Shipping network design in a growth market: The case of Indonesia

Transportation Research Part E xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Transportation Research Part E journal homepage: www.elsevier.com/locate/tre

Shipping network design in a growth market: The case of Indonesia Ningwen Tua, Dimas Adiputrantob, Xiaowen Fub, Zhi-Chun Lia, a b

⁎

School of Management, Huazhong University of Science and Technology, Wuhan 430074, China Institute of Transport and Logistics Studies, University of Sydney, Australia

AR TI CLE I NF O

AB S T R A CT

Keywords: Shipping network design OD demand estimation Social welfare Hub port International gateway Indonesia

This paper investigates the design issues of a shipping network when cargo demand increases rapidly. A gravity-type model for origin-destination (OD) demand estimation is first presented and calibrated based on the current cargo volumes of the Indonesian maritime market. A model for maximizing total social welfare is then proposed to design the shipping network with cargo demand levels forecasted for future years. The results show that for the Indonesian maritime market, a hub-and-spoke network with fully connected hub ports is better than a network with sequentially connected hub ports in terms of total social welfare. The optimal choices for the international gateway and domestic hub ports vary as cargo demand increases over time. The results suggest that a progressive policy can be promising for infrastructure investments in developing countries: government planning and regulations may be introduced in early years to enhance infrastructure utilization and economic return. With increased demand the market may be liberalized to promote healthy competition.

1. Introduction The Republic of Indonesia, which consists of approximately 17,500 islands, is an archipelago country in Southeast Asia. According to BPS Statistics Indonesia1 (2015), Indonesia is divided into 34 administrative provinces over five main islands and four archipelagos. The country shares land borders with Malaysia, East Timor and Papua New Guinea, and marine boundaries with Singapore, Philippines and Australia. As the world’s largest archipelago country, marine shipping is a major transportation mode for Indonesia. The Indonesian president, Joko Widodo, has declared twice that he wants to transform the country into a strong maritime nation, confirming the nation’s policy priority of developing the maritime sector. However, Indonesia is still relying on neighboring countries for the distribution and logistics services of international trades. Currently, 90% of Indonesian international cargoes are transshipped through hub ports in Singapore or Malaysia (Bahagia et al., 2015). In order to develop its own international gateway ports and to achieve the vision of an economically strong maritime nation, the Indonesian government has initiated several maritime programs. One program, called Pendulum Nusantara, was proposed by the state corporation PELINDO 2 in 2012. This program plans to develop six main hub ports connected with regular shipping services, as depicted in Fig. 1 (Lino, 2012). The Pendulum Nusantara program also includes the Sorong-West Pacific Hub Port Development Project, which aims to develop the Port of Sorong into an international gateway in the West Pacific, connecting East Asia to Oceania, as shown in Fig. 2. The implementation, which started in January 2016, is expected to be finished by 2018 (Desfika, 2016). The

⁎

1

Corresponding author. E-mail address: [email protected] (Z.-C. Li). BPS Statistics Indonesia (Badan Pusat Statistik Indonesia) is a national statistics office directly under the President of the Republic of Indonesia.

http://dx.doi.org/10.1016/j.tre.2017.10.001 Received 15 January 2017; Received in revised form 17 August 2017; Accepted 2 October 2017 1366-5545/ © 2017 Elsevier Ltd. All rights reserved.

Please cite this article as: Tu, N., Transportation Research Part E (2017), http://dx.doi.org/10.1016/j.tre.2017.10.001

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

Fig. 1. Proposed main routes for “Pendulum Nusantara”. Source: Indonesian Ministry of National Development Planning, 2014.

Fig. 2. Sorong-West Pacific hub port development project. Source: Indonesian Ministry of National Development Planning, 2014.

2

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

Fig. 3. Geographic locations of 24 strategic ports. Source: Indonesian Ministry of National Development Planning, 2014.

development of the new Sorong port alone is anticipated to cost approximately IDR3.5 trillion (about US$245 million).2 Another major program is the Maritime Highway Initiative, which is an ambitious plan that consists of the development of 24 strategic ports (5 hubs and 19 feeder ports) throughout the nation, government-backed regular short sea shipping routes and the procurement of new vessels to be used on those routes. The initiative also plans to upgrade regional hubs, Belawan or Bitung, into international hubs that connect Indonesia's domestic network to international network. The Indonesian government has allocated IDR700 trillion (approximately US$49 billion) to the Maritime Highway Initiative over a course of 5 years. Fig. 3 depicts the geographic location of the strategic ports and also one possible scheme of how the feeder ports may connect to the proposed hub ports (Indonesian Ministry of National Development Planning, 2014). With great stakes for such mega-projects, a careful plan and policy assessment is needed before significant investments are made into the related infrastructure. However, the above two programs, if carried out independently, may lead to network and capacity redundancy, which would be very inefficient for a developing country, such as Indonesia. For example, no consensus has been reached with respect to the optimal development strategy for the Sorong port. The port serves as an international gateway hub as well as a domestic hub in the Pendulum Nusantara program, yet it only serves as a feeder port in the Maritime Highway Initiative. Moreover, two separate international hub developments in the country’s east region could cause unintended rivalry that might reduce the capacity utilization of both hubs. An inefficient domestic shipping network could also reduce the operational efficiency and economic benefits linked to international commodity trade (Halim et al., 2012). The Indonesian government thus needs to develop a strategic plan for the shipping network improvement that is beneficial for the country’s overall welfare. Although the literature on shipping network design is well developed, few studies have analyzed the case of Indonesia despite the huge investments involved. Cargo flow between origin-destination (OD) pairs is the most significant input data for network formulation (Bell et al., 2011, 2013; Meng and Wang, 2011; Wang et al., 2015a,b; Zheng and Yang, 2016). Data from different official resources for Indonesia are, however, not entirely consistent. In certain cases, official figures from provincial statistic agencies are inconsistent with national data. For example, the national statistics record “Statistical Yearbook of Indonesia 2015” (BPS, 2016a) reports the amount of unloaded cargo in Maluku province in 2015 to be over 1.7 million gross tons. However, Maluku’s own province record, “Maluku Dalam Angka 2015” (BPS, 2016b) states that the amount of unloaded cargo in 2015 was slightly less than 0.8 million gross tons. Such inconsistency in official records poses the challenges of determining actual cargo volumes. Although data on loaded and unloaded cargoes (gross ton) are available at the province level (BPS Statistics Indonesia, 2015), such data are aggregates of many destinations for a given port. They may not be directly usable for network design, which requires port-to-port or province-toprovince level data. Another challenge is the consideration of shipping network design in a fast-growing market. The globalization of the world’s economies contributes significantly to the fast growth of international trade and maritime shipping volumes (Tavasszy et al., 2011; Jiang et al., 2015; Tran and Haasis, 2015). According to Faisal (2015), the GDP of Indonesia has achieved a rapid growth in the past decade and is expected to further accelerate in the years to come (Indonesian Ministry of National Development Planning, 2014). The 2 IDR denotes the Indonesian Rupiah. As of 1 Dec 2016, US$1.0 approximates IDR14285.7. This exchange rate will be used for currency conversion hereafter unless specified otherwise.

