Transport Policy 38 (2015) 40–51

Contents lists available at ScienceDirect

Transport Policy

http://d0967-07

n CorrE-m

rdoblehhecro45

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

Container port competitiveness and connectivity: The Canary Islandsmain ports case

Beatriz Tovar a,n, Rubén Hernández b, Héctor Rodríguez-Déniz c

a Infrastructure and Transport Research Group (EIT), Department of Applied Economics, University of Las Palmas de Gran Canaria, FCEE D.2.20, 35017 LasPalmas de Gran Canaria, Spainb Department of Logistics, Innovation Center, Industrial Engineering Department, Mahidol University, 25/25 Phuttamonthon 4 Road, Salaya, NakhonPathom73170, Thailandc Statistics Division, Department of Computer and Information Science, Linköping University, 581 83 Linköping, Sweden

a r t i c l e i n f o

Article history:Received 10 April 2014Received in revised form27 October 2014Accepted 2 November 2014Available online 18 December 2014

Keywords:Container portsPort competitivenessPort connectivityPort accessibilityCanary Islands portsTransport networks

x.doi.org/10.1016/j.tranpol.2014.11.0010X/& 2014 Elsevier Ltd. All rights reserved.

esponding author.ail addresses: btovar@daea.ulpgc.es (B. Tovar)@hotmail.com (R. Hernández),9@student.liu.se (H. Rodríguez-Déniz).

a b s t r a c t

The Canary Islands' economy is extremely dependent on sea transport. Since accessibility and con-nectivity are major determinants of international transport costs, the analysis of their main ports' con-nectivity is crucial for keeping costs under control. Since different port authorities manage the majorports of the Canary Islands, they could be tempted to compete for transshipment cargoes, instead ofworking together to facilitate supply chain integration that would increase their competitive standing.

The aim of the paper is twofold. First, the infrastructure and superstructure endowment of the mainCanarian ports and their accessibility, by evaluating site and situation factors, is documented. Secondly,the connectivity of the main Canarian ports is assessed by means of graph theory. This provides im-portant measures that define a port's competitiveness, and its potential to achieve or keep regional orglobal hub status, and also to follow its evolution. A brief review of papers measuring port connectivitybased on graph theory is included to illustrate the current approaches in port network analysis, and tojustify our methodological framework. A sub-network of 53 ports directly related with Las Palmas andTenerife ports has been selected for this purpose.

Our findings are mainly related to the connections among the nodes in the sample network, and tothe position that the targeted ports hold. Additionally, some policy recommendations, regarding how toimprove the connectivity and competitiveness of the Canarian ports, are also enumerated. Previousanalysis indicates that, at present, the Las Palmas port is the only regional hub in the Canaries. BothCanarian port authorities should differentiate themselves by specializing in certain valued added servicesand increasing traffic in these services. This would avoid the danger of a destructive competition be-tween them to attract transit traffic. In summary they should be proactive in maintaining and improvingthe main Canarian ports' connectivity.

& 2014 Elsevier Ltd. All rights reserved.

1. Introduction

As interchangers between sea and land transportation modes,ports constitute a central element within any transportation sys-tem, and by extension for the economy as a whole. The sig-nificance of a port is even greater for the economy in island re-gions where practically the totality of goods enter and leavethrough ports. This is the case for the Canary Islands whosetourism-based economy is highly dependent on the sea transportthat satisfies the needs of the population and the millions of

,

tourists visiting the islands each year. Significant levels of con-nectivity contribute to the competitiveness of ports (Lee et al.,2014), and generate network effects that contribute to the econ-omy (Laird et al., 2005). Moreover, and as Martinez and Hoffmann(2007) show, connectivity is one of the main determinants of in-ternational transport costs, so improving port connectivity iscrucial for keeping transport costs under control.

The Canary Islands form one of Spain's 17 autonomous com-munities, and are one of the outermost regions of the EuropeanUnion. Located in the Atlantic Ocean and consisting of seven is-lands, they are situated 115 km from the northwest African coastand are at an average distance of 1750 km from Madrid. TheCanary Islands main ports, Las Palmas and Tenerife, are located inthe islands of Gran Canaria and Tenerife, respectively; togetherthey accounted for more than 88% of the total freight moved to

B. Tovar et al. / Transport Policy 38 (2015) 40–51 41

and from other geographical areas in 2013. They are managed bydifferent port authorities.1

Both centrality and intermediacy play a role in defining themerits of a port as a transshipment hub. Ideally such ports need tobe both central to the market they serve and intermediary to theshipping lanes linking markets (McCalla, 2008). Las Palmas port iscurrently a transshipment hub and the Mediterranean ShippingCompany (MSC), one of the leading global shipping lines world-wide, has channeled part of its operations in the region through it.On the other hand, Tenerife port, which has the same situationfactors, plans to improve its site factors, in order to be capable ofassuming more international container traffic. Currently, it falls farshort of Las Palmas' port records.

The advantages of having good port connectivity were accruedto the Canary Islands once one of their ports became an interna-tional hub. Due to the global nature of the transshipment business,where global operators are seeking their own strategies and ob-jectives, it seems sensible to ask whether both port authoritiesshould cooperate. This might avoid placing themselves in a weakerposition to negotiate with shipping companies, due toovercapacity.

In the actual context of intense competition and decliningfreight rates Asgari et al. (2013) have introduced the idea of co-operation as a potential substitution for competition. They statecooperation can take the form of (i) horizontal cooperation be-tween/among the ports, (ii) vertical cooperation between the portsand the shipping company, and (iii) full cooperation among all ofthese stakeholders. Moreover, it has been increasingly recognizedthat some degree of coordination among ports can enable in-creases in efficiency and in supply chain integration, which farfrom reducing competition, in fact increases the ports' competitivestandings. Moreover, the building of cooperative relationshipswith other ports allows them to provide a range of incentives toshippers and operators, in order to attract trade volumes (Merk,2013).

Therefore, and trying to shed light on the aforementioned ar-gument the aim of the paper is twofold. First, it aims to documentthe port infrastructure and superstructure endowment of both themain Canary Island ports, and to give the first qualitative ap-proximation of their sea accessibility through an evaluation oftheir site and situational factors. The second objective is to com-plete the previous analysis by providing the first direct evaluationof the main Canarian port connectivity based on graph theory.These are important measurements in establishing the competi-tiveness of a port (Low et al., 2009), and in measuring its potentialfor achieving or keeping hub status, either regionally or globally.These let us follow future evolution. To this end, in the ensuingsection, a survey on the current state of literature measuring portconnectivity, based on graph theory, is presented.

The paper is structured as follows. In Section 2 we provide abrief but comprehensive review of studies of port connectivityusing graph theory techniques. Moreover, the Canary Islands mainports are described. Section 3 describes the data and presents themethodological issues. Results, discussion and policy implicationsare presented in Section 4. Finally, Section 5 presents the mostrelevant conclusions and directions for future research.

1 For a detailed analysis of the port management model in Spain, see Ro-dríguez-Álvarez and Tovar (2012) and Tovar and Wall (2014).

2. Background

2.1. Graph theory and port connectivity

Although the introduction of the maritime connectivity con-cept is relatively recent in the literature, it has rapidly gainedpopularity. An increasing number of studies analyze the influenceof maritime connectivity on different aspects such as maritimetransport cost (i.e. Kumar and Hoffmann, 2002; Wilmsmeier et al.,2006; Martinez and Hoffmann, 2007; Márquez-Ramos et al., 2011;Duval and Utoktham, 2011), port/regional competitiveness (i.e.;Yeo et al., 2008; Wang and Cullinane, 2008; Verhetsel and Sel,2009; Freire and Pais, 2011; Yeo et al., 2011) logistics connectivity2

(i.e. Notteboom, 2004; Kronbak and Cullinane, 2011) and maritimesecurity (i.e. Bichou, 2004; Angeloudis et al., 2007). Relatively afew studies have analyzed port connectivity using graph theory(see Table 1).

