Global Airline Alliances Transportation Research 2022

Contents lists available at ScienceDirect
Transportation Research Part A
journal homepage: www.elsevier.com/locate/tra
Codesharing network vulnerability of global airline alliances
Richard Klophaus a,⁎, Oriol Lordan b
a Competence Center Aviation Management (CCAM), Worms University of Applied Sciences, 67549 Worms, Germany
b Universitat Politècnica de Catalunya – Barcelona Tech., Department of Management, Colom 11, TR6, 3.17, 08222 Terrassa, Spain
A R T I C L E I N F O
Keywords:
Global airline alliances
Codesharing
Network vulnerability
Complex networks
A B S T R A C T
Global airline alliances provide connectivity based on codesharing agreements between member
airlines. An alliance member exit leads to the deletion of routes (if not operated by other
members) which affects network connectivity. The paper measures the vulnerability of the co-
desharing network (CN) of Star Alliance, SkyTeam and oneworld, respectively, by applying the
theory of complex networks. A normalized CN vulnerability metric is proposed. Using airline
schedules data, a ranking of member airlines according to their share in the overall CN vulner-
ability is derived. The results for CNs are compared with the ones for the respective total network
(TN) that includes routes with and without codesharing. The findings show that oneworld is the
most vulnerable global airline alliance, SkyTeam ranks second followed by Star Alliance. The
proposed graph theory approach might become a building block for a more comprehensive
measurement of real world airline networks.
1. Introduction
The restructuring of airline activities into branded global alliances has been one of the major traits of this industry since Star
Alliance was founded in 1997, followed by the formation of oneworld in 1999 and SkyTeam in 2000. Global alliances provide
network connectivity based on codesharing agreements between member airlines. Codesharing is an interline partnership where one
carrier sell tickets by placing its designator code on another carrier's flights. The airline selling seats is referred to as the marketing
carrier and the airline providing the flight is referred to as the operating carrier. While codesharing dates back to the 1960s, it only
became common in the early 1990s. The launch of the modern era of global airline alliances began with the large-scale codesharing
agreement between Northwest Airlines and KLM in 1989. The aim is that the route network of a global alliance appears to be an
extension of each partner’s network. Codesharing in combination with coordinated flight schedules allows the provision of con-
tinuous services for passengers connecting between airlines. However, alliance members may take advantage of the route networks of
partner airlines even without codesharing, e.g., due to interline agreements between individual airlines that cover connecting flights.
A codeshare agreement usually requires an interline agreement.
At present, Star Alliance has 28 member airlines, SkyTeam 20, and oneworld 14 (cf. Appendix A) which together have a share of
around 60% in worldwide air traffic. Extensive codesharing among global alliances allows airlines to offer routes without operating
them which is cost-efficient. Avoiding overlapping operations also implies less competition. The drawback is a dependency on partner
airlines. A member exit leads to the deletion of routes (if not operated by other alliance members) which affects network connectivity.
In 2014, US Airways and TAM left Star Alliance after these two carriers merged with airlines from oneworld. oneworld on its part lost
Malév after the financial collapse of this former Hungarian flag carrier in 2012. In early 2016, Qatar Airways threatened to withdraw
from oneworld should fellow member American Airlines continue to push the US government to restrict market access for the Gulf
https://doi.org/10.1016/j.tra.2018.02.010
Received 11 August 2017; Received in revised form 13 February 2018; Accepted 19 February 2018
⁎ Corresponding author.
E-mail addresses: [e-mail] (R. Klophaus), [e-mail] (O. Lordan).Transportation Research Part A 111 (2018) 1–100965-8564/ © 2018 Elsevier Ltd. All rights reserved. T

carriers. An exit of partner airlines has negative consequences for global alliances, e.g., in the form of sunk costs due to alliance-
specific investments or the risk that former alliance members use confidential information to their competitive advantage. Further, it
implies a decrease in network coverage. The assessment of the (potential) damage to the codesharing network (CN) of global airline
alliances is the subject matter of the present paper.
Not all member exits have the same impact because some airlines contribute more to the CN of a global alliance than others.
Therefore, it is an important issue for the managing bodies of an alliance how to accurately assess the impact of an exit of a given
member airline and similarly, how to develop a CN with appropriate partner selection. This paper studies the CN vulnerability of
global alliances to member exits. We propose measures that can be instrumental in assessing the dependency of an alliance on a
member’s route network and can also serve to develop a more resilient CN. The results for CNs are compared with the ones for the
respective total network (TN) that includes routes with and without codesharing.
Research on airline alliances and codesharing among airlines includes several studies on the effects of airline alliances on traffic
volumes and airfares (Oum et al., 1996; Park, 1997; Park and Zhang, 1998; Brueckner, 2001; Brueckner et al., 2011; Zou et al., 2011).
