Understanding Port Choice Determinants and Port Hinterlands: Findings from an Empirical Analysis of Spain

Abstract

In the present work, the determinants of port choice regarding container cargoes from specific hinterland regions are analyzed, based on an empirical study of Spain. Previous work has been extended by including novel explanatory variables for the market shares of ports in hinterland locations. Discrete choice theory is the methodological approach used here. More specifically, a nested logit model is proposed. As potential explanatory variables, the model includes maritime connectivity to specific overseas regions and intermodal connectivity of the port to specific hinterland locations. The empirical analysis is based on detailed Spanish customs data. The analysis shows that all variables hypothesized to influence the market share of a port in a specific hinterland region (i.e., road distance to the hinterland region, maritime distance, maritime connectivity of the port, and intermodal connectivity of the port) indeed influence significantly its market share, with the signs as expected. The findings add to the understanding of port competitiveness in specific regions with three conclusions: First, port hinterlands are relational, in the sense that they depend on the overseas origin or destination of the cargo; Second, the analysis suggests that ports that predominantly handle transhipment cargoes may have a “transhipment orientation,” which is an impediment for reaching hinterland markets; Third, intermodal connectivity is a determinant of the market share of a port in a certain hinterland region.

1 Introduction and Literature Review

The analysis of what hinterlands are served by ports has attracted scholarly attention since the nineteenth century. Various approaches have been used to advance in the understanding of port hinterlands. These include conceptual contributions, such as the widely cited work on port regionalization (Notteboom and Rodrigue 2005; Monios and Wilmsmeier 2012), work on dry ports (Roso et al. 2009) and extended gates (Veenstra et al. 2012), as well as numerous case studies (Van den Berg and De Langen 2011; Bask et al. 2014) and modeling of port competition and hinterlands (De Borger et al. 2008; Luo et al. 2012). A fourth important stream of research enriches the understanding of port hinterlands through the analysis of port choice. The present paper is part of this fourth research stream. A comprehensive literature review on port choice is provided by Martínez Moya and Feo Valero (2017). The aim of the work presented herein is to deepen the understanding of factors that may influence port choice and determine the market shares of ports in certain hinterlands. Three factors that have not been analyzed in detail previously are the focus of this study: maritime connectivity, intermodal connectivity, and transhipment orientation.

Shipment data are used (i.e., data that show the origin and destination of container shipments as well as the ports that were used) to test the relevance of factors that influence port choice. This method was selected as detailed shipment data allow precise statistical tests.

The shipment data does not include information about who has made the shipment routing decision. This may be partly because actor roles in port choice are ambiguous, in the sense that there is often no single actor that determines cargo routing; For instance, shippers may outsource the organization of maritime transport to third-party logistics providers but de facto continue to make port choice decisions. Given the absence of data on actors, a behavioral approach is not taken here: no attention is paid to potential differences in port choice decisions between different actors, even though De Langen (2007) and Tongzon (2009) do demonstrate differences in port choice decisions between forwarders and shippers.

Regarding port choice factors, the relevance of distance and shipping costs is well established (Luo and Grigalunas 2003; Nir et al. 2003; Veldman and Bückmann 2003; Malchow and Kanafani 2004). Additional variables that have been shown to influence port choice include reliability (Luo and Grigalunas 2003) and inland container transport balance (Veldman and Bückmann 2003), as well as indicators of the service quality of a port (Luo and Grigalunas 2003; Fan et al. 2010). The latter may, for instance, include the maximum vessel size that can be handled in the port. In addition to these variables, research has established the relevance of the type of cargo in port choice. Hinterlands are commodity specific, in the sense that, due to differences in values of time, the hinterland of a port is not the same for waste paper than for electronics (Malchow and Kanafani 2004).

Even though the research on port choice has led to a much better understanding of the issue and the ability of ports to serve specific hinterlands, there is no established integrated theory that encompasses all previous insights. Instead, research is at a stage where additional potential variables of port choice are proposed and tested. Three additional factors that may influence port choice are discussed herein: maritime connectivity, intermodal connectivity, and transhipment orientation.

