Analysis of mode choice variables in short-distance intermodal freight transport using an agent-based model

Transportation Research Part A 61 (2014) 100–120

Contents lists available at ScienceDirect

Transportation Research Part A journal homepage: www.elsevier.com/locate/tra

Analysis of mode choice variables in short-distance intermodal freight transport using an agent-based model Vasco Reis ⇑ Centre for Urban and Regional Systems, Instituto Superior Técnico, University of Lisbon, Portugal

a r t i c l e

i n f o

Article history: Received 20 December 2012 Received in revised form 6 December 2013 Accepted 3 January 2014

Keywords: Mode choice Short distance Intermodal transport Freight Agent-based modelling

a b s t r a c t Medium to long-distance intermodal transport has been strongly promoted by the European Commission and national governments as a solution for ensuring the sustainability of the freight transport sector. However, so far, intermodal transport has revealed limited capacity for competing against road transport. New solutions aimed at expanding the limits of its competitiveness are needed. Some successful cases of short-distance intermodal transport reveal untapped market opportunities. The literature on mode choice fails to explain these successes. The research has focused mainly on long-distance services, and the findings are not necessarily transferable to the short-distance transport. This paper presents the results of research aimed at testing this assumption. A new agent-based model to simulate the transport operations and behavioural reactions of transport agents was developed, applying mode choice variables that are consensually referred to as pivotal in the mode choice process: price, transit time, reliability and flexibility. The use of the model was to ascertain the performance of competing transport modes (intermodal and road) under different demand scenarios. Applications of the model to a short-distance transport service show that only price could explain the Freight Forwarder choice for intermodality. The evidence produced by this research suggests that the mode choice process for short-distance transport services may be governed by other decision variables and that current intermodality-oriented policy options should be revised, as they exclude a potential market segment. Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction At the European Union (EU) level, intermodal transport1 has received strong political support over the last two decades (Konings et al., 2008). By 1995 the European Commission (EC) had already recognised the unsustainable growth of road transport and called for a significant modal shift (European Commission, 1995). Later, in 1999, an important advancement was made with the Treaty of Amsterdam, which, for the first time, included sustainability as a key objective in the EU’s development paradigm. The EU’s first sustainable development strategy (European Commission, 2001a), launched at the 2001 Gothenburg Summit, reinforced the need to limit the growth of road transport and called for a modal shift towards rail or waterborne transport. That same year the EC’s 2001 White Paper on Transport proposed a set of measures to attain this objective (European Commission, 2001b). Throughout the decade, other policy documents – including the 2006 Mid-Term Review of the 2001 White Paper

⇑ Address: DECivil – Instituto Superior Técnico, Av. Rovisco Pais 1, 1049-001 Lisbon, Portugal. Tel.: +351 218 418 321. E-mail address: [email protected] Intermodal transport is defined as ‘‘the movement of goods (in one and the same loading unit or vehicle) by successive modes of transport without handling of the goods themselves when changing modes’’ (United Nations, 2001). 1

0965-8564/$ - see front matter Ó 2014 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.tra.2014.01.002

