An evolution of the Global Container Shipping Network: port connectivity and trading community structure (2011–2017)

Abstract

Port connectivity and trading community structure are two fundamental network characteristics that complement one another in explaining the evolution of maritime transport networks. Although port connectivity has been widely studied in the literature, the investigations on trading community structures are rather limited. To better fill this gap, this paper aims to provide a more complete picture of the Global Container Shipping Network (GCSN)’s evolution, based on our earlier works in MEL. In doing so, the GCSN, representing a snapshot of trade at the end of each quarter, from Q3/2011 to Q3/2017, is first constructed. The connectivity of ports and their respective trading communities are then extracted by the Container Port Connectivity Index and the Louvain algorithm, respectively. With our proposed framework, related players would be able to understand the growth of GCSN, as well as the impacts of maritime occurrences on the network of container shipping. Our computational results indicate that port connectivity and trading community structure gradually evolve according to the economic conditions that change over time and the evolution of GCSN could be well explained by these two explanatory variables. In this regard, ports in East Asia consistently dominate others in terms of both inbound and outbound connectivity, led by Shanghai and other major ports of mainland China. Furthermore, the formation of trading communities largely depends on trading patterns—rather than geographical locations—which is evident from the insolvency and mergers of communities in the North American region right after the expansion of the Panama Canal in 2016.

1 Introduction

International trade is one of the most important global economic activities, whose value topped 30% of the world GDP in 2019—a huge leap from the baseline of 13% in 1970, according to the World Bank. As most of international trade is typically carried by sea—about 80% in terms of trade volume (UNCTAD 2020), or even higher for major economies that rely on maritime transportation for import and export, such as China (Zhang et al. 2018)—maritime transport could rightfully be regarded as the backbone of the global economy. The role of maritime transport, in a global trade perspective, becomes more prominent as international trade flows are expected to increase.

In order to accommodate global economic growth, especially in emerging markets, the maritime industry has undergone a series of technological transformations over the last two decades (Mengqiao et al. 2015; Haralambides 2019). Vessel sizes, for example, have tripled since 2000, to achieve greater economies of scale (Gharehgozli et al. 2017), while port infrastructure has also been enhanced in terms of both efficiency and accessibility (Ducruet et al. 2020). When combining these revolutions with economic conditions that forcefully change from time to time, the structure of the global shipping network—and so its topology—is dynamically evolving. This evolution is of paramount importance not only to researchers who mainly focus on the changes of network topology, but also to practitioners that aim to develop a more sustainable service network with higher coverage and revenues (Jiang et al. 2015; Cheung et al. 2020).

To this end, a wide range of assessment frameworks have been devised to help address the abovementioned issues, many of which are based on the concept of complex network theory, or network science. In such a frame of reference, ports and services provided by liners have formed a network of ocean shipping, and network-based measures have played an important role in explaining the dynamics of the evolving network (Ducruet 2020; Álvarez et al. 2021). Ducruet et al. (2020), for instance, explored the spatio-temporal evolution of the Global Container Shipping Network (GCSN) based on the movement of vessels between ports from 1977 to 2016, by average clustering coefficient, average shortest path length, average eccentricity, and degree distribution. Mengqiao et al. (2015) focused more on the evolution of GCSN, from 2001 to 2012, based on aggregated container traffics at the regional level. The evolution of unweighted maritime transport networks from 1890 to 2000 was also analyzed by Kosowska-Stamirowska et al. (2016). Their results confirm that the GCSN evolves in terms of connectivity over time. East Asia was found to be the most significant maritime network actor, dominating others in terms of trade activities. This is consonant with the development of the maritime industry in this period.

To better gain insights into the evolution of network connectivity at a more granular level, the United Nations Conference on Trade and Development (UNCTAD) has established the Port Liner Shipping Connectivity Index (PLSCI)—one of the multi-faceted port connectivity measures that includes both network topology and economic information, namely port degree and the Liner Shipping Connectivity Index (LSCI)¹—to assess the competitiveness of ports within the GCSN since 2006. According to the reported PLSCI values (publicly available at www.unctadstat.unctad.org/wds), ports in East Asia have experienced a significant connectivity growth, especially those in mainland China, due to the expansion of intra-regional trade: almost all of the 50 best connected port pairs by the LSCI are in Asia (UNCTAD 2020). The PLSCI is also found to be positively correlated with the Container Port Connectivity Index (CPCI), another prominent measure that shares similar characteristics with the PLSCI—but with a more complicated connectivity identification (Jarumaneeroj et al. 2023). More specifically, the connectivity of a port by the CPCI is derived not only from local connectivity to its immediate neighbors (port degree) but also from that of the ‘neighbors-of-neighbors’, which better represents the movement of container vessels from port to port along the service loops (Bartholdi et al. 2016). Accordingly, the best connected ports, as measured by the CPCI, are not necessarily those with the most links but rather the ones with good connections with other well-connected ports.

