Sources of efficiency changes at Asian container ports

Introduction

Container ports and terminals form an essential component of the modern economy. Containerization since the middle of the twentieth century has largely reduced the transportation cost of international trade, resulting in dramatically growing demand for container transport. Physical expansion and efficiency improvement have been the two major approaches to enlarge container port capacity to cope with escalating trade volumes (Le-Griffin and Murphy, 2006). Yet in places where port expansion is constrained by a limited supply of land and increasing environmental concerns, improving port efficiency is more feasible and effective. It is then critical to assess the potential sources of port efficiency gains over time for governments and port operators to devise strategies accordingly. In particular, governments are able to optimize the collocation of the coastal resources to enhance the competitiveness of hinterlands while port operators may benchmark their performance with comparable ports to identify areas for improvement.

The Belt and Road Initiative, also known as One-Belt One-Road, is a strategic project that has been a major topic of discussion in countries along the twenty-first-century Maritime Silk Road of One-Belt One-Road. “One-Belt” denotes the Silk Road Economic Belt and includes countries such as Indonesia, Malaysia, Philippines, Singapore, Thailand, etc. It is believed that the Silk Road Economic Belt will serve as a vehicle to create economic prosperity for the world. Ports along the Silk Road Economic Belt is not only affecting the local economy but also reverberating across the worldwide economy.

Port has more to offer to businesses than just a maritime facility. Therefore, an efficient port will be substantially enhancing many economic activities around the port from trade to shipping, and from transport to warehousing. Previous research has not sufficiently addressed efficiency changes in large Asian container ports compared with those in developed regions. Few studies have tried to analyze the sources to which efficiency gains and losses can be attributed (Cullinane et al., 2002; Yip et al., 2011). In this study, we aim to fill these gaps by estimating the efficiency changes of major Asian container ports, many of which are along the Silk Road Economic Belt, for the period from 2000 to 2007. In addition to estimate the overall efficiency change, we intend to decompose the overall change into components related to technical, scale and technological efficiency and thereby derive policy implications to governments, port authorities and operators. Finally, a series of media reports have been published recently on the decline of Hong Kong Port (Heaver, 2017; Grinter, 2018), motivating us to examine in detail the performance of Hong Kong Port, and benchmark it with its main competitors. In this way, this study will provide valuable findings such that ports along the Silk Road Economic Belt will have possible options to enhance their efficiency.

We will first review the concept and methodologies of calculating efficiency changes before describing the data and scope of this study. We will then present the result of analysis and discuss its implications for port authorities and operators by providing an overview of efficiencies for all ports and examining in detail the sources of inefficiencies. The performance of Hong Kong Port, as well as its comparison with its major competitors, will be addressed. Finally, the limitations of this study and areas for further research will also be discussed.

Literature review

Widely used methods to calculate productivity include index number approach, traditional regression methods, corrected original least squares (COLS), stochastic frontier analysis (SFA), and data envelop analysis (DEA). Derived from DEA framework, recent studies have increasingly utilized the Malmquist productivity index (MI) method to measure productivity of container ports and terminals (Choen et al., 2009; Song and Cui, 2014; Ding et al., 2015). The MI method shares many of the advantages of DEA method, and is particularly useful to evaluate the productivity change of decision-making unit (DMUs) between two time periods. Furthermore, it has the advantage of decomposing overall productivity change into various components. In the following, we discuss in detail the intuition of each approach, as well as their advantages and limitations.

The index number approach attempts to capture the ability of DMUs to combine inputs and produce outputs. The total factor productivity (TFP) is the most widely used measure in the index number approach. Though easy to calculate, such index has limitation in distinguishing efficiency changes from the effects of scale economies and input substitution (Choen et al., 2009).

Original least squares (OLS) estimation is another approach where a regression line is fitted into the existing data, representing the productivity of each DMU derived from the observed data. Yet this approach relies on the assumption of an optimal production or cost function and is therefore inaccurate as decisions of DMUs are not always optimized. Corrected original least squares (COLS) estimation is an improved method where the regression line in OLS is shifted to enclose all the data. The shifted line represents the efficiency frontier and the relative efficiency of each individual port can be measured against the frontier. However, the weakness of this approach includes its dependence on a priori production or cost function and its sensitivity to the frontier used (Liu, 2010).

