurred from independent operations. Travel demand may shift from services by individual TSPs to
the more convenient and economic services offered by the CMP. Such changing in demand distribution is mostly induced and impacted
by travel expenses. Specifically, the greater the discount promoted by the CMP, the more significant the travel demand distribution
changes. To capture such pattern, fare discounts are employed to delineate the change in travel demand distribution. Let γ denote the
fare discount offered by the CMP, then the cumulative distribution function of demand for path k can be written as Dk (γ). In general,
Dk (γ) becomes flatter for multimodal paths, i.e., k ∈ KM , as these paths attract more travel demand, and the demand distribution turns
more dispersed with the increase of the upper limit. To illustrate this, let us consider the case of a uniform distribution. The shape of the
cumulative distribution function curve will deviate from Fig. 5(a) to Fig. 5(b) as the upper limit increases. Conversely, Dk (γ) of paths
solely operated by individual TSPs, i.e., k ∈ K/KM , shifts to the left due to decreasing travel demand, e.g., from right to left in Fig. 5.
We continue to use the illustrative example depicted in Fig. 3 to demonstrate the issue arising from demand changing. Assuming
that the CMP implements a 50 % reduction in the trip fare, i.e., γ = 0.5, the demand distribution undergoes corresponding modifi­
cations, as presented in Table 3. Compared with Table 2, the demand distribution of paths independently operated by TSPs (X-Y and Y-
Z) has transitioned from Unif[0, 30] to Unif[0, 15]. While the process of realistic altering can involve more intricate factors, in this
context, we will straightforwardly multiply the upper limit of the uniform distribution by γ to illustrate the possible failure of stability.

11
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

Fig. 5. Illustration of change in cumulative distribution function curve.

Table 3
Attributes of paths considering demand changing.
Paths Operator Fare ($) Demand distribution (without CMP) Demand distribution (with CMP)

X-Y A 5 Unif[0, 30] Unif[0, 15]

Y-Z B 10 Unif[0, 30] Unif[0, 15]
X-Y-Z CMP 7.5 / Unif[0, 20]

Recall that without cooperating with the CMP, the maximal payoffs of two TSPs are πA = $55.15, πB = $105.63. Considering the
demand changing, we calculate the optimal revenue-sharing rule by Eq. (9), which is θ*A = 0.3, θ*B = 0.7. The total revenue equals
$142.67 with π *A = $46.45 and π *B = $96.22, respectively. Apparently, in this case, the CMP does not yield advantages for both TSPs.
The revenues shared from the CMP cannot offset the revenue loss on their independently operating paths due to changing demand.
Hence, the coalition is unstable, and TSPs will cease cooperation with the CMP. In that case, the CMP cannot provide multimodal
service. The travel demand distribution will revert to the state in Table 2, where TSPs benefit more through independent operation.

5.2. Subsidization scheme

Facing the potential instability issue arising from demand variation, we resort to subsidization to stabilize the coalition (Bachrach
et al., 2009; Liu et al., 2018). Since the demand for paths that TSP independently operate may decrease, we design the subsidization
scheme in the following way:
Remark 3. The amount of subsidy for TSP i equals the difference between the revenues collected from their independently operating
paths before and after demand changes, i.e., S = ΔπSi .
For instance, in the above case with γ = 0.5, before introducing the CMP, TSP A receives $55.15 by independently operating service
link 1. With the CMP, the travel demand for OD pair X-Y decreases, resulting in the revenues collected on this path dropping to $31.78.
Then the subsidy for TSP A equals the difference, i.e., $23.37. As a result, TSP A finally receives $69.82, which is clearly greater than
$55.15. This ensures that TSP A has no incentive to break the cooperation, as such the stability is achieved.
It should be noted that in certain circumstances, subsidies are not required. When demand distribution does not change signifi­
cantly, the revenues shared from the CMP will compensate for the profit loss on TSPs’ independently operating paths. In the above
example, if the fare discount is set to γ = 0.8, then demand distribution of path X-Y and Y-Z follows uniform distribution of U[0, 24]
after the introduction of the CMP. Under this demand pattern, the payoffs for TSP A and TSP B will be $88.27 and $136.81, respec­
tively. In this case, the coalition has been stabilized even without subsidization. We will conduct sensitivity analysis to investigate the
impacts of demand elasticity through numerical studies in Section 6.2.3.
Overall, in light of the changing demand distribution, we maintain our proposed revenue-sharing rule to maximize the total
revenue. By subsidizing TSPs externally when necessary, we can stabilize the cooperation.

6. Numerical study

In this section, we present two numerical studies to demonstrate the efficiency and stability of the proposed revenue-sharing rule
and resource allocation scheme. First, we validate the efficiency of the proposed revenue-sharing rule by comparing it to two typical
revenue-sharing rules applied in practice using the Great Bay Area (GBA) intercity transport network in China. Next, to derive
additional managerial insights, we use a more flexible urban transport network (Nguyen Dupuis network) to investigate two market
structure scenarios: (1) two TSPs of comparable size; (2) a large TSP and a small TSP. The benefits of cooperating with the CMP for
different individual companies are compared. Further, considering demand elasticity, we conduct sensitivity analysis on the demand
distribution variation and provide corresponding managerial insights.

