Game Theoretic Models for Competition in Public Transit Services
Eddie Y. S. Chan
1
and Janny M. Y. Leung
2
1
Quintiq, Penang, Malaysia
2
Department of Systems Engineering and Engineering Management, The Chinese University of Hong Kong,
Shatin, New Territories, Hong Kong
Keywords:
Game Theory, Competition, Public Transit.
Abstract:
As metropolitan areas grow, the need to travel by the populace has increased the burden on the transport
systems, leading to increased traffic congestion and environmental concerns. In this paper, we discuss some
game-theoretic models that can be used to investigate the competitive situation when several service providers
offer public transit services. The competition among the operators can be modelled by a class of games called
potential games, and we discuss mathematical programmes that can be used to find the Nash equilibria for
these games. By examining the equilibrium solutions, we can investigate the relative merits and tradeoffs
for different structures of the transit networks, and the interplay between the services offered and the overall
ridership of the system.
1 INTRODUCTION
Metropolitan areas have accounted for the majority
of increases in population and economic growth in
recent decades. China’s phenomenal economic de-
velopment has been fuelled by growth in the major
cities, many of which has over 5 million in popula-
tion. Metropolitan areas account for over half of the
population, and a significant majority of the GDP, of
the United States. As the geographical size and pop-
ulation of major metropolitan areas have increased,
much economic activity remain focussed in the cen-
tral business districts of the metropolises, thus the av-
erage travel distances for work have not decreased
as expected. The average commuting distance for
London is over 10 kilometres. The need to travel
by the populace has placed significant burden on the
transport systems of metropolitan areas, leading to in-
creased traffic congestion and attendant safety and en-
vironmental concerns.
Development of transport infrastructure and pub-
lic transit services have not kept apace with the
swell and sprawl of metropolitan areas, with seri-
ous congestion occurring in central business districts
and insufficient coverage in peripheral areas. In
metropolises where public transit services are pro-
vided by private firms in a relatively free market, op-
erators tend to focus on high-profit routes and outly-
ing smaller communities are under-served. In Hong
Kong, the already congested Central business district
is often jammed with half-empty double-decker buses
from all the bus operators, while bus services to satel-
lite communities in the New Territories are very in-
frequent and expensive.
In this paper, we discuss some game-theoretic
models that can be used to investigate the competitive
situation when several service providers offer public
transit services, and study the impact on the total set
of services offered to the public and the resultant level
of ridership of the system. The competition among
the operators can be modelled by a class of games
called potential games. We discuss mathematical pro-
grammes that can be used to find the Nash equilibria
for these games. By examining the equilibria solu-
tions, we can examine the relative merits and tradeoffs
for different structures of the transit networks, and the
interplay between the services offered and the overall
ridership of the system. We hope that our modelling
and analysis may provide some insight on the types
and bundling of routes being offered by operators, and
the locations for transportation interchanges and hubs.
2 BACKGROUND
Whilst not explicitly acknowledged, concepts of
game theory have been pervasively used in traffic
studies. As Fisk (1984) pointed out, the famous
133
Y. S. Chan E. and M. Y. Leung J..
Game Theoretic Models for Competition in Public Transit Services.
DOI: 10.5220/0005218201330139
In Proceedings of the International Conference on Operations Research and Enterprise Systems (ICORES-2015), pages 133-139
ISBN: 978-989-758-075-8
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
Wardrop’s (1952) user-equilibrium principle is essen-
tially the condition for a Nash (1950) game-theoretic
equilibrium among road-users, since no driver can re-
duce his/her travel time by switching to a different
route choice. Wardrop’s principle has been a cor-
nerstone in road traffic research for decades. For an
overview of traffic equilibrium models, see Patriksson
and Labbe (2004).
Other researchers have developed specific game-
theoretic models for transport-related problems. Bell
(2000) investigates network reliability by studying a
zero-sum game between a cost-minimising driver and
a demon that sets the link costs. This game is a con-
cept game in the sense that the demon is not a real
player, and is used to explore the worst-case scenar-
ios faced by the driver. Other researchers have also
explored concept games among road users. James
(1998) studies a game among n road-users where
any player’s utility of using the road segment de-
creases when there are more users. Levinson (2005)
also studies congestion by investigating a game where
the players (drivers) choose their departure times.