Stochastic frontier analysis (SFA) is a parametric and stochastic approach to estimate productive efficiency. The major breakthrough of SFA compared to regression methods is that SFA calculates the inefficiency of DMUs based on distribution assumptions, so different entities can have different inefficiencies (Yip et al., 2011; Merkel, 2018). In common with the COLS approach, SFA relaxes the assumption that the behavior of DMUs is optimized. As in COLS, the degree of (in)efficiency of individual DMUs can be measured against the frontier. The main advantage of the SFA method is that it considers statistical noise, and hence, it is possible to test the validity of certain assumptions and hypotheses and that there is great flexibility in specifying the production technology in functional form. However, similar to COLS, the drawbacks include the need to impose a priori structure when constructing the frontier functional form. Furthermore, the estimation results using SFA is sensitive to the assumptions concerning the distribution of the inefficiency terms.

Data envelopment analysis (DEA) is a mathematical programming approach to estimate efficiency. Similar to COLS and SFA, this approach also maps out a production frontier based on inputs and outputs information, and the relative efficiency of each DMU is estimated from its distance to the frontier. The strength of this method is that no priori structural assumptions is required. The drawbacks of this method lie in that the accuracy of it is sensitive to outliers, and that it does not consider the measurement error so that it is not possible to test the statistical significance of the estimated efficiency. A growing number of studies have used the DEA method to evaluate port efficiency (Panayides et al., 2009). Roll and Hayuth (1993) is probably the first study which applies the Charnes, Cooper, and Rhodes (CCR) DEA model with the assumption of constant returns to scale (CRS) to the port sector. It uses a hypothetical example of 20 container ports to generate simulated results. Martínez -Budria et al. (1999) uses the Banker, Chames and Cooper (BCC) DEA model assuming variable returns to scale (VRS) to analyze 26 Spanish ports using input and output data during the years 1993-1997. Tongzon (2001) argues that to restrict the scope of analysis to a limited number of ports and a specific type of cargo is necessary for the multiplicity of ports and cargo handled. It uses both CCR DEA and DEA-additive models to analyze the efficiency of 4 Australian and 12 other international container ports for the year 1996. Valentine and Gray (2001) again use a CCR DEA model to compare the efficiency of 31 world-class container ports with different ownerships and organizational structures in 1998. Cullinane et al. (2005) investigate the relationship between privatization and efficiency by analyzing 25 container ports during 1992-1999. Nguyen et al. (2016) used bootstrapped DEA to measure efficiency of Vietnamese ports.

The Malmquist productivity index (MI), based on DEA models, is one of the prominent indices for measuring the relative productivity change of DMUs over time. Using this method, the estimated efficiency changes are decomposed into frontier shift effects (due to technological advancement) and catch-up effects. The catch-effects are further separated into pure technical efficiency effects and scale efficiency effects, indicating the extent to which an operator catches up with the best practice in the field and optimize the scale of its operations to meet the demand side. This index has been used to measure efficiency changes in other regulated infrastructure sectors such as electricity (Hjalmarsson and Veiderpass, 1992), natural gas (Price and Weyman-Jones, 1996), and airports (Abbot and Wu, 2002). For the port sector, this type of research has been conducted for countries like Yugoslavia and Mexico (Nishimizu and Page, 1982; Estache et al., 2004; Choen et al., 2009) have also applied MI to 98 world scale container ports and major national gateway ports. Ding et al. (2015) applied MI to coastal secondary ports in China. The following section will discuss in detail the methodology of calculating the MI.