3

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

cargo throughputs at Indonesian ports will grow in both international and national markets (BPS, 2015). This implies that an optimal design for the current market may not be the best choice for the future. In addition, as infrastructure investments related to both ports and ships have been quite limited in previous years, a strategic plan for Indonesia needs to take many cost items into consideration instead of focusing on a particular port or on particular shipping companies. Such challenges probably explain why few studies are available in the public domain. In addition, as reviewed in the following section, most previous studies on shipping network design have focused on the optimization at corporate level to minimize costs or maximize profits. Although they provide valuable business insights and managerial recommendations, these studies cannot be directly used to provide recommendations for public policies that aim beyond individual organizations. In light of the above, this paper addresses the future development issues of the ports and shipping networks in Indonesia at a strategic level. Such a study contributes both in modeling methodology and policy evaluation. On the methodology side, an integrated model that combines the OD cargo demand estimation and the shipping network design is proposed. The proposed model accounts for the interactions among the government, carriers, and shippers, which are formulated as three interrelated submodels, i.e., the shipper’s route choice model, the carrier’s profit maximization model and the government’s social welfare maximization model. The profit maximization model of carriers aims to optimize the cargo shipping fee, vessel type and vessel speed. The social welfare maximization model of the authority is to determine the optimal locations of hub and gateway ports for creating an efficient hub-and-spoke (HS) network. The OD demand estimation is used to calibrate the parameter in the doubly constraint model and to estimate the OD cargo demand matrix for Indonesia's shipping market. Alternative hub port connection patterns are also explored and compared such that the optimal network configuration can be identified for a specific shipping market. In general, such an integrated model can be calibrated with limited industry data to model the strategic behaviors of multiple stake-holders in realistic market conditions. Therefore, it can be a useful tool for governments to evaluate alternative government policies and investment plans. Indeed, our analysis provides rich managerial and policy insights. For the Indonesian market considered, our analysis shows that the network with fully connected hub ports would be better than that the case with sequentially connected hub ports in terms of total social welfare. Ports of Bitung and Sorong are the best choice for the international gateways at the current low level of cargo demand. However, as the cargo demand increases and reaches a threshold, the optimal gateway scheme contains only the Port of Bitung and the Port of Sorong is better to serve as a domestic hub port. For industrial policy in general, our study calls for a comprehensive analysis of all stake-holders’ decisions incorporating demand growth patterns and market dynamics. Investments in ports and fleets should be planned accordingly in view of the resultant shifting network patterns. The remainder of this paper is organized as follows. Section 2 provides a brief review of the related literature and highlights the contributions of the paper. Section 3 summarizes the market conditions, input data compilations and the method to estimate the OD cargo demand matrix. Section 4 discusses the details of model formulation for HS network design. In Section 5, the proposed models are applied to the Indonesian maritime sector. The last section concludes the paper and provides recommendations for further studies. 2. Literature review HS networks offer many benefits, such as traffic flow consolidation, large network coverage and simplified operations. They have been widely adopted by airlines, shipping companies, telecommunication systems and logistics operations (Hendricks et al., 1995, 1999; Zhang, 1996; Brueckner and Zhang, 2001; Hsu and Hsieh, 2005; Chong et al., 2006; Takano and Arai, 2008; Homsombat et al., 2011; Meng and Wang, 2011; Adler et al., 2014). Hubs serve as transshipment points and replace large quantities of direct connections with fewer indirect connections. O'Kelly (1987), one of the first to study the HS networks, developed several heuristics to solve this kind of problems. Several studies have examined waterborne networks based on the HS structure. Meng and Wang (2011) combined the HS and multiport-calling operations to study the liner shipping service network design problems, but the candidate shipping lines were predetermined. Zheng and Yang (2016) proposed a mixed-integer linear programming model to design an HS network for the Yangtze River and supported the trends of cargo concentration. In Imai et al. (2009), the multiport-calling and the HS network were compared and applied to the problem of the Asia-Europe and Asia-North America trade lanes. Hsu and Hsieh (2005) applied a two-objective model to decide on whether to route a shipment through a hub or directly to the destination for a simple network. Chong et al. (2006) proposed a heuristic procedure to solve the problem of scheduling and routing in a hybrid HS network, which included direct delivery. Takano and Arai (2008) applied a new algorithm to solve the p-hub median problem for containerized cargo transport networks. As HS networks have been extensively used and studied, and have been chosen for the Indonesian Maritime Highway program, we examine the implications of such a network configuration for the Indonesian market. We consider the 24 “strategic ports” included in the Maritime Highway program and compare alternative hub schemes for the shipping network. A large number of studies on shipping networks have been conducted. Tran and Haasis (2015) reviewed the relevant literature and concluded that network design can be carried out with alternative objectives, such as cost minimization, sailing and dwelling time minimization, shipping distance minimization, travel and transit time minimization, profit and/or revenue maximization, shipping volume maximization and other alternative objectives adopted in the industry and/or imposed by government agencies in the case of container liner shipping. Network efficiency indicators have also been used, such as Nagurney-Qiang measures, which reflect the weighted average of shipping volume per unit cost (Nagurney and Qiang, 2008). Cost minimization or profit maximization models have been widely applied to shipping problems. For example, the global-scale simulation models, namely the World Container Model (Tavasszy et al., 2011), Container World (Sinha-Ray et al., 2003), GloTram-2 (Smith et al., 2011), and the models developed for the regional and national levels by Halim et al. (2012), considered the cost minimization problems. Wang et al. (2015b) studied a profit-based container assignment problem and optimized the cargo shipping fee and the number of containers to transport. Shintani 4

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

et al. (2007) addressed the problem of designing the container liner shipping network based on a profit maximization model and considered a variety of operating costs. In this paper, we consider three types of stakeholders, namely the shipper, the carrier and the government, which are usually treated separately by most of previous studies. The interactions between these three stakeholders can be described as a multi-hierarchical system. The government aims to optimize the locations of the hubs so as to maximize the total social welfare. The carriers seek to maximize the net profit by optimizing the service quality and the shipping fee for the operating network, while the shippers minimize their total shipping cost given carriers’ service decisions. Such an integrated model, unlike most studies that focused on individual organization’s decisions, is better positioned for public policy analysis which usually targets on the maximization of the wellbeing of the whole society. As an important starting point for any network design, estimation of OD demand is of critical importance for the Indonesian maritime sector. Munizaga and Palma (2012) argued that an OD demand matrix is a fundamental prerequisite in a transport analysis for both research and policy planning purposes. However, they also emphasized that it is common to have major data gaps for reliable estimation of cargo movements. This is a major challenge in the case of container shipping in Indonesia, as there is no reliable cargo movement data at the province-to-province or port-to-port level. Many studies have tried to estimate an OD matrix or point-to-point trade volumes through using gravity type models or other mathematical solutions. To estimate a full OD matrix with limited data, Levine et al. (2009) formulated an optimization model with which a disaggregated OD demand representation can be generated with aggregated data and a gravity model. They used a case study to identify the movement of containers shipped from international ports to final destinations in the U.S. However, this method cannot be directly applied to the case of Indonesia due to significant differences in data availability and aggregation level. Luo and Grigalunas (2003) developed a gravity model to distribute U.S. containerized imports to states based on population. However, distance factor was not included in the specification. Firdaus and Widyasanti (2010) applied a general gravity model to measure the trade volumes between the domestic regions within Indonesia. Banitya (2013) applied a gravity model to identify the variables that affect Indonesia’s domestic trade and confirmed that the gross domestic regional product (GRDP) and the transport distance are important determinants of trade volume. In addition, Sheffi (1985) also proposed a doubly constrained model, which has been widely used in transportation demand forecast and trip distribution analysis (Abrahamsson and Lundqvist, 1999; Boyce and Zhang, 1998; De Grange et al., 2010). Therefore, according to the findings of previous studies, in this paper the doubly constrained model and a gravity function similar to the specification in Silva and Tenreyro (2005) will be combined to estimate the OD demand matrix for a maritime network. As different port facilities and vessels are needed for different types of cargoes (Zhuang et al., 2014), we focus on container shipping only. Compared to dry bulk and tanker transport, container shipping plays a more important role in trade and economic growth (Lau et al., 2013) and many studies have focused on containerized cargos (Shintani et al., 2007; Bensassi et al., 2015; Wang et al., 2015b; Márquez-Ramos, 2014; Wang et al., 2016). From 1990 to 2010, the container trades grew at an average rate of 8.2% per year (Unctad, 2011). Consistent with global trend of increased shipments of containerized cargo, the container shipping demand of Indonesia has also experienced a substantial growth. Syafi’i et al. (2005) showed that the average annual growth of Indonesian container shipping demand was 14.7% from 1990 to 2002. Siahaan et al. (2013) stated that the handling of containers at ports of eastern Indonesia experienced an average growth rate of 17.22% from 2000 to 2012. Faisal (2015) also indicated that the containerized cargo of Indonesia will increase with the positive GDP growth in a long period. In addition, bulk goods shipping mostly uses point-to-point networks, thus the associated planning is relatively straightforward. The following section first introduces the input data used for the analysis, followed by the specification of the models. 3. OD demand estimation and parameter calibration 3.1. Market definition and ports considered OD cargo demand matrix is the most important input for the design of shipping networks. In order to generate the OD matrix, we first introduce some basic profiles about the Indonesian ports. Both the Pendulum Nusantara and the Maritime Highway programs planned in Indonesia involve a large number of ports. Therefore, we consider shipping networks based on the 24 strategic ports that have been identified to receive funding and development support from the programs. Table 1 provides a list of the strategic ports, and Table 1 Strategic ports. Port Name

Abbrev.

Status

Port Name

Abbrev.

Status

Malahayati Belawan Batam Teluk Bayur Jambi Palembang Panjang Tanjung Priok Tanjung Emas Tanjung Perak Pontianak Sampit

P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P11 P12

Feeder Hub Hub Feeder Feeder Feeder Feeder Hub Feeder Hub Feeder Feeder

Banjarmasin Balikpapan Samarinda Pantoloan Makassar Kendari Bitung Tenau Kupang Ternate Ambon Sorong Jayapura

P13 P14 P15 P16 P17 P18 P19 P20 P21 P22 P23 P24

Feeder Feeder Feeder Feeder Hub Feeder Hub Feeder Feeder Feeder Hub Feeder

5

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

Table 2 Indonesian province-port pairs. No.

Province Name

Port

No.