Compared to the PLSCI, and other port connectivity measures in the literature, the CPCI seems more appropriate, as it produces two separate scores for a port, thus supporting a more detailed analysis in terms of both inbound and outbound connectivity. Besides, each of these scores could be further disaggregated into elements according to the five statistics² comprising the LSCI which, in turn, enables related key players to comprehend why a port has become important and by means of which factors. In this regard, Jarumaneeroj et al. (2023) found that, among the five LSCI components, ‘the largest capacity of ships calling’ was the main contributor for a port to claim higher rating.

Although the CPCI adequately explains economic roles played by each port based on both network topology and economic information, it unfortunately fails to capture the evolving trading patterns that seem to concurrently develop together with port growth. Considering that port connectivity is impacted by the evolving trading patterns and, at the same time, impacting on the arrangement of trading communities (Fugazza and Hoffmann 2017), a proper evolutionary study should therefore be conducted based on both perspectives. Nonetheless, investigations on the latter are rather limited, especially in the weighted shipping networks (Zhang et al. 2018; Kosowska-Stamirowska 2020; Álvarez et al. 2021).

In light of this gap, this paper aims to provide a more complete picture of GCSN evolution, from 2011 to 2017, based on two different information pieces—namely, port connectivity and trading community structure—each complementing the another in explaining the GCSN. In doing so, the GCSN is first defined as a directed weighted network representing a snapshot of trade at the end of each quarter, similar to that of Bartholdi et al. (2016). The connectivity of ports and their respective trading communities at different time periods are then determined by the CPCI and the Louvain algorithm (Blondel et al. 2008), one of the community detection algorithms that gradually creates communities based on the incremental change in modularity value. With our proposed framework, related players would be able to understand the growth of GCSN, as well as the impacts of maritime occurrences on the network of container shipping. This is especially useful for the development of policies—e.g., to properly invest in ports with greater potential or to establish trade agreements/trade zones with major trade partners within and across trading communities—that will help enhance the strategic roles of ports in a more sustainable fashion.

The rest of this paper is organized as follows. A brief review of port connectivity and trading community structure is provided in Sect. 2, followed by a discussion of the evolution of port connectivity and trading communities in Sect. 3. The impacts of major economic phenomena on the evolution of GCSN are then provided in Sect. 4; and, lastly, Sect. 5 concludes all the work and possible future research directions.

2 Port connectivity and trading community structure

2.1 Port connectivity

Port connectivity is a broad concept that concerns the quality of port connections within the network of ocean shipping (Jiang et al. 2015), resembling the identification of influential nodes, or node centrality, in the social science literature (Freeman 1978). In particular, a port with higher connectivity is typically the one with greater capability to access (or being accessed by) other ports within the network (Calatayud et al. 2017). A wide range of contemporary port connectivity measures have been recently proposed—mostly based on the concept of centrality measures (Ducruet 2020), where a port (node) customarily derives its connectivity (centrality) from the following: (i) the number of direct connections to its immediate neighbors (degree centrality), (ii) the average number of connections from such a port to the rest of the network (closeness centrality), (iii) the number of shortest paths between all pairs of ports that passes through such a port (betweenness centrality), and (iv) the centrality of ports to which it connects (eigenvector centrality).

In addition to the four traditional centrality measures mentioned above, researchers have further included other important network characteristics into the development of more informative connectivity measures. Tang et al. (2011), for example, proposed the Network Connectivity Index for the evaluation of port choice behavior, where the connectivity of a port is derived from its attributes and the number of origin–destination pairs the port serves. Bartholdi et al. (2016), on the other hand, introduced the CPCI, a variant of eigenvector centrality measures, which integrates trade information with network topology, for the assessment of inbound and outbound connectivity of container ports worldwide. Jarumaneeroj et al. (2023) later extended this work by decomposing the CPCI into elements according to the components of LSCI: (i) number of liner services calling, (ii) number of liner companies, (iii) number of ships, (iv) combined capacity of ships in TEUs, and (v) the largest capacity of ships calling. With this decomposition approach (see Appendix 1 for the computation of CPCI and its decomposition framework), they successfully identified the determinants of port connectivity, and so explanations for changes in port ranking as a result of major economic phenomena in the maritime industry, including the expansion of Panama Canal in June 2016 and the bankruptcy of Hanjin shipping in February 2017.