Methodology

Based on information on the inputs and outputs of DMUs in two periods, the MI method can determine whether the variation of performance is due to technical efficiency change (TEC) and/or technological change (TC). Following Estache et al. (2004), the MI calculated for year t and t + 1 can be calculated as the following:

Here, the D0t(x0s,y0s) represents the distance from the period s observation to the period t technology. When M₀ is greater than 1, it indicates productivity growth, and when M₀ is smaller than 1, it indicates productivity deterioration. The MI can be decomposed into two components: the technical efficiency change (TEC) and the shift of productivity frontier due to technological change (TC):

The first component measures how close the DMU is to the frontier in year t + 1 compared with year t, and therefore, with TEC greater than 1, the DMU has moved closer to the frontier in year t + 1 than in year t, and vice versa. The second component captures the change in technology between the two periods. If TC is greater than 1, it indicates technological progress, and vice versa.

Given that the MI measure is derived from DEA, its decomposition also depends on the assumptions on returns to scale. The TEC and TC indices above are calculated under the assumption of CRS, that is, assuming that all DMUs are already operating at the optimal scale. As illustrated in Grilo and Santos (2015), the TEC calculated under the assumption of CRS can be further decomposed into pure technical efficiency change (PTEC) and scale efficiency change (SEC) under the assumptions of VRS:

PTEC captures changes in technical efficiency resulting from improvements in operations and management practices, that is, the least inputs are used in producing the outputs. SEC captures the productivity change associated with the movements of DMUs inside the frontier and assess whether the movements are in the right direction to attain the CRS point, which is the optimal scale point.

In this study, we use the MI method to investigate the overall efficiency changes at major Asian container ports from 2000 to 2007. We further look into the contributions from improvements in pure technical efficiency, scale efficiency and technological progress for each port. The performance of Hong Kong Port will be benchmarked with the overall statistics and with its major competitors. Finally, a statistical test will be run to reveal the correlation between overall efficiency change, that is MI, and PTEC, SEC or TC, which indicates the contribution of each source to the overall efficiency improvement/deterioration.

Data and scope

Ideally, all activities and resources involved in container port operations should be taken into account when calculating efficiency. This decision of which input and output variables to be included, however, largely depends on the availability and quality of data.

The specification of inputs in the literature is not unified. Some studies consider labor and capital as input variables (Liu, 1995; Coto-Millan et al., 2000; Estache et al., 2002; Cullinane and Song, 2003). Some others specify inputs based on the infrastructure and machineries of the ports, that is, quay length, terminal area, number of cargo handling equipment and storage capacity (Tongzon and Heng, 2005; Cullinane et al., 2002; Cullinane and Song, 2006; Sun et al., 2006). Though important for port operations, labor inputs may not be necessary or suitable for determining productivity for several reasons. First, a fairly close relationship exists between the number of workers in a container terminal and the number of gantry cranes. The labor input can thereby be derived by a function of the facilities of the terminal (Notteboom et al., 2000). Second, many operations during cargo handling are outsourced to third-party logistics firms, making the port’s statistics data on labor less reliable. Finally, in the era of containerization, many port operations are standardized. The efficiency differences caused by labor are not very significant (Liu et al., 2006). In contrast, the infrastructure and machineries inputs reflect a more accurate configuration of the ports (Notteboom et al., 2000). In this case, given the characteristics of container ports operations and the limitation of information, total container berth length (in meter), container terminal area (in square meter), and landside container crane capacity (in ton) are selected as the inputs for analysis. Other input factors, such as berth working hours, berth waiting time and other equipment, are not included from the consideration of both data availability and avoidance of the problem of multicollinearity.

The specification of outputs in the literature is more unified. Though some recent studies have started to incorporate multiple outputs (Barros, 2005; Rodriguez-Alvarez et al., 2007), the annual container throughput (in TEU) is still the most widely accepted indicator for container transport activities. In addition, to reduce the impact of severe output fluctuation that might have been caused by external shocks (e.g. labor dispute), the average of three successive years’ throughput is used as the final output (e.g. the average value of 1998, 1999 and 2000’s output is used as the output data for 2000)[1].