6.1. Validation and comparison

The GBA encompasses two Special Administrative Regions, namely Hong Kong and Macao, along with nine cities located in
Guangdong Province. To fully leverage the composite advantages of these cities, foster in-depth integration within the region, and

12
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

Fig. 6. Great Bay Area intercity railway lines.

stimulate coordinated regional economic development, the GBA has made significant advancements in its intercity transportation
system, particularly in the realm of railway system, which has greatly enhanced the convenience for commuters traveling between
cities. According to an annual report published by the Guangzhou Municipal Planning and Natural Resources Bureau (2023), the
overall commuting population in GBA in 2022 reached 1.18 million. The typical trajectories for this cohort of travelers would likely
involve three stages: (1) commuting via bus from their place of residence to the railway station; (2) utilizing the railway system to
reach the desired destination city; and (3) utilizing another bus service to commute from the railway station to their respective
workplaces. Currently, the railway transport system in the GBA is managed by Guangzhou Metro Group and Guangdong Intercity
Railway Operation, while bus lines are operated by local bus companies in different cities. Mobile payment methods, such as Alipay,
have gained widespread adoption and become the predominant means of payment for railway and bus travel in the GBA. Additionally,
electronic payment cards like Yang Cheng Tong can be used by travelers to access transport services across various cities in the GBA.
These advanced payment methods have facilitated the development of an integrated ticketing system, enabling the provision of
transfer discounts. The CMP can further leverage the advantages of these payment systems to benefit travelers, especially those who
frequently transfer between different travel modes. The CMP can coordinate bus timetables with the railway schedule, allowing for
seamless transfers and providing combined tickets or monthly travel bundles. This integration also facilitates a fair distribution of
aggregate revenues among the cooperating TSPs. For clarity in the following numerical tests, we assume that there are three players in
this intercity transport system: the railway company, the bus company, and the CMP.
The numerical study on the cooperation between railway and bus companies involves the examination of several railway lines and
bus lines, encompassing both urban and coach bus services, as seen in Fig. 6. In this scenario, we assume the CMP will operate several
bus-railway-bus routes connecting cities with large travel demand. An example of such a route would be the Guangzhou-Foshan line.
Real-world data is utilized to determine the network parameters, including the trip pricing and travel demand associated with each
service link. Detailed network data is provided by Table 7 and Table 8 in Appendix F.
Using this realistic intercity network, we compare our proposed optimal revenue-sharing rule θ* with two typical revenue-sharing
rules: mileage-based revenue-sharing rule, denoted as θM , and fare-based revenue-sharing rule, denoted as θP . Mileage-based revenue-
sharing rule decides the proportion that one TSP can obtain from the multimodal path according to the length of its links. The
mathematical formulation is presented as follows:
∑
δlk ml
θM = ∑l∈Li , ∀i ∈ I (17)
l∈L δlk ml
ik

where ml denotes the length of link l.

Fare-based revenue-sharing rule is quite similar, except that the determinant factor is the current fare of links. The mathematical
formulation is presented as follows:
∑
l∈Li δlk pl
θPik = , ∀i ∈ I (18)
Pk

We continue to use the example in Fig. 3 (with attributes listed in Table 2) to better illustrate these two revenue-sharing rules. Under
mileage-based revenue-sharing rule, the proportions for TSP A and B are θM 4 M
A = 4+6 = 0.4,θB = 0.6; Under fare-based revenue-sharing
rule, the proportions for TSP A and B are θFA = 5+10
5
= 13, θFB = 23.
Note that under these two rules, there exists multiple Nash equilibria in the second stage. We select the one that maximizes the total
revenue of two companies for comparison. The results are presented in Table 4.

13
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

Table 4
The revenues of two TSPs under different revenue-sharing rules.
Railway company Bus company Total

Independent operation Shared from the CMP Total Independent operation Shared from the CMP Total

Without CMP 74952.03 / 74952.03 11,097 / 11,097 86049.03

θ* 74910.36 7214.42 82124.78 10625.44 3645.85 14271.29 96396.07
θM 74910.36 7737.13 82647.49 10625.44 3118.54 13743.98 96391.47
θP 74910.36 9031.59 83941.95 10,661 1737.5 12398.5 96340.45

1. The unit of figures in the table is ¥Chinese Yuan/hour.

As shown in Table 4, regardless of the revenue-sharing rule, both companies obtain higher profits after cooperating with the CMP.
Among these three revenue-sharing rules, θ* achieves the highest total revenue, thereby verifying its efficiency. In this case, we notice
that θM and θP yield near-optimal solutions, which suggests that they could be alternative revenue-sharing rules in practice with a lack
of demand information. For the railway company, θP brings the highest profit, whereas it is the most adverse revenue-sharing rule for
the bus company. This reveals the fact that the most beneficial revenue-sharing rules are likely to vary across for different TSPs. If the
CMP lacks the ability to dictate the revenue-sharing rule, the rule negotiated by TSPs largely depends on their market power, which
will raise equity concerns and sacrifice the profits of small-sized TSPs. In practice, since the majority of multimodal transport systems
are led by the government, including this GBA case, it is more plausible that the CMP offers a cooperative contract with the pre-
designed revenue-sharing rule. TSPs have the option to either accept or reject the contract but lack bargaining power with the
CMP regarding the revenue-sharing rule. As long as the cooperation ensures more profits, TSPs would be inclined to share their
mobility resources. Therefore, we can regard the CMP has full control over the authority to determine revenue-sharing rules. In order
to maximize the aggregate revenues of all TSPs, the CMP will choose θ* over other alternative revenue-sharing rules, which also
prevents TSPs from suffering from unfair revenue-sharing rules to some extent.
In this numerical experiment, the railway company dominates the bus company overwhelmingly in terms of market power,
embodied in travel demand and available mobility resources. This results in the railway company’s total revenue being significantly
greater than that of the bus company, which is consistent with reality. In order to obtain a comprehensive understanding of the impacts
of cooperating with the CMP under a variety of scenarios, e.g., under different market structures, we utilize a more flexible urban
network for further examination in Section 6.2.

6.2. Exploration of managerial insights

We use the Nguyen Dupuis network (Fig. 7) to examine the role that CMP plays in different scenarios. Consider there are two TSPs
operating in this urban network, which could be taxi, ridesourcing, metro, etc. Without a CMP, they operate their services indepen­
dently, and travelers transfer at nodes to use services from two TSPs with separate payment and unpredictable waiting time. With the
CMP, we assume that the CMP provides travelers with seamless services between four long-distance OD trips (as shown in Fig. 7) by
operating several multimodal paths after TSPs share mobility resources with it. Detailed network data is provided by Table 9 and
Table 10 in Appendix G.