Pedersen (2003) investigates road safety by a game
where players choose the behavioural level of driv-
ing aggression. All these games study the competi-
tion among road users. Holland and Prashker (2006)
give an excellent review of recent literature on non-
cooperative games in transport research.
Surprisingly, studies on the competitive situation
amongst public transit operators have received little
attention from transport researchers. Castell et al.
(2004) modelled a Stackelberg game between two au-
thorities (one determining flow, and the other capaci-
ties) in a freight transport network, which is different
to a passenger transit network since the route choice
is not determined by the transportee (freight). Ac-
cording to Holland and Prashker (2006), the “small
number of such games is surprising, considering that
NCGT [non-cooperative game theory] seems a natu-
ral tool for analysing relations between authorities. ...
Trends such as tendering and privatisation, that have
a vital role on today’s transport agenda, also seem apt
to be modelled through games between authorities”.
Surprisingly, there has been very little research along
this line.
Some researchers have studied games between au-
thorities and travellers. Fisk (1984) investigates a
Stackelberg game between the authority that sets traf-
fic signals and all travellers who then finds the user-
equilibrium solution. Chen and Ben-Akiva (1998)
investigates a similar game in a dynamic setting.
Reyniers (1992) studies a game between the railway
operators who sets the capacities for different fare
classes and the passengers who chooses which class
to use. Hollander et al. (2006) studies a game be-
tween the parking authority and travellers to explore
the incentives for public transport ridership. All these
games, however, only involve one operator/authority.
Only few researchers have investigated games with
several operators and passengers. Van Zuylen and
Taale (2004) studies a game with two authorities (one
for urban roads and one for ring roads) and the driv-
ing public. Fernandez and Marcotte (1992) and Fer-
nandez et al. (1993) investigated traffic equilibrium
models involving car users, bus users and bus opera-
tors and presented algorithms for finding the equilib-
rium solutions; their models are general and consid-
ered traffic congestion effects. The focus of our paper
is on the strategic competition among public transit
operators.
3 MODELS
We have made some preliminary investigation into
the strategic gaming situation among competing pub-
lic transit service providers. In our first-cut model,
we assume that all the operators have the same cost
and price structure, and that the total ridership be-
tween each origin-destination pair is equally divided
among all the operators that service that particular
route. In this setting, a player of the game is the ser-
vice provider, and its strategy is the set of routes that
it chooses to offer service. Each player tries to max-
imize its total profit, and a Nash equilibrium occurs
when no player can improve its profit by unilaterally
changing the set of routes it services.
We can show that this game can be modelled as
what is known as a potential game (first introduced
by Rosenthal, 1973) where the equilibrium can be
computed by solving an auxiliary mathematical pro-
gramme with a potential function as a surrogate ob-
jective. The solvability for the Nash equilibrium al-
lows us to make some comparisons between the com-
petitive equilibrium and a centralised monopolistic
profit-maximizing operator and draw some insights.
3.1 Basic Competitive Model
We consider a game among n players (service
providers) and m possible routes (origin-destination
pairs). Player is strategy consists of a subset of the
routes S
i
M = {1, 2,··· , m}. For each route j, let k
j
be the number of players who choose to offer service
on the route, i.e., k
j
= |{i : j S
i
}|. Note that k
j
is
endogenously determined. Let a
j
denote the revenue
from the total ridership on route j, assumed exoge-
nously determined, and let δ
j
be the cost of operat-
ICORES2015-InternationalConferenceonOperationsResearchandEnterpriseSystems
134
ing the route. Each player (service provider) that of-
fers service on this route earns an identical payoff of
p
j
(k
j
) which depends on the total number of players
(k
j
) who choose to serve that route:
p
j
(k
j
) =
(
a
j
k
j
δ
j
, k > 0,
0, k = 0.