Roughly defined, graph theory is a branch of mathematicsconcerned with how networks can be encoded and their proper-ties measured (Rodrigue et al., 2006). As far as we are aware, thefirst paper to use the twin geographical concepts of centrality andintermediacy is Fleming and Hayuth (1994). They identify bothcharacteristics as spatial qualities that enhance the traffic levels oftransportation hubs, and hence indicate which places are strate-gically located. However, their analysis is qualitative because theyapproached both concepts via port throughput, instead of calcu-lating them by constructing a network.

The first empirical studies that have attempted to measure portconnectivity based on graph theory appear in the 2000s. This isprobably due to the commencement in 2001 of the installation ofAutomatic Identification System (AIS) equipment in ships andports. This has facilitated easier access to the data demands en-tailed by the maritime network construction. Table 1 provides anoverview of studies using this technique, and it can be seen that, inall of them, a maritime network was built using data basicallyfrom two sources: Containerization International (e.g. McCallaet al., 2005; Cullinane and Wang, 2009 and 2012; Wang and Cul-linane, 2014) and Lloyd's Marine Intelligence Unit (e.g. Ducruetet al., 2010a; Kaluza et al., 2010; Gonzalez et al., 2012).

Regarding the graph definition itself, there is no unanimity onthe circumstances in which a pair of ports should be consideredlinked, so as to take into account the complexity of the maritimenetworks. Two graph configurations have appeared to deal withthis particular problem, and while some authors do not refer tothis issue many others define their networks according to thisframework, and perform parallel studies for both models. Thesetwo representations are referred to as Graph of Direct Links (GDL)and Graph of All Links (GAL). The GDL represents the sequence ofports a vessel calls at within a liner service (Fig. 1), and results in asimple approximation of the topology of the network; this is be-cause two ports will be connected, only if they are called at con-secutively. GAL, however, includes the consecutive links betweenports on a specific route, and also completely connects the ports ofthe service, by making the assumption that ports in the sameservice are linked either directly or indirectly. This latter approachis expected to be a better representation of liner shipping.

These studies can be grouped in terms of the method used tobuild the network, and these may adopt important differences intheir respective structures. The first set of studies uses a networkrepresentation that only includes direct successive calls betweenports, i.e. GDL, (McCalla et al., 2005; Ducruet et al., 2010a; Kaluza

2 Whereas port connectivity is focused on the connectivity among ports, lo-gistics centered connectivity goes a step further and in its analysis includes otherareas influenced by a port such as foreland and hinterland.

Table 1Studies of port connectivity using graph theory.

Author Dataa Sourceb Graph typec,d,e Main measures Goal

(A) (B) (C)

McCalla et al. (2005) 1994, 90 Caribbean ports 2002, 88 Caribbeanports SN

CIY UW N/S GDL β coefficient Analyze the container shipping network and the emergence of transshipmenthubsDegree centrality

Wang and Cullinane(2008)

2004, World Wide top 10 container portsWFFCCV

SR, AX UW, W(TEUs,NM)

UD N/S Principal eigenvector To assess the competitive position of a port through accessibility

Hu and Zhu (2009) 2006, 878 WWide ports CS CIO UW, W(CL) D GDL,GAL

Degree centrality To study the worldwide maritime transportation network from a complexnetwork perspectiveBetweenness centrality

Clustering coefficientNearest neighborsdegree

Cullinane and Wang(2009)

2004, World Wide top 10 container portsWFFCCV

CIY, AL W(TEUs) UD N/S Port accessibility index The formulation of an index of port accessibility

Ducruet et al. (2010a) 1996, 975 WWide ports 2006, 1240 WWideports WFFCCV

LMIU W(TEUs) N/S GDL Degree centrality To explore the properties of liner shipping networks and their influence onthe evolution of port hierarchiesBetweenness centrality

Vulnerability

Ducruet et al. (2010b) 1996 and 2006, 23 Northeast Asian portsWFFCCV

LMIU W(TEUs) N/S GDL,GAL

Degree centrality To investigate how ports are positioned in the Northeast Asian network be-tween 1996 and 2006Betweenness centrality

VulnerabilityRelative diversity index

Ducruet et al. (2010c) 1996, 307 Atlantic Ports 2006, 351 AtlanticPorts WFFCCV

LMIU W(TEUs) UD GDL,GAL

Degree centrality To verify to what extent the hub-and-spoke strategies of ports and oceancarriers have modified the structure of a maritime network, based on theAtlantic case

Betweenness centralityVulnerabilityBisecting K-meansclustering

Kaluza et al. (2010) 2007, 951 Ports CS, GCS, BCS LMIU W(TEUs) D GDL Degree centrality To study the spread of invasive species through a complex network of globalcargo ship movementsBetweenness centrality

Cluster coefficientNode strength

Wang and Cullinane(2011)

2010, 18 East Asia 2010, 21 East–West routeContainer ports LSN

AL, CIO W(TEUs) D N/S Principal eigenvector To assess the effects of traffic consolidation on the container port industryBetweenness centralityMultiple linkage analysisHierarchical clustering

Ducruet and Notteboom(2012)

1996, 975 WWide ports 2006, 1240 WWideports WFFCCV

LMIU W(TEUs) UD GDL,GAL

Degree centrality Analyzes the relative position of ports in the global network through in-dicators of centralityBetweenness centrality

Vulnerability

Ducruet and Zaidi (2012) 1996, 975 WWide ports 2006, 1240 WWideports WFFCCV

LMIU UW UD GDL,GAL

Degree centrality To discover whether are there bridges (hubs) and communities (clusters) inmaritime networkTopological decompos.

Cullinane and Wang(2012)

2010, 18 East Asia 2010, 21 East–West routecontainer ports LSN

AL, CIO W(TEUs) D N/S Multiple linkage analysis To determine an alternative basis for the establishment of a port hierarchywithin a networkHierarchical clustering

Gonzalez et al. (2012) 2008, 584 WWide ports 2010, 861 WWideports WFFCCV

LMIU W(TEUs) N/S N/S Degree centrality To look at changes in the maritime network prior to and after the crisisBetweenness centralityVulnerability

Pais et al. (2012) Mar 08–Feb 09, 1268 Mar 09–Feb 10, 1768 Mar10–Feb 11, 1692 World Wide ports CS and GCS

LMIU W(TEUs) UD N/S Degree centrality To search the way general and containerized traffic has evolved between 2008and 2011Betweenness centrality

Wang and Cullinane(2014)

2010, 18 East Asia 2010, 21 East–West routecontainer ports LSN

AL, CIO W(TEUs) D N/S Multiple linkage analysis To assess the effects of freight traffic consolidation on the container portindustryHierarchical clustering

B.Tovaret

al./Transport

Policy38

(2015)40

–5142

Presen

tstudy

Octob

er20

12,5

3W

orld

Wideports

CS

AL

W(TEU

s)UD

GDL

Deg

reecentrality

Evaluatetheaccessibility

andco

nnectivity

ofthemainCan

arianports

Betwee

nnesscentrality

Vulnerab

ility

Port

accessibility

index

aLS

N:Liner

shippingnetworks,W

FFCCV:W

orld

Flee

tof

Fully

CellularCon

tainer

Vessels,C

S:co

ntainer

ship,G

CS:

general

cargosh

ip,a

ndBCS:

bulk

cargosh

ip.

bCIY:Con

tainerizationIntern

ational

Yearbo

ok,A

L:Alphaliner,A

X:Axsmarine,

SR:Se

arates,C

IO:Con

tainerizationIntern

ational

Onlin

e,an

dLM

IU:Lloy

d's

MarineIntellige

nce

Unit.

c(A

)Ed

geweigh

t.UW:unw

eigh

ted,W

(C):

weigh

tedviaC,T

EUs:

20-ftEq

uivalen

tUnits,

CL:

no.

ofco

ntainer

lines,a

ndNM:nau

ticalmile

s.d(B)Ed

gedirection

.UD:undirected,D

:directed,a

ndN/S:not

specified

.e(C)Networkrepresentation

.GDL:

Graphof

DirectLinks,G

AL:

Graphof

AllLinks,a

ndN/S:not

specified

.