6.2.1. Different market structures

We start by investigating two market structure scenarios that differ in the market power of two TSPs: (1) two TSPs of comparable
size; (2) a large TSP and a small TSP. This serves to characterize the impacts of cooperation with the CMP on different individual

Fig. 7. Nguyen Dupuis network.

14
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

Table 5
The revenues of two TSPs with and without a CMP.
Without CMP With CMP

Independent operation Independent operation Shared from the CMP Total

The first market Total 1048.73 914.22 340.75 1254.97

TSP A 501.31 429.83 208.16 637.99
TSP B 547.42 484.39 132.59 616.98
The second market Total 453.33 391.56 161.44 553.00
TSP A (large) 368.92 318.32 82.83 401.15
TSP B (small) 84.41 73.24 78.61 151.85

1. The unit of figures in the table is ¥ Chinese Yuan.

entities. We use our proposed rule θ* to partition the revenues of multimodal paths and the fare discount is set at 0.9, i.e., γ = 0.9.
Table 5 summarizes the results.
The implementation of the CMP in both market structures leads to an enhancement in the profitability of both TSPs. The second
market structure exhibits a slightly higher rise in total revenue (22.0 %) compared to the first market structure (19.7 %). This trend is
consistent throughout the other two demand scenarios analyzed. This might result from the expanded customer base that CMP brings
to the small-sized TSP. Unlike the first market structure, where two TSPs compete with each other with comparable operational ca­
pabilities (available mobility resources, cost management, etc.), the small-sized TSP is inherently disadvantaged in competitiveness
against its larger counterpart due to its constrained investment capacity. Collaboratively engaging in the sharing of mobility resources
with the CMP would be a judicious decision, as it would enable the small-sized TSP to sustain profitability by prioritizing collaboration
over rivalry.

Fig. 8. The revenues of two TSPs under different demand scenarios.

15
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

Moreover, in the second market structure, TSP A and TSP B increase their profits by 8.7 % and 79.9 %, respectively. This finding
implies that the proposed scheme also favors the small-sized TSP over the larger one. This is because the small TSP is characterized by
limited capacity and higher operational expenses, resulting in larger Lagrange multipliers and first-order derivatives. Consequently,
these factors contribute to a larger proportion in Equation (9). Hence, within the context of the second market structure, the imple­
mentation of the CMP aligns with the principles of antitrust legislation and contributes to the augmentation of TSP B’s market in­
fluence, enabling it to engage in competition with TSP A to a certain degree.

6.2.2. Different demand scenarios

Next, we test our proposed scheme under different demand scenarios. These scenarios differ in the relative proportions of the
available supply compared to demand. In the low, medium, and high scenarios, the average capacity-to-expected travel demand ratio
of all service links is 2.24, 1.50, and 1.12, respectively. The results are presented in Fig. 8.
When examining three different demand scenarios, it is shown that the implementation of the CMP yields the most benefits to two
TSPs when the level of demand is approximately equal to the level of supply. In these two market structures, the total revenue ex­
periences a significant increase of 26.3 % and 34.1 % respectively. Intuitively, when travel demand is low, there exists a sufficient
supply of mobility resources to accommodate the need. While the CMP has the potential to stimulate new demand, the magnitude of
this increase is very low when compared to the existing supply. Whereas when demand is high, the mobility resources are in short
supply. Although TSPs have the willingness to serve the new demand that the CMP brings, they face limitations in their ability to meet
this demand due to the binding capacity constraints. Both scenarios improve total revenue less significantly compared to the medium
demand level, where the CMP attracts sufficient new demand and two TSPs are able to serve it.
When comparing two market structures, it is evident that the disparity in the increase of total revenue between the second market
structure and the first market structure is more noticeable when the demand is at a medium level (34.1 % compared to 26.3 %) as
opposed to a low level (22.0 % compared to 19.7 %) and a high level (25.9 % compared to 19.9 %). This suggests that the antitrust
implications of the CMP are more pronounced when there is a balance between supply and demand.

6.2.3. Demand elasticity

In light of the discussion on demand elasticity in Section 5, we perform a sensitivity analysis on the fare discount parameter γ to
accommodate changes in demand distribution within the first market structure. Under uniform distribution, we assume a fare discount
γ changes the demand distribution of those paths independently operated by TSPs from [0, U] to [0, γU]. Note that while the demand
variation in practice may be more intricate, a downward trend is assured; therefore, for the sake of illustration, we employ a simple
form here. The results of the sensitivity analysis are presented in Table 6.
As γ decreases from 1.00 to 0.70, the demand distribution undergoes a more pronounced change, resulting in reduced profitability
for both TSPs. Subsidies are necessary when the demand distribution is strongly impacted, specifically when the value of γ is less than
or equal to 0.8. Intriguingly, although a higher fare discount attracts more travelers using the multimodal services provided by the
CMP, it leads to a decrease in the demand for independently operated service links, as well as the decline in the proportion of revenue
shared from the CMP. This is due to the considerably low multimodal service fare, diminishing the benefits of sharing resources with
the CMP.
From the above experiments, we derive the following policy implications:

• First, our proposed revenue-sharing rule favors small-sized TSPs more, and in return, these TSPs are more willing to cooperate with
the CMP. In an oligopoly market with less competition, the implementation of the CMP provides greater benefits to all of the
entities.
• Second, the enhancement of revenues is most pronounced when the level of demand is almost equal to the level of supply, without
being excessive or insufficient.
• Third, the fare discount influences travel demand and further determines the revenues of TSPs. Therefore, it is imperative to ex­
ercise careful consideration when determining the multimodal service fare discount. An excessively high fare discount drastically
alters the demand distribution, leading to low revenues and even requiring subsidies for TSPs. On the contrary, an insufficiently low