V. Reis / Transportation Research Part A 61 (2014) 100–120

101

on Transport (European Commission, 2006), the 2007 EU’s Freight Transport Agenda (European Commission, 2007) or the 2009 EC’s Communication on the Future of Transport (European Commission, 2009) – advocated intermodal transport as a means of curbing the unsustainable growth of the transport sector, largely driven by road transport, without jeopardising the economic, social and sustainable development of the EU. Recently, the 2011 White Paper on Transport (European Commission, 2011) underpinned the relevance of intermodal transport for achieving the long-term sustainability of the EU’s transport system. Notwithstanding, at the EU level over 75% of freight (measured in tonne-kilometres) is still transported by road (Cloodt, 2012), intermodal transport accounts for only approximately 5% of total freight (Savy, 2009). The reasons for the difficulties facing intermodal transport are well documented in the literature (Blauwens et al., 2006a; Button, 2010; Frémont and Franc, 2010). Firstly, there is the inherent complexity of producing intermodal transport services (Woxenius, 1998), since it involves streamlining five different types of flows – physical, logical, contractual, financial and relational (Reis, 2010) – between multiple transport agents. Secondly, there are multiple external barriers, including: inadequate regulatory framework (Slack, 2001), absence of an intermodal liability regime (Asariotis, 1999) and lack of integration between the transport networks (Leinbach and Capineri, 2006). These complexities and the barriers contribute to hampering efficiency, raising production costs and reducing market opportunities for intermodal services (Rich et al., 2011). Thirdly, narrow policy options focused on the promotion of medium to long-distance intermodal transport. By way of example, the 2011 White Paper on Transport foresees a modal shift of around 30% from road freight to other modes such as rail or waterborne transport by 2030, but only on distances above 300 km. Captivity by road transport below this threshold is implicitly assumed (European Commission, 2011). Improving the intermodal transport market positioning requires an increase in competitiveness and the entry into new markets. Considering that, EU wide, around 50% of freight transport demand is for distances up to 400 km (EUROSTAT, 2012), short-distance services may represent an untapped opportunity. Successful cases of short distance intermodal transport reinforce the likely existence of such opportunities (Macharis et al., 2010 or the case presented in this paper). The promotion of short-distance intermodal transport faces, however, significant challenges. Empirical evidence sets the limits of competitiveness at around 400 km (Tsamboulas, 2008). Also, there is a scarcity of literature on the mode choice behaviour for short-distance intermodal transport. The available literature mainly describes investigations concerning medium to long-distance services (e.g., Janic, 2007 or Tsamboulas et al., 2007) and the EU co-funded research projects primarily use case studies on long distance services and occasionally medium-distance ones.2 The demonstration that mode choice behaviour is independent of distance is still to be made. Therefore, the validity of deploying the findings to short-distance services is questionable. The paper presents an investigation aimed at assessing whether mode choice variables used in medium to long-distance transport services can be used to explain agent behaviour in short-distance transport cases. If the hypothesis proves false then we may contend that a gap in the literature concerning the mode choice variables over the short distance may exist. Such a gap may imply that further research is required. A running intermodal transport service was used as a case study. The performance of the intermodal transport service was compared against a potential road transport service for a set of mode choice variables, including: price, transit time, reliability and flexibility. These variables are consistently reported in the literature as being explanatory for mode choice processes. The performance was assessed using a new micro-simulation model based on agent technology. Agent technology has been successfully applied in the study of transport-related phenomena, since it allows for the customisation of transport agents (e.g., shipper, freight forwarder or carriers) and isolation of their interactions (e.g., physical or information). This paper is organised as follows. Section 2 briefly reviews the state of the art of transport modelling, agent-based modelling and mode choice. Evidence is given that agent-based modelling can satisfactorily simulate the dynamics of a transport service and that a reduced set of variables is invariably mentioned as key. Section 3 explains the fundamental assumptions and design principles of the simulation model. In Section 4, the case study is described and the simulation model is calibrated. The results of the simulations are analysed in Section 5. Finally, Section 6 summarises the main conclusions of the research and recommends directions for further research in this area.

2. Literature review 2.1. Freight transport modelling The first European freight transport models date from the early 1970s. Over the years, multiple dedicated freight transport models have been proposed (e.g., Ben-Akiva et al., 2013; Chow et al., 2010; De Jong et al., 2004; Liedtke, 2009; Tavasszy, 2006). It is important to note that the four-step model, initially developed for modelling passenger transport, has been widely and successfully adapted to model freight transport (De Jong et al., 2004). The typical structure of a freight transport model thus consists of four modelling steps: production and attraction of freight, trip distribution, mode choice and trip assignment (Tavasszy et al., 2012). The types of freight transport models developed for each of the four steps are briefly reviewed, and examples provided, in the following. 2 A search on the EC’s TRIP web portal yielded a total of 134 research projects dealing with intermodal topics. No case study or demonstrator was found concerning short distance services (search on: 03rd July 2013).