1.1 Maritime Connectivity

Maritime connectivity has been increasingly analyzed using advanced data and data analysis methods (UNCTAD 2018; Ducruet et al. 2010; and Bartholdi 2016). Wilmsmeier and Hoffmann (2008) demonstrated the relevance of liner shipping connectivity as determinant of freight rates. While, initially, liner shipping connectivity was developed as a countrywide performance indicator, the bilateral maritime connectivity between countries has also been added later, and in addition, scholars have developed maritime connectivity indicators for specific ports (Bartholdi 2016).

The relevance of maritime connectivity in understanding port hinterlands has been analyzed. Anderson et al. (2009) showed that shipment time, likely to be good proxy for maritime connectivity, was a determinant of port choice; Halim et al. (2016) developed a model which suggests that maritime connectivity is a relevant component for location decisions of distribution centers; Wang et al. (2016) developed an integrated approach to port connectivity, including both maritime and intermodal connectivity. However, none of the aforementioned works empirically assess the relevance of maritime connectivity for port choice decisions to and from specific hinterland locations. Such assessment is carried out in the present work, based on region-specific port connectivity data.

1.2 Intermodal Connectivity

Intermodal connectivity has been increasingly used in the analysis of port hinterlands. The relevance of intermodalism for the expansion of port hinterlands has been acknowledged for decades (Hayuth 1982; van Klink and van den Berg 1998). More recently, the development of inland ports (sometimes termed dry ports) and their relations with seaports have also been studied (Garcia-Alonso et al. 2018; Roso et al. 2009; Van den Berg and De Langen 2011; Monios and Wilmsmeier 2012). Along with the deepening of the understanding of the role of inland ports and intermodal services, empirical work on the relation between port choice and intermodal connections has emerged¹ (Tavasszy et al. 2011); For instance, Ferrari et al. (2011) showed, in the case of the Ligurian ports, that intermodal connections influenced the size of port hinterlands, while Chen et al. (2016) identified infrastructure bottlenecks in the Malaysian intermodal port hinterland connections. However, none of these studies used shipment data to assess the effect of hinterland connectivity on port choice. The inclusion of hinterland connectivity in such a port choice framework is a novelty. Fan et al. (2010) developed a model for port choice that included estimated costs of intermodal rail transport, but did not include the intermodal connections between a hinterland region and the various ports. In the present work, empirical data were collected on the intermodal connectivity of ports, in line with the approach taken by De Langen and Sharypova (2013) and De Langen et al. (2017).

1.3 Transhipment Orientation

A theoretically novel and empirically unexplored issue introduced in the present study is what we have termed the potential “transhipment orientation” of a port. The core argument is that a dominant share of transhipment (i.e., moves of a container from one ship to another²) may be at the expense of services to hinterlands. The advancing hub-and-spoke networks in container shipping have led to the emergence of transhipment hubs. The function of these hubs differs radically from traditional ports in that they function as a critical part of liner shipping networks. This has important consequences; For instance, the location logic of transhipment facilities is completely different: various such facilities are located on small islands (e.g., Malta) with neglectable volumes destined for local and regional hinterlands (Baird 2006). In addition, shipping lines are more actively involved in terminal operations in such hub ports, through their own terminal facilities or equity stakes in terminals (Notteboom et al. 2017). Finally, the productivity of transhipment terminals differs from gateway terminals (where flows to the hinterland are dominant): productivity measures in moves per crane or per ship are typically higher because of the lower complexity of operations (Morales Sarriera et al. 2013).

Three theoretical reasons for a negative side effect on hinterland services of transhipment orientation are provided herein. First, the terminal design of a transhipment terminal is different from the design of a gateway terminal (Monaco et al. 2009). In general, in transhipment terminals, less attention is given to handling hinterland cargo. This may lead to relatively low service levels at hinterland gates. Given the limited hinterland volumes, scale economies are absent, which may have a negative impact on hinterland service levels.