Fig. 1. Example of GDL and GAL configurations.Source: Own elaboration.

B. Tovar et al. / Transport Policy 38 (2015) 40–51 43

et al., 2010). The second set uses both GDL and GAL (e.g. Hu andZhu, 2009; Ducruet et al., 2010b; Ducruet and Zaidi, 2012). Avariety of strategies are employed, in order to characterize therelationship between each pair of connected ports; these includethe direction of the itineraries and the addition of weights, whichare also present in the literature.

2.2. The Canary Islands' main ports

The Canary Islands have an excellent strategic position, andmajor maritime routes linking the northern and southern hemi-spheres cross through their geographical area; these maritimelines channel a significant percentage of worldwide trade. Themain Canarian ports are currently ranked among the top 10Spanish ports, in terms of cargo movement. Moreover, looking atthe worldwide rankings, in terms of container throughput, LasPalmas port is ranked 87 in the top 200 ports worldwide andTenerife is ranked 189 (CI-Yearbook, 2011).

A comparison between the two major Canarian ports, in termsof container traffic throughput and the number of linked portsworldwide, shows Las Palmas to be the most important port in theCanary Islands (see Table 2). The Las Palmas port transhipmenttraffic, which is mainly container operations, now accounts fornearly 69% of the total number of handled containers. Tenerifeport, on the other hand, merely focuses its container traffic on thedomestic market.

Following McCalla (2008), it has been noticed that two fun-damental geographical factors define the relative merits of oneport over another: the site characteristics of a port and its geo-graphical situation. The in situ characteristics can be alteredthrough investment and management; however, the situationfactors are more important since relative location cannot bemanaged.

Table 3 below shows the centrality of the main Canarian ports,and it can be stated that, unlike cases such as Singapore or Mar-saxlokk, the main Canarian ports do not have a multi-scale cen-trality and intermediacy. Neither of them have an important rolein terms of global centrality; however, in terms of hemisphericcentrality, both are almost equidistant from north Europe, south-east Africa and the Mediterranean. At a regional level, the two

Table 2Container handling at Canarian ports 2008–2011.Source: Las Palmas port and Tenerife port Yearbooks 2008–2011 and Puertos delEstado statistics 2013.

Year Domestic T/shipment Total TEUs Numberof linkedportsTEUs % TEUs %

Las Pal-masport

2008 441,028 35.72 793,746 64.28 1,234,774 4482009 353,484 37.52 588,655 62.48 942,139 3762010 380,572 35.99 676,742 64.01 1,057,314 4332011 383,975 31.16 848,246 68.84 1,232,221 n/a

Tenerifeport

2008 357,723 97.80 8,046 2.20 365,769 1802009 308,724 96.37 11,620 3.63 320,344 1952010 315,348 95.19 15,928 4.81 331,276 1962011 319,825 96.83 10,455 3.17 330,280 261

Table 3Main Canarian ports' site and situation factors (centrality) 2013.Source: Distances from http://www.searates.com, and personal communications with terminal operators and websites.

Situation factors (centrality)

Scale Sailing distance from Las Palmas port Tenerife portTo Nautical miles Nautical miles

Global (east–west) Central America (Manzanillo Panama) 3,849 3,789East Coast North America (New York) 3,017 2,983East Coast South America (Santos) 3,707 3,700Far East (Singapore) 7,612 7,629South Africa (Cape Town) 4,406 4,416

Hemispheric (north–south) Mediterranean (Marsaxlokk) 1,736 1,757North Europe (Rotterdam) 1,740 1,737South East Africa (Abidjan) 1,984 2,003

Regional Canary Islands (S.C. Tenerife) 50 –

Canary Islands (Las Palmas) – 50North East Africa (Dakar) 844 863South Spain (Algeciras) 702 719

Site characterstics

Terminal Las Palmas port Tenerife port

Name OPCSA La Luz CAPSA TCTQuay length (m) 1,780 1,099 480 697Draught (m) 14 to 18 11 to 18 14 16Ship to shore quay cranes 8 5 4 3Area (ha) 33.2 16.1 10.5 15.28Management terminal operator MSC-JPMorgan-OPCSA Grupo Boluda Grupo TCB OHL-DavilaMax. vessel can be accommodated 10 4/5 3/4 2Average moves per hour and crane 28–30 28–30 28–30 28–30

B. Tovar et al. / Transport Policy 38 (2015) 40–5144

main Canarian ports are well located with regard to both southSpain and northeast African ports. Regarding site characteristics,two container terminals can be identified at the Tenerife port,CAPSA y TCT; both of them have similar characteristics in terms ofdraught and quay cranes, and both are operated by national en-tities. There are also two major terminals in Las Palmas port: LaLuz, operated by Grupo Boluda and OPCSA; this second portterminal, which is part owned by the Mediterranean ShippingCompany (MSC), is the largest terminal with a quay length of1,780 m and 8 quay cranes. Both terminals, offer draughts of up to18 m and can operate even with 11,000 TEUs container vessels.MSC is the only global operator in the Canarian ports. Las Palmasport currently serves, through its transshipment hub role, onetransshipment foreland market; i.e. the Mediterranean Sea. It canbe concluded that, although both the main Canarian ports have asimilar centrality, they greatly differ in their site characteristics.

In this section, we have presented a qualitative approximationof the site and situation factors of the main Canarian ports. Inorder to provide a more detailed understanding, we will continuemeasuring both ports' connectivity through different measuresbased on graph theory. We perform our analysis using a variety ofpopular measures, in order to assess the connectivity (maritimedegree), centrality/accessibility (betweenness, port accessibilityindex) and inter-port dependency (vulnerability) of each port inthe sample. All of these graph-based methodologies are wide-spread in the current literature (see Table 1) on maritimenetworks.

3. Data and methodology

3.1. Database

In order to measure port connectivity, we collated the linercontainer traffic data from October 2012. Since the containershipping companies' statistical data on traffic flows handled by

each pair of ports is supposed to be highly strategic for competi-tion, obtaining the required data is the most difficult aspect of anynetwork construction process. Under those terms, and consideringa high correlation between the existing shipping capacity and theactual market demand, and following what other authors such asWang and Cullinane (2011) and Gonzalez et al. (2012) have alsoconsidered, the transportation capacity between the pair of portsis used in the present paper as a faithful representation of theactual traffic flows in the construction of the network. Our net-work has been built using data from the Alphaliner database.

The processes to select those ports that make up the finalnetwork have had several stages. Firstly, using both the main Ca-narian ports 2011 annual reports, those ports with higher cargomovement with Las Palmas' and Tenerife's ports were identified;subsequently, and in order to include only the more relevant ports,feedback from several port container terminal managers wasconsidered. Finally, the top performing ports worldwide, in termsof container throughput, were also considered. The final sample ismade up of 53 ports, representing 36 countries and coveringtwelve different geographical areas as is shown in Fig. 2.