Each player i selects a strategy to maximise his to-
tal profit π
i
(S
1
,S
2
,· ·· , S
n
) =
jS
i
p
j
(k
j
). A pure
Nash (1950) equilibrium is a set of strategies
{S
1
,S
2
,· ·· , S
n
} such that each player cannot unilat-
erally improve his total profit, that is,
π
i
(S
1
,· ·· , S
i1
,S
i
,S
i+1
,· ·· , S
n
)
π
i
(S
1
,· ·· , S
i1
,S
i
,S
i+1
,· ·· , S
n
), S
i
Following Rosenthal (1973), we can show that
the Nash equilibrium for this game can be obtained
by solving the following auxiliary mathematical pro-
gramme:
(G0) : Max
m
j=1
k
j
y=1
p
j
(y)
s.t.
n
i=1
x
j
i
= k
j
, j = 1,...,m, (1)
x
j
i
{0,1}, i = 1,. .. ,n, j = 1,. .. ,m, (2)
where x
j
i
is 1 if player i offers service on route j. We
note that this is not a straightforward binary linear
programme, since the k
j
values are also variables, so
the objective is not a linear function.
Proof. We will show, by contradiction, that the solu-
tion of (G0) yields a Nash equilibrium. Let {x
,k
}
be an optimal solution to (G0) such that the associ-
ated strategy combination is not a Nash equilibrium.
Then for some player l, there is a strategy
ˆ
S
l
such that
j
ˆ
S
l
\S
l
p
j
(k
j
+ 1) >
jS
l
\
ˆ
S
l
p
j
(k
j
)
where S
l
is the strategy used by l as indicated by the
values of x
j
i
. Consider the new values { ˆx
j
i
,
ˆ
k
j
} asso-
ciated with player l changing to the pure-strategy
ˆ
S
l
(with the rest of the players not changing their strate-
gies). The objective value of (G0) for this solution
is:
m
j=1
ˆ
k
j
y=1
p
j
(y)
=
m
j=1
k
j
y=1
p
j
(y) +
j
ˆ
S
l
\S
l
p
j
(k
j
+ 1)
jS
l
\
ˆ
S
l
p
j
(k
j
)
|
{z }
>0
>
m
j=1
k
j
y=1
p
j
(y)
which contradict with the optimality of {x
j
i
,k
j
}.
We note that the result also holds when there is a
limit to the number of routes offered by the operators.
Equivalently, a solution of (G0) can be found by
solving the following integer programme (G1). Let
y
jk
=
(
1, if k
j
= k,
0, otherwise.
We also define
P
j
(k) =
(
k
z=1
p
j
(z), for k > 0,
0, if k = 0.
(G1) : Max
m
j=1
n
k=0
P
j
(k)y
jk
s.t.
n
i=1
x
j
i
=
n
k=0
ky
jk
, j = 1,· ·· , m, (3)
n
k=0
y
jk
= 1, j = 1,· ·· , m, (4)
x
j
i
{0,1}, i = 1,...,n, j = 1, ·· · ,m. (5)
Note that
P
j
(k) = p
j
(1) + p
j
(2) + ··· + p
j
(k)
= (a
j
δ
j
) + (a
j
/2 δ
j
) + ··· + (a
j
/k δ
j
)
Therefore, P
j
(k) is at a maximum at
k
j
= argmax{h : p
j
(h) > 0}.
Observation 1:
Operators will continue to ”enter the market” for a
route until it is no longer profitable.
Therefore, as long as the marginal profit is non-
negative, additional operators will offer service on a
route. This leads to better service for the customers,
but reduces the profit for each operator. We compare
this to a monopolistic setting below.
3.2 Basic Profit-Maximising Model
Consider the case when there is only one operator
which wants to maximize the total profit. The opti-
mal choice of routes to offer service is given by the
solution to:
(G2) : Max
m
j=1
n
k=0
kp
j
(k)y
jk
GameTheoreticModelsforCompetitioninPublicTransitServices
135
such that
n
i=1
x
j
i
=
n
k=0
ky
jk
, j = 1,· ·· , m, (6)
n
k=0
y
jk
= 1, j = 1,· ·· , m, (7)
x
j
i
{0,1}, i = 1,. .. ,n, j = 1,...,m, (8)
y
jk
{0,1}, k = 0, .. ., n, j = 1,. ..,m.(9)
Observation 2:
To maximize total profit, the optimal solution is to
have at most one operator per route.
Proof. Since for k 6= 0,
kp
j
(k) = k(a
j
/k δ
j
) = a
j
kδ
j
the objective is maximised when k
j
= 1 as long as
a
j
> δ
j
.