102

V. Reis / Transportation Research Part A 61 (2014) 100–120

2.1.1. Models for production and attraction of freight These models aim to estimate the amount of freight produced and consumed in every zone within a spatial context. Several types have been developed. Trend and time series models use historical data to build the correlation function and to extrapolate future values. Multiple methods have been proposed for building the correlation functions (such as the growth factor method or autoregressive correlations) (Garrido, 2000). A similar method – Zonal Trip Rate Models – relies on collected data on traffic volumes leaving or entering the zones in question. Economic models (such as input–output and related models) rely on regional economic activity to estimate the production and consumption of goods in each zone (Cascetta, 1996, 2009). System Dynamics methods have also been used to link the amount of produced and consumed goods at regional level (zones) to changes in the economy (e.g., GDP), society (e.g., demography) or governments (e.g., public policies) (Fiorello et al., 2010). One example of a system dynamics model is the ASTRA – Assessment of Transport Strategies Model.3 2.1.2. Models for trip distribution These models determine the flow of goods between each pair of zones. The results are normally presented in a table format designated as an Origin–Destination Matrix (OD Matrix). Gravity Models are the most commonly used methods to determine the flows. The flow Xij between a pair OD is the function of the product between the production at the origin – Oi – and the attraction of destination – Dj – divided by the impedance to moving the freight between Oi and Dj. Production and Attraction are normally measured using the above Models for Production and Attraction of Freight. Impedance is normally calculated by a transport cost function. Transport cost functions may cover a wide range of impedance factors, including: distance, tolls, taxes, transhipment operations, etc. These methods require a limited amount of data and the transport cost functions allow the assessment of transport policies. Conversely, the scope for including explanatory variables is rather limited and the number of calibration parameters is also limited (Erlander and Stewart, 1990). The SMILE model, developed by Tavasszy and his colleagues (1998), deploys a Gravity Model. Economic Models are other methods to determine the distribution of the flows. These methods, described in the previous section, use economic data to determine the supply and demand for freight in each zone, (output of the Models for Production and Attraction of Freight), and respective distribution between zones. These methods have the advantage of being linked to the regional economy and land development. They can provide information on the transport and land use interactions and, in specific circumstances, be used to assess policy effects. However, they require a substantial amount of data with a level of detail that is often difficult to obtain (Cascetta, 2009). The multi-regional input–output model proposed by Cascetta (1996) is an example of an economic model. 2.1.3. Models for mode choice These models allocate freight flows (between each pair of zones) to the available transport services (supply). The transport services can be either single-modal (e.g., road, train or sea) or intermodal (e.g., road and train, or road and sea). A wide set of models are available. Economic Models are normally grounded in the theory of the firm. They build the transport agent’s cost function, in which the available transport services are considered as one of the inputs. Demand functions, based on the costs function, can then be derived. Oum (1989) presents a model using neoclassical economy. Disaggregated Modal Split Models represent the shipping firm’s decision-making process. They are grounded in the assumption that shipping firms are rational and will opt for the transport solutions that maximise their benefits or utility. Utility functions are then built for typologies of firms, normally using the Multinomial Logit or the Nested Logit methods. These methods require a substantial amount of data which may not be readily available. Typical sources include: surveys of companies or transport companies and available statistics on freight flows. Ben-Akiva and De Jong (2013) present an aggregated–disaggregated–aggregated freight transport model in which logistic decisions are made at a disaggregated level. Blauwens et al. (2006b) presents a model that deploys an inventory-theoretic framework to calculate the total logistics costs. Aggregated Modal Split Models estimate the average market share of the transport services. Most models, rather than modelling the decision making of individual firms, rely on available statistics (modal share for a number of zones) to infer the utility functions, normally in the form of the Binomial or Multinomial Logit Models, of each transport service. The validity interval of the utility functions is therefore limited to source zone flows. These models have reduced data requirements. However, since they work with average values, they provide little information on the causal effects underlying the results. An example of this application can be found in Blauwens and Voorde (1988). 2.1.4. Models for trip assignment The purpose of these models is to allocate vehicle trips to the transport services. Many models do not include this step and most other models include only truck assignments (De Jong et al., 2004). Also, passenger and freight trips are often assigned jointly, since many transport infrastructures (e.g., road or rail) are shared by both types. These models assume that the agents behave rationally and choose the paths that result in lower costs. Accordingly, the equilibrium of a network is obtained when no vehicle can change a route without increasing its costs. Limited capacity of the road network and therefore congestion problems can be accounted for through the introduction of penalties (such as an 3

More information at: http://www.astra-model.eu (accessed on 1st June 2013).

V. Reis / Transportation Research