Scope of analysis
For our analysis of the port competition–performance link, we focussed on large-scale, global hubs, or national gateway ports, in 2004. Before the US housing market began to decline in 2006, leading to the global economic meltdown two years later, the selected year was the most dynamic for global shipping (annual growth of 14.4% in world container port traffic [World Bank 2015]), and it was thus suitable for testing our inverted-U hypothesis.
We have also added an analysis of the year 1991, as a contrast group against the year 2004. The global port sector experienced vast technological/institutional restructuring between these two periods, which caused greater market instability (Cheon et al. 2010). Comparing these two periods—which represent varying levels of environmental stability in the global container port market—provides us with a clearer picture for examining our hypothesis.
The idea of size-localized competition suggests that organizations in different size groups face fundamentally different mechanisms (e.g. minor fishing ports usually do not compete against major container ports), and that competition influences survival and performance (Hannan and Freeman 1989). Since economies of scale prevail in the port sector (Turner et al. 2004), smaller ports may face significant disadvantages in competition. We thus focussed on large-scale, global hubs or national gateway ports (Cheon et al. 2010) for constructing the comparison set of functionally similar organizations (138 world ports in 2004 and 98 ports in 1991).
Measuring PSCP
Measurement
Equation (1) presents the proposed measure of competitive pressure, hereafter called ‘hinterland market accessibility’ (HMA) (Cheon et al. 2008). HMA captures the extent to which the impact of each port reaches hinterland markets, discounted by any competitive threats that competing ports generate. Because the index measures the degree to which market demands support a port, given multiple competitors, the HMA of a port is inversely related to the intensity of competition:
$$\mathop {HMA_{i} }\limits_{i \ne k} = {{\sum\limits_{J} {\left( {GDP/D_{i,J}^{\alpha } } \right)} } \mathord{\left/ {\vphantom {{\sum\limits_{J} {\left( {GDP/D_{i,J}^{\alpha } } \right)} } {\sum\limits_{J} {\sum\limits_{k = 1}^{m} {\left( {Q_{k} /D_{k,J}^{\alpha } } \right)} } }}} \right. \kern-0pt} {\sum\limits_{J} {\sum\limits_{k = 1}^{m} {\left( {Q_{k} /D_{k,J}^{\alpha } } \right)} } }},$$
(1)
where GDP
J
is the gross domestic product of economic zone J (J = 1,2,…, N); D
i,J
is the distance between port i (i = 1, 2,…, m) and economic zone J; D
k,J
is the distance between port k and economic zone J; Q
k
is the total throughput of port k \((k \ne i)\); and a is the distance impedance parameter.
In Fig. 2, since the hinterland markets related to port i consist of multiple, often legally defined economic zones, J
1, J
2, J
3,…J
N,, such as countries, states, and cities, the total market opportunities faced by port i are the additive aggregation of the hinterland market opportunities generated from the economic zones. The hinterland market opportunities can thus be expressed as a gravity model: \(\sum\nolimits_{j} {\left( {GDP_{J} /D_{{i,J}}^{\alpha } } \right)}\), in which α is the distance impedance parameter. Let this scale value be a unique value for hinterland size of port i: i.e. port i’s unique geographic market domain, determined by the distinctive location of each port and the economic zones that are universally available to all ports.
The numerator of Eq. (1), hinterland size, should be discounted by the degree of threat from competing ports serving the overlapping hinterlands. The denominator of Eq. (1) assesses the port-specific degree of aggregated threats from all other ports across the economic zones considered. The aggregated threats of port i arise from (i) the aggregated size (and number) of competing ports, k, \(\sum\nolimits_{{k = 1}}^{m} {Q_{k} }\) and (ii) the distance of each competitor, k, from the hinterland economic zones, J \(\left( {\sum\nolimits_{k = 1}^{m} {1/D_{k,J}^{\alpha } } } \right)\). HMA essentially reflects the concepts of market domain overlap, market commonality rivals in a spatial dimension, and mass dependence competition (Barnett and Amburgey 1990).