Second, transhipment terminals may relatively often be bypassed by container ships. Shipping lines do not always adhere to their published schedules. This leads to the so called blank sailings, where either an entire service or a specific port call may be canceled.³ Such blank sailings are not uncommon (Mongelluzzo 2018). Such a port-specific blank sailing occurs when the port is congested or ships omit a port to make up for delays. If a port needs to be bypassed, shipping lines may often select a transhipment port because transhipment operations can be shifted to other ports against relatively limited costs compared with shifts of cargo between gateway ports. There are no publicly available data to assess whether or not transhipment ports are more affected by blank sailings; this is a relevant area for further research. It is noted that even a fairly limited risk of a blank sailing may be a reason for a shipper to use a different port for (a part of) shipments, especially of time-sensitive goods.

Third, terminals that principally serve hinterlands are commercially active in attracting cargo volumes in these areas. Increasingly, gateway terminals develop hinterland services (Franc and Van der Horst 2010). Transhipment terminals may have a commercial focus on transhipment cargoes and more limited commercial activities in the hinterland. This also may reduce the volumes to/and from port hinterlands compared with ports without a transhipment orientation.

For these theoretical reasons, in the present paper it is tested whether such a transhipment orientation does have an effect on the market shares of a port in various hinterland regions.⁴

In conclusion, the aim of the present work is to expand the understanding of factors that influence port choice and, thus, port hinterlands. The empirical analysis is done based on data from Spain. The case of Spain has attracted considerable interest over the years, probably due to the fierce competition between a large number of ports in Spain and the interesting feature of Spain as a peninsula, with ports on the north, south, and east coasts, as well as the availability of detailed shipment statistics.

The remainder of this manuscript is organized as follows: Sect. 2 details the new factors that may influence port choice and presents the theoretical model and its variables; Sect. 3 details the empirical analysis based on data from Spain; Sect. 4 discusses the findings; and Sect. 5 concludes and relates the findings to the emerging literature on strategies of port authorities and terminal operators to improve hinterland access.

2 Port Choice Model

The literature review of port economics, management, and policy, by Pallis et al. (2010), identifies seven main research themes, one of which is termed competition and competitiveness. This theme includes port choice analysis, a topic of growing interest in literature, where discrete choice theory stands out as one of the most widely used methodological approaches by researchers. Since the 1980s, discrete choice theory has been increasingly used to analyze the choices made in the transportation field, firstly for passengers and later for goods. Specifically for the port choice analysis, Malchow and Kanafani (2004) were pioneers in proposing this methodological approach. Lagoudis et al. (2017) and Paixao Casaca et al. (2010) found that discrete choice modeling (DCM) was applied in around 20% of the papers published on the topic between 1981 and 2009. More recently, Martínez Moya and Feo Valero (2017) highlighted that most of the articles analyzing port choice topic from the discrete choice theory perspective proposed multinomial logit (MNL) models.

Following this approach, the probability of port p to be chosen to channel shipment s from province h to the world region wr can be expressed as (1):

$$P_{phwr} = \left( {\frac{{e^{{U_{phwr} }} }}{{\mathop \sum \nolimits_{p = 1}^{p = P} e^{{U_{qhwr} }} }}} \right),$$

(1)

where P_phwr is the probability of choosing port p among all the possible ports, p = 1, …, P, and can be interpreted as the market share of port p with respect to all ports considered.

Using common notation, Port p will be chosen only if the utility derived from its choice satisfies (2):

$$P_{p,h,wr} = {\text{Prob}} \left( {U_{p,h,wr} > U_{q,h,wr} } \right) \forall p \ne q.$$

(2)

The utility of a port p, U_p,h,wr is known by the decision-maker but not by the researcher, who must consider it with a random component. Hence, utility has to be broken down into two parts: (i) the observed part (representing the modeled effect of variables considered), V_p,h,wr, and (ii) the unobserved part (or error term), ε. Based on the analysis made in the previous section, components of the observed part, based on the attributes considered in this paper, can be obtained from (3):

$$V_{p,h,wr} = {\text{ASC}}_{p} + \alpha_{1} {\text{RD}}_{p,h} + \alpha_{2} {\text{MD}}_{p,wr} + \alpha_{3} {\text{MC}}_{p,wr} + \alpha_{4} {\text{IC}}_{p,h} .$$

(3)