Table 6
The revenues of two TSPs under different γ (The first market structure).
TSP A TSP B Total

γ Independent Shared from the Total Subsidy Independent Shared from the Total Subsidy
operation CMP operation CMP

1.00 465.94 261.69 727.63 0 526.16 175.86 702.02 0 1429.65

0.95 448.18 235.09 683.27 0 506.02 153.39 659.41 0 1342.68
0.90 429.83 208.16 637.99 0 484.39 132.59 616.98 0 1254.97
0.85 410.78 181.37 592.15 0 461.87 112.99 574.86 0 1167.01
0.80 391.05 156.21 547.26 0 438.67 93.04 531.71 108.75 1078.97
0.75 370.94 130.56 501.5 0 414.74 74.80 489.54 132.68 991.04
0.70 349.94 103.95 453.89 151.37 390.00 59.56 449.56 157.42 903.45

1. The unit of figures in the table is ¥ Chinese Yuan.

16
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

discount becomes ineffective in enticing a sufficient number of travelers to use the integrated multimodal services provided by the
CMP. Despite the fact that it ensures substantial financial gains for TSPs, travelers’ experiences may not improve significantly,
hence suggesting a lower level of customer surplus.

7. Conclusion

In this paper, we develop a two-stage hierarchical game theoretical model to capture the strategic behaviors of self-interested TSPs
when cooperating with the CMP. The first stage applies cooperative game theory to modeling the coalition among TSPs and solves the
revenue-sharing problem for the CMP. The second stage explores TSPs’ resource allocation strategies in a non-cooperative game. The
two stages interact with each other through the revenue-sharing rule and the resource allocation scheme. Our main contribution is
designing a revenue-sharing and resource allocation scheme for cooperative multimodal transport systems that achieves efficiency and
stability at the same time. This multimodal transport system can be set either within the context of urban transport at the city level or
intercity transport at the regional level. The performance of the proposed revenue-sharing rule and resource allocation scheme is
evaluated through an illustrative example and numerical studies. We further extend our model to account for demand distribution
variation caused by the emerging multimodal services provided by the CMP. Subsidization scheme is designed to stabilize the coa­
lition. By investigating the proposed scheme under various market structures and demand scenarios, we derive managerial insights for
policymakers.
Several directions are worthy of further investigation. First, in this study, we assume that each service link’s fare is fixed and
characterize the travel demand price elasticity by trip fare discount in a simplified way. In practice, however, TSPs may modify link
fares to achieve higher profits in response to the implementation of the CMP. For future research, we will consider service fares as TSPs’
decision variables and incorporate demand price elasticity in a more sophisticated way in the non-cooperative game. This will lead the
second stage problem to an equilibrium problem with equilibrium constraints (EPEC), which calls for more effort to solve. Second, the
proposed revenue-sharing rule requires TSPs to share private demand information with the CMP, despite the fact that some TSPs may
be unwilling to do so. We have compared two typical revenue-sharing rules with our proposed rule through numerical studies and
found they perform well. However, the comparison lack theoretic analysis, and whether they can be universally satisfactory alter­
natives to the optimal revenue-sharing rule in all cases remains ambiguous. Hence, designing a heuristic revenue-sharing rule that does
not require private information from TSPs and can be readily applied in practice without solving complicated models warrants further
investigation. Finally, it is worth noting that the current model assumes that the CMP does not share the aggregate revenues. However,
if the CMP’s share and its strategic behavior are incorporated, the revenue-sharing rule should be modified accordingly to ensure
stability and efficiency. Exploring this aspect could present an interesting avenue for future research.

CRediT authorship contribution statement

Xiaoshu Ding: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis. Sisi Jian: Writing
– review & editing, Writing – original draft, Supervision, Project administration, Methodology, Investigation, Funding acquisition,
Formal analysis, Conceptualization, Resources.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

Acknowledgement

This study is supported by the Hong Kong Research General Council – General Research Fund (No. 16204822) and the National
Natural Science Foundation of China (No.72101064).

Appendix A. Proof of Lemma 1

Given that the cost function f( • ) is a convex increasing function, it implies that the objective function of P1 is concave. The feasible
region is a convex set defined by a set of linear constraints in (4). Therefore, a solution must exist for this concave programming, which
indicates that the Lagrange multipliers in (9) always exist. Then the existence of the revenue-sharing rule θ* defined by Eq. (9) is
guaranteed. This completes the proof.■

Appendix B. Proof of Lemma 2

From the left-hand side to the right-hand side, given any vector w ∈ W(θ), we define the resource allocation strategy xi for TSP i as
follows: xik = wk for all k ∈ KM S
i and xik = xik (w) for all k ∈ Ki where xik (w) is solved from P4. According to the definition of w, no TSP
would deviate from xi , ∀i ∈ I because the revenues from independent operation has been maximized and they cannot receive more
revenues by allocating more resources to multimodal paths on the condition that others’ allocation strategies remain unchanged. On

17
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

the contrary, if one TSP chooses to decrease the resources allocated to multimodal paths, this will finally lead to a Nash equilibrium
with xM i = 0. Therefore, TSPs’ resource allocation strategies with respect to w, i.e., x(w) = (x1 , x2 , ⋯, xi )i∈I , forms a Nash equilibrium,
i.e., x(w) ∈ X (θ).
In the other direction, let x* ∈ X (θ) and x*ik , k ∈ KM
i be the corresponding resources allocated to multimodal paths by TSP i. Then,
for any multimodal path k, we have x* = x*ʹ , ∀i,iʹ ∈ I . The reason for this is that allocating more resources than the minimum amount
ik ik k
determined by other TSPs jointly cooperating the multimodal path cannot serve more travelers, only to increase the operational cost.
Hence, this is useless to improve the payoff. Hence, we can define wk = x*ik for any k ∈ KM and obviously, such w(x* ) belongs to W(θ). ■