Operationalization
The first issue when operationalizing HMA is to consider variations in demand across multiple economic zones. We apply two alternative approaches in order to compare the results and check the measures’ validity. The first, which is simpler and less data-intensive, uses countries’ economic data. We apply GDP data for calculating the aggregated sizes of the economic hinterlands for each port, and we use capital city locations (usually found in major economic conurbations) for measuring the distance between ports and the economic zones. We call this the HMA-Capitals model. While the assumption behind this is that capital cities are where major economic activities occur, this could be a source of error for larger countries’ economic activities. We adopt a second, more data-intensive approach to partly resolve this problem, using all global cities larger than 500,000 people. We name this the HMA-Cities model. We then assign each country’s GDP, using the ratio of city populations, based on the formula: \(GDP_{j} = GDP_{c} \left( {P_{j} /\sum\nolimits_{l = 1}^{m} {P_{l} } } \right),\) where GDP
j
is the GDP assigned to city j; GDP
c
the GDP of country c, with which city j is territorially affiliated; P
j
the population of city j; and P
l= is the population of all cities within country c with >500,000 people.
The number of countries and cities considered for the 1991 and 2004 GDP data was 186 and 776, respectively, including most countries except small island ones. To calculate HMA, we initially consider the 257 largest container ports, which produced more than 90% of the 245 million 20-foot equivalent units (TEUs) handled by 533 container ports in the world in 2004 (CI 2004). The idea of scale-based selection suggests that smaller organizations will follow different mechanisms to deal with competitive pressure from scale gaps (Dobrev and Carroll 2003). We have thus removed the smallest ports (which produce the remaining 10% of world TEUs) for building our initial set of port samples engaged in competition.
We have measured distances along the geodesic curve using ArcGIS. The distance parameter α also had to be defined. The shape of the distance parameter can largely depend on the extent to which competition is local or global in a space dimension (Vogel 2008). In the case of local competition (high α), a port competes only with its direct neighbours, which is similar to the dyadic view of competition (e.g. Baum and Korn 1999). Under global competition (low α), a port competes with all other ports in the global port industry. This is also the approach early organizational ecologists have employed.
The level of α may also change depending on other distance-related factors which affect the spatial immobility of cargo movement between ports and economic zones. These include higher fuel prices in world oil markets, creating a high α. However, since few empirical studies have been conducted on the spatial scope of port competition, we also adopt a series of sensitivity analyses, attempting to avoid choosing between two extreme cases (global vs. local). We test a variety of cases with α values between 0.1 and 10 to confirm the consistency of data patterns with different scales of distance parameters, and to validate whether the results are consistent for certain ranges of distance parameters.
Variables for institutional competitive pressure
Because we differentiate between organization-specific competitive pressure and macro-level institutional competitive pressure, we have also collected data on several variables for the national-level institutional environment. For example, previous literature has discussed special restrictions on domestic participation of foreign suppliers of cargo-handling services, and how compulsory some port services are for incoming ships (Fink et al. 2002). Since limited information is available on these regulations, we have collected data on six variables for proxying national institutional environments: (1) number of administrative procedures for warehouse-building; (2) days needed to complete procedures for warehouse-building (from the World Bank [2005]); (3) customs procedures burden; (4) trade barriers prevalence; (5) government regulation burdens; and (6) national-level competition intensity in most industries (from the Global Competitiveness Reports [GCRs]; Schwab and Porter 2008).
Measuring port performance
We adopt data envelopment analysis (DEA) for measuring port performance. DEA translates Pareto efficiency into the relative efficiency of decision-making units (DMUs) based on non-parametric mathematical programming (Charnes et al. 1978). We have adopted the output-oriented DEA-CCR and DEA-BCC models, formalized in Eq. (2) (Cooper et al. 2004), based on the observation that ports are throughput maximizers (Tongzon 1995). In these models, DMUs on the efficient frontier have an efficiency score of 1. Efficiency scores of sub-optimal DMUs, measured relative to efficient DMUs, have scores >1.