The scope of this research is for Asia and a total number of 23 containers ports will be investigated (see Figure 1 for their geographic locations). We first identify 27 ports whose throughput in 2008 were at least two million TEUs, and they contributed to more than 80 per cent of the total throughput of Asian container ports in that year. Then for Asian countries not covered by these ports, we will include their largest container ports as well. However, the final sample size is reduced due to data availability. Specifically, data on port input and output are mainly collected from Containerization International Yearbook, whose information is collected directly from the terminal operating organizations. Other data sources include China’s Port Yearbook and port authorities’ websites, etc. Output and input data are available for 28 ports for year 2007, but this number decreases to 23 ports for year 2000. Therefore, the final sample will be restricted to the 23 Asian container ports that have data for both years.

The 23 investigated container ports are represented by 14 countries/regions, Singapore, China, Hong Kong (China), Korea, Taipei (China), Malaysia, Thailand, Japan, India, Indonesia, Sri Lanka, Philippines, Pakistan and Brunei, and therefore shall have different policies, management structures and regulatory characteristics. Their output variables (throughput in TEU) and input variables (berth length in meter, terminal areas in square meter, crane capacity in ton) are presented in Table I, for year 2000 and year 2007 respectively. The output throughput for 2000 is the average number of the throughput of 1998, 1999 and 2000. Similarly, the output for 2007 is the average of that in 2005, 2006 and 2007. The summary statistics of the input and output variables are reported in Table II. In the following section, the MI analysis will be implemented with the 23 ports that have data on both years.

Findings

Table III has summarized the efficiency changes and their sources at the investigated ports from 2000 to 2007 using the MI method. Overall, the major Asian container ports have improved their efficiency by 14.3 per cent (with average MI = 1.143). The sources of efficiency gains/losses can be attributed to:

• a 41 per cent increase in pure technical efficiency (average TEC = 1.41);

• a 47.5 per cent increase in scale efficiency (average SEC = 1.475); and

• a 30.5 per cent decrease in technological efficiency (average TC = 0.695).

Nine ports appear to have improved their efficiency while the remaining fourteen have retrogressed. Busan Port is the best-performing port according to our analysis, with MI = 2.805. This increase in MI mainly comes from its improvement in technical efficiency (TEC = 2.439) and scale efficiency (SEC = 1.428), while its TC score is less than 1 (TC = 0.805). It is worth noting that all ports have a TC index that is smaller than 1, which indicates deterioration in technological improvement, that is, no innovation in technologies. This result contradicts the general belief that most large Asian container ports have rapidly improved their container handling, managerial and security technologies. In addition, in light of Choen et al. (2009), ports like Singapore and Hong Kong have made strategic and aggressive capital investment in the most cut-edging technologies, yet their respective TC is only 0.666 and 0.305.

Moreover, a series of media reports have been published recently on the decline of Hong Kong Port (Heaver, 2017; Grinter, 2018). From our analysis, Hong Kong Port ranks as 21st with its MI = 0.292, TEC = 1.000, SEC = 0.957 and TC = 0.305. This indicates decreased efficiency with deterioration in all three sources. All these indexes are below the average. Compared with its major competitors, the port of Singapore and Shanghai also lag behind the regional average in almost all indexes (except for Shanghai’s TC). Yet both ports are performing better than Hong Kong in technological development (TC). As for another major competitor, the port of Kaosiung, it has a better ranking of 9th and a higher TC figure of 0.711. In this sense, the port of Hong Kong may first want to improve its technological efficiency, which involves both capital investment and technology utilization. After that, it has to investigate and improve all three areas (technical, scale and technological).

Table IV has summarized the correlations between MI and its three sources. First, the pure technical efficiency seems to have the strongest impact on a port’s overall efficiency improvement (correlation = 0.85). Such pure technical efficiency gains are achieved by rationalizing port inputs to generate the maximum output, which usually involves port reforms, strategic long-term planning, and efforts to catch up with the best practices in the industry (Gosasang et al., 2018). SEC and TC seem to have a smaller impact on the overall efficiency gains and losses, with correlations equaling 0.21 and 0.24, respectively. This implies the limitation of simply adjusting port operations scales and adopting the state-of-art technologies to improve port overall efficiency. From Table V, which reports the summary statistics of the (in)efficiencies, it is worth noting that TC has the smallest variance with a standard deviation of 0.18. This implies that ports may easily assimilate with each other in the use of technologies. In this sense, capital investment in new technologies is to be strategic, yet not likely to create sustainable competitive advantage.