Appendix C. Proof of Lemma Proposition 1

We start with deriving the KKT conditions for P3 and P4 as follows:

KKT conditions of P3 KKT conditions of P4

∑ ∑
Pk Dk (xik ) − fik (xik ) −
ʹ
l∈Li δlk μil + φik = 0 ∀k ∈ KSi Pk Dk (xik ) − fik (xik ) −
ʹ S S
l∈Li δlk μil + φik = 0 ∀k ∈ KSi
∑
θik Pk Dk (xik ) − fʹik (xik ) − l∈Li δlk μil − σik + φik = 0 ∀k ∈ KMi
(∑ ) (∑ )
∀l ∈ Li μSil
∑ ∀l ∈ Li
μil k∈Ki δlk xik − cl =0 k∈KSi δlk xik − cl + k∈KM
i
δlk wk =0
σik (xik − wk ) = 0 ∀k ∈ KM
i φSik xik = 0 ∀k ∈ KSi
φik xik = 0 ∀k ∈ Ki
∑ ∑ ∑
k∈Ki δlk xik ≤ cl ∀l ∈ Li k∈KSi δlk xik ≤ cl − k∈KM
i
δlk wk ∀l ∈ Li
xik ≤ wk ∀k ∈ KM
i
xik ≥ 0 ∀k ∈ KSi
xik ≥ 0 ∀k ∈ Ki

φik ≥ 0 ∀k ∈ Ki φSik ≥ 0 ∀k ∈ KSi

μil ≥ 0 ∀l ∈ Li μSil ≥ 0 ∀l ∈ Li
σik ≥ 0 ∀k ∈ KM
i

When calculating the first order derivative of E[min(dk , xik ) ] to derive the stationary conditions, we first rewrite it as follows:
∫ xik ∫∞
E[min(dk , xik ) ] = tgk (t)dt + xik gk (t)dt
0 xik

∫ xik
= tDk (t)x0ik − Dk (t)dt + xik [1 − Dk (xik ) ]
0

∫ xik
= xik Dk (xik ) − Dk (t)dt + xik [1 − Dk (xik ) ]
0

∫ xik
= xik − Dk (t)dt
0

where gk ( • ) denotes the probability density function and it equals the first order derivative of the cumulative distribution function
Dk ( • ), i.e., gk (t) = ∂D∂kt(t).
With the above reformulation, we can calculate the first derivative as follows:
∂E[min(dk , xik ) ]
= 1 − Dk (xik ) = Dk (xik )
∂xik

The rest of the stationary condition derivation is standard and thus omitted here.
{ }
In the KKT conditions of P3, the dual variables μi = {μil : l ∈ Li }, σ i = σik : k ∈ KM
i , φi = {φik : k ∈ Ki } are the Lagrange multipliers
for the capacity constraints, the upper bound constraints imposed by w, and the non-negativity constraints for xi , respectively. To
distinguish with the KKT conditions of P3, let μSi , φSi denote the Lagrange multipliers of P4.
( )
First, suppose we have w ∈ W(θ), then an optimal solution to P3 for TSP i is given by xi = xSi ,w , that is, TSP i allocates exactly wk
to multimodal path k ∈ KM S S S
i . We can define the Lagrange multipliers of P4: μil = μil ,φik = φik ,∀l ∈ Li ,k ∈ Ki . For a multimodal path k, if
wk > 0, we must have φik = 0 by complementary slackness condition. Then, from stationarity condition, we can infer that
ʹ (x ) − ∑ δ μ − σ = 0, which leads to θ P D (x ) − fʹ (x ) = ∑ δ μ + σ ≥ ∑ δ μ = ∑ δ μS .
θik Pk Dk (xik ) − fik ik l∈Li lk il ik ik k k ik ik ik l∈Li lk il ik l∈Li lk il l∈Li lk il
To prove the reverse implication, let w be a feasible vector of public signals and let μSi , φSi be the Lagrange multipliers for an optimal
solution of P4 associated with w. Suppose condition (16) is satisfied by μSi ,φSi , then we want to prove that w ∈ W(θ). In other words, P3
( )
admits a solution xi = xSi , xM
i with xM S M
i = w, where xi and xi are the resource allocation strategy of TSP i for independent operated
paths and multimodal paths, respectively. To achieve this, we only need to find a feasible set of dual variables that satisfy KKT

18
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

conditions of P3. We define the Lagrange multipliers as follows: μil = μSil ,∀l ∈ Li ,φik = φSik ,k ∈ KSi . For a multimodal path k, if wk = 0,
then we can choose arbitrary values σ ik ≥ 0, φik ≥ 0 such that the stationarity condition is guaranteed, i.e.,
ʹ (w ) − ∑ δ μS − σ + φ = 0. If w > 0, we have φ = 0 by complementary slackness. Then, we must choose σ
θik Pk Dk (xik ) − fik k l∈Li lk il ik ik k ik ik

such that θik Pk Dk (xik ) − fik ʹ (w ) − ∑ δ μS − σ = 0. By dual feasibility, we have σ ≥ 0, which is equivalent to
k l∈Li lk il ik ik
∑
θ P D (w ) − fʹ (w ) −
ik k k k ik k δ μ ≥ 0. This follows the hypothesis that condition (16) is satisfied.
l∈Li lk il
Overall, we can always find a certain set of Lagrange multipliers satisfying all KKT conditions, indicating that xM
i = w is the optimal
solution to P3. This completes the proof.■

Appendix D. Proof of Lemma Proposition 2

The KKT conditions of grand coalition’s problem P1 are given by:

( ) ∑ ʹ ( I ) ∑ δ μI + φI = 0
Pk Dk xIk − î∈Ik fîk xk − l∈L lk l k
∀k ∈ K
( ∑ )
μIl I
k∈K δlk xk − cl = 0
∀l ∈ L
φIk xIk = 0 ∀k ∈ K
∑ I
k∈K δlk xk ≤ cl
∀l ∈ L
xIk ≥ 0 ∀k ∈ K