Province Name

Port

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

Aceh Bali Banten Bengkulu Gorontalo Jakarta Jambi Jawa Barat Jawa Tengah Jawa Timur Kalimantan Barat Kalimantan Selatan Kalimantan Tengah Kalimantan Timur Kalimantan Utara Kepulauan Bangka Belitung Kepulauan Riau

P1 P10 P8 P4 P19 P8 P5 P8 P9 P10 P11 P13 P12 P14 P15 P6 P3

18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34

Lampung Maluku Maluku Utara Nusa Tenggara Barat Nusa Tenggara Timur Papua Papua Barat Riau Sulawesi Barat Sulawesi Selatan Sulawesi Tengah Sulawesi Tenggara Sulawesi Utara Sumatera Barat Sumatera Selatan Sumatera Utara Yogyakarta

P7 P22 P21 P20 P20 P24 P23 P2 P17 P17 P16 P18 P19 P4 P6 P2 P9

their locations are displayed in Fig. 3. To obtain the loaded and unloaded cargos for 24 strategic ports, each province is allocated to one strategic port. Following the same approach as in Faisal (2015), we assume that each province is associated with a strategic port within that province. However, if a province does not have any strategic ports, then it is allocated to the closest strategic port from its capital. This is clearly a simplification because there are many factors in addition to distance that determine shippers’ port choice, such as port charge, service quality, specialized facilities (e.g. cold chain), and the availability of liner services. For example, Bensassi et al. (2015) analyzed the relationship between logistics and trade of Spain, and concluded that the number, size and quality of logistics facilities positively influence export flows and cargo volumes. In practice, the same hinterland market may be served by multiple ports and there may be spill-over effects between ports. Márquez-Ramos (2014) estimated an augmented gravity model using exports from 19 Spanish regions to 45 countries, and found significant regional spillovers effects of port services. However, fully modeling such factors’ influences on cargo flow would require rich data at port/regional level, which is very difficult for the Indonesian market. For example, in order to estimate import OD matrix, Wang et al. (2016) compiled Port Import/Export Reporting Service data, Surface Transportation Board waybill data, and landside transport mode share data for the ports considered. When detailed and reliable data are available, more reasonable modeling approaches can be adopted, such as a multinomial logit specification that is commonly used in traffic assignment. In this current paper, however, we are forced to use this simplified approach. Data for loaded and unloaded cargoes at the ports are gathered from the provinces they serve. To avoid data inconsistency problems, all of the data used for cargo and container movements should come from the same source. Indonesia’s national records are used because they are more detailed and cover all regions. The province-port pairs are shown in Table 2. Both domestic and international cargo demand should be considered in the design of Indonesian shipping network. According to Lazuardi et al. (2017), two biggest ports, ports of Tanjung Priok (Jakarta) and Tanjung Perak (Surabaya), now serve as the domestic hub ports in Indonesia. These two ports have the functions of aggregating and transshipping both the international and domestic commodities. Currently, approximately 90% of the Indonesian international trades are transshipped through the hub ports in Singapore or Malaysia (Bahagia et al., 2015). The Port of Singapore has been used as a primary transit hub by Indonesia for its international trade, whereas the Port of Tanjung Priok has also been used as an international gateway for the trade with Oceania (Lazuardi et al., 2017). Reflecting such industry reality, the ports of Singapore and Tanjung Priok are considered as international gateways in the initial shipping network. The transportation of the international cargos can be divided into two categories: one is shipments via the international network and the other is the shipments via the Indonesian domestic network. In this paper, we focus on the domestic network for which the optimal national (Indonesia) government policy needs to be developed. The corresponding cargo volume is the sum of the international and domestic cargo demands. The loaded/unloaded international cargoes of foreign regions and Indonesian provinces can be gathered directly from the BPS Statistics Indonesia (2015). In our model, international shipping to and from other markets are aggregated to six regions, namely ASEAN, North and East Asia, the Americas, Oceania, Europe, and Africa and rest of Asia. The correspondingly representative hub ports are ports of Singapore, Shanghai, Los Angeles, Melbourne, Rotterdam, and Dubai. Using these regional major hubs to represent cargo destinations allows us to focus on network modeling within Indonesia, which is the main target of Indonesian government investment plans and public policies. 3.2. Specifications of cargo volume Although we focus on container shipping, the available data on cargo movements are mostly in gross tons. This requires a conversion of gross ton cargo to container volume. BPS Statistics Indonesia (2015) reported that in 2014, a total of 1,550,271,403 gross tons of cargos were shipped through the sea and the World Bank (2016) reported that a total of 11,900,763 TEUs (twenty-foot equivalent unit) were shipped in Indonesia. Leonardi and Browne (2010) indicated that 1 TEU carries approximately 10 tons of 6

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

Fig. 4. OD demand matrix.

cargos, which implies that container shipping accounts for approximately 7.7% of Indonesia’s sea shipping. However, PELINDO 2 (2015) stated that there were 599,425,593 gross tons of cargos and approximately 5,710,000 TEUs handled in Port of Tanjung Priok in 2015. This implies that container shipping accounts for approximately 9.5% of Tanjung Priok’s sea shipping. PELINDO 3 (2015) also reported that there were 463,851,457 gross tons of cargos and approximately 3,100,000 TEUs moving into Port of Tanjung Perak in 2015, which suggests a market share of 6.7% for container shipping in terms of tonnage. Based on these observations, the ratio of 7.7% for the overall Indonesia is between the ratio of 9.5% for Port of Tanjung Priok and that of 6.7% for Port Tanjung Perak, which means that it is feasible to transform the tonnage traffic to the container traffic by 7.7% in the model calibration. Loaded and unloaded cargo tonnage at each province is gathered from the BPS Statistics Indonesia (2015), for which the same conversion rule with ratio 7.7% is used. 3.3. OD demand estimation When both the total traffic volume generated at origin ports and the total traffic volume attracted to destination ports are known, we can adopt a doubly constrained gravity model and the likelihood estimation of parameters to estimate a reliable OD cargo demand matrix for Indonesian shipping network. A general OD demand matrix pattern is shown in Fig. 4. In Fig. 4, n is the total number of ports and qij denotes the cargo demand between OD pair (i, j). The column (O1,O2,…,On − 1,On )′ represents the total cargo volume originating at each origin port and the row (D1,D2,…,Dn − 1,Dn ) represents the total cargo volume attracted to each destination port (i.e., the sum of all elements in each row/column must be equal to the total traffic generation or attraction). The doubly constrained model has been widely used in transportation demand forecast and trip distribution analysis (see Sheffi, 1985), and is specified as follows:

qij = Kij Oi Dj exp(−γuij ) , ∀ i,j,

(1)

subject to

∑

qij = Oi , ∀ i,

(2)

j

∑

qij = Dj , ∀ j,

(3)

i

where the OD cargo demand qij is proportional to the total cargo volume originating at the origin port i, Oi , and the total cargo volume attracted to the destination port j, Dj . qij is dependent on the service level between that OD pair, exp(−γuij ) , which is a negative exponential function of the OD travel time uij . Kij is a constant, and γ is an unknown parameter to be estimated. Eqs. (2) and (3) are the conservation constraints. After some algebraic operations, the doubly constrained model (1)–(3) can further be written as

qij = Ai Bj Oi Dj exp(−γuij ) , ∀ i,j,

(4)

where

Ai =

Bj =

1 , ∀ i, ∑j Bj Dj exp(−γuij )

(5)

1 , ∀ j. ∑i Ai Oi exp(−γuij )

(6)

Substituting Eqs. (5) and (6) into (4), one can indicate that the resultant OD demand qij satisfies constraints (2) and (3). In Eq. (4), γ is an unknown parameter and needs to be calibrated. In this paper, the maximum likelihood method can be adopted. The route choices for cargo transport are assumed to follow a multinomial distribution, and the maximum likelihood function L is defined as follows. 7

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

max

L=

γ

qij (γ) ⎞Qij Q! Πi,j ⎛⎜ ⎟ , Πi,j Qij ! ⎝ T ⎠

(7)

where Qij is the number of cargos observed to be transported from port i to port j and Q is the total number of the observed cargos in the system, i.e., Q = ∑i,j Qij . The observed cargo demand in this paper is obtained from the BPS Statistics Indonesia (2015) and Silva and Tenreyro (2005). T is the total travel demand to be distributed, i.e., T = ∑i Oi = ∑j Dj . After rewriting Eq. (7) as the logarithm function and deleting the constant parts, the model to calibrate the parameter γ can be stated as follows (Boyce and Zhang, 1998).

max

lnL (γ) =

γ

∑ i,j

qij (γ) ⎞ Qij ln ⎛⎜ ⎟, ⎝ T ⎠

(8)

subject to Eqs. (4)–(6). Boyce and Zhang (1998) showed that the likelihood function (8) is quite flat and the variation in the value of γ could result in negligible changes in lnL (γ) . Therefore, the function itself does not provide decisive guidance. If the travel cost uij remains temporarily fixed, then we can use the condition, ∑i,j (Qij / Q) uij = ∑i,j (qij (γ)/ T ) uij , as the termination rule for calibrating the value of γ . The left-hand and right-hand sides of the equation represent the observed and estimated average travel cost of the system, respectively. The step-by-step procedure for estimating the value of γ in Eq. (8) is shown as below. Step 1. Choose an initial value for parameter γ , represented as γ (1) , set the iteration counter to τ = 1. Step 2. Determine {qijτ (γ (τ) )} in terms of Eqs. (4)–(6) by using furness iteration method (see Sheffi, 1985). Step 3. If |

∑i,j (qijτ (γ(τ) ) / T ) uij ∑i,j (Qij / Q) uij

−1| < ε (ε is a pre-specified precision), stop; otherwise, update the value of γ as follows and return to Step

2.