We note that while the monopolistic profit-
maximising solution is to assign only one operator per
route, to determine an equitable allocation of routes
to operators is not a simple task but requires solving a
non-linear optimisation problem.
Observation 3:
The competitive equilibrium solution has as many as
possible operators offering service on each route un-
til it is no longer profitable to do so, whereas the
centrally-controlled (total profit maximizing) solution
has at most one operator per route.
3.3 Model with Player - and Route -
Dependent Operating Cost
In reality, each operator might have different operat-
ing costs on different routes. Hence we extend our
model to a congestion game with player- and route-
specific payoff functions which admit a potential. In
contrast to Milchtaich’s work [13], our model is not
a singleton game; players can choose more than one
route in their strategy sets.
As with the basic model, the revenue is shared
among all operators of a route, that is, the revenue
to each player i serving route j is r
j
(k
j
) = a
j
/k
j
for
k
j
> 0, where k
j
is the number of players offering
service on route j. Because of the cost structure,
the payoff to player i offering service on route j is
p(i, j,k
j
) = a
j
/k
j
δ
i j
, which depends on player i,
route j and the total number of player choosing that
particular route k
j
. We next show that a Nash equi-
librium of this problem can be found by solving the
following auxiliary problem:
(GD1) : Max
m
j=1
k
j
y=1
r
j
(y)
m
j=1
n
i=1
δ
i j
x
j
i
s.t.
n
i=1
x
j
i
= k
j
, j = 1,...,m, (10)
x
j
i
{0,1}, i = 1,. .. ,n, j = 1,. .. ,m, (11)
Proof. Let x
be an optimal solution of (GD1), and
let (S
1
,S
2
,· ·· , S
n
) be the corresponding strategies of
the players. If this is not an equilibrium solution, then
for some player l, there exist another strategy
ˆ
S
l
such
that
π
i
(S
1
,· ·· , S
l1
,
ˆ
S
l
,S
l+1
,· ·· , S
n
) >
π
i
(S
1
,· ·· , S
l1
,S
l
,S
l+1
,· ·· , S
n
),
that is,
j
ˆ
S
l
\S
l
a
j
(k
j
+ 1)
δ
l j
!
jS
l
\
ˆ
S
l
a
j
k
j
δ
l j
!
> 0 (∗∗)
Let ( ˆx,
ˆ
k) be the solution of (GD1) corresponding to
(S
1
,· ·· , S
l1
,
ˆ
S
l
,S
l+1
,· ·· , S
n
). Then
m
j=1
ˆ
k
j
y=1
r
j
(y)
m
j=1
n
i=1
δ
i j
ˆx
j
i
=
m
j=1
k
j
y=1
r
j
(y)
m
j=1
n
i=1
δ
i j
x
j
i
+
j
ˆ
S
l
\S
l
a
j
(k
j
+ 1)
δ
l j
!
jS
l
\
ˆ
S
l
a
j
(k
j
)
δ
l j
!
| {z }
> 0 by (**)
>
m
j=1
k
j
y=1
r
j
(y)
m
j=1
n
i=1
δ
i j
x
j
i
Hence, we have another solution to (GD1) with a bet-
ter objective value contradicting the optimality of x
.
Similar to the basic model, (GD1) can be formu-
lated as an equivalent integer programme:
(GD2) : Max
m
j=1
n
k=0
R
j
(k)y
jk
m
j=1
n
k=0
δ
i j
x
j
i
s.t.
n
i=1
x
j
i
=
n
k=0
ky
jk
, j = 1,· ·· , m, (12)
n
k=0
y
jk
= 1, j = 1,· ·· , m, (13)
x
j
i
{0,1}, i = 1,. .. ,n, j = 1,··· , m, (14)
y
jk
{0,1}, k = 0, 1,. .. ,n, j = 1,...,m,(15)
ICORES2015-InternationalConferenceonOperationsResearchandEnterpriseSystems
136
where R
j
(k) =
k
y=1
r
j
(k) = a
j
+ a
j
/2 + ··· + a
j
/k.
Observation 4:
As for the basic model, operators will continue to en-
ter the market for a route until it is no longer prof-
itable.
The number of operators for route j is
k
j
= max{h : a
j
/h δ
[h] j
> 0},
where [·] is a permutation of {1,2,· ·· , n} such that
δ
[1] j
δ
[2] j
·· · δ
[n] j
.