φIk ≥ 0 ∀k ∈ K
μIl ≥ 0 ∀l ∈ L

[ ( )]
Similar with the derivation process presented in the proof of Proposition 1, the first order derivative of E min dk , xNk is equal to
( N)
1 − Dk xk . The remaining parts of the stationary conditions are straightforward to obtain.
{ } { }
Let μI = μIl : l ∈ L , φI = φIk : k ∈ K denote the Lagrange multipliers for the grand problem. The primal and dual variables xI , μI ,
φI also satisfy the KKT conditions of P4 where wk = xIk for k ∈ KM . In order to prove efficiency of θ* , we only need to show that when xI
functions as the public signal, i.e., wk = xIk for k ∈ KM , condition (16) will be satisfied with μI under θ* . If this condition holds, then
according to Proposition 1, the grand coalition solution xI corresponds to one of the Nash equilibria associated with θ* , which aligns
with the definition of efficiency.
( ) ∑ ʹ ( I ) ∑ δ μI = 0. Together with the
For xIk > 0, we have φIk = 0. Then by stationary condition, we have Pk Dk xIk − î∈Ik fîk xk − l∈L lk l
∑
δ μI +fik
l∈Li lk l
ʹ (xI ) ( I) ∑ ( I) ( I) ʹ ( I)
* * I
revenue-sharing rule θik = ∑ δ μI +∑ fʹ xI , we have θik Pk Dk xk = l∈Li δlk μl + fik xk , leading to θik Pk Dk xk − fik xk =
k ʹ
l∈L lk l î∈Ik îk
( k )
∑ I I M I *
l∈Li δlk μl . Note that we have defined wk = xk for all k ∈ K , so condition (16) is satisfied with μ . We can conclude that θ defined in
(9) implements the grand solution as a Nash equilibrium, thus is efficient. This completes the proof.■

Appendix E. Proof of Proposition 3

Let N be an arbitrary subset of I and suppose all the other TSPs i ∈ I\N follow the centralized resource allocation decision xIi . We can
regard N as a merged single TSP with mobility resources from all the member TSPs in set N. The maximal payoff of this new TSP can
obtain through cooperation with the CMP is denoted as πN . Suppose the revenue-sharing proportion for the new TSP N is the sum­
∑
mation of the proportions of its member, i.e., θ*Nk = i∈N θ*ik , then the new revenue-sharing rule is still an efficient one because the
resource allocation scheme of the grand coalition remains unchanged. Hence, for the new merged TSP, the optimal resource allocation
( ) ∑ ∑ { [ ( )] ( )}
strategy is still xIN = xIi i∈N , then we have πN = k∈K i∈N θ*ik Pk E min dk , xIk − fik xIk .
If this smaller coalition N leaves the grand coalition, it means that the newly merged TSP operates independently. Then, its payoff
V(N) is equal to the optimal objective value of P3 with all wk equals 0 because it does not cooperate with other TSPs in the set I\N on
multimodal paths. Obviously, this solution is bounded by the optimum of P3 with wk = xIk , ∀k ∈ KM because the former solution is a
feasible but not necessarily optimal one to the latter problem under θ* . Note that πN is the payoff under the Nash equilibrium of the
∑ ∑ { [ ( )] ( )}
latter optimum, therefore, we can conclude that V(N) ≤ πN = k∈K i∈N θ*ik Pk E min dk , xIk − fik xIk and θ* is stable. This com­
pletes the proof. ■

19
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

Appendix F. Attributes of the Great Bay Area network

Table 7
Attributes of service links in the GBA intercity transport network.

O D Length (km) Fare (¥) Operational cost (¥) Capacity Demand distributionUnif [0, U] TSP

1 5 13 6 3 2400 [0,2500] Railway

2 4 36 25 24 800 [0,900] Railway
3 4 40 60 20 1200 [0,2000] Railway
4 7 40 40 20 1200 [0,1500] Railway
5 6 66 15 10 500 [0,800] Railway
5 8 90 55 45 150 [0,900] Railway
8 9 47 26 22 400 [0,500] Railway
8 10 48 20 18 200 [0,500] Railway
1 3 66 40 33 60 [0,200] Bus
1 5 27 21 13 60 [0,200] Bus
2 4 55 54 27 120 [0,200] Bus
3 4 54 45 27 120 [0,200] Bus
4 7 51 65 25 60 [0,200] Bus
1. The city mapping to each number: 1-Guangzhou; 2-Huizhou; 3-Dongguan; 4-Shenzhen; 5-Foshan; 6-Zhaoqing; 7-HongKong; 8-Zhongshan; 9-Zhu­
hai; 10-Jiangmen.
2. Cost function is linear and the operational cost in the table denotes the average cost of serving one traveler.
3. Demand distribution of railway links is estimated from the hourly commuting data of the commute monitoring big data platform by Baidu Map
(Baidu Map, 2023). All of the bus links listed in the table are coach bus links, whose demand distribution is determined by the coach capacity and its
frequency.
4. For bus services connecting railway stations and actual origins or destinations, e.g., home or workplace, we assume they have the same attributes:
The length is 2 km, the trip fare is ¥2 with an operational cost per person of ¥1, the capacity is 300 people, and the travel demand follows a uniform
distribution of [0, 500].

Table 8
Attributes of multimodal paths in the GBA intercity transport network.