γ (τ + 1) = γ (τ)

τ (τ) ⎛ ∑i,j (qij (γ )/ T ) uij ⎞ . ⎜ ∑ (Qij / Q) uij ⎟ i,j ⎝ ⎠

(9)

The estimated and observed values are compared to produce a new value of γ for the next iteration. According to the data of base year and the procedure presented as above, the value of parameter γ can be calibrated. Applying the calibrated γ to model (4)–(6), one can estimate the OD cargo demand matrixes for future years, which are the input data for the network design problem. 4. Shipping network design model The shipping network design in this paper aims to determine an optimal HS network by locating domestic hub ports and international gateway ports. Domestic hubs have the function to agglomerate the cargos in the domestic network, whereas the international gateways agglomerate the export/import cargos before leaving/entering the domestic network. Without loss of

Fig. 5. The structure of the proposed model.

8

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

generality, the following basic assumptions are made for simplifying the modeling process. A1. The feeder ports, candidate domestic hubs and international gateways are pre-determined. A2. Three stakeholders are considered in the shipping network to be designed, namely the authority (the government), the shippers, and the carriers. The authority aims to determine the optimal locations of hub and gateway ports in the HS network. The carriers seek to optimize the cargo shipping fee, the vessel type and vessel speed to maximize the net profit. The shippers choose their routes with minimum shipping cost. The decision processes and interactions between these three stakeholders are summarized in Fig. 5. A3. Each domestic feeder port could be linked to at least one domestic hub. Two scenarios with different connection patterns among domestic hub ports are considered. The domestic hub ports could be fully connected to each other or connected in a specific sequence from the west to the east of Indonesia, as indicated in Fig. 3. Each representative international port of the regions is directly linked to a particular gateway. Therefore, the domestic network and the international network can be separately treated, and we can focus on the Indonesian domestic shipping network for which the optimal government policy can be designed. A4. An elastic demand function is adopted to capture the responses of shippers to the level of carriers’ service qualities and cargo shipping fee (Li et al., 2010; Wang et al., 2015b; Sheng et al., 2017). For presentation purposes, some notations used in this paper are defined as follows P A a Rij H G H G

set of all Indonesian ports, which consists of hub ports and feeder ports set of shipping legs a shipping leg which is defined as the direct linkage between ports, a ∈ A set of routes between OD pair (i, j) set of all hub ports set of all gateways set of candidate hub ports, H ⊆ P set of candidate gateways, G ⊆ P

4.1. Shippers' route choices A shipping network involves three types of participants: ports, carriers and shippers. Ports are the interfaces between ground transportation and waterborne transportation, where many cargo handling activities can be performed. Carriers provide transport services, whereas shippers aim to minimize the generalized shipping cost, which includes the cargo shipping fee paid to carriers and the time cost consumed during cargo transportation (Hsu and Hsieh, 2005, 2007; Talley, 2014). The transport time on route r ∈ Rij , represented as tr , includes the voyage time on the sea and handling time at the origin port, a , is determined by the transshipment ports and destination port. The voyage time (in days) of each vessel operating on leg a, tsail nautical distance and the ship’s velocity as follows:

da , ∀ a ∈ A, 24va

a tsail =

(10)

where da , measured in nautical mile, is the shipping distance of leg a. va is the speed of the vessels operating on leg a (in knot). Port handling time can be separated into two categories, namely the handling time at origin/destination ports and the handling time at hub ports. The handling time at origin/destination ports is the loading or discharging time, while the handling time at transshipment hubs is the sum of discharging and loading time, which are all proportional to the handling rate and transshipment volume. To simplify the description of the model, we calculate the handling time at ports for each leg. If k and l are, respectively, defined as the tail node and head node of leg a, the port handling time (in days) of each vessel operating on leg a can be measured by a tport =

sa sa , ∀ a = (k ,l) ∈ A, + k 24σ up 24σ ldown

(11)

l k where sa (in TEU) is the vessel size of leg a. σup (TEUs/h) is the number of containers that can be loaded per hour at port k, and σdown is the number of containers that can be discharged per hour at port l. Therefore, the shipping/transport time on route r ∈ Rij , tr , can be expressed as follows.

tr =

∑

a a (ttail + tport )δar , ∀ r ∈ Rij,i,j ∈ P ,

(12)

a∈A

where δar = 1 means route r contains leg a, and δar = 0 , otherwise. Let Ur be the total shipping cost per container traveling on route r. The minimum shipping cost Uij between OD pair (i, j) can then be expressed as

Uij = min Ur , ∀ i,j ∈ P ,

(13)

r ∈ Rij

9

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

where

Ur = fr + κtr , ∀ r ∈ Rij,i,j ∈ P ,

(14)

where fr is the cargo shipping fee per TEU traveling from port i to port j on route r. κ is the value of the shipping time. To capture the responses of shippers to the carriers’ services and cargo shipping fee, an exponential elastic demand function (Li et al., 2010) is used and specified as follows.

qij = qij0exp(−θUij ) , ∀ i,j ∈ P , where

qij0

(15)

is the potential cargo demand between OD pair (i, j), which can be obtained by the OD estimation model presented in

Section 3. If neither port i or j is a gateway port, then qij0 represents only the domestic demand. If one of the two ports, port i or j, is a gateway, qij0 is the sum of the domestic demand and the international demand passing through that gateway. θ is the demand dispersion parameter that reflects the demand sensitivity to the minimum shipping cost. Once the optimal shipping route for OD pair (i, j) is determined as r ∗, the shipping demand is then assigned to that route, i.e.,

qij, r = r ∗, qr = ⎧ ⎨ ⎩ 0, otherwise,

∀ r ∈ Rij,i,j ∈ P ,

(16)

where qr is the cargo demand on route r between OD pair (i, j). 4.2. Carriers' decisions on cargo shipping fee, vessel size and velocity Carriers determine the service quality (vessel type and velocity) and cargo shipping fee for their operating network to attract shippers to maximize their net profit. The profit of carriers is defined as the difference between the total revenue from the shippers’ cargo shipping fee and the total operating cost on all of the vessels. The total operating cost for each vessel contains three categories of costs, namely sailing-related costs, port-related costs and container-related costs. Sailing-related costs include the fuel cost during voyage and the daily charter and operating cost. The daily fuel cost per vessel is estimated to be proportional to the cube of vessel velocity and the fuel efficiency of the vessel, the vessel capacity and the fuel price (Corbett et al., 2009; Wang et al., 2015a). The specified expression for daily fuel cost of each vessel operating on leg a, CFa , (in US$) is as below.

CFa = λ μa sa va3,

(17)

where va (in knot) is the velocity of each vessel operating on leg a. sa (in ton) is the capacity of the vessel operating on leg a. μa (in ton1/2/knot3) is the fuel efficiency parameter that changes with vessel size (Notteboom and Vernimmen, 2009). λ is the fuel price, measured in US$ per ton of fuel. The daily capital and operating cost per vessel is influenced by several factors, such as crew, ship size, insurance policy and maintenance (Hsu and Hsieh, 2007; Tran, 2011). We can approximate the daily capital and operating cost for each vessel operating on leg a as follows.

CCa = α1saα2,

(18)

where α1 > 0 (in US$) is the capital cost parameter, sa (in ton) is the ship size, and 0 < α2 ⩽ 1 is the factor modeling the effect of scale economy due to vessel size. Port-related charges mainly include port dues and the stevedoring costs for containers at ports. Port dues which are paid for pilotage, towage and berth occupancy (Hsu and Hsieh, 2007; Tran, 2011), are determined by the vessel capacity. Port dues of the vessel operating on leg a, CUa , are computed as follows (Tran, 2011).

CUa = ω1sa + ω2,

(19)

where sa (in TEU) is the vessel size. ω1 and ω2 (in US$) are relative parameters. The stevedoring costs are paid for container loaded and unloaded at ports, and increase with the number of transshipment during a voyage. The total stevedoring costs per TEU at all ports on route r, CP,r , are specified as

CP,r = βi + 2

∑

βh δhr + βj , ∀ r ∈ Rij,i,j ∈ P ,

(20)

h∈H

where βi , βh and βj are, respectively, the fee charged per container at origin port i, hub port h and destination port j. δhr = 1 means hub port h ∈ H is on route r, and δhr = 0 otherwise. Additionally, container-related costs are the lease costs for the usage of containers during transportation, which are determined by the unit rent cost, the total quantity of containers used by the vessels and the whole transportation duration. Based on the above specifications, the carriers’ profit maximization model that determines the optimal cargo shipping fee, vessel size and speed is specified as follows.

max f,v,s

π(f,v,s) =

∑ ∑ i,j ∈ P r ∈ Rij

qr (fr −CP,r )−

∑

a a a [CUa Fa + CFa Fa tsail + (CL qa + CCa Fa)(tsail + tport )],

(21)

a∈A

where f, v and s are the vectors of cargo shipping fee, vessel velocity and ship size, respectively. qa = ∑i,j ∈ P ∑r ∈ Rij qr δar is the total 10