Alternatively, the alternative specific constants (ASC) can be replaced by the port throughput and the transhipment orientation of ports, as shown in (4):

$$V_{p,h,wr} = \alpha_{1} {\text{RD}}_{p,h} + \alpha_{2} {\text{MD}}_{p,wr} + \alpha_{3} {\text{MC}}_{p,wr} + \alpha_{4} {\text{IC}}_{p,h} + \alpha_{5} {\text{PS}}_{p} + \alpha_{6} {\text{TS}}_{p} ,$$

(4)

where RD_p,h is the road distance between port p and province h (in km); MD_p,wr is the maritime distance between port p and a world region (in nautical miles); MC_p,wr is the maritime connectivity of port p with a certain world region, based on vessel calls and their capacity; IC_p,h is a dummy regarding the existence of intermodal connectivity between port p and province h; PS_p is the total container throughput of the port, based on the idea that there are scale economies in port operations, leading to higher productivity and/or lower costs in larger ports; TS_p reflects the transhipment orientation, which, based on the theoretical arguments provided above, is added for ports with high transhipment to test whether this affects negatively their shares in hinterland regions.

As discussed above, the majority of discrete choice models in ports have employed MNL models. Their main advantage is their simplicity. It is assumed that the error term is independently and identically distributed and follows the Gumbel distribution (McFadden 1973). This assumption implies that the introduction (or improvement) of any additional alternative will have the same impact on the probability of the rest to be chosen, which is too restrictive. The most straightforward way to overcome that restriction is by applying a nested logit (NL) model (Anderson et al. 2009; Veldman et al. 2013; Cantillo et al. 2018). The NL model allows us to relax the assumption of independence from irrelevant alternatives (IIA). To solve this problem, the alternatives (ports) are grouped in nests with some degree of similarity among them (e.g., to be located on the same coastline) in such a way that the hypothesis of IIA remains valid within each nest (a group of ports on the Atlantic coast and a group on the Mediterranean coast). This approach is also deployed in the empirical analysis of the present work.

To test the significance of the new variables, information is required on the market shares of ports in specific hinterland regions and regarding specific overseas destination areas. The estimation of these data, as well as the results of the statistical tests, are addressed in the next section.

3 Empirical Analysis Based on Data for Spain

The addressed case study focuses on Spanish containerized exports channeled by sea through the main Spanish ports. Data about flows were obtained from the Foreign Trade Statistics of the Customs and Excise Duties Department of the Spanish Tax Agency. They are freely available and provide information about exports composition and volume (both in Euros and tonnes), their origin (at the level of Spain’s provinces), and the country of destination. As Escamilla-Navarro et al. (2010) pointed out, this data source is particularly reliable for the analysis of extra-EU maritime traffic⁵ and useful in delimiting the hinterland of ports. The data have already been successfully used for this purpose in previous port hinterland analyses, most recently in Moura et al. (2017, 2018).⁶

The ports usually considered when analyzing port choice questions in Spain are the main four ports (islands excluded): Algeciras, Barcelona, Bilbao, and Valencia. This is because these are the only Spanish ports competing for a contestable hinterland. The rest only attract traffic from their closest territories, and their inclusion could distort our results (for instance, the role of distance). Nevertheless, the author’s intention was to expand the analysis and include more ports in the study. Based on their container traffic, the ports of Castellón and Vigo were the most relevant potential additions. Vigo was included in the analysis as it is located far away from the rest. Hence, it is expected not to distort the results and to be of interest. The port of Castellón was excluded because of its proximity to Valencia and the mainly captive local hinterland it serves. The Portuguese ports were also excluded in the analysis because the Spanish traffic channeled through them is insignificant (Santos and Soares 2017). Based on these considerations, a database was constructed dividing all maritime shipments according to departure and destination region. The number of observations was determined by the number of provinces in Spain (47⁷), by the number of ports considered (5), and by the number of destination country regions (4), essentially covering the areas north, east, south, and west of Spain. This led to 940 observations for 1 year.