Finally, a key area of concern could be the ports selected for thestudy. The selected network may seem too broad, given that LasPalmas and Tenerife are regional ports where only one of them hasa modest proportion of intercontinental transshipment. We re-plicate all the calculation with two different network configura-tions, in order to test the influence of different network sizes andcompositions on our results. The first alternative network wasbuilt by removing the large Asian hubs, Shanghai and Singapore,and Los Angeles port to check the influence of the most significantand/or remote hubs on the Canarian ports connectivity. In thiscase, the final sample consists of fifty ports. The second alternativewas built by keeping those ports located in the Atlantic Ocean, sothe final sample is made up of 42 ports. As we observed no sub-stantial difference between the results from our network and thetwo reduced alternative options hereafter, we only report the re-sults referred to in our original network (see Fig. 2).

Fig. 2. International maritime routes and ports in the simple.Source: Las Palmas and Tenerife port authorities' websites.

Table 4Relevant characteristics of node connectivity indicators.

Measure Weighting Scope Purpose

Degree No Local Connectivity/accessibilityBetweenness No Global Connectivity/centrality/

intermediacyPort accessibility index TEUs Global Economical accessibility/

attractiveness

B. Tovar et al. / Transport Policy 38 (2015) 40–51 45

It is important to highlight that, unlike other authors' papers,such as Kaluza et al. (2010) and Pais et al. (2012), among others,our sample includes any shipping company and any vessel movingcontainers regardless of size. Finally, using Alphaliner allows us toobtain the information needed (containerized cargo), avoid con-sidering vessels which have purposes other than stevedoring ac-tivities when entering a port; this is in contrast to Ducruet et al.(2010c).

The data needed some preprocessing, in order to proceed withthe analysis. Each shipping service in the sample provided in-formation related to the origin, destination, intermediate ports ofcall and the vessel capacity in TEUs. Therefore, each segment of theservice, including the intermediate stops, was considered as anindividual directed path comprising origin, destination and thecapacity of the link between them. Thus the merging of all theseindividual sub-networks, and the aggregation of the vessel capa-cities for the links for the same origin–destination pair, resulted inthe complete network; i.e. a GDL.

3.2. Methodology

The analysis of transport networks and the development ofindicators to measure connectivity using graph theory is not arecent issue (e.g. Garrison and Marble, 1960; Kansky, 1963). Inspite of that, the study of the connectivity of ports is still a novelfield of study, although a significant number of methods of ana-lysis have been used to shed light on the complexity of shippingnetworks. These methodologies can be classified, according totheir scope, into network-related and port-related. Network-re-lated indicators are used as a first approximation of the structureof the network and for descriptive purposes; and they include thenumber of nodes and edges, average path length and networkdensity. Conversely, results from port-related indicators are usefulto compare and highlight differences between individual ports,thus allowing a fine-grained analysis of the network under study.In the present work, both types of indicators were employed toexamine the network, and to assess the connectivity potential ofeach of the ports.

3.2.1. Network-related IndicatorsThe average shortest path length (APL) measures the efficiency

of the routes that connect any two nodes in the network. Thesmaller the APL is, the more likely it is that the two nodes can belinked through a small number of steps. Network density indicateshow saturated the network is, in terms of edges and nodes. It is

given by dividing the actual number of edges by the maximumpossible number of edges. The diameter is the longest distancebetween any two nodes in the network, and is a measure of net-work compactness. The network average clustering coefficient(ACC; Watts and Strogatz, 1998) denotes the presence of regionswith a high edge density among its members; i.e. clusters.

With regard to the global structure of the network, a number ofauthors also analyze the presence or absence of the scale-free andsmall-world phenomena. The most notable characteristic of ascale-free network is the presence of hubs; i.e. a few nodes thatexhibit a great connectivity potential compared with the rest ofthe elements in the network. Numerous works (e.g. Hu and Zhu,2009; Ducruet et al., 2010a; see Table 1) study these types ofnetworks that present a power law relationship within the degree(i.e. number of connections of each port), and usually correspondsto a GDL graph representation. On the other hand, small-worldnetworks (Watts and Strogatz, 1998) are uniformly distributednetworks. The nodes have about the same number of connections,and are ubiquitous in many real-world applications; for example,social networks and road maps. Networks with small average pathlength (APL) and high clustering coefficient (ACC) values are ex-pected to exhibit small-world behavior. Note that this effectusually appears when a GAL network is set up, due to the increaseof network density.

3.2.2. Port-related indicatorsAmong the variety of port-related measures, which assess the

connectivity of individual ports, we focus particularly on degree,betweenness and the port accessibility index; these are com-plementary measures that characterize different aspects of thenode connectivity, as is described in Table 4. They offer us a morecomplete picture than they give when used in isolation.

The maritime degree represents the number of connectionsthat a port has, and has become a standard approach for mea-suring the connectivity potential of every node in the network, as

B. Tovar et al. / Transport Policy 38 (2015) 40–5146

it is strongly correlated to the port throughput in TEUs. Degree canbe formalized for a port i as

Deg iA A

( )2j

ij ji∑=+

where Aij is the adjacency matrix, in which A 1ij = if the port i isconnected to port j, and 0 otherwise.

The betweenness centrality quantifies the prominence of anode, in terms of connectivity within a network (Freeman, 1977;Brandes, 2001), by computing how frequently a node lies on theshortest path between any other two nodes. It is given by

C vv

( )( )

Bs v t V

st

st∑ σ

σ=

≠ ≠ ∈

where stσ is the number of minimum length paths connectingnodes s V∈ and t V∈ , and v( )stσ is the number of such pathswhich some v V∈ lies on. Ports with high levels of betweennessare strategically placed close to the main lines of maritime trans-port, and therefore they are in a privileged, central position incomparison with the rest of their peers.

Accessibility and connectivity are key elements in transportnetworks, since they measure the capacity either of a node to bereached or the capacity to reach other nodes. However, the lit-erature has been somewhat restricted to topological-based indiceslike betweenness,3 and thus fails to take advantage of factors suchas TEUs in the network analysis. Cullinane and Wang (2009) de-veloped a measure of economic accessibility based on the max-imum transportation capacity (TEUs/day) between ports, in orderto assess the nodal accessibility of each element in a shippingnetwork. They define the daily maximum transportation capacitybetween each pair of ports using a weighted graph (valued graphconnection matrix, L), which is calculated as follows:

L V TcF

365ijk

kijk∑= = ×

where Tckij is the average vessel capacity in TEUs and Fk is theannual frequency of the liner service k between ports i and j. If weare to consider not only direct connections between nodes, butalso indirect paths of length n, the valued accessibility matrix willbe given by

T L L L Ln2 3= + + + … +

The value n should be equal to the diameter of the network (thedistance between the farthest nodes-3 in the present case). Finally,the sum across the rows of T corresponds to the desired indicator.Thus, accessibility becomes a measure of port competitiveness interms of container movements or, in other words, how attractive aport is as a cargo hub.

Finally, the vulnerability of a port is defined as the share of itstotal traffic that is shared through its most relevant link. Thismeasure is intended to identify ports that depend on a few con-nections, and to reveal hub–subsidiary relationships betweenports. Ports with high levels of degree are expected to be lessvulnerable, and due to their diversification are able to maintaintheir status in the shipping network.