$${\text{max}}\phi - \varepsilon \left( {\sum\limits_{{j = 1}}^{m} {s_{i}^{ - } + \sum\limits_{{r = 1}}^{s} {s_{r}^{ + } } } } \right),{\text{ subject to }}\sum\limits_{{j = 1}}^{n} {\lambda _{j} x_{{ij}} + s_{i}^{ - } = x_{{i0,}} {\mkern 1mu} \quad i = 1,2, \ldots m;} {\text{ }}\sum\limits_{{j = 1}}^{n} {\lambda _{j} y_{{rj}} - s_{r}^{ + } = \phi y_{{r0,}} } {\mkern 1mu} \quad r = 1,2, \ldots s;{\text{ }}\lambda _{j} ,s_{i}^{ - } ,s_{r}^{ + } \ge 0{\mkern 1mu} \quad \forall i,j,r;{\text{ }}\sum\limits_{{j = 1}}^{n} {\lambda _{j} = 1;{\text{ for the DEA-BCC model}}}$$
(2)
where ϕ
0 is the relative efficiency of DMU0; \(s_{i}^{ - }\) is the input slack variable; \(s_{r}^{ + }\) is the output slack variable; and ε is a ‘non-Archimedean’ element, which should be smaller than any positive real number.
Three main input factors for container production considered were as follows: total container berth length (metres), container terminal area (metres2), and total capacity of container cranes (tonnage), as has been used in the previous port studies (e.g. Tongzon and Heng 2005). As an output variable, we have selected the container volume handled (total TEUs) of a port. We acquired port input/output data from Cheon et al. (2010). We also selected the year 2004 for the competition–performance link analysis, and for constructing port competitive pressure data that would match our input/output data.
Control variables
We consider several control variables in estimating the impact of competition on port performance. These include the following:
-
Port Size (‘Size’) Researchers have found that economies of scale have prevailed in the port sector over the years and these have significantly influenced port performance (Turner et al. 2004). We control for size effects by including a measure of overall capital assets, proxied by the natural logarithm of linear combinations of container berth numbers, container berth average depths, and crane numbers.
-
Network Externalities (‘Network’) A port’s connectedness influences its performance; well-connected ports attract higher container volumes, because shipping networks are more valuable to shipping lines and shippers when ports are connected with other local liner services and spoke ports (McCalla 2003). We control for port connectedness by including the number of direct liner services in ports.
-
Global Terminal Operator (‘GTO’) Two dozen GTOs have emerged in the last twenty years; they have been critical sources for the transformation of the structure of the global port sector (Cheon 2009). These specialized entities usually adopt effective investment/management programmes for port infrastructures and superstructures. Since ports involved with GTOs are expected to perform better, we include the percentage of TEUs handled by GTOs in ports’ total container production volume.
-
Hinterland Infrastructure (‘Infra’) Ports’ surface infrastructure condition is crucial to port performance (Clark et al. 2004; Turner et al. 2004). If ports’ hinterland transportation networks are unfavourable to cargo movement, shippers/carriers may choose other ports. We consider this effect by using a national-level variable: the percentage of paved roads in the total road network.
-
Dummy for China Factor (‘China’) Chinese ports are unusual in terms of their expansion and performance improvement during the last decade, their emerging economic hinterlands, and their new, well-designed container terminals. The dummy variable China can capture the difference in performance between Chinese ports and those of other countries.
Statistical model
As shown in Eq. (3), we specify port performance (DEA) as a quadratic function of competitive pressure (HMA) and the vector of control variables (X) including macro-level institutional competitive pressure for port i, in addition to an error term (u
i
):
$$DEA_{i} = \beta_{1} HMA_{i}^{2} + \beta_{2} HMA_{i} + \beta X_{i} + u_{i},$$
(3)
where β1 represents a parameter for the quadratic relationship between competitive pressure and port performance.
We specify both HMA and DEA models in this way; that is, as the index values of HMA and DEA increase, the actual intensities of competitive pressure and port performance decrease. Therefore, we expect statistically significant, positive values for β1 in 2004 to support the inverted-U hypothesis, and values that do not differ from zero for β1 and β 2 for 1991 as the contrast group. Table 2 summarizes the descriptive statistics of the variables.
Table 2 Descriptive statistics and correlation coefficients