The road distance between port p and hinterland h in kilometers (RD_p,h) was calculated for all the province–port pairs using Google Maps. The distance was calculated from each port to the provincial capital city, where the main density of population and economic activity are concentrated. When the port is located at the capital city, the internal distance of provinces was considered proportional to the square root of their area (in km²), following Garcia-Alonso and Marquez (2017), as can be seen in (5):⁸

$$d_{ii} = 0.66 \sqrt {\frac{{{\text{area}}_{h} }}{\varPi }} .$$

(5)

The maritime distance between port p and a world region in nautical miles (MD_p,wr) was calculated by taking one reference port⁹ and including this distance, calculated with the website www.sea-distances.org. Table 1 presents the distances in nautical miles from the five Spanish ports to the four world regions.

Table 1 Distances in nautical miles from Spanish ports to four main world regions

Maritime connectivity of port p with a certain world region (MC_p,wr) is expressed as the sum of the vessel capacities of ships that call in port p and also in the world region in question,¹⁰ based on data drawing on Lloyd’s List ship movements (Lam and Yap 2008; Mohamed-Chérif and Ducruet 2016; Arvis et al. 2018. In the present model, MC_p,wr was normalized per port: the percentage that each port represents in each zone was calculated.

The dummy regarding the existence of intermodal connectivity between port p and hinterland h (IC_p,h) was calculated based on data on intermodal connections, in a matrix of 47 regions and 5 ports, and has the value 1 if there is an intermodal connection and 0 if this is not the case. In this approach, the value of intermodal connectivity is 1 if there is a service to the origin/destination province. Only rail transport increases intermodal connectivity in Spain, as the country has no inland waterways for containers. In this approach, the value of intermodal connectivity is only 1 if there is a service to the origin/destination province. The imperfection of this method acknowledged, as it does not consider link quality and it could be argued that a province is also intermodally connected when there is a service from a neighboring region to a port. However, given the complexity of expressing link quality (De Langen et al. 2016) as well as accounting for indirect intermodal connections, this issue was not addressed in the present work.

Port size (PS_p) is the variable for the total container throughput of the port, based on the idea that there are scale economies in port operations leading to higher productivity and/or lower costs in larger ports. This information was provided by the Spanish national entity governing ports (Puertos del Estado 2018).

Transhipment (TS_p) is a variable that reflects the transhipment orientation. As detailed in Sect. 2, the theoretical logic is broadly that if both shipping lines and the terminal operator are focused on transhipment operations, this is at the expense of the quality of services for containers to/from the hinterland. In Spain, this leads to a dummy, which is 1 for Algeciras and Valencia (with a transhipment percentage of 92% and 61% of the total container volume handled in 2015, respectively, thus higher than a cut-off level of 50%) and 0 for the other Spanish ports, as these do not have high shares of transhipment. Maritime connectivity was based on data from ship movements from Lloyd’s List Intelligence¹¹ (Ducruet 2015, 2017, and World Bank).

4 Results and Discussion of the Findings

The model results are presented in Tables 2 and 3. On the one hand, Table 2 presents the two specifications of the model, (3) and (4). Both have a similar adjustment, but model (4) was considered to fit better than (3). Although (4) is slightly weaker in terms of log likelihood, all the variables are significant in this specification and its hit rate is higher. Table 3 presents a so-called confusion matrix, used to describe the performance of the models by comparing their results with the already known port choice carried out. In both specifications, (3) and (4), the best fit is for Vigo and the worst for Valencia. This is due to the fact that the hinterland of Vigo is mainly captive and limited to a smaller number of provinces, while Valencia is the port with largest hinterland reach in Spain (Garcia-Alonso et al. 2016; Grecco et al. 2017), followed by Barcelona (the second worst fit in terms of the confusion matrix).

Table 2 Results of analysis

Table 3 Confusion matrix

As can be seen in Table 2, all variables are significant with the expected sign. Maritime connectivity and transhipment orientation have the strongest influence on port choice. These results support the emphasis of policy-makers and port developers on maritime and intermodal connectivity, showing that both types of connectivity are important when operators choose a port. The main findings of the study are summarized below:

• Maritime distance significantly affects the market share of a port for a specific world region, which shows that hinterlands are relational; For instance, Valencia presents a higher share (across regions) of cargoes to and from Asia, while Bilbao presents a higher market share of cargoes to and from Northern Europe. This shows the hinterland of a port is relational, in the sense that it depends of the overseas origin/destination of the cargo; For instance, the hinterland of the port complex of Los Angeles and Long Beach covers, say, as much as 60% of the whole US territory for goods to/from North East Asia, while it may be as limited as 15% of the US territory for goods to/from Europe and Asia.