Kleymann and Seristö (2001) analyzed the trade-off between alliance benefits and risks. Douglas and Tan (2017) examined whether
the formation of global airline alliances resulted in an increase in profitability for the founding members. Garg (2016) presented a
model based approach to select strategic alliance partners. Different reasons for a company to leave an inter-firm co-operation are
discussed by Sroka and Hittmár (2013). The welfare effects of codesharing agreements have been investigated by Hassin and Shy
(2004) and, more recently, Adler and Hanany (2016).
There is an increasing and extensive literature of transport vulnerability studies. This paper measures the (potential) damage for
the CNs of global airline alliances caused by member exits. Such alliances constitute an intermediate level of air transport networks
between individual airline networks and the industry network (Lordan et al., 2014a). Mattsson and Jenelius (2015) provide an
overview of recent research on vulnerability and resilience of transport systems. While they point out that there is no commonly
agreed definition of transport system vulnerability, they conceptualize vulnerability as the susceptibility of transport systems to
infrequent events that can result in considerable network degradations. In the context of the present paper, the infrequent event is a
member exit having an adverse impact on the CN of a global airline alliance. The study of air transport networks includes the
topological analysis of global (Guimerà and Amaral, 2004; Guimerà et al., 2005) and regional (Bagler, 2008; Zhang et al., 2010) route
networks. Vulnerability is investigated for global (Lordan et al., 2014b) and regional (Chi and Cai, 2004; Du et al., 2016) networks.
Lordan et al. (2015) examine the robustness of alliance airline route networks based on the assumption of unweighted networks only
considering operating flights. Hence, the differing economic relevance of a given route operated by one member to other alliance
members is disregarded. Weights could be based on the number of flight frequencies or seat capacities, and also by distinguishing
between domestic, continental and intercontinental routes. The consideration of codesharing as an indicator of route relevance from
the perspective of partner airlines represents a basic weighting scheme to enhance the practical meaning of the vulnerability mea-
sures. CNs are subsets of the respective TNs which consist only of operated routes that have a marketing flight number by at least one
other carrier belonging to the same alliance. Our paper only looks at codesharing between member airlines of the same global
alliance. In spite of this, the industry showcases other codesharing partnerships. There is codesharing between aligned and non-
aligned airlines (e.g., Qantas and Emirates) and between carriers belonging to the same holding company (e.g., Lufthansa and
Eurowings). Codesharing across global alliances is rather unusual. One example is Aeroflot and Finnair on the Helsinki-Moscow-
route.
In this paper, CN vulnerability of real world networks is analyzed building on the theory of complex networks (Estrada, 2011;
Estrada and Knight, 2015). More specifically, CN vulnerability is measured using the concept of normalized average edge be-
tweenness (Mishkovski et al., 2011). The proposed method to measure CN vulnerability relates to work using a graph theory ap-
proach to develop strategies to increase the resilience of air traffic networks to disruptive events, such as extreme weather events,
strike action or terrorist threats (Dunn and Wilkinson, 2015). It might also be valuable for a more comprehensive study of route
networks that include other network indicators such as hubness and size (Roucolle et al., 2017).
The proposed methodology provides a normalized measure of the vulnerability of a given CN to (potential) member exits. Data
comes from the OAG airline schedules database. One result of applying this measure is that oneworld is the most vulnerable CN,
SkyTeam ranks second and Star Alliance is the most robust CN. Further, the paper indicates a positive relation between network
robustness and route overlaps among members of global airline alliances. We also rank member airlines according to their con-
tribution to the overall CN vulnerability. Our paper shows that the size of a carrier’s scheduled operation is not strictly related to the
carrier’s importance for the vulnerability of an airline alliance route network. Finally, a comparison with results for TNs as un-
weighted alliance route networks illustrates the importance of bringing out relevant routes in future analysis of airline route net-
works.
2. Methodology
A codesharing network (CN) contains airports (nodes) connected by codeshared routes (edges), i.e., two airports are linked if an
alliance member is operating flights between them with a designator code of another carrier from the same alliance. The proposed
metric to assess CN vulnerability extends the graph theory concept of average edge betweenness as introduced by Boccaletti et al.