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

cargo flow on leg a. Fa = int(qa / sa) represents the round-off value of the annual number of voyages for vessel sa on leg a. CL is the lease cost per container per day. CUa , CFa , and CCa are, respectively, the daily port due, daily fuel cost and daily capital and operating cost for the vessel on specific leg a. The right-hand side of the objective function (21) contains two parts: the first part represents the difference between the total revenue from shippers’ cargo shipping fee and the total stevedoring costs at all operating routes in the network, and the second part represents the rest cost of all vessels operating on legs, including the total port dues, the fuel costs, the container rental costs and the capital and operating costs of vessels during the whole year’s voyage. 4.3. Authority’s decision for the design of HS network Based on the decisions of shippers and carriers, the government determines the optimal locations of domestic hubs and international gateways from the given sets of candidate ports to maximize the total social welfare. The total social welfare is the sum of the producer surplus (carriers’ net profit) and the consumer surplus in the whole shipping network. The network design problem can be formulated as the following maximization model.

max Φ(f (x,y),v (x,y),s (x,y)) = π(f (x,y),v (x,y),s (x,y)) + x,y

∑

qij (U (f (x,y),v (x,y),s (x,y)))/θ,

i,j ∈ P

(22)

subject to

∑

xh = N,

(23)

h∈H

∑

yg = M, (24)

g∈G

1, port h is set to be a hub, xh = ⎧ ⎨ ⎩ 0, otherwise,

∀ h ∈ H,

1, port g is set to be a gateway, yg = ⎧ ⎨ ⎩ 0, otherwise,

(25)

∀ g ∈ G,

(26)

where x and y are the decision variables. π , which is a function in x and y through f, v and s, is determined by the carrier’s profit maximization model (17)–(21). q and U, which are functions in y and x through f, v and s, are determined by the shipper’s route choice model (10)–(16). The objective function (22) includes the carriers’ network profit and the consumer surplus. Constraints (23) and (24) state that N new domestic hubs and M new gateways will be built in the network. Constraints (25) and (26) define 0–1 decision variables. In order to solve the 0–1 integer programming problem (22)–(26), we develop a heuristic solution algorithm as follows. Step 1. Initialization. Set Φ∗ = +∞ as the upper bound of the objective function Φ in Eq. (22). Step 2. First loop operation (determining optimal locations of gateway ports). Given the set of candidate gateways G , check all possible gateway schemes sequentially with one scheme at a time. Let G (1) denote the initial gateway scheme. Set the scheme counter to τ = 1. Step 3. Second loop operation (determining optimal locations of domestic hub ports). Given the set of candidate domestic hubs H , check all possible hub schemes sequentially with one scheme at a time. Let H (1) denote the initial hub scheme. Set the scheme counter to ξ = 1. Step 3.1. If all possible domestic hub schemes are checked, then go to Step 4. Step 3.2. Solve the carriers’ profit maximization model (17)–(21) to obtain the optimal cargo shipping fee, velocity and ship size solutions f (ξ) , v (ξ) and s(ξ) . Solve the shippers’ route choice model (10)–(16) to generate the route flow q(ξ) = {qr(ξ) } and the shipping cost U (ξ) = {Ur(ξ) } . Then, compute the objective value Φ(ξ) for the current hub scheme H (ξ) with the fixed gateway scheme G (τ) . Step 3.3. Termination check for the second loop operation. If Φ(ξ) > Φ∗, then put f ∗ = f (ξ) , v ∗ = v (ξ) , s∗ = s(ξ) , q∗ = q(ξ) , Φ∗ = Φ(ξ) , {x∗,y ∗} = {x (ξ),y (ξ) } , and ξ = ξ + 1, and go to Step 3.1. Otherwise, set ξ = ξ + 1 and go to Step 3.1. Step 4. Termination check for the first loop operation. If all possible gateway schemes are checked, then terminate the algorithm and output the optimal solution {f ∗,v ∗,s∗,x∗,y ∗,q∗} and the corresponding objective function value Φ∗. Otherwise, set τ = τ + 1, and go to Step 3. 5. Case study for Indonesian maritime market 5.1. Parameter calibrations and specifications The OD demand estimation model and the three-hierarchical model for shipping network design are applied to the Indonesian maritime market. Data for the GRDPs and the loaded and unloaded cargos in 2013 are collected for all provinces. After getting the observed cargo volume {Qij} , the parameter γ can be calibrated as the value of 1.059, with an average error of less than 10−3 between the estimated OD cargo demand and the observed demand data. The future OD cargo demand matrix can then be estimated using the 11

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

Table 3 Indonesian GDP for 2014 and its forecasts for 2019 and 2024 (trillion IDR). Year GDP

2014 10,542

2019 13,778

2024 19,324

Note: “IDR” stands for Indonesian currency and US$1.0 approximates IDR14285.7 on 1 December 2016.

gravity model (4)–(6) with γ value of 1.059. Besides, data for the 2014 GRDPs of provinces are also available. We first forecast the GRDPs and the cargo volumes for years 2014, 2019 and 2024 based on the observed data. According to BPS Statistics Indonesia (2015), Indonesia’s GDP in 2014 was approximately IDR10542 trillion (approximately US$737 billion) with an annual growth rate of 5.21%. BPS Statistics Indonesia (2015) forecasted an average annual growth of 5.50% over the next 5 years (i.e., 2014–2019), whereas the Indonesian Ministry of National Development Planning (2014) forecasted an average annual growth of 7.00% over the sequential period of 5 years when the proposed government development initiatives are applied nationwide (i.e., 2019–2024). The GDP forecasts for specific future years are summarized in Table 3. The annual GRDP growth rates of provinces in 2014 are obtained from BPS Statistics Indonesia (2015). With reference to the national annual growth and the regional annual growth in 2014, the growth rate of GRDP for each province from 2014 to 2019 can be predicted to be proportional to the growth rate of the national GDP (BPS Statistics Indonesia, 2016c). The cargo volume at the province level is also predicted to be proportional to the GRDP, as trade volume is significantly influenced by the overall economy. Voyage distance data are compiled for port-pairs from searates.com and Google Earth. According to the previous analysis, ports of Tanjung Priok and Tanjung Perak are two domestic hub ports in Indonesia, and the ports of Singapore and Tanjung Priok are the international gateways for the cargo transportation of Indonesia. Port of Singapore is mainly serving the international trade with Europe, Asia, Africa and the Americas. However, the Indonesian government is eager to develop its own international gateway ports to replace the Port of Singapore, so that the international cargos would be transshipped at the country’s own gateways. In our analysis, three ports may serve as the candidate international gateway ports, including the ports of Belawan, Bitung and Sorong. Furthermore, the development of the HS network for Indonesia is another policy target. We will discuss seven candidate ports for the domestic hubs, including the ports of Belawan, Batam, Tanjung Priok, Tanjung Perak, Makassar, Bitung and Sorong. A port can be a domestic hub and an international gateway at the same time. We consider all possible schemes with different numbers of new gateways and hubs. If a port is not chosen as a hub, it is then set to be a feeder port. Two scenarios with different connection patterns among hub ports are discussed. The values of the other relevant parameters are collected from previous studies. According to Lazuardi et al. (2017) and Kalem (2015), three representative vessels (i.e., small vessel, middle vessel, and big vessel) are adopted in this paper for the Indonesian shipping network. The details about the vessels operating on legs and the corresponding fuel efficiency are summarized in Table 4. The bunker fuel price is set to be λ = 375US$/ton (Tran, 2011), α1 is assumed to be 40US$/day·tonα2 and α2 = 0.6257 (Tran, 2011). The coefficient θ is assumed to be 0.0003. The lease cost per container, cL , is assumed to be US$4.5/TEU/day and the rate of depreciation or time cost related to the shipment, κ , for each container is assumed to be US$20/TEU/day (Bell et al., 2013). Wijnolst et al. (2000) showed that the port due is computed as CDa = 2.365sa + 204.828 per vessel. The handling charge (US$/TEU) per movement of container at port p, βp , the handling rate (TEUs/crane/hour) and the number of the cranes at each port are given in Table 5 (Lazuardi et al., 2017). The solution algorithms were coded in Matlab and run on a Thinkpad X1 computer with an Intel (R) Core (TM) i5 CPU (2.4-GHz) and 8 GB of RAM. 5.2. Discussions 5.2.1. The optimal hub and gateway schemes for two scenarios In the case study of Indonesian shipping network design, we investigate two scenarios: Scenario 1 – the domestic hub ports are fully connected to each other, and Scenario 2 – the domestic hub ports are connected sequentially from the west to the east of Indonesia. The corresponding results are discussed as follows. Tables 6 and 7 display how the optimal domestic hub schemes of two scenarios change with three cargo demand levels, respectively, for 2014, 2019 and 2024 when only the Indonesian domestic market is considered. In Table 6, the domestic hub ports are connected to each other. It can be noted that there is a difference in the optimal hub schemes for 2014, 2019, and 2024. Specifically, the optimal hub schemes include five ports for both 2014 and 2019, namely Belawan (port 2), Tanjung Priok (port 8), Tanjung Perak (port 10), Makassar (port 17), and Sorong (port 23). The corresponding total social welfares (including both consumer surplus and carriers’ profits) are US$12.477 billion and US$ 16.221 billion, respectively. However, in 2024, the optimal network needs six hub ports, namely ports 2, 8, 10, 17, 23 and Bitung (port 19), with a total social welfare of US$ 22.691 billion. These results indicate that for the only Indonesian domestic market, the total number of optimal hub ports may increase with growing cargo demand. In Table 7, the hub ports are connected by a specific sequence (i.e., from west to east). It can be observed that for all the three demand levels, the Table 4 Vessel types and the corresponding value of fuel efficiency. Vessel type (TEU) Fuel efficiency (10−5 ton1/2/knot3)