It is worth remarking that, apart from connectivity, centrality measures have proven themselves useful as vulnerability measures in the study of ocean shipping networks at both regional and global levels (see Ducruet et al. 2010; Viljoen and Joubert 2016; Wu et al. 2019 for more details). They have also been widely adopted in the evolutionary study of various transportation networks, including air transport networks (Allroggen et al. 2015; Zhang et al. 2017), rail transport networks (Cats 2017; Bangxang and Jarumaneeroj 2018; Yang and Chen 2018; Meng et al. 2020), and foreland transport networks (Martínez-Moya and Feo-Valero 2020).

2.2 Trading community structure

Community structure is another fundamental characteristic of a network that helps provide insights into the interaction of nodes within the network—where a collection of nodes with dense and strong connections among themselves, but with sparser and weaker connections to the others, is typically referred to as a community, or a cluster, in the social science literature (Newman 2004a). It is well known that a community detection problem is an NP-complete problem (Brandes et al. 2008) and the most widely used community detection algorithms are based on the modularity maximization approaches that aim to find the partitions of nodes in such a way that the modularity value is maximized (Zhang et al. 2018).

A prominent example of the modularity optimization approaches is the spectrum-based method (see Jarumaneeroj 2014 for the detailed implementation) that iteratively divides a group of nodes into communities based on the signs of principal eigenvector’s elements (Newman 2004b; Leicht and Newman 2008). With this algorithm, Kaluza et al. (2010) found that the network of ocean shipping resolved into 12 communities. Kölzsch and Blasius (2011), later, revealed two main trading communities within the ocean shipping network, namely the trans-Pacific and the trans-Atlantic, along with nine smaller communities that were scattered across the globe. Unlike Kölzsch and Blasius (2011), Bartholdi et al. (2016) found only eight trading communities, presumably because of different network settings. However, their identified community structure represented well the patterns of global trade, with idiosyncrasies that could be explained from the strategic functions of ports in their respective communities.

While promising, the community structure found by the spectrum-based method might be, however, locally optimal since it is only capable of dividing a group of nodes into at most two subgroups at a time; and, once assigned, the communities of nodes would not be subsequently changed. This might accordingly prevent the improvement of the modularity value in later iterations. To address these issues, Blondel et al. (2008) introduced an interesting algorithm that gradually created communities based on the concept of node-aggregation rather than network partition, called the Louvain algorithm for community detection (see Appendix 2 for the detailed implementation). Technically speaking, the Louvain algorithm first assigns each node a unique community and then iteratively builds a larger community by aggregating a neighbor node from another community in such a way that the modularity value is maximized. Compared to the spectrum-based method, the Louvain method is superior in terms of both algorithmic performance and implementation, as it has fewer restrictions, and we may also place a limit on the number of communities for ease of comparison across timeline (Aynaud et al. 2013).

3 An evolution of the Global Container Shipping Network

3.1 The construction of Global Container Shipping Network

The data used in this research are based on the movement of container vessels between ports, provided by a commercial service provider, namely www.BlueWaterReporting.com, at the end of each quarter since Q3/2011 to Q3/2017. For each of these 25 data sets, the GCSN is constructed—similar to that of Bartholdi et al. (2016)—where each node and link in the GCSN denote a unique container port and trade intensity between a port pair, as computed by the LSCI. These networks are proven to be strongly connected, with about 501 ports and 2727 links on average.

3.2 An evolution of port connectivity

We compute the inbound and outbound connectivity of ports, as measured by the CPCI, over the period of study. We find that the ranking of ports changes dynamically due to the changes in all five LSCI components. Furthermore, the top 10 ports with the highest inbound and outbound connectivity during this time period are consistently located in Asia, with Shanghai ranked highest, followed by other big Chinese ports on the East coast of China, as illustrated in Fig. 1.

Fig. 1

Changes in the ranking of top 10 highest-scoring ports in terms of both inbound and outbound connectivity from Q3/2011 to Q3/2017

Notwithstanding this, the ranking of the port of Hong Kong has declined significantly, from the first place in Q3/2011 to the third and fourth place at the end of Q3/2017, in terms of inbound and outbound connectivity, respectively.³ The decline of Hong Kong is especially interesting as it could be explained by three main economic factors that have affected the attractiveness of Hong Kong over this time period—namely, (i) the growth of nearby mainland China ports that forms a multi-port system with greater competition (Liu et al. 2020; Li et al. 2022), (ii) the relocation of liner shipping companies from the port of Hong Kong to Singapore, and (iii) the air draft restriction at Tsing Ma bridge (Grinter 2017; Mooney 2017).