O D Route Fare (¥) Demand distributionUnif [0, U] O D Route Fare (¥) Demand distributionUnif [0, U]

1 5 1–1-5–5 10 [0,500] 1 2 1–3-3–4-2–2 129 [0,200]

1 4 1–3-3–4-4 104 [0,400] 4 5 4–4-3–3-1–5 125 [0,400]
1 2 1–3-4–4-2–2 114 [0,200] 4 2 4–4-2–2 29 [0,500]
1 2 1–3-3–4-4–2 158 [0,200] 4 3 4–4-3–3 64 [0,500]
1. Each number represents the same city as in Table 7.
2. We assume all the multimodal paths operated by the CMP are bus-railway-bus paths. In each route, the first consecutive stop pair with the same
number represents the transfer from urban bus to railway, and the second pair represents the transfer from railway to bus. For instance, on route
1–1–5–5, 1–1 denotes the local bus line in Guangzhou from the origin to the railway station, 1–5 is the railway line from Guangzhou to Foshan, and
5–5 denotes the local bus line in Foshan from the railway station to the destination.
3. Demand distribution of multimodal paths is estimated from the hourly commuting data of the commute monitoring big data platform by Baidu Map
(Baidu Map, 2023).

Appendix G. Attributes of the Nguyen Dupuis network

Table 9
Attributes of links in the Nguyen Dupuis network under the first market structure.

O D Length (km) Fare (¥) Operational cost (¥) Capacity Demand distributionUnif [0, U] TSP

1 5 1 8 4 33 [0,60] A
1 12 1 5 2 25 [0,40] A
4 5 1 7 3 34 [0,30] A
4 9 1.4 2 1 27 [0,30] A
5 9 1 2 1 21 [0,20] A
9 10 1 7 3 45 [0,20] A
12 8 2.2 6 2 41 [0,40] A
12 6 1 4 1 20 [0,60] A
6 7 1 6 2 50 [0,30] A
10 11 1 8 4 42 [0,30] A
(continued on next page)

20
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

Table 9 (continued )
O D Length (km) Fare (¥) Operational cost (¥) Capacity Demand distributionUnif [0, U] TSP

13 3 1 4 1 37 [0,40] A
7 8 1 3 1 24 [0,20] A
11 2 1 6 2 30 [0,30] A
11 3 1 3 1 26 [0,20] A
8 2 1 5 2 50 [0,60] A
1 5 1 3 1 12 [0,40] B
1 12 1 4 1 20 [0,60] B
4 5 1 6 2 39 [0,40] B
4 9 1.4 5 2 49 [0,20] B
5 6 1 5 2 40 [0,60] B
9 13 1.4 4 1 44 [0,40] B
12 8 2.2 4 2 30 [0,20] B
12 6 1 8 3 29 [0,50] B
6 7 1 3 1 44 [0,60] B
6 10 1 2 1 20 [0,30] B
13 3 1 8 4 46 [0,20] B
7 11 1 6 3 39 [0,30] B
11 2 1 7 2 32 [0,40] B
11 3 1 5 2 34 [0,40] B
8 2 1 3 1 44 [0,40] B
1. Cost function is linear and the operational cost in the table denotes the average cost of serving one traveler.

Table 10
Attributes of paths in the Nguyen Dupuis network under the first market structure.

O D Route Fair (¥) Demand distributionUnif [0, U]

1 2 1(A)-12(A)-12(B)-8(B)-2(B) 12 [0,5]
1 2 1(A)-12(A)-12(B)-8(B)-8(A)-2(A) 14 [0,15]
1 2 1(A)-12(A)-6(A)-7(A)-8(A)-8(B)-2(B) 21 [0,5]
1 3 1(A)-5(A)-9(A)-9(B)-13(B)-13(A)-3(A) 18 [0,5]
1 3 1(A)-1(B)-5(B)-6(B)-7(B)-11(B)-11(A)-3(A) 20 [0,15]
1 3 1(A)-12(A)-6(A)-6(B)-10(B)-10(A)-11(A)-3(A) 22 [0,15]
4 2 4(A)-4(B)-5(B)-6(B)-7(B)-7(A)-8(A)-2(A) 22 [0,5]
4 2 4(A)-9(A)-10(A)-11(A)-11(B)-2(B) 24 [0,10]
4 2 4(A)-4(B)-5(B)-6(B)-10(B)-10(A)-11(A)-2(A) 27 [0,15]
4 3 4(A)-9(A)-9(B)-13(B)-13(A)-3(A) 10 [0,25]
4 3 4(A)-5(A)-9(A)-9(B)-13(B)-3(B) 21 [0,10]
4 3 4(A)-4(B)-9(B)-9(A)-10(A)-11(A)-3(A) 23 [0,10]
1. To distinguish the service links of TSP A and TSP B, we add the operating TSP after each node number. For instance, on path 1(A)-12(A)-12(B)-8(B)-
2(B), 1(A)-12(A) denotes the link between node 1 and node 12 operated by TSP A, 12(B)-8(B) denotes the link between node 12 and node 8 operated
by TSP B, and 12(A)-12(B) is a virtual transfer link.