750 15.90

12

1600 8.84

3750 5.41

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

Table 5 Handling rates and handling charges for the movement of containers at ports. Port

Handling charges (US$/TEU. movement)

Belawan Betam Tanjung Priok Tanjung Perak Makassar Bitung Sorong Other ports

25.93 20.00 25.59 25.59 29.63 32.59 30.74 20.00

Cargo handling rate

Number of cranes

Loading (TEU/crane/hour)

Unloading (TEU/crane/hour)

18 16 20 20 20 18 18 16

20 18 22 22 22 20 20 18

3 2 4 4 4 3 3 2

Table 6 Optimal domestic hub schemes with hub ports fully connected to each other when only domestic market is considered. Year

Optimal domestic hub scheme

Total social welfare (billion US$)

2014 2019 2024

[2, 8, 10, 17, 23] [2, 8, 10, 17, 23] [2, 8, 10, 17, 19, 23]

12.477 16.221 22.691

Table 7 Optimal domestic hub schemes with hub ports sequentially connected when only domestic market is considered. Year

Optimal domestic hub scheme

Total social welfare (billion US$)

2014 2019 2024

[2, 8, 10, 17, 19] [2, 8, 10, 17, 23] [2, 8, 10, 17, 23]

12.431 16.181 22.670

optimal hub schemes contain five ports. At lower demand level (i.e., 2014), the five hubs are ports 2, 8, 10, 17, 19, while at higher demand levels (i.e., 2019 and 2024), Port of Sorong (port 23) outperforms the Port of Bitung (port 19) and the optimal domestic hubs change to ports 2, 8, 10, 17 and 23. In addition, it can also be noted that ports of Belawan, Tanjung Priok, Tanjung Perak and Makassar are usually chosen as the domestic hubs because the regions served by these ports have large population size, high GRDPs, and large port throughputs. This is consistent with the fact that ports of Tanjung Priok and Tanjung Perak are currently serving as the domestic hubs. Tables 8 and 9 indicate the optimal Indonesian network configurations and the corresponding total social welfare when both domestic and international cargo demands are considered. In Table 8, the domestic hub ports are connected to each other. It can be noted that the optimal choice of the international gateway ports and domestic hub ports vary with growing demand. Specifically, at the demand levels of 2014 and 2019, developing two international gateways (ports of Bitung and Sorong) and six domestic hubs (ports of Belawan, Batam, Tanjung Priok, Tanjung Perak, Makassar and Bitung) is the best choice for the Indonesian shipping network, with the total social welfare of US$ 19.361 billion and US$ 25.503 billion, respectively. However, with growth in cargo demand, the corresponding total number of the optimal gateways decreases while the number of domestic hub ports increases. In 2024, the optimal gateway scheme contains only the Port of Bitung (port 19), while Port of Sorong (port 23) changes to the seventh domestic hub port. It indicates that at the higher demand level, Port of Sorong is better to act as a domestic hub than as a gateway. Table 9 shows the optimal schemes when domestic hub ports are connected by a specific sequence. In this scenario, the optimal scheme with one international gateway (Port of Bitung) and five domestic hub ports (ports of Belawan, Tanjung Priok, Tanjung Perak, Makassar and Sorong) is always the best choice for three demand levels. Comparing Tables 6 and 7, or Tables 8 and 9, it can be observed that the total social welfare of the network with fully connected hub ports (Scenario 1) is always higher than that of the network with sequentially connected hub ports (Scenario 2). Thus, the former network structure is beneficial for the whole system in terms of the social welfare. However, the optimal gateway and domestic hub

Table 8 Optimal network structure with hub ports fully connected to each other when both international and domestic markets are considered. Year

Optimal gateway scheme

Optimal domestic hub scheme

Total social welfare (billion US$)

2014 2019 2024

Bitung+Sorong Bitung+Sorong Bitung

[2, 3, 8, 10, 17, 19] [2, 3, 8, 10, 17, 19] [2, 3, 8, 10, 17, 19, 23]

19.361 25.503 36.160

13

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

Table 9 Optimal network structure with hub ports sequentially connected when both international and domestic markets are considered. Year

Optimal gateway scheme

Optimal domestic hub scheme

Total social welfare (billion US$)

2014 2019 2024

Bitung Bitung Bitung

[2, 8, 10, 17, 23] [2, 8, 10, 17, 23] [2, 8, 10, 17, 23]

19.292 25.390 35.978

Table 10 Total social welfare for the initial Indonesian network at three demand levels. Year Total social welfare (billion US$)

2014 19.154

2019 25.160

2024 35.612

schemes of the network with fully connected hub ports are more variable with the growing demand than those with sequentially connected network. It can be observed that if the hub ports of the network are connected in a sequence, the current optimal network configuration may also be suitable for the next decade. Furthermore, Table 10 shows the results of the original network for three demand levels. It can be confirmed that developing new international gateways and domestic hubs can improve the total social welfare of Indonesia.

5.2.2. Sensitivity analysis of model parameters As our network configurations are not entirely endogenous, additional tests are carried out to test whether they are robust to other factors. The proposed model is applied to compute the total social welfare of Indonesian shipping network for two scenarios, with the ports of Belawan (port 2), Tanjung Priok (port 8), Tanjung Perak (port 10), Makassar (port 17) and Bitung (port 19) served as the domestic hubs and the ports of Tanjung Priok and Bitung served as the international gateways. Fig. 6 reports the impact of handling rates σ on the total social welfare for the Indonesian network. The horizontal axis represents the decreasing/increasing number of cranes at ports at the same time, and the value of zero denotes that the number of cranes at each port takes the initial value of Table 5. The more the cranes, the faster the stevedoring speed. Improved handling rates (i.e., adding additional cranes) at ports reduce the dwelling time at the origin port, transshipment hubs and destination port, thus reducing the associated operation costs. Reduced transport time also reduces shippers’ time cost and attracts more cargo demand. These benefits are confirmed as shown in Fig. 6, where the total social welfares at different cargo volumes increase with the handling rate. Fig. 6 also shows that the total social welfare of the network with fully connected hub ports is higher than that of the network with sequentially connected hub ports for 2014, 2019 and 2024. The social welfare differences are presented in Fig. 7. Note that at three demand levels, although the social welfare differences between Scenario 1 and Scenario 2 decrease for some specific handling rates, the differences are always positive. Therefore, the network structure with fully connected hub ports may perform better than the other kind of network structure, regardless of the handling rate. Fig. 8 illustrates the changes of the system’s total social welfare with different values of α2 (0 < α2 ⩽ 1), which is a factor describing the scale economy associated with vessel size when computing capital and operating costs. It is noted that as α2 increases (i.e., equivalently, the scale economy decreases), the total social welfare marginally decreases regardless of the connection patterns among hub ports. This implies significant welfare rises when large ships replace small ships, but such a benefit diminishes at large vessel sizes. In order to further identify the difference between two kinds of networks, Fig. 9 displays the total social welfare

Fig. 6. Effects of the handling rate (the number of cranes) on total social welfare of Indonesian maritime network: fully connected hub ports (Scenario 1) vs. sequentially connected hub ports (Scenario 2).

14

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

Fig. 7. Effects of the handling rate (the number of cranes) on difference in total social welfare of Indonesian maritime network: fully connected hub ports vs. sequentially connected hub ports.

Fig. 8. Effects of scale economy factor α2 on total social welfare of Indonesian maritime network: fully connected hub ports vs. sequentially connected hub ports.

Fig. 9. Effects of scale economy factor α2 on difference in total social welfare of Indonesian maritime network: fully connected hub ports vs. sequentially connected hub ports.

difference between two network structures with different values of α2 . It can also been seen that at the three demand levels, the welfare differences are positive and increase with α2 . Although the sequentially connected hub ports (Scenario 2) can save more cost through agglomeration of traffic volumes at hubs and thus larger vessels can be used, the longer transportation time will decrease the cargo demand, and increase the total operating cost. However, such a cost saving advantage for Scenario 2 diminishes in future years with growing demand. Therefore, the welfare difference at a high demand level in 2024 increases faster than that at a low demand level (e.g., Years 2014 and 2019). Fig. 10 displays the impact of the fuel price λ on the total social welfare at the demand level of 2014. It is noted that the total social welfare shows a decreasing trend with the increasing fuel price for both Scenarios 1 and 2. This is because the rise of the fuel 15

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

Fig. 10. Effects of fuel price λ on total social welfare of Indonesian maritime network: fully connected hub ports vs. sequentially connected hub ports.