4 NETWORK DESIGN
Using this initial framework, we have also explored
the impact of the network structure on the profit for
the service providers. We consider a service area with
t townships and compared the equilibrium solution
for a network structure where direct point-to-point
services are offered between every pair of townships,
to the equilibrium for a hub-and-spoke network where
every route between any two townships involves an
interchange via a central hub. The first one is a com-
plete network with t nodes and
t1
u=1
= t(t 1)/2
links, one for each origin-destination pair. In the sec-
ond network structure, the routes offered are between
a township and the central hub, and the total ridership
from each origin (to all destinations) is consolidated
into the ridership from the origin to the central hub.
That is, the second is a hub-and-spoke network with
t + 1 nodes and t links. Figure 1 illustrate these two
networks with t = 6. For each service provider, the
profit from a route depends on the operating cost of
offering the service, the total revenue due to the rid-
ership and the number of competitors also servicing
that route.
Figure 1: Network structures.
4.1 Hub-and-Spoke vs Complete
Network
We consider a simple case where all origin-
destination pairs have the same demand (that is, a
j
=
a for all j) and all operators have the same cost for
all routes (that is, δ
i j
= δ for all i and j). We assume
there is no loss of ridership between i and j whether
the route is direct or via a central hub. Therefore, all
demand from a node i is aggregated to the demand
on the link from the node to the hub. This is illus-
trated in Figure 2. For fair comparison between the
Figure 2: Demand on links.
two networks, the revenue on a route between a hub
and an origin/destination is half that of a direct route
between an origin and a destination. For the simplistic
case where the ridership between every pair of town-
ships are the same and all fixed operating costs are the
same, the overall profits depends on the ratio of rider-
ship to route operating cost. For a complete network
on t nodes, the total profit in a competitive equilib-
rium is:
π
C
= t(t 1)/2(a δ
k
) where k = ba/kc.
For the hub-and-spoke network, the total profit in a
competitive equilibrium is:
π
H
= t((t 1)a/2 δh) where h = b(t 1)a/2c.
Therefore, π
H
is greater than π
C
if
b
a(t 1)
2δ
c
t 1
2
b
a
δ
c.
It ca be observed that if the ratio a/δ is integer-valued
and t is odd, we have π
C
= π
H
. When this is not the
case, rounding has an impact, but not always favour-
ing one network structure over the other, as illustrated
in Figure 3.
GameTheoreticModelsforCompetitioninPublicTransitServices
137
Figure 3: Rounding Effects.
5 SUMMARY AND FUTURE
EXTENSIONS
In this paper, we investigate the competition among
operators of public transit using the framework of
congestion games. The Nash equilibria can be found
by solving an auxiliary integer programme. Using
this framework, we can draw some insights regard-
ing competition vs. centralization, and the impact of
network structure on the profit of the operators.
The assumption that the network structure and the
total service bundle being offered will not affect the
overall ridership is perhaps too restrictive and unre-
alistic. This assumption implies that all passenger
who would travel from town X to town Y with di-
rect service will still travel even if the trip involves
an interchange via a central hub. Clearly, the con-
venience level, travel time and possibly travel cost
will not be the same for the two trips. Also, appar-
ently with wider service coverage and greater num-
ber of operators servicing a particular route (higher
frequency), it will attract more ridership. A more re-
alistic model would allow for ridership to depend on
origin-destination pairs, on the network infrastructure
and also on the set of transit services available. This
may lead to a bi-level model where not only do op-
erators compete with each other but the passengers
preferences and patronage depends on the set of ser-
vices offered by the transit operators. The upper-level
model would represent the strategic game among the
service providers as they select the services to be of-
fered to maximise their individual profit. The lower-
level is the game between the public and the operators
as a group, in that the public may be diverted to other
forms of transport (e.g. taxis, private vehicles) if the
availability and service quality (e.g. interchanges re-
quired, circuitous routes, travel time) of the basket of
services offered by the operators are too low. The two
levels of the bi-level problem are interlinked since the
choice of the public to utilise public transit or not
would affect the potential ridership of the system and
thus impact the potential profit of the operators.