Prior to the emergence of mainland China ports, Hong Kong was recognized as a gateway to the southeast region of China (Luo et al. 2010). Every product manufactured from producers in this region was normally shipped to Hong Kong and, later, distributed to the rest of the world. This also applied for the materials required by manufacturers that were imported through the port of Hong Kong. The reason was primarily due to the port’s efficiency which, at the time, was far higher than the efficiency of the nearby ports in Shenzhen and Guangzhou. Furthermore, the restrictions on international trade in Hong Kong were also relatively fewer when compared to those of mainland China. However, over the last two decades, maritime transport in China grew significantly, with better port infrastructure and less restrictive international trade laws. According to the Organization for Economic Co-operation and Development (OECD), between 2002 and 2012, the average annual growth rate of ports in Guangzhou and Shenzhen increased notably by 57% and 20%, while that of Hong Kong stagnated at only 2%. Subsequently, Hong Kong gradually lost its key strategic position as a gateway and suffered badly from the loss of throughput to the ever-growing ports in Shenzhen and Guangzhou, as demonstrated in Table 1.

Table 1 Total port throughput, including both loaded and empty TEUs, at highly rated Asian ports from 2014 to 2017 (adapted from the World Shipping Council)

In addition to fierce competition among ports within the same region, the role of Hong Kong port as a global transhipment hub was weakened by the port of Singapore, as many shipping service providers decided to move their Asian-based headquarters from Hong Kong to Singapore. The major French shipping line, CMA CGM, for example, relocated its regional office from the port of Hong Kong to the port of Singapore in early 2017. Later on, the Ocean Network Express (ONE), the newly formed Japanese shipping conglomerate, followed CMA CGM’s move by establishing its global headquarters also in Singapore. The strategic moves of these major players evidently affected not only the momentum of trade flows but also the operations of small service providers, who might soon need to relocate to Singapore due to the loss of transhipment opportunities. Ironically, the most important client of a port—or equivalently the largest carrier calling there—is also the greatest threat to the hub port, as such a company could relocate to a competing port which, in turn, increases economic pressure on smaller liner shipping companies to follow its move (Jarumaneeroj 2014).

Regarding the air draft restriction at Tsing Ma bridge, vessels with a height of more than 53 m—or equivalently those with capacities of more than 18,000 TEUs—cannot pass underneath the bridge. This restriction is especially an issue for larger vessels to call at the ports of Hong Kong and Shekou, as they need to navigate through the Ma Wan channel, where the Tsing Ma bridge is located⁴ (Mooney 2017). As a result, the port of Hong Kong has lost about 400,000 TEUs of cargo in 2014; and it is estimated that the accumulated throughput loss could rise to 4.2 million TEUs by 2030⁵ (Grinter 2017).

It is evident that each of the abovementioned factors has deteriorated the relative prominence of Hong Kong, unlike other highly rated ports, such as Shanghai and Singapore, that seem to consistently improve in terms of LSCI components over time. The largest capacity of ships calling at Hong Kong, for instance, increased slightly from 18,270 in Q3/2014 to 20,568 TEUs at the end of Q3/2017, while the largest capacity of ships calling at Shanghai and Singapore, increased remarkably from 18,270 and 15,908 TEUs in Q3/2014 to 21,413 TEUs at the end of Q3/2017. Similar stagnated growth patterns are also observed in the remaining LSCI components of Hong Kong, which have gradually led to the decline of the once most connected port in the world.

3.3 An evolution of trading community structure

3.3.1 Trading communities

We have applied both the Louvain and the Spectrum-based algorithms to the GCSNs. It turns out that the Louvain algorithm outperforms the other algorithm in terms of modularity value in all of the 25 GCSN snapshots (see Table 2 in Appendix 3). Although the number of communities, together with their respective members, may vary from time to time, a total of eight major trading communities are frequently revealed in these snapshots, including (i) Oceania, (ii) Asia–Pacific and trans-Pacific, (iii) South Asia and Middle East, (iv) Europe and North Africa, (v) East Coast of North America, Caribbean, and Northeast Coast of South America, (vi) West and South Africa, (vii) Pacific Coast of America, and (viii) South-Eastern Latin America, as shown in Fig. 2.

Fig. 2

An illustration of eight major trading communities detected by the Louvain algorithm at the end of Q1/2016

Apart from the eight backbone trading communities, the Louvain algorithm occasionally detects a number of minor trading communities that seem to be geographically localized, such as (i) Japan, (ii) Norway, (iii) Southern Europe and Mediterranean, (iv) West Africa, (v) Indonesia, Australia, and Papua New Guinea, and (v) Micronesia, Marshall Islands, and Palau—presumably because of the evolving trading patterns. A small community of Japan, for example, appears in five out of the 25 GCSNs, while it is included as a part of the Asia–Pacific and trans-Pacific community in the remaining snapshots (see Fig. 3 for the illustration). Further investigations reveal that, during these periods, the internal trade within Japan is relatively denser, and there are more seasonal trade routes that have been added into the GCSNs, such as Toyama-North Coast and Kanazawa-North Coast services.