3 While the use of weighted graphs is widespread in the literature (see Ta-ble 1), the actual computation of accessibility/centrality measures has in manycases been based on unweighted algorithms. For example, the popular tools Gephi(e.g. Gonzalez et al., 2012; Pais et al., 2012) and Tulip (e.g. Ducruet et al. 2010a,2010b, 2010c; Ducruet and Notteboom, 2012) have their implementations of be-tweenness centrality based on Brandes' algorithm, a very fast method for the cal-culation of unweighted betweenness (Brandes, 2001). These tools are open sourcesoftware, and their source code is freely available on the internet.

4. Results and discussion

4.1. Analysis of the Canarian main ports' network

In line with most studies on transport networks, we representour maritime network as a graph. In this particular case, un-directed and TEUs-weighted edges are the preferred strategy, inorder to represent the one-to-one relationship between each pairof ports. We obtained a network with n 53= nodes and m¼625edges (network density¼0.45). Some interesting findings can berevealed, if we take a closer look at the resulting network; thepresence of the small-world phenomena being the most notice-able. In this case, an extremely low average path length (1.58) andnetwork diameter (3) are found to be the case. Most nodes in thenetwork are directly connected or share a common hub, while anynode can be accessed in a maximum of three steps. Also, the valueof the network average clustering coefficient (ACC¼0.71) indicatesthat our network is highly clustered. Additionally, the degreedistribution of the sample seems to follow a power law(y x141.25 1.73= − , R 0.822 = ), which is one of the representativefeatures of the so-calledscale-free networks. As a result of thisevidence, we claim that our maritime network is small-world andscale-free. Similar conclusions are obtained by, among others,Guimerà et al. (2005) and Kaluza et al. (2010) for their respectivepapers on worldwide air transportation network and the globalcargo shipping network.

Table 5 presents the results on degree, betweenness and vul-nerability ranked for each port in our network, from better toworse. Regarding betweenness, Hamburg port along with Shang-hai, Asia's largest gateway, hold joint first positions for both in-dicators, highlighting their statuses as major ports in their re-spective markets. The second position, held by Shanghai, is due tothe bias the sample has towards the European market, which isderived from the fact that we have constructed a Canarian portbased network. This also explains why eight of the top 10 ports areEuropean.

Regarding the Spanish ports, the results are as expected asAlgeciras, Valencia and Barcelona are listed among the top 10 inthe ranking for both indicators. As for the Canarian ports, thedifferences in the ranking between Las Palmas port and Tenerife'sport are noticeable; Las Palmas' port has an important number ofdirect connections (degree) and is ranked 24th in the betweennessassessment. Tenerife port holds the lowest positions in the rankingfor both indicators.

Fig. 3 lets us visualize the main Canarian port shipping networkthrough two indicators, degree centrality and betweenness cen-trality. This figure was created using Gephi software and a “force-directed” algorithm to distribute nodes. Force-directed algorithmsare used to generate a compact display of a network whereby themost relevant nodes are located in the center of the graph, whilethe less important ones are left on the periphery. The relativeposition of each port related to degree centrality and betweennesscentrality is represented by a hierarchy of size and a color scale,respectively. As expected, nodes like Shanghai, Hamburg andRotterdam are the best connected in the network. On the secondlevel, as expected again, we found important European ports, suchas Le Havre, Algeciras, Valencia and Barcelona due to their pivotalhubs and/or gateway roles; New York was also included here. Oneinteresting case to be considered is Jebel Ali; this port, rankedamong the top 10 ports in terms of container throughput world-wide (CI-Yearbook 2011). It is the major port on the Rotterdam–

Shanghai route, but it has a low ranking in terms of connectedports. Hence, Jebel Ali port is assessed high in terms of between-ness centrality with a lower degree. Regarding the Canarian ports,when analyzing the structure of a maritime network based on the

Table 5Port-based measures 2012.Source: Own elaboration.

Port Degree Port Betweenness Port Vulnerability

Hamburg 45 Hamburg 1 Lisbon 0.116Shanghai 44 Shanghai 0.919 Algeciras 0.117Le Havre 41 Jebel Ali 0.765 Manzanillo-Pan 0.121Rotterdam 41 Rotterdam 0.666 Lome 0.133Algeciras 40 Le Havre 0.559 Douala 0.133New York 38 Valencia 0.513 Las Palmas 0.135Valencia 38 Algeciras 0.508 Valencia 0.141Lisbon 36 New York 0.488 Tangier 0.146Tangier 36 Barcelona 0.484 Kingston 0.156Barcelona 35 Tangier 0.374 Ngqura 0.163Genoa 34 Lisbon 0.340 Shanghai 0.166Miami 32 Durban 0.336 Miami 0.168Singapore 32 Genoa 0.326 Tema 0.168Santos 31 Miami 0.312 Rotterdam 0.171Bremerhaven 30 Cape Town 0.302 Guayaquil 0.173Felixstowe 30 Abidjan 0.239 Felixstowe 0.173Manzanillo-Pan 30 Manzanillo-Pan 0.237 Le Havre 0.182Durban 29 Santos 0.210 Hamburg 0.190Rio de Janeiro 29 Felixstowe 0.205 Freeport 0.197Las Palmas 28 Bremerhaven 0.203 Bremerhaven 0.199Cape Town 27 Singapore 0.185 Montreal 0.204Jebel Ali 27 Rio de Janeiro 0.171 Abidjan 0.207Cartagena 26 Cartagena 0.139 New York 0.208Abidjan 23 Las Palmas 0.115 Genoa 0.213Dakar 23 Los Angeles 0.093 Montevideo 0.219Lome 23 Freeport 0.078 Walvis Bay 0.220Buenos Aires 22 Kingston 0.072 La Guaira 0.226Los Angeles 22 Lome 0.059 Cape Town 0.230Montevideo 22 Callao 0.053 Santos 0.232Freeport 21 Dakar 0.048 Cartagena 0.233Tema 21 Valparaiso 0.040 Port Said 0.240Callao 19 Manzanillo-Mex 0.037 Dakar 0.245Kingston 19 La Spezia 0.036 Barcelona 0.248Douala 18 Tema 0.035 Southampton 0.254La Spezia 17 La Guaira 0.034 Puerto Cabello 0.258Ngqura 17 Leixoes 0.025 San Antonio 0.271Valparaiso 17 Douala 0.024 Rio de Janeiro 0.272Leixoes 16 Port of Spain 0.024 Buenos Aires 0.273Port Said 16 Buenos Aires 0.023 Durban 0.288San Antonio 15 Montevideo 0.023 Port of Spain 0.290Walvis Bay 15 Port Said 0.023 La Spezia 0.297Guayaquil 14 Walvis Bay 0.020 Manzanillo-Mex 0.302Manzanillo-Mex 14 Montreal 0.018 Colon 0.306Port of Spain 14 Guayaquil 0.018 Callao 0.309La Guaira 13 San Antonio 0.018 Valparaiso 0.332Southampton 13 Colon 0.017 Tampico 0.333Montreal 12 Ngqura 0.015 Singapore 0.367Colon 11 Southampton 0.011 Leixoes 0.417Puerto Cabello 11 Puerto Cabello 0.011 Damietta 0.468Damietta 10 Damietta 0.002 Jebel Ali 0.484Tenerife 9 Tenerife 0.001 Tenerife 0.542Tampico 3 Port Sudan 0.000 Los Angeles 0.574Port Sudan 1 Tampico 0.000 Port Sudan 1

B. Tovar et al. / Transport Policy 38 (2015) 40–51 47

Atlantic case Ducruet et al. (2010c) state “… other clusters arecharacterized by very local cycles such as the Canary cluster, al-though the latter have a peculiar structure as they connect vir-tually all Atlantic regions despite their smaller size”. Fig. 3 showsthe important differences between both ports, in terms of degreeand betweenness. Las Palmas is much better connected than Te-nerife, and is better positioned to compete for transit cargo thansome of its potential competitors; e.g. Dakar. Moreover, Ducruetet al. (2010a) hypothesize that hub ports will have both a highdegree and betweenness due to their role as inter-regional pivotsin the global network. Fig. 3 seems to confirm that hypothesis,showing that the main hubs in our network present thischaracteristic.