• Maritime connectivity also significantly and positively influences the market shares of ports in hinterland. This is an important insight, as maritime connectivity depends both on the port and the world region, whereas container throughput depends only on the port.

• Intermodal connectivity, that is, intermodal services, also positively influences the market share of a port in a hinterland region.

• Port size is also significant and represents the positive effect of scale economies in ports.

• Road distance has a significant negative effect on the market share of a port. This is in line with previous studies.

• The dummy variable—transhipment—is also significant, with the expected negative sign. This is an important finding as it suggests that a high share of transhipment volume may lead to an orientation on transhipment, with adverse effects on hinterland flows.

The present work proves that there are no clear boundaries between the hinterlands of different ports, that various ports have market shares in contestable hinterlands, and that hinterlands differ per overseas destination. Nevertheless, the findings provide a basis for assessing regions where a port either has a significant market share or can aspire to develop such a market share, based on current¹² maritime connectivity and hinterland transport infrastructure (Ferrari et al. 2011), based on a utility function that shows the utility for port users of importing/exporting through port p to/from hinterland h.

5 Conclusions and Further Research

A model with novel variables was tested for the market shares of a port in a certain hinterland. The empirical analysis, based on detailed shipment data from Spain, confirmed the expected effect of all tested variables: road distance (−), maritime distance (−), maritime connectivity (+), intermodal connectivity (+), port size (+), and transhipment orientation (−).

Three limitations of the study need to be acknowledged: First, the method used to calculate intermodal connectivity is imperfect and could be improved by adding a way to assess the quality of hinterland links, as well as the indirect intermodal connectivity of a province, through the availability of intermodal connections in a neighboring province. Second, this paper suggests the detrimental effects of a transhipment orientation. This is a relevant contribution but calls for additional analysis given that the current method to treat transhipment through a dummy is imperfect, as it requires a cut-off value, which in this paper was arbitrarily set at 50%. Furthermore, a transhipment orientation may be better regarded as a risk for transhipment ports rather than as a given. There certainly is scope for agency by the port community in a transhipment port to improve services to the hinterland. Third, like in the previous work on port choice, the imperfect data availability prevents an all-encompassing test on port choice, where all potentially relevant variables are included; For instance, in the present study, data on the shipment type (e.g., commodity group or shipment value) were not included as a potential determinant of port choice, even though it was found to be relevant in previous studies (e.g., Malchow and Kanafani 2004).

These findings are relevant for the emerging literature on strategies of port authorities and terminal operators to improve hinterland access. The model results show that, from the variables mentioned above, maritime connectivity has the strongest effect on the market share of a port in the hinterland. Thus, a strategy by both the port authority and the wider port community to explicitly pursue increases in maritime connectivity may be worth considering. One element of such a strategy may be pricing (Van den Berg et al. 2017). More research on potential tools to increase maritime connectivity and their effectiveness is warranted.

A second variable that affects the market share of a port and can be influenced by a port authority and other actors in the port community is intermodal connectivity. Thus, strategies of both the port authority and terminal operators to actively improve such connectivity may be sensible, and indeed have been developed (Van den Berg and De Langen 2011; Monios and Wilmsmeier 2012; Shi and Li 2016).

The third variable is the transhipment orientation. This variable also has a strong influence on the market shares of a port. Given the fact that this concept is novel, both conceptually and empirically, more research is required to establish whether there is indeed a trade-off between a hinterland and a transhipment orientation, or alternatively, whether an and–and strategy can be pursued. One relevant issue is whether port authorities, in ports with predominantly transhipment traffic, can effectively include hinterland service levels in concession granting procedures; For instance, the inclusion of hinterland services in the assessment (scoring) of proposals may be an option. Alternatively, minimum hinterland service levels (e.g., average turnaround times of trucks) may be included in concession contracts. This is a relevant topic in the emerging stream of research on concessions in ports (Theys et al. 2010; Ferrari et al. 2018).