Fig. 3

The isolation of Japan community from the Asia–Pacific and trans-Pacific community in Q3/2016

A similar explanation could be drawn for the formation of the Southern Europe and Mediterranean community that appears in 11 out of the 25 GCSN snapshots. As illustrated in Fig. 4, this community is well determined by the geography of the Mediterranean Sea, mainly because of denser internal trade at its five communal anchors, namely Valencia, Port Said, Piraeus, Marsaxlokk, and Genoa.

Fig. 4

The isolation of the Southern Europe and Mediterranean community in Q2/2016

3.3.2 Unique communal members

While it is intuitive that ports belonging to the same community should be geographically close to each other, we find that there are some communal members that are peculiarly located far away from the remaining. Figure 5, for instance, shows the relatively dispersed locations of Montreal and Pointe-a-Pitre that have been recognized as a part of the Europe and North Africa community in Q3/2016, despite their geographical locations in the Americas. Likewise, Portsmouth has been included in the East Coast of North America, Caribbean, and Northeast Coast of South America community, despite its geographical location on the South Coast of England. This finding is, notably, not an idiosyncrasy but rather evidence that emphasizes the economic roles played by these ports from a global trade perspective.

Fig. 5

Unique communal members that are peculiarly located in two backbone trading communities in Q3/2016

In the case of Montreal, Canada, the port is recognized as a communal member of the Europe and North Africa community due to its function as a gateway for exporting food and forestry products to European countries via the ports of Lisbon, Sines, Southampton, and Antwerp. The amount of goods traded with these European ports are markedly intense—accounting for about 88% and 76% in terms of import and export, in 2016 (Ortiz-Ospina et al. 2018). Likewise, Pointe-a-Pitre, an overseas department of France in the Caribbean Sea, is considered a communal member of the Europe and North Africa community, despite its geographical location in the Middle American subregion. This is mainly because of its frequent trade connections with France, Morocco, and Spain. The main exported products of Guadeloupe to these European countries are agricultural crops—such as bananas, sugar, and rum—whereas most of the imported goods are heavy machinery and consumer products—accounting for about 37% and 42% in terms of export and import, in 2016, respectively (Ortiz-Ospina et al. 2018). Finally, Portsmouth is assigned to the East Coast of North America, Caribbean, and Northeast Coast of South America community, due largely to its function as a gateway for importing tropical fruits from Central America to Europe.

3.3.3 Linking ports between communities

In each community, there are generally communal anchors that significantly contribute to the modularity value, such as Hong Kong, Shanghai, Busan, Long Beach, and Seattle in the Asia–Pacific and trans-Pacific community, or Rotterdam, Algeciras, and Antwerp in the European community. While these communal anchors are typically highly ranked ports with one dedicated communal assignment, we find that there are also some well-connected ports that belong to more than one trading community, such as Singapore and Port Said of Egypt.

Singapore is one of the major ports that consistently appears in the top 10 highest-scoring ports, as measured by the CPCI ever since Q3/2011. In line with Bartholdi et al. (2016), we find that Singapore is a communal member of the Asia–Pacific and trans-Pacific community in Q3/2011. However, the membership of Singapore seems to lean toward another backbone trading community, namely the South Asia and Middle East community in later periods. To be precise, among the 25 GCSNs, Singapore appears 16 times in the South Asia and Middle East community—almost twice the times it appears in the Asia–Pacific and trans-Pacific community. A more in-depth study reveals that, as a global transhipment hub, Singapore does not have dense connections with other well-connected ports within either region. Unlike other communal anchors that are well connected with other ports within the same community, Singapore connects ports in the service loops that span multiple geographical regions with few direct services to each other. Subsequently, the communal membership of Singapore is defined mostly by its degree and the communities of ports to which it connects. A similar reasoning also holds for Port Said of Egypt that has been recognized as a part of three different trading communities over the 25 GCSN snapshots—three times in the Europe and North Africa community, 16 times in the South Asia and Middle East community, and six times in the geographical localized community in Southern Europe and the Mediterranean Sea.

Apart from trading patterns, these two ports share one interesting similarity, namely their geographical locations that help strengthen their positions as global transit hubs. More specifically, Singapore is situated on the tip of the Malay peninsula, next to the Strait of Malacca—the main channel that connects the Indian Ocean with the Pacific Ocean. Singapore could be, therefore, regarded as one of the ideal transit locations for shipping companies that ply the routes between Asia and Europe. Likewise, Port Said—one of the top 50 highest-scoring ports—serves as one of major transit hubs in most of Asia–Europe services due largely to its geographical location at the northern end of the Suez Canal—the waterway that provides the shortest trade routes between Asia and Europe.