Fig. 11. Effects of fuel price λ on difference in total social welfare of Indonesian maritime network: fully connected hub ports vs. sequentially connected hub ports.

price will increase the operating cost of carriers. In order to compensate the extra cost, the carriers will raise the shipping fee and adjust the service decisions, which may lead to a decrease in the shipping demand and thus in the total social welfare. It can also be noted in Fig. 10 that the difference in the social welfare between the network with fully connected hub ports and the network with sequentially connected hub ports increases with increasing fuel price. Fig. 11 shows the social welfare difference for three demand levels. It can be seen that the difference is positive, and the higher the cargo demand, the larger the social welfare difference. Therefore, the network with fully connected hub ports would be better than that with sequentially connected hub ports in terms of social welfare, regardless of the fuel price. 6. Concluding remarks As a major archipelago country with a huge population, substantial investment in Indonesia’s maritime sector is needed to promote the nation’s trade and economic growth. Despite the huge stakes involved, there may be significant inconsistency between strategic government plans, such as the Pendulum Nusantara and Maritime Highway Initiatives. Large investments in the maritime sector usually take a very long time to finish, during which rapid growth in cargo volumes is expected for Indonesia. All of these issues call for careful planning by the country’s maritime sector and government policy. However, few studies have investigated the optimal design of the shipping network in Indonesia and even less is known about how government policies and investment plans should adapt to changing demand over time. This paper aims to fill these gaps in research and policy planning by conducting a comprehensive study of shipping network design. To overcome the severe shortage of detailed cargo flow data, a doubly constrained gravity model combined with the parameter estimation procedure is first applied to the Indonesian markets to calibrate a full OD cargo demand matrix for the current and future markets. An integrated model, which characterizes the shippers’ route choice, the carriers’ profit maximization and the government’s total social welfare maximization, is then developed. The proposed model considers both domestic and international cargo flows to determine the optimal domestic hubs and international gateways for the Indonesian maritime market. Some important findings and new insights have been obtained. First, the network with fully connected hub ports performs better than the network with sequentially connected hub ports, in terms of the total social welfare. Second, the optimal international gateway and domestic hub scheme will change with network configuration and cargo demand. Specifically, when the domestic hub ports are fully connected to each other, two international gateways (ports of Bitung and Sorong) are the best choice for relatively low demand levels (e.g., in 2014 and 2019). The corresponding domestic hub scheme contains ports of Belawan (port 2), Batam (port 3), 16

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

Tanjung Priok (port 8), Tanjung Perak (port 10), Makassar (port 17) and Bitung (port 19). With a rapid growth in cargo demand, however, Port of Sorong changes to act as a domestic hub port at high demand levels (e.g., in 2024). When the domestic hub ports are connected in a specific sequence, one additional gateway port (Port of Bitung) and five domestic hub ports (Belawan, Tanjung Priok, Tanjung Perak, Bitung and Sorong) is the best choice for the next decade. Third, ports of Belawan, Tanjung Priok, Tanjung Perak and Makassar have good potential to become the domestic hubs because they serve regions that have large population size, high GRDPs, and large cargo throughputs. In addition to these detailed recommendations for shipping network design in Indonesia, our study also reveals the importance of considering market dynamics and network structures in strategic planning and government policy. One major policy decision is whether government intervention and planning should be imposed. Government intervention may avoid duplicate investments and thus increase the utilization and return of infrastructure investment. However, market-based mechanisms tend to be more responsive and efficient, bringing competition and innovations in the long term. Our analysis suggests that the optimal decision may evolve dynamically with market conditions. For the case of the Indonesian maritime sector, if the hub ports are fully connected to each other, two international gateways and six domestic hubs are needed for relatively low demand level. With increased traffic volumes, one international gateway port changes to act as the domestic hub port. These modeling results suggest that a progressive policy may be promising for developing countries that are usually short of capital. In the early stages, when demand is relatively low, government intervention and planning can avoid duplicate investments and promote operational efficiency. However, as traffic volume and demand increase over time, it may be optimal to liberalize the maritime sector and promote healthy competition between ports and regions. We also highlight the interactive dynamics between port operations and shipping networks. As shown in our sensitivity tests in Figs. 6–9, the increased handling rates and the scale economy of large ships will affect the network configuration and thus the throughput and transshipment volumes at ports. Therefore, government policy and planning should be both long term and comprehensive. Finally, because the OD cargo flows are one of the key determinants of optimal shipping networks, it is important for government to compile more detailed data. Compared to advanced economies, statistical agencies in developing countries often compile less detailed data. This can be an expensive mistake, as more infrastructure investments and associated plannings are often needed in developing countries. Although we have tried to conduct a comprehensive study using real market data, some simplifying assumptions and model calibrations have been imposed due to the lack of some critical data. The network configuration is not entirely endogenous and we have not considered more complex hybrid networks. Although our modeling results suggest that a progressive government policy is promising, in reality government interventions often give rise to corruption and bureaucracy, especially in developing countries. These issues, however, are difficult to model in quantitative analysis. Our study is a step toward better planning and policymaking, yet more advanced studies should be carried out when more detailed data are available. Such efforts are particularly important for the maritime sector, in which infrastructure development involves substantial investment over extended periods. Acknowledgments We are grateful to three anonymous referees for their helpful comments and suggestions on an earlier version of the paper. The work described in this paper was jointly supported by grants from the National Natural Science Foundation of China (71525003, 71222107), the Major Program of National Social Science Foundation of China (13 & ZD175), Huazhong University of Science and Technology (5001300001), the Fundamental Research Funds for the Central Universities (2017KFKJXX003), University of Sydney and the Sydney Southeast Asia Center, Australia. References Abrahamsson, T., Lundqvist, L., 1999. Formulation and estimation of combined network equilibrium models with applications to Stockholm. Transp. Sci. 33 (1), 80–100. Adler, N., Fu, X., Oum, T.H., Yu, C., 2014. Air transport liberalization and airport slot allocation: the case of the Northeast Asian transport market. Transp. Res. Part A 62, 3–19. Bahagia, S.N., Sandee, H., Meeuws, R., 2015. State of Logistics Indonesia 2015. State Secretariat for Economic Affairs SECO Switzerland, Jakarta. Banitya, R.R., 2013. Gravity model of trade in indonesian market. Paper presented at the 6th Conference of Indonesian Students Association in Korea, Daejeon, July 7, 2013. Bell, M.G.H., Liu, X., Angeloudis, P., Fonzone, A., Hosseinloo, S.H., 2011. A frequency-based maritime container assignment model. Transp. Res. Part B 45 (8), 1152–1161. Bell, M.G.H., Liu, X., Rioult, J., Angeloudis, P., 2013. A cost-based maritime container assignment model. Transp. Res. Part B 58 (3), 58–70. Bensassi, S., Márquez-Ramos, L., Martínez-Zarzoso, I., Suárez-Burguet, C., 2015. Relationship between logistics infrastructure and trade: evidence from Spanish regional exports. Transp. Res. Part A: Policy Pract. 72, 47–61. Boyce, D.E., Zhang, Y., 1998. Parameter estimation for combined travel choice models. In: Lundqvist, L., Mattsson, L.G., Kim, T.J. (Eds.), Network Infrastructure and the Urban Environment. Springer-Verlag, Berlin, Heidelberg, pp. 177–193. Brueckner, J.K., Zhang, Y., 2001. A model of scheduling in airline networks: how a hub-and-spoke system affects flight frequency, fares and welfare. J. Transp. Econ. Policy 35 (2), 195–222. BPS – Statistics Indonesia, 2015. Statistical Yearbook of Indonesia 2015, catalogue no. 1101001. BPS – Statistics Indonesia, Jakarta. BPS – Statistics Indonesia, 2016a. Statistical Yearbook of Indonesia 2015, catalogue no. 1101001. BPS – Statistics Indonesia, Jakarta. BPS – Statistics Indonesia, 2016b. Maluku in Figures 2015, catalogue no. 1102001.81. BPS – Statistics Maluku, Ambon. BPS – Statistics Indonesia, 2016c. [Seri 2010] Laju Pertumbuhan Produk Domestik Regional Bruto per Kapita Atas Dasar Harga Konstan 2010 Menurut Provinsi, 20112014, viewed April 3, 2016. < https://www.bps.go.id/linkTableDinamis/view/id/964 > . Chong, L., Kennedy, D., Chan, W.M., 2006. Direct shipping logistic planning for a hub-and-spoke network with given discrete intershipment times. Int. Trans. Oper. Res. 13 (1), 17–32. Corbett, J.J., Wang, H., Winebrake, J.J., 2009. The effectiveness and costs of speed reductions on emissions from international shipping. Transp. Res. Part D 14 (8),