References

Algaba, E., Fragnelli, V., Llorca, N., Sánchez-Soriano, J., 2019. Horizontal cooperation in a multimodal public transport system: The profit allocation problem. Eur. J.
Oper. Res. 275 (2), 659–665.
Álvarez-SanJaime, Ó., Cantos-Sanchez, P., Moner-Colonques, R., Sempere-Monerris, J.J., 2021. The effect of cooperative infrastructure fees on high-speed rail and
airline competition. Transp. Policy 112, 125–141.
Atasoy, B., Schulte, F., Steenkamp, A., 2020. Platform-based collaborative routing using dynamic prices as incentives. Transp. Res. Rec. 2674 (10), 670–679.
Bachrach, Y., Elkind, E., Meir, R., Pasechnik, D., Zuckerman, M., Rothe, J., Rosenschein, J.S., 2009. The cost of stability in coalitional games. In: Algorithmic Game
Theory: Second International Symposium, SAGT 2009, Paphos, Cyprus, October 18-20, 2009. Proceedings 2. Springer Berlin Heidelberg, pp. 122–134.
Bahinipati, B.K., Kanda, A., Deshmukh, S.G., 2009. Revenue sharing in semiconductor industry supply chain: Cooperative game theoretic approach. Sadhana 34,
501–527.
Baidu Map (2023). Commute monitoring big data platform. https://jiaotong.baidu.com/tongqin/.
Clark, D.J., Jørgensen, F., Mathisen, T.A., 2014. Competition in complementary transport services. Transp. Res. B Methodol. 60, 146–159.
Cohen, M.C., Zhang, R., 2022. Competition and coopetition for two-sided platforms. Prod. Oper. Manag. 31 (5), 1997–2014.
Ding, X., Qi, Q., Jian, S., Yang, H., 2023. Mechanism design for Mobility-as-a-Service platform considering travelers’ strategic behavior and multidimensional
requirements. Transp. Res. B Methodol. 173, 1–30.
Engevall, S., Göthe-Lundgren, M., Värbrand, P., 2004. The heterogeneous vehicle-routing game. Transp. Sci. 38 (1), 71–85.
Guangzhou Municipal Planning and Natural Resources Bureau. (2023). 2022 Guangzhou Transportation Development Annual Report. Guangzhou Government.
https://www.cnbayarea.org.cn/attachment/0/11/11924/1122615.pdf.
Heilker, T., Sieg, G., 2018. A duopoly of transportation network companies and traditional radio-taxi dispatch service agencies. Eur. J. Transp. Infrastruct. Res. 18 (2).
Hu, X., Caldentey, R., Vulcano, G., 2013. Revenue sharing in airline alliances. Manag. Sci. 59 (5), 1177–1195.
Jian, S., Liu, W., Wang, X., Yang, H., Waller, S.T., 2020. On integrating carsharing and parking sharing services. Transp. Res. B Methodol. 142, 19–44.
Jiang, C., Zhang, A., 2014. Effects of high-speed rail and airline cooperation under hub airport capacity constraint. Transp. Res. B Methodol. 60, 33–49.
Kimms, A., Çetiner, D., 2012. Approximate nucleolus-based revenue sharing in airline alliances. Eur. J. Oper. Res. 220 (2), 510–521.

21
X. Ding and S. Jian Transportation Research Part C 164 (2024) 104666

Liu, L., Qi, X., Xu, Z., 2018. Simultaneous penalization and subsidization for stabilizing grand cooperation. Oper. Res. 66 (5), 1362–1375.
Lozano, S., Moreno, P., Adenso-Díaz, B., Algaba, E., 2013. Cooperative game theory approach to allocating benefits of horizontal cooperation. Eur. J. Oper. Res. 229
(2), 444–452.
Najmi, A., Rashidi, T.H., Waller, T., 2023. A multimodal multi-provider market equilibrium model: A game-theoretic approach. Transportation Research Part C:
Emerging Technologies 146, 103959.
Nash, J., 1951. Non-cooperative games. Ann. Math. 286–295.
Nash, J., 1953. Two-person cooperative games. Econometrica 128–140.
Ni, L., Chen, C., Wang, X.C., Chen, X.M., 2021. Modeling network equilibrium of competitive ride-sourcing market with heterogeneous transportation network
companies. Transportation Research Part C: Emerging Technologies 130, 103277.
Niyato, D., Vasilakos, A.V., Kun, Z., 2011. Resource and revenue sharing with coalition formation of cloud providers: Game theoretic approach. In: 2011 11th IEEE/
ACM International Symposium on Cluster, Cloud and Grid Computing. IEEE, pp. 215–224.
Özener, O.Ö., Ergun, Ö., 2008. Allocating costs in a collaborative transportation procurement network. Transp. Sci. 42 (2), 146–165.
Pantelidis, T.P., Chow, J.Y., Rasulkhani, S., 2020. A many-to-many assignment game and stable outcome algorithm to evaluate collaborative mobility-as-a-service
platforms. Transp. Res. B Methodol. 140, 79–100.
Pantelidis, T.P., Chow, J.Y., Cats, O., 2023. Mobility operator service capacity sharing contract design to risk-pool against network disruptions. Transportmetrica A:
Transport Science 1–30.
Peleg, B., Sudhölter, P., 2007. Introduction to the theory of cooperative games. Springer Science & Business Media.
Rasouli, S., Timmermans, H., 2012. Uncertainty in travel demand forecasting models: literature review and research agenda. Transportation Letters 4 (1), 55–73.
Rosenthal, E.C., 2017. A cooperative game approach to cost allocation in a rapid-transit network. Transp. Res. B Methodol. 97, 64–77.
Schlicher, L., Lurkin, V., 2022. Stable allocations for choice-based collaborative price setting. Eur. J. Oper. Res. 302 (3), 1242–1254. https://doi.org/10.1016/j.
ejor.2022.01.036.
Schmeidler, D., 1969. The nucleolus of a characteristic function game. SIAM J. Appl. Math. 17 (6), 1163–1170.
Shapley, L. S. (1953). A value for n-person games.
Singh, C., Sarkar, S., Aram, A., Kumar, A., 2011. Cooperative profit sharing in coalition-based resource allocation in wireless networks. IEEE/ACM Trans. Networking
20 (1), 69–83.
Wang, H., Yang, H., 2019. Ridesourcing systems: A framework and review. Transp. Res. B Methodol. 129, 122–155.
Wu, Y., Zhu, Q., Huang, J., Tsang, D.H., 2014. Revenue sharing based resource allocation for dynamic spectrum access networks. IEEE J. Sel. Areas Commun. 32 (11),
2280–2296.
Xu, G., Zhong, L., Wu, R., Hu, X., Guo, J., 2022. Optimize train capacity allocation for the high-speed railway mixed transportation of passenger and freight. Comput.
Ind. Eng. 174, 108788.
Zhu, Z., Xu, A., He, Q., Yang, H., 2021. Competition between the transportation network company and the government with subsidies to public transit riders.
Transportation Research Part E: Logistics and Transportation Review 152, 102426.

22