4 Effects of major economic phenomena on the evolution of GCSN

It could be seen that port connectivity and trading community structure gradually evolve according to the economic conditions that change from time to time and the evolution of GCSN could be well explained by these two explanatory variables. In this section, we will explore how major economic phenomena—namely, the bankruptcy of Hanjin shipping and the expansion of Panama Canal—affect port connectivity and trading community structure of the GCSN.

4.1 The bankruptcy of Hanjin shipping

Hanjin shipping was once the seventh-largest shipping service provider before it declared bankruptcy on February 17th, 2017—the largest ever bankruptcy for a shipping line in terms of capacity. During the receivership,⁶ Hanjin’s operations were in a state of great disrepair, as its vessels were not allowed to call at a number of container ports worldwide, including its two main majority-owned terminals in Seattle and Long Beach, due to the uncertainty in the payment of port access fees and risk of liquidation (Su et al. 2019; Song et al. 2019).

While the bankruptcy of Hanjin affected trade intensity of ports at which Hanjin called—and thus their connectivity ratings due to the changes in their LSCI link weights—the carrier’s default seems to have had no impact on the formation of trading communities. In particular, we find that the inbound rating of Long Beach and the outbound rating of Seattle declined greatly during the incidence, from the 21st place and the 53rd place in Q2/2016 to the 31st place and the 77th place in Q4/2016, respectively. However, they are still recognized as communal members of the Asia–Pacific and trans-Pacific community, mainly because of the redundancy of services between themselves and large Asian ports in the same trading community (Su et al. 2019). To be precise, Hanjin was not a sole player that transported goods from Asian-based manufacturers to Long Beach nor from Seattle back to large Asian ports. The collapse of Hanjin therefore affected only the intensity of trade among ports, but not the GCSN structure. Furthermore, as all services originated from Seattle were destined to ports in the Asia–Pacific and trans-Pacific community, as illustrated in Fig. 6, it is less likely that the communal membership of Seattle would change due to the insolvency of Hanjin. Similar to Seattle, it could be seen from Fig. 6 that the majority of services destined to Long Beach originated from ports in the Asia–Pacific and trans-Pacific community, and only about 19.91%, 11.49%, and 14.90% of inbound trade originated from the communal members of other communities in Q2/2016, Q3/2016, and Q4/2016. Since the contribution of these links to the modularity value is relatively small, when compared to that of connections within the Asia–Pacific and trans-Pacific community itself, the communal membership of Long Beach thence remained unchanged.

Fig. 6

Inbound connections to Long Beach and outbound connections from Seattle between Q2/2016 and Q4/2016

4.2 The expansion of Panama Canal

The Panama Canal is one of the most important waterways, providing access between the Pacific Ocean and the Atlantic Ocean since 1914. But, due to the increase in container vessel sizes that were far too large for its earlier infrastructure, the Panama Canal Expansion project was initiated and later completed in May 2016. By adding a new lock with greater width and depth, the canal now accommodates larger vessels with a capacity of up to 14,000 TEUs. Since the beginning of its commercial operations on June 26th, 2016, the expanded canal is currently handling container vessels with nearly the designed size and capacity, accounting for about 14.1 million TEUs in 2017—or equivalently 14% increase in terms of throughput within a year of operations.

The expansion of Panama Canal is clearly a major economic development in the maritime industry that immensely impacted both connectivity of ports and their communal memberships—especially those in the United States and Central American region, due to the changes in trade connections that shifted to the canal after its expansion (Bhadury 2016; Park et al. 2020). In terms of port connectivity, we find that the inbound connectivity ratings of many US West Coast ports have been severely affected from the introduction of direct trade routes between the US and large Asian ports, as all-water routes from the Pacific Ocean to the Atlantic Ocean via the expanded canal are now more attractive when compared to the intermodal routes via the US West Coast ports (Woo et al. 2018). Although the inbound ratings of major US East Coast and Gulf Coast ports tended to rise after the expansion, we find that the outbound connectivity ratings of these ports either stagnated or declined. For example, the inbound connectivity rating of Houston rose from the 185th in Q2/2016 to the 88th in Q3/2016, but its outbound connectivity rating scarcely changed, i.e., from the 245th in Q2/2016 to the 241st in Q3/2016. This asymmetricity in port connectivity rating is notably interesting from the global trade perspective as it indicates the existence of direct trade routes from large Asian ports to the US, via the Coastal ports, that subsequently include some intermediate ports for transhipment before returning—leading to an asymmetrical jump in only the inbound connectivity.