17

Transportation Research Part E xxx (xxxx) xxx–xxx

N. Tu et al.

593–598. De Grange, L., Fernández, E., De Cea, J., 2010. A consolidated model of trip distribution. Transp. Res. Part E 46 (1), 61–75. Desfika, T.S., 2016. Pembangunan Pelabuhan Sorong Dimulai Januari 2016, Berita Satu, November 17, viewed March 15. < http://www.beritasatu.com/ekonomi/ 322774-pembangunan-pelabuhan-sorong-dimulai-januari-2016.html > . Faisal A., 2015. Designing National Freight Maritime Network in Indonesia: A Supporting Study for Maritime Highway Initiatives (Kebijakan Tol Laut) in Some Future Scenarios. Master thesis. Faculty of Technology, Policy, and Management, Delft University of Technology, viewed 10 March 2016. < http://repository.tudelft.nl/ view/ir/uuid%3A73ea38f6-3f5c-44bf-9551-5330c1979306/ > . Firdaus, M., Widyasanti, A.A., 2010. Indonesian interprovincial trade: what can be revealed from a gravity modeling? Paper presented at the 10th Indonesian Regional Science Association International Conference, Surabaya, July 28–29, 2010. Halim, R.A., Tavasszy, L.A., Seck, M.D., 2012. Modeling the global freight transportation system: a multi-level modeling perspective. In: Proceedings of the 2012 Winter Simulation Conference, pp. 1–13. Hendricks, K., Piccione, M., Tan, G., 1995. The economics of hubs: the case of monopoly. Rev. Econ. Stud. 62 (1), 83–99. Hendricks, K., Piccione, M., Tan, G., 1999. Equilibria in networks. Econometrica 67 (6), 1407–1434. Homsombat, W., Lei, Z., Fu, X., 2011. Development status and prospects for aviation hubs—a comparative study of the major airports in Southeast Asia. Singapore Econ. Rev. 56 (4), 573–591. Hsu, C.I., Hsieh, Y.P., 2005. Direct versus terminal routing on a maritime hub-and-spoke container network. J. Mar. Sci. Technol. 13 (3), 209–217. Hsu, C.I., Hsieh, Y.P., 2007. Routing, ship size, and sailing frequency decision-making for a maritime hub-and-spoke container network. Math. Comput. Model. 45 (7/ 8), 899–916. Imai, A., Shintani, K., Papadimitriou, S., 2009. Multi-port vs. hub-and-spoke port calls by containerships. Transp. Res. Part E 45 (5), 740–757. Indonesian Ministry of National Development Planning, 2014. Pengembangan Tol Laut Dalam Rpjmn 2015–2019 dan Implementasi 2015. Directorate of Transportation, Jakarta, viewed February 29, 2016. Jiang, J., Lee, L.H., Chew, E.P., Gan, C.C., 2015. Port connectivity study: an analysis framework from a global container liner shipping network perspective. Transp. Res. Part E 73, 47–64. Kalem, H., 2015. Container liner-shipping network design using a path formulation model for Indonesia. Master thesis. Erasmus University Rotterdam. Retrieved from < http://hdl.handle.net/2105/31037 > . Lau, J., Ng, A., Fu, X., Li, K., 2013. The evolution and research trend of container shipping. Marit. Policy Manage. 40 (7), 654–674. Lazuardi, S.D., van Riessen, B., Achmadi, T., Hadi, F., Mustakim, A., 2017. Analyzing the national logistics system through integrated and efficient logistics networks: a case study of container shipping connectivity in Indonesia. Appl. Mech. Mater. 862, 238–243. Lino, R.J., 2012. Transporting Light to the Nation: Indonesia Maritime Infrastructure, World Export Development Forum 2012, at Jakarta, October 15. Li, Z.C., Lam, W., Wong, S.C., Fu, X., 2010. Optimal route allocation in a liberalizing airline market. Transp. Res. Part B 44 (7), 886–902. Leonardi, J., Browne, M., 2010. A method for assessing the carbon footprint of maritime freight transport: European case study and results. Int. J. Logist. Res. Appl. 13 (5), 349–358. Levine, B., Nozick, L., Jones, D., 2009. Estimating an origin-destination table for us exports of waterborne containerised freight. Transp. Res. Part E 45 (4), 611–626. Luo, M., Grigalunas, T.A., 2003. A spatial-economic multimodal transportation simulation model for us coastal container ports. Marit. Econ. Logist. 5 (2), 158–178. Meng, Q., Wang, S., 2011. Liner shipping service network design with empty container repositioning. Transp. Res. Part E 47 (5), 695–708. Márquez-Ramos, L., 2014. Port facilities, regional spillovers and exports: empirical evidence from Spain. Pap. Reg. Sci. 95 (2), 329–351. Munizaga, M.A., Palma, C., 2012. Estimation of a disaggregate multimodal public transport origin–destination matrix from passive smartcard data from Santiago, Chile. Transp. Res. Part C 24 (9), 9–18. Nagurney, A., Qiang, Q., 2008. An efficiency measure for dynamic networks modeled as evolutionary variational inequalities with application to the Internet and vulnerability analysis. Netnomics 9 (1), 1–20. Notteboom, T.E., Vernimmen, B., 2009. The effect of high fuel costs on liner service configuration in container shipping. J. Transp. Geogr. 17 (5), 325–337. O'Kelly, M.E., 1987. A quadratic integer program for the location of interacting hub facilities. Eur. J. Oper. Res. 32 (3), 393–404. PELINDO 2, 2015. Annual Report 2014, Jakarta. PELINDO 3, 2015. Annual Report 2014, Surabaya. Sheffi, Y., 1985. Urban Transportation Networks: Equilibrium Analysis with Mathematical Programming Methods. Prentice-Hall, Inc, Englewood Cliffs, New Jersey. Sheng, D., Li, Z.C., Fu, X., Gillen, D., 2017. Modeling the effects of unilateral and uniform emission regulations under shipping company and port competition. Transp. Res. Part E 101, 99–114. Shintani, K., Imai, A., Nishimura, E., Papadimitriou, S., 2007. The container shipping network design problem with empty container repositioning. Transp. Res. Part E 43 (1), 39–59. Siahaan, D., Jinca, M.Y., Wunas, S., Pallu, M.S., 2013. Container sea transportation demand in eastern Indonesia. Int. Refer. J. Eng. Sci. 2 (9), 19–25. Silva, J.M.C.S., Tenreyro, S., 2005. The log of gravity. Rev. Econ. Stat. 88 (4), 641–658. Sinha-Ray, P., Carter, J., Field, T., Marshall, J., Polak, J., Schumacher, K., Song, D., Woods, J., Zhang, J., 2003. Container world: global agent-based modelling of the container transport business. In: Proceedings of 4th International Workshop on Agent-Based Simulation. Smith, T., Barrett, M., Parker, S., Eoin, O., 2011. Description of GloTraM-2, a Prototype Model to Calculate the Future Energy Emissions and Costs of the Global Shipping Industry. Transport and Logistic Group, Delft University of Technology. Syafi'i, Kuroda, K., Takebayashi, M., 2005. Forecasting the demand of container throughput in Indonesia. Memoirs Constr. Eng. Res. Inst. 47, 227–231. Takano, K., Arai, M., 2008. A genetic algorithm for the hub-and-spoke problem applied to containerized cargo transport. J. Mar. Sci. Technol. 14 (2), 256–274. Talley, W.K., 2014. Maritime transport chains: carrier, port and shipper choice effects. Int. J. Prod. Econ. 151, 174–179. Tavasszy, L., Minderhoud, M., Perrin, J.F., Notteboom, T., 2011. A strategic network choice model for global container flows: specification, estimation and application. J. Transp. Geogr. 19 (6), 1163–1172. Tran, N.K., 2011. Studying port selection on liner routes: an approach from logistics perspective. Res. Transp. Econ. 32 (1), 39–53. Tran, N.K., Haasis, H.D., 2015. Literature survey of network optimization in container liner shipping. Flex. Serv. Manuf. J. 27 (2), 139–179. Unctad, 2011. Review of maritime transport. United Nations, New York. Wang, H., Gearhart, J., Jones, K., Frazier, C., Nozick, L., Levine, B., Jones, D., 2016. Estimation of an origin–destination table for US imports of waterborne containerized freight. Transp. Res. Rec.: J. Transp. Res. Board 2548, 35–42. Wang, K., Fu, X., Luo, M., 2015a. Modeling the impacts of alternative emission trading schemes on international shipping. Transp. Res. Part A 77, 35–49. Wang, S., Liu, Z., Bell, M.G.H., 2015b. Profit-based maritime container assignment models for liner shipping networks. Transp. Res. Part B 72, 59–76. Wijnolst, N., Waals, F., Bello, F., Gendroneau, Y., Kempen, D.V., 2000. Malacca-Max 2: Container Shipping Network Economy. DUP, Delft. World Bank, 2016. Container port traffic (TEU: 20 foot equivalent units). World Bank, viewed 10 March 2016. < http://data.worldbank.org/indicator/IS.SHP. GOOD.TU > . Zhang, A., 1996. An analysis of fortress hubs in network-based markets. J. Transp. Econ. Policy 30, 293–308. Zheng, J., Yang, D., 2016. Hub-and-spoke network design for container shipping along the Yangtze River. J. Transp. Geogr. 55, 51–57. Zhuang, W., Fu, X., Luo, M., 2014. A game theory analysis of port specialization-implications to the Chinese port industry. Marit. Policy Manage. 41 (3), 268–287.

18