Limitations of the study

Apart from the sensitivity to the frontier used and the inability to test for significance of the DEA method, this paper also has several other limitations. First, in addition to inputs, other exogenous factors may influence the efficiency of container ports, such as the transport networks of the operators, the regulations of the regional governments and the general economic trends that affect demand for container transportation. The method used in this study, however, is unable to quantify and separate the efficiency changes from these exogenous factors. Following this, it can be unfair to penalize port operators with deteriorating SEC indexes as external demand is usually driven by the economic sizes and strengths of port hinterlands. According to Estache et al. (2004), it will be more appropriate to assess port efficiency changes using MI net SEC (MI* = MI/SEC). We follow this approach and report the bet MI of major Asian container ports in Table VI. Once ruling out the effect of pure scale economies and focusing on technical and technological improvements, Port Kobe seems to have outperformed other ports while Port Muara experienced the least productivity improvement during our sample period.

In addition, the DEA model used here presumes that any port different enough from the others to be the outlier and therefore identify it as the best practice. Yet this port may still be inefficient to some extent. In this sense, this study may not be able to help port operators to fully exploit efficiency gains and other research results shall be used in combination for port authorities/operators to devise proper strategies.

Furthermore, the basic requirement for reliable port efficiency analysis and benchmarking is the appropriate selection of homogenous DMUs. While our choice of limiting the sample to include only large Asian container ports may to some extent ease the concern, there is uninventable heterogeneity in terms of the tasks and objectives of the ports, market conditions they face, other institutional or cultural factors, etc. In this case, one should be cautious when interpreting and generalizing the findings of this study.

Another important limitation is that port throughputs are not decomposed into direct shipment and transhipment. The development of short-sea shipping around the Malacca Straits is phenomenal. However, most official statistics do not report the direct transhipment and transhipment separately, which makes further analysis impossible at this stage.

Finally, the data set includes container ports during 2000-2007. Either the impact from the Asian financial crisis or the economic downturn from the end of 2007 was not covered by the current study. While focusing on a period when external market is relatively stable is beneficial to efficiency analysis, given that estimated efficiency deterioration can result from shrinking market demand, instead of real deterioration in technical, scale, or technological efficiency, we do acknowledge this as a limitation of our study. Future studies on the efficiency dynamics of Asian container ports are needed, particularly for the post-2007 period.

Conclusions

This study has investigated the overall efficiency changes at 23 major Asian container ports from 2000 to 2007. It also decomposes the sources of such changes into purely technical, scale and technological efficiency gains by using the Malmquist index equation from Estache et al. (2004). Overall, these ports have improved their efficiency by 14.3 per cent with improvement in technical and scale efficiency but with deterioration in technological development. The port of Hong Kong seems to underperform in all aspects investigated. In particular, it may first improve its technological adoption as this aspect lags behind its major competitor the most (Choy et al., 2016).

In addition, this study reveals that the pure technical efficiency may have had the most direct effect on the overall efficiency improvement given a high correlation between the two (0.85). Scale and technological effect seem to have less direct impact with relatively low correlation with the overall efficiency improvement (<0.25). Furthermore, the fact that technological efficiency has little variance among the ports investigated suggests that this source may not bring substantial competitive advantage. Given the fact that port authorities and operators are given little control over the demand side, it may be more appropriate to assess port overall efficiency by looking at the total efficiency change net scale effect. After all, the Malmquist index is just one type of indicators and should be used in combination with others in decision-making.

Finally, container ports are important infrastructure that support their countries’ economic development. Although most of these ports are located along the Silk Road Economic Belt, their efficiencies vary from port to port across countries. Efficient ports will help to reduce shipping and trade costs in the region and efficiency growth of these ports will benefit the worldwide economy.