Maritime shipping services have a degree of geographicalflexibility because shipping lines can modify their routes according

to internal circumstances, such as strategic alliances and financialcapacities, and to external factors related to maybe technologicalimprovement or economic downturn. As a consequence, maritimenetworks change as time passes, and also a port's position in thenetwork may change. So, another important aspect to analyze isthe vulnerability of a port's position to changes in the network.Table 5 also shows a port's vulnerability calculated as the share ofthe largest flow link within the total port traffic. Regarding Ca-narian ports, Las Palmas' port has a better position than Tenerife'sport because its traffic is better distributed. On the other hand,Tenerife's dominant link is Las Palmas' port (54.2%) and 91% of itstraffic is concentrated on only five links, four of them located inthe Iberian Peninsula. Finally, Fig. 4 shows the relationship be-tween vulnerability and degree. As expected, the higher the de-gree the lower the vulnerability. That is, ports with a higher degree

Fig. 3. Graph visualization of the main Canarian port shipping network 2012.Source: Own elaboration using Gephi software and a “force-directed” algorithm.

Fig. 4. Port vulnerability versus degree 2012.Source: Own elaboration.

Table 6Port accessibility index 2012.Source: Own elaboration (container throughput data from Alphaliner database).

Rank Port Port accessibility index TEUs throughoutput 2012

1 Shanghai 1 67,822,0922 Singapore 0.9287 30,727,8213 Rotterdam 0.6414 28,920,9234 Hamburg 0.6198 26,030,7845 Jebel Ali 0.5987 17,216,5116 Felixstowe 0.5079 20,293,1297 Le Havre 0.4724 23,512,2568 Bremerhaven 0.4710 21,550,9509 Valencia 0.3387 19,191,757… … … …11 New York 0.2694 12,618,667… … … …20 Santos 0.1551 10,594,720… … … …22 Manzanillo-Pan 0.1321 9,473,682… … … …34 Las Palmas 0.0440 3,284,804… … … …51 Tenerife 0.0023 326,23952 Port Sudan 0.0017 118,24953 Tampico 0.0001 13,345

B. Tovar et al. / Transport Policy 38 (2015) 40–5148

have their traffic more evenly distributed among their links thanthe others. In spite of that, there are some exceptions like Jebel Aliand Singapore, which have a great dependency on Shanghai: 48%and 37% of their total traffic, respectively. The relationship be-tween both indicators again shows the weaker position of Tener-ife's port when compared to Las Palmas' port.

Results also provide information regarding economic accessi-bility which is measured by means of the port accessibility indexdefined in Section 4.2. Table 6 below presents the port accessibilityresults.4 The ranking is led by the two major Asian ports, Shanghaiand Singapore; next are Rotterdam and Hamburg, the two most

4 We remark that the PAI, unlike degree and betweenness takes into accountweighted links (TEUs).

important European ports in terms of container throughput. Va-lencia, the first Spanish port ranked in the list, is located amongthe top 10, whereas Las Palmas' port is ranked 31st and Tenerife'sport is again among the lowest rated ports in the sample. Table 6also shows the container throughput in TEUs for each of the portsin 2012; by including this information, the relationship betweenthroughput and the positions that ports hold in the index iscompared. As Cullinane and Wang (2009) state, the accessibilityranking ports hold should be generally consistent with their an-nual throughput rankings, and those ranked higher in terms ofthroughput can be considered to be more competitive.

The ports in the sample are ranked, with some exceptions,

B. Tovar et al. / Transport Policy 38 (2015) 40–51 49

according to what Cullinane and Wang (2009) state. Shanghai andSingapore are both jointly located in first position in the Port Ac-cessibility Index presented above, and are the ports with thehighest level of throughput; those with lowest throughput areranked at the bottom. A consultation of the Canarian ports showsthe prevalence of Las Palmas' port. It is important to rememberthat the sample is biased in favor of the European area. Those portsin our sample, which should be positioned in the top of theranking, are affected by this fact; this explains why six out of thetop nine ports are Europeans.

4.2. Policy implications

Previous analysis indicates that, at present, the Las Palmas portis the only regional hub in the Canaries. Since different port au-thorities manage the Canary Islands' main ports, they could betempted to compete for transshipment cargoes.

To attend to transit cargoes a port must have suitable facilitiesand infrastructure, which imply huge investments. Moreover, thedemand which port terminals face is characterized by a degree ofuncertainty that varies depending on the traffic mix. Rodríguez-Álvarez et al. (2011) and Tovar and Wall (2012) have analyzed thisphenomenon using data on three5 port terminals located in LasPalmas' port, and concluded that the terminal whose volume oftransshipment is the most considerable faces greater demandvariability than the other two. They concluded that demandvariability affects firm input choices and costs. Tovar and Wall(2014) reached the same conclusion using a panel data set of 26Spanish port authorities observed over the period 1993–2007.

As load centers, whose competitive strategy is based largely onan intermediary role, hubs ports are in an unstable and delicateposition. This is because, once infrastructure has been built, it hasno alternative use; and it is also because this kind of traffic is veryvolatile and totally depends on the strategies of shippers withregard to their service. Likewise, inability within a region to gen-erate sufficient transshipment traffic will increase rivalry foravailable business. Rivalry for transshipment business is intense,even for ports that have established leading positions as loadcenters. These rivalries create substantial pressure on transship-ment pricing. Strong competition and fear of underutilization ofterminal infrastructure and equipment could exert downwardpressure on tariffs and port terminals. Carriers use their bargainingpower with the aim of achieving significant discounts on portdues.

Until 2003, Spanish ports were subject to legislation that leftthem with no control over pricing policies. Rodríguez-Álvarez andTovar (2012) investigated the efficiency of Spanish port authoritiesover the period 1993–2007 and found evidence of inefficient be-havior, with inefficiency worsening from 2003. Overinvestment incapacity is likely to be the main driver of this result. As a con-sequence of not having control over pricing policies, the compe-titive strategy of ports rested on the manipulation of those vari-ables under their control, the most relevant of which was invest-ment policy. This led to a massive increase in infrastructural in-vestment during the period 1993–2008, resulting in significantexpansions of port capacity over this period (Garcia et al., 2010).Following Castillo-Manzano and Asensio-Flores (2012) this processwas for two reasons eased. First, because the port system suc-ceeded in attracting new sources of funding: the European Re-gional Development Fund, revenue from the management of realestate and the greater involvement of the private sector in the

5 As we have seen in Section 2, currently there are two terminals in Las Palmas'port, although until 2010 there were three. In this year, one of them stopped op-erating due to financial problems.

owning and management of the infrastructure. Secondly, becausethe port system succeeded in spectacularly increasing the amountof debt that the Spanish port authorities still owe to financialinstitutions.

Recent legislation has introduced important changes, and the33/2010 Law and 2/2011 Decree have endowed port authoritieswith increased autonomy and control over policy (Tovar and Wall,2014). In their recent paper, Castillo-Manzano and Asensio-Flores(2012) talked about Iberian Peninsula ports, and warned of theexistence of a potential risk for port tariff price wars. They point tothe lack of cooperation between Spain and Portugal in the field oftransport, the huge capacity of many ports and the greater free-dom of Spanish port authorities to set tariffs as factors underlyingthis situation. They continue stating, “This war will lead to a furtherdrag on many ports’ economic results, especially the medium- andsmall-sized ports that, generally-speaking have smaller margins forlowering their tariffs. In this context, and despite all the ruthlesscompetition to corner the international container transshipmentmarket, there might be the possibility, and the need, for a greatercooperation”.