The expansion of Panama Canal also affected the arrangement of trading communities, as illustrated in Fig. 7. In particular, before the expansion of Panama Canal, it could be seen that the US West Coast and the US East Coast ports were separately assigned to different trading communities—where Long Beach and Oakland were part of the Asia–Pacific and trans-Pacific community, while ports in the West Coast of Central America and those in the East Coast of North and Central America belong to either the Pacific Coast of America community or the East Coast of North America, Caribbean, and Northeast Coast of South America community. The Panamanian ports at different canal entrances are also recognized as parts of different communities, where the port of Balboa, at the Pacific entrance, is categorized as a communal member of the Asia–Pacific and trans-Pacific community, whereas the ports of Colon and Manzanillo, at the Atlantic entrance, are assigned to the East Coast of North America, Caribbean, and Northeast Coast of South America community, respectively.

Fig. 7

Trading community structures before and after the expansion of Panama Canal

However, shortly after the expansion, i.e., in Q3/2016, trading community structure in this region changed significantly, where a small community on the Pacific Coast of America was split and largely merged with the East Coast of North America, Caribbean, and Northeast Coast of South America community, due to changes in trading patterns which shifted to the canal. All of the Panamanian ports at both entrances have now become communal members of the same trading communities—although the connectivity of Balboa greatly declined, as more goods are directly shipped through the ports of Colon and Manzanillo, along with those on the US East Coast and Gulf Coast, via the expanded canal.

According to Fig. 7, there are also some ports that merged with the Asia–Pacific and trans-Pacific community, after the insolvency of the Pacific Coast of America community, such as the Mexican port in Ensenada. The reason to this is due to the roles of these Mexican ports as transit points between ports in East Asia and the Panama Canal before the expansion. Since the intensity of trade connections between these Mexican ports and the ports in East Asia intensified, they have now become a lot closer and eventually recognized as a part of the same community.

In addition to the merger with other trading communities, from Fig. 7, it can be seen that there is a newly formed community on the West Coast of South America, which joined the East Coast of North America, Caribbean, and Northeast Coast of South America community in later periods, as shown in Fig. 8. This small trading community seems to be a transitory community that appeared only once during this period, due to an unusual change in global trade caused by a sudden drop in trade intensity between its communal members and those of other backbone communities. Trade intensity of Callao, a communal anchor of the West Coast of South America community, for instance, notably declined in terms of both import and export from Q2/2016 to Q3/2016; but, these figures have bounced back in later periods as container traffic became more stable and steadily shifted to the Panama Canal after Q4/2016.

Fig. 8

The inclusion of the newly formed community on the West Coast of South America in the East Coast of North America, Caribbean, and Northeast Coast of South America community (Q4/2016)

5 Conclusions

Port connectivity and trading community structure are two fundamental characteristics of the GCSN that have recently drawn attention from both academic and practitioner communities. While there is a number of port connectivity measures in the literature, which capture well the GCSN topology, they have failed to recognize the importance of trade, thus leading to inadequate connectivity ratings or misinterpretations of changes in port connectivity over time. Considering that port connectivity is impacted by the evolving trading patterns and, at the same time, it impacts the arrangement of trading communities, the evolutionary study of GCSN should, therefore, be conducted based on both perspectives. Nonetheless, investigations on the latter are rather limited, especially in the weighted shipping network with trade information.

To better fill this gap, we have devised a more complete picture of GCSN evolution based on both pieces of information, where the GCSN is first defined as a weighted network representing a snapshot of trade at the end of each quarter, since from Q3/2011 to Q3/2017. Once the GCSN is constructed, port connectivity and trading communities are then extracted, by the CPCI (Bartholdi et al. 2016) and the Louvain algorithm (Blondel et al. 2008), respectively.

With this evolutionary framework, we find that the connectivity of ports, as well as the formation of trading communities within the GCSN, gradually evolves as a result of major economic revolution over the period of study. For instance, we find that the expansion of Panama Canal has greatly benefited ports on the East Coast of North and Central America, such as Savannah and Houston, providing a better, cost-effective, access from large Asian ports to the North American region without a need of intermodal transportation over the US West Coast ports. However, before returning, these services typically call at a number of transit ports, including the ports of Colon and Manzanillo at the Atlantic entrance of the canal, for transshipment, leading to an asymmetrical connectivity rating—this is similar to the case of Long Beach and Seattle in the trans-Pacific service loops identified by Bartholdi et al. (2016). We also find that the existence of these trading patterns further affects the formation of trading communities in this region, as more container traffic has been channeled through, and circulated within, this region, thanks to the expanded canal.