(2007) for the graph G:∑=∈b G E b( ) 1| | l El
(1)
R. Klophaus, O. Lordan Transportation Research Part A 111 (2018) 1–102

where |E| is the number of edges and b l is the edge betweenness of the edge l defined as∑=≠b n ln( )li jijij (2)
where nij (l) is the number of geodesics (shortest paths) from node i to node j that contain the edge l, and n ij is the total number of
shortest paths between i and j. If N represents the number of nodes of a network, then the b(G) values for a complete graph and a path
graph are= = +b G b G N N( ) 1 and ( ) ( 1)6complete path (3)
and, hence, b(Gcomplete ) ≤ b(G) ≤ b(G path ). G is more robust than G', if b(G) < b(G'). The normalized average edge betweenness of a
network is defined as (Mishkovski et al., 2011)= −− = −−+b G b G b Gb G b Gb G( ) ( ) ( )( ) ( )( ) 11norcompletepath complete N N( 1)6 (4)
where b nor(G) ranges from 0 (i.e., the most robust network) to 1 (i.e., the most vulnerable network). Thus, b nor(G) is a normalized
measure of network vulnerability. The contribution of a member airline to the overall vulnerability of a CN can then be calculated as
the relative difference of the normalized average edge betweenness, that is= ′ −D b G b Gb G( ) ( )( )member nor nornor (5)
where G' is the graph obtained from G (i.e., the entire CN) after removing the edges of the exiting member airline which are not
operated by any other member. A positive value of D member implies that the CN becomes more vulnerable. The higher the value of
Dmember the more negatively affected is the CN by the exit of the respective airline. A negative value of D member would mean that a
member exit is actually decreasing the CN vulnerability, i.e., the alliance is more robust without this airline.
3. Results
The vulnerability of the three global alliances is analyzed using OAG airline schedules data for the week ending September 11,
2016. In Fig. 1 we rank member airlines of Star Alliance, SkyTeam and oneworld based on their contribution to the codesharing
network (CN) of the respective alliance, i.e., according to their D member value. ALL (with value 0) refers to the entire CN without any
exit.
As D member is calculated with the normalized average edge betweenness b nor (G) of a specific alliance, this metric measures the
relative impact on network vulnerability caused by the removal of an airline from this alliance. As the value of b nor(G) can vary
between alliances, the absolute impact on network vulnerability can be quite different across alliances for similar values of D member.
Taking this into account, the relative impact of American Airlines (AA) on oneworld’s CN is comparatively larger than the one of
United Airlines (UA) on Star Alliance and Delta Airlines (DL) on SkyTeam. This is illustrated in Fig. 1 by the length of the bars
representing UA, DL, and AA in relation to the bars of the other carriers of Star Alliance, SkyTeam, and oneworld, respectively.
Table 1 provides the values of the average edge betweenness b(G'), the normalized average edge betweenness b nor (G'), and
relative difference of the normalized average edge betweenness D member for each member airline. For ALL, b(G') and b nor (G') equal b
(G) and b nor (G), respectively, as ALL stands for the CN without any member removal. The b nor (G) value ALL = 0.00724 for one-
world’s entire CN is larger than the respective values for SkyTeam and Star Alliance which makes it the most vulnerable among the
three CNs. While the values of D member and bnor(G') represent a one-to-one mapping, i.e., a higher (lower) value of D member is strictly
related to a higher (lower) value of b nor (G'), this is not the case for the relation between D member and b(G'). For example, the b nor (G') of
the Star Alliance members United Airlines (UA) and Air China (CA) are 0.00508 and 0.00464, respectively, while the values for b(G')
have a reverse order (277.7 and 312.6). The average edge betweenness does not account for the change in the number of airport
nodes of a CN resulting from a member exit. This is the reason why D member is computed as the relative difference of the normalized
average edge betweenness of a CN with and without a given member airline.
CNs contain only routes operated by one member airline of a global airline alliance network that also has codesharing (i.e., a
marketing flight number) by at least another carrier from the respective alliance. Codesharing is an indicator for relevance of a given
alliance route. This can also be seen as a basic weighting scheme assigning the weight “1” to all codeshared routes and “0” to all non-
codeshared routes. The results for CNs are now related to the respective total network (TN) that includes routes with and without
codesharing.
Table 2 provides the values for nodes N, edges E, shared nodes S N, and shared edges SE of each airline belonging to one of the
three TNs. SN and SE stand for the number of airport and route duplicates in a TN with other member airlines out of the total alliance
nodes and edges. All non-shared nodes and edges of an airline, i.e., all airports and routes not operated by any other alliance member,
disappear from the TN if this airline leaves the alliance. For ALL, S N and S E stand for all duplicates among its members out of the total
alliance nodes and edges. 39.0% of all 1207 weekly scheduled airports operated by Star Alliance in September 2016 are duplicates,
33.4% out of 1058 at SkyTeam and only 27.3% out of 942 at oneworld. oneworld has the lowest S E percentage with 2.6%, while SE
percentages for Star Alliance and SkyTeam are 5.8% and 9.0%, respectively. S N and SE for ALL are the lowest for oneworld which are
R. Klophaus, O. Lordan Transportation Research Part A 111 (2018) 1–103