The potential game framework is a non-
cooperative game framework. Operators may con-
sider to cooperate when their resources are limited
(which is mostly true). One operator may offers ser-
vices on a part of the network that serve as feeder
links to the service provided by another operator,
and vice versa. Developing a cooperative game-
theoretic model may help us to compare and contrast
the equilibrium solutions of both setting, the coopera-
tive game and non-cooperative settings. In a coopera-
tive game setting, it would also be interesting to inves-
tigate what is the appropriate profit-sharing scheme to
induce higher profits or more comprehensive services
for the public.
By investigating these extensions, we may ob-
tain further insights into the relationship among the
network infrastructure, competitive situation between
operators and the impact on the type and level of ser-
vices offered to the public. These relationships could
further guide us in decision making on possible in-
frastructural investments and incentives to offer both
operators and the riding public, which is very helpful
for the government authorities and to ensure a public
transit system that well-serves the public and benefits
the community in terms of costs, convenience, qual-
ity, environmental impact and other concerns being
designed.
ACKNOWLEDGEMENTS
This research is partially supported by a GRF
grant from the Research Grants Council of the
Hong Kong Special Administrative Region, China
(CUHK414313).
REFERENCES
Bell, M. G. H. (2000). A game theory approach to measur-
ing the performance reliability of transport networks,
Transportation Research: Part B, 34, 533-545.
Castelli, L., G. Longo, R. Pesenti and W. Ukovich
(2004). Two-player noncooperative games over a
freight transportation network, Transportation Sci-
ence, 38(2), 149-159.
Chen, O. J. and M.E. Ben-Akiva (1998). Game-theoretic
formulations of interaction between dynamic traffic
control and dynamic traffic assignment, In Trans-
portation Research Record: Journal of the Trans-
portation Research Board, No. 1617, National Re-
search Council, Washington, DC., pp. 179-188.
Fernandez, E. and P. Marcotte (1992) Operators-Users
Equilibrium Model in a Partially Regulated Transit
System, Transportation Science, 26(2): 93-105.
ICORES2015-InternationalConferenceonOperationsResearchandEnterpriseSystems
138
Fernandez L., J. E., P. Marcotte, S. Mondschein, J. Vera and
A. Weintraub (1993). Solution Approaches to the Bus
Operator Problem, Transportation Research, Part B,
27B(1): 1-11.
Fisk, C. S. (1984). Game theory and transportation systems
modelling, Transportation Research B, 18B (4/5),
301-313.
James, T. (1998). A game theoretic model of road usage,
Proceedings of the 3rd IMA International Conference
on Mathematics in Transport and Control, 401-409.
Hollander, Y. and J. N. Prashker (2006). The applicability
of non-cooperative game theory in transport analysis,
Transportation, 33, 481-496.
Levinson, D. (2005). Micro-foundations of congestion and
pricing: a game theory perspective, Transportation
Research: Part A, 39, 691-704.
Monderer, D. and L. S. Shapley (1996). Potential Games,
Games and Economic Behaviour, 14, 124-143.
Nash, J. (1950). Equilibrium points in n-person games,
Proceedings of the National Academy of Sciences
36(1):48-49.
Patriksson, M. and M. Labb
´
e (eds.) (2004) Transportation
Planning: State of the Art, Kluwer Academic Publish-
ers.
Pedersen, P. A. (2003). Moral Hazard in traffic games, Jour-
nal of Transport Economic Policy 37 (1), 47-68.
Reyniers, D. (1992). Crowding levels and fare classes in
public transport, In Griffiths, J. (ed.) Mathematics in
Transport planning and Control, Oxford University
Press, pp. 319-327.
Rosenthal, R. W. (1973). A class of games possessing
pure-strategy Nash equilibria, International Journal
of Game Theory, 2, 65-67.
Van Zuylen, H.J. and H. Taale (2004) . Urban network
with ring roads: a two-level three player game, In:
Proceedings of the 83rd Annual Meeting of the Trans-
portation Research Board, Washington DC.
von Stackelberg, H. F. (1934). Marktform und Gle-
ichgewicht (Market and Equilibrium), Vienna.
Wardrop. J. G. (1952). Some theoretical aspects of road
traffic research”, Proceedings of the Institution of
Civil Engineers, Part II, 1, 325-378.
GameTheoreticModelsforCompetitioninPublicTransitServices
139