It may be worth noting that, while our findings are based only on a few major economic phenomena, due to limited data sets, they are all well explained based on both network and global trade perspectives. We expect that, as more information is becoming available, we would be able to systematically analyze the impacts of recent major economic phenomena on the GCSN, including occurrences that impede global economic growth—such as the China–United States trade frictions since 2018 and the spread of novel SARS-CoV-2 in 2020—or the ones that facilitate trade flows across regions, i.e., the China’s Belt and Road Initiative. With this piece of information, we should be able to better comprehend the dynamics of port connectivity and trading community structure, which will be useful not only for academic and practitioner communities but also for the policymakers in describing and evaluating the impacts of such policies in a more systematic fashion.

Appendices

Appendix 1: The computation of CPCI and its decomposition

Let G = (V,E) be the GCSN, where V and E denote a set of container ports and a set of links connecting pairs of ports with trade intensity as computed by the LSCI. The CPCI could be determined by the solution to Eqs. (1) and (2).

$$\lambda x = L^{T} y \Rightarrow \lambda^{{2}} x = L^{T} Lx,$$

(1)

$$\lambda y = Lx \Rightarrow \lambda^{{2}} y = LL^{T} y,$$

(2)

where L, λ, x, and y denote the LSCI matrix, the principal eigenvalue of L, the inbound CPCI vector, and the outbound CPCI vector, respectively.

In the connectivity decomposition approach (Jarumaneeroj et al. 2023), the CPCI is decomposed into components—each reflecting its contribution to the overall port importance score. In doing so, Eqs. (1) and (2) are replaced by Eqs. (3) and (4), as follows.

$$\lambda x = \left( {\sum\limits_{i} {c_{i} } } \right)^{T} y \Rightarrow \lambda^{{2}} x = \, \left( {\sum\limits_{i} {c_{i} } } \right)^{T} \left( {\sum\limits_{i} {c_{i} } } \right)x$$

(3)

$$\lambda y = \left( {\sum\limits_{i} {c_{i} } } \right) x \Rightarrow \lambda^{{2}} y = \, \left( {\sum\limits_{i} {c_{i} } } \right)\left( {\sum\limits_{i} {c_{i} } } \right)^{T} y,$$

(4)

where C_i denotes the ith component of the LSCI in matrix form, i.e., C1 denotes the number of liner services calling, C2 denotes the number of liner companies, C3 denotes the number of ships, C4 denotes the combined capacity of ships in TEUs, and C5 denotes the largest capacity of ships calling.

Appendix 2: The Louvain algorithm for community detection

The Louvain Algorithm is a hierarchical algorithm used to cluster the communities within a network so that the modularity value, defined by Eq. 5, is maximized (Blondel et al. 2008).

$$Q=\frac{1}{W}\sum_{ij}\left({w}_{ij}-\frac{{w}_{i}^{out}{w}_{j}^{in}}{W}\right)\delta \left({c}_{i},{c}_{j}\right),$$

(5)

where ${w}_{ij}$ denotes the weight of an edge connecting nodes $i$ and $j$, ${w}_{i}^{out}=\sum_{j}{w}_{ij}$ and ${w}_{j}^{in}=\sum_{i}{w}_{ij }$ are output strength of node $i$ and input strength of node $j$, respectively, and $W={\sum }_{i}{\sum }_{j}{w}_{ij}$.

Initially, the Louvain algorithm assigns each node a unique community. Then, it will iteratively verify whether moving node $i$ from its current community to that of neighbor node $j$ would help increase the overall modularity value. If the modularity gain from such a move, defined by $\Delta Q$ in Eq. 6, is positive, node $i$ should be moved to a new community; else, it should remain in the same community as before. This process continues in this fashion until no modularity gain is found. And, once terminated, the Louvain algorithm would return a community structure with the greatest modularity value.

$$\Delta Q=\left[\frac{{\sum }_{in} +2{k}_{i,in}}{2m}-{\left(\frac{{\sum }_{tot} +{k}_{i}}{2m}\right)}^{2}\right]-\left[\frac{{\sum }_{in} }{2m}-{\left(\frac{{\sum }_{tot} }{2m}\right)}^{2}-{\left(\frac{{k}_{i}}{2m}\right)}^{2}\right],$$

(6)

where ${\sum }_{in}$ is the sum of all the weights of links inside the community that node $i$ is moving into, ${\sum }_{tot}$ is the sum of all the weights of links to nodes in the community that node $i$ is moving into, ${k}_{i}$ is the weighted degree of node $i$, ${k}_{i,in}$ is the sum of the weights of links between node $i$ and other nodes in the community that node $i$ is moving into, and m is the sum of the weights of all links in the network.

Appendix 3: The modularity values of 25 community structures identified by the Louvain and Spectrum-based algorithms

See Table 2.

Table 2 The modularity values derived from both algorithms