The previous discussion offers important lessons; because itshows a scenario similar to what the Canarian port authoritiescould suffer if they do not cooperate. The question then arises as towhat extent the two Canarian port authorities can differentiatethemselves. For example, they might specialize in certain valuedadded services and increase traffic in these services, in order toavoid the danger of a destructive competition to attract transittraffic. As Wilmsmeier and Monios (2013) state “…a small shift inthe traffic to another port may undermine the viability of the portbusiness”.

It is well known that maritime networks are dynamic in char-acter. Due to the high dependence of the Canarian economy on seatransport, and the well-known relation between connectivity andtransport costs, both port authorities should be proactive inmaintaining and improving Las Palmas' and Tenerife's port con-nectivity. To do this, both port authorities should take advantage ofthe existing instruments that the new port legislation has madeavailable, in order to create a favorable climate to attract privateinitiatives. These initiatives would serve for the provision of portservices and attract private investment in port infrastructure.However, it would avoid incurring an infrastructure overcapacityfor transit cargo that could place the Canary port system in a verycomplicated financial situation.

One of the main instruments available for ports relies on thepossibility of tailoring their policies to differentiate themselvesaccording to their needs. In fact, the new Act 33/2010 creates aflexible pricing model, so that each port authority is able to adaptto the economic reality at any moment. This law could contributeto facilitating the necessary coordination between both port au-thorities because as a recent paper (Yip et al., 2014) suggests thepreferences of port authorities over inter-port competition is in-fluenced by factors such as service differentiation among com-peting ports, market size and service quality differentials.

5. Conclusion

After performing a qualitative approximation of site and si-tuation factors of the main Canarian ports, it is certain that the LasPalmas port transshipment feature differs from the ongoing realityat the Tenerife port. The latter merely focuses its container trafficon the domestic market, and its transshipment record has noteven reached 5% of the containers handled yearly in the periodunder consideration. Meanwhile transshipment operations in LasPalmas' port have reached a rate close to 69% of the total numberof containers handled.

B. Tovar et al. / Transport Policy 38 (2015) 40–5150

For quantitative approximations, we have utilized graph theory.Both network-related and port-related indicators were employed,in order to examine the network as a whole and to assess theconnectivity potential of each of the ports. Previous analysis in-dicates that Las Palmas' port is the only regional hub in the Can-aries, but its transshipment role in the network is outperformed byits main competitors in terms of transshipment traffic. Tangiers,Algeciras and Valencia are in a better position than Las Palmas'port.

Recent literature acknowledges that inter-port coordinationcan be seen as a strategic response to a competitive environment(e. g. Castillo-Manzano and Asensio-Flores, 2012; Asgari et al.,2013). This generates a whole that is larger than the sum of itsparts, and improves information transfer, supply chain integrationand the effective allocation of infrastructure (Merk, 2013). For thisreason, and once the Canary Islands enjoy good port connectivity,the Canarian government should encourage inter-port coordina-tion between both the Canarian port authorities.

Along these lines, the Basque government recently put forwarda proposal for a possible cooperation protocol among the Bilbao,Pasaia and Bermeo port authorities. The main objective of thisagreement is to establish a framework for collaboration that wouldcontribute “to improving the competitiveness of the Basque portsystem and further develop and utilize their potential”. This ob-jective is to be pursued by coordinating actions to improve ports,by facilitating joint optimization strategies, by exploiting synergiesto streamline flows and by promoting collaborative managementto make it more efficient and sustainable.

The Basque Competition Authority has issued a negative reportin relation to the latter proposal. It notes that this “may violate therules of competition with varying degrees of intensity in thoseareas that affect the commercial or economic work of the ports”.While we recognize that it is necessary to create market condi-tions that do not violate competition rules, it is also important toavoid destructive competition for transshipment cargo betweenport authorities. This is especially true if they are providing ser-vices to an archipelago or small set of islands, where one of themhas already reached the necessary maritime connectivity. We be-lieve that the present study could shed light on the previous dis-cussion, and contribute to the recent literature defending inter-port coordination as a strategic response to a competitiveenvironment.

From the methodological viewpoint, future research shouldfocus on using variants of the standard techniques (e.g. Rodríguez-Déniz et al., 2013), in order to take advantage of as many variablesas possible (vessel frequency, TEUs and geographical distance).

It could be also interesting to focus on building different sub-networks based on the most important competitors that the Ca-narian ports have (e.g. Tangiers, Algeciras). This would shed lighton the positions of the Las Palmas and Tenerife ports within thosesubnetworks, and would have the aim of defining strategies thatcould help to improve their competitiveness in transshipmenttraffic in the region. Both port authorities should be proactive inmaintaining and improving port connectivity at Las Palmas andTenerife.

Acknowledgements

We wish to acknowledge that a temporary trial version of theAlphaliner website database, with full access to the informationneeded to carrying out this research, was provided by Alphalinercommercial. This research was partially funded with GrantHAR2010-17408 fromMinisterio de Ciencia e Innovación (Spain). APrevious version of this paper has been presented in the

International Association of Maritime Economists Annual Con-ference, IAME2013, celebrated in Marseille, France.

References

Asgari, N., Zanjirani Farahani, R., Goh, M., 2013. Network design approach for hub ports-shipping companies competition and cooperation. Transp. Res. A 48 (2013), 1–18.

Angeloudis, P., Bichou, K., Bell, M.G.H., 2007. Security and reliability of the linercontainer-shipping network: analysis of robustness using a complex networkFramework. In: Bichou, Bell, Evans (Eds.), Risk Management in Port Operations,Logistics and Supply-Chain Security. Informa Maritime & Transport, London,ISBN: 9781843116554.

Bichou, K., 2004. The ISPS code and the cost of port compliance: an initial logisticsand supply chain framework for port security assessment and management.Marit. Econ. Logist. 6 (4), 322–348.

Brandes, U., 2001. A faster algorithm for betweenness centrality. J. Math. Sociol. 25(2), 163–177.

CI-Yearbook, 2011. Containerisation International Yearbook 2011. Informa, UK.Castillo-Manzano, J.I., Asensio-Flores, J.P., 2012. Competition between New Port

Governance Models on the Iberian Peninsula. Transp. Rev. 32, 519–537.Cullinane, K., Wang, Y., 2009. A capacity-based measure of container port facility.

Int. J. Logist. Res. Appl. 12 (2), 103–117.Cullinane, K., Wang, Y., 2012. The hierarchical configuration of the container port

industry: an application of multiple linkage analysis. Marit. Policy Manag. 39(2), 169–187.

Ducruet, C., Notteboom T., Banos A. , Ietri D., Rozenblat C., 2010a. Structure anddynamics of liner shipping networks. In: Proceedings of the 2010 AnnualConference of the International Association of Maritime Economists (IAME2010), Lisbon.

Ducruet, C., Lee, S.-W., Adolf, K.Y., 2010b. Centrality and vulnerability in linershipping networks: revisiting the Northeast Asian port hierarchy. Marit. PolicyManag. 37 (1), 17–36.

Ducruet, C., Rozenblat, C., Zaidi, F., 2010c. Ports in multi-level maritime networks:evidence from the Atlantic (1996–2006). J. Transp. Geogr. 18 (4), 508–518.

Ducruet, C., Notteboom, T., 2012. The worldwide maritime network of containershipping: spatial structure and regional dynamics. Glob. Netw. 12 (3), 277–423.

Ducruet, C., Zaidi, F., 2012. Maritime constellations: a complex network approach toshipping and ports. Marit. Policy Manag. 39 (2), 151–168.

Duval, Y., Utoktham, C., 2011. Trade facilitation in Asia and the Pacific: which po-licies and measures affect trade cost the most?. Asia-Pacific Research andTraining Network on Trade Working Paper Series, No. 94, January 2011.

Fleming, D.K., Hayuth, Y., 1994. Spatial characteristics of transportation hubs:centrality and intermediacy. J. Transp. Geogr. 2 (1), 3–18.

Freeman, L.C., 1977. A set of measures of centrality based on betweenness. Socio-metry 40 (1), 35–41.

Freire, M., Pais, C., 2011. El Puerto Exterior en Ferrol y la Cuarta Revolución en elTráfico Contenerizado. Chapter 5: Estructura del Tráfico de Contenedores. EntePúblico de Puertos del Estado, Madrid.

Garcia, L., Cherchi, E., Ortuzar, J.D., Pinto, J.A., 2010. Distribución interportuaria delos contenedores: análisis del caso español. In: Proceedings of the XVII Con-greso Panamericano de Ingeniería de Tránsito, Transporte y Logística, Lisboa.

Garrison, W.L., Marble, D.F., 1960. Connectivity of the inter-state highway system.Pap. Reg. Sci. Assoc. 6, 121–137.

Gonzalez, F., Freire, M., Pais, C., 2012. Maritime degree, centrality and vulnerability:port hierarchies and emerging areas in containerized transport (2008–2010). J.Transp. Geogr. 24, 33–44.

Guimerà, R., Mossa, S., Turtschi, A., Amaral, L.A.N., 2005. The worldwide air trans-portation network: anomalous centrality, community structure, and cities'global roles. Proc. Natl. Acad. Sci. USA 102 (22), 7794–7799.

Hu, Y., Zhu, D., 2009. Empirical analysis of the worldwide maritime transportationnetwork. Physica A 388 (10), 2061–2071.

Kaluza, P., Kölzsch, A., Gastner, M., Blasius, B., 2010. The complex network of globalcargo ship movements. J. R. Soc. Interface 7 (48), 1093–1103.

Kansky, K.J., 1963. Structure of Transportation Networks: Relationship betweenNetwork Geometry and Regional Characteristics. University of Chicago De-partment of Geography. University of Chicago, Chicago (Research Paper No. 84).

Kronbak, J., Cullinane, K., 2011. Captive and contestable port hinterlands: modellingand visualization using GIS. In: International Handbook of Maritime Economics.Edward Elgar Publishing, Cheltenham, pp. 362–384.

Kumar, S., Hoffmann, J., 2002. Globalization, the maritime Nexus. In: Grammenos,C.Th (Ed.), Handbook of Maritime Economics and Business. LLP, London.

Laird, J.J., Nellthorp, J., Mackie, P.J., 2005. Network effects and total economic impactin transport appraisal. Transp. Policy 12, 537–544.

Lee, C.B., Wan, J., Shi, W., Li, K., 2014. A cross-country study of competitiveness ofthe shipping industry. Transp. Policy 35, 366–376.

Low, J.M.W., Lam, S.W., Tang, L.C., 2009. Assessment of hub status among Asianports from a network perspective. Transp. Res. A 43 (6), 593–606.

Márquez-Ramos, L., Martínez-Zarzoso, I., Pérez-García, E., Wilmsmeier, G., 2011.“Special issue on Latin–American research” maritime networks, servicesstructure and maritime trade. Netw. Spat. Econ. 11 (3), 555–576.

Martinez I., Hoffmann J., 2007. Costes de Transporte y Conectividad en el ComercioInternacional entre la Unión Europea y Latinoamérica, Comercio Internacional yCostes de Transporte ICE 834, Enero-Febrero 2007.

B. Tovar et al. / Transport Policy 38 (2015) 40–51 51

McCalla, R., 2008. Container transhipment at Kingston, Jamaica. J. Transp. Geogr. 16(3), 182–190.

McCalla, R., Slack, B., Comtois, C., 2005. The Caribbean basin: adjusting to globaltrends in containerization. Marit. Policy Manag. 32 (3), 245–261.

Merk, O., 2013. In: The Competitiveness of Global Port Cities: Synthesis Report.OECD. http://www.oecd.org/regional/portcities.

Notteboom, T., 2004. A carrier's perspective on container network configuration atsea and on land. J. Int. Logist. Trade 1 (2), 65–87.

Pais, C., Gonzalez, F., Freire, M., 2012. General cargo and containership emergentroutes: a complex networks description. Transp. Policy 24, 126–140.

Rodrigue, J.P., Comtois, C., Slack, B., 2006. The Geography of Transport Systems.Routledge, New York, NY.

Rodríguez-Álvarez, A., Tovar, B., 2012. Have Spanish port sector reforms during thelast two decades been successful? A cost frontier approach. Transp. Policy 27,73–82.

Rodríguez-Álvarez, A., Tovar, B., Wall, A., 2011. The effect of demand uncertainty onport terminal costs. J. Transp. Econ. Policy 45 (2), 303–328.

Rodríguez-Déniz, H., Suau-Sanchez, P., Voltes-Dorta, A., 2013. Classifying airportsaccording to their hub dimensions: an application to the US domestic network.J. Transp. Geogr. 33, 188–195.

Tovar, B., Wall, A., 2012. Economies of scale and scope in service firms with demanduncertainty: an application to a Spanish port. Marit. Econ. Logist. 14, 362–385.

Tovar, B., Wall, A., 2014. The impact of demand uncertainty on port infrastructurecosts: useful information for regulators? Transp. Policy 33, 176–183.

Verhetsel, A., Sel, S., 2009. World maritime cities: from which cities do containershipping companies make decisions? Transp. Policy 16, 240–250.

Wang, Y., Cullinane, K., 2008. Measuring container port accessibility: an applicationof the Principal Eigenvector Method (PEM). Marit. Econ. Logist. 10, 75–89.

Wang Y., Cullinane K., 2011. Network flow analysis of freight traffic consolidation inthe container port industry. In: Proceedings of the 2011 Annual Conference ofthe International Association of Maritime Economists (IAME 2011), Santiago.

Wang, Y., Cullinane, K., 2014. Traffic consolidation in East Asian container ports: anetwork flow analysis. Transp. Res. A 61, 152–163.

Watts, D.J., Strogatz, S., 1998. Collective dynamics of ‘small-world’ networks. Nature393 (6684), 440–442.

Wilmsmeier, G., Hoffmann, J., Sánchez, R., 2006. The impact of port characteristicson international maritime transport costs. Res. Transp. Econ. 16, 117–140.

Wilmsmeier, G., Monios, J., 2013. Counterbalancing peripherality and concentra-tion: an analysis of the UK container port system. Marit. Policy Manag. 40 (2),116–132.

Yeo, G.-T., Roe, M., Dinwoodie, J., 2008. Evaluating the competitiveness of containerports in Korea and China. Transp. Res. A 42 (6), 910–921.

Yeo, G.-T., Roe, M., Dinwoodie, J., 2011. Measuring the competitiveness of containerports: logisticians' perspectives. Eur. J. Mark. 45 (3), 455–470.

Yip, T.L., Liu, J.J., Fu, X., Feng, J., 2014. Modeling the effects of competition on seaportterminal awarding. Transp. Policy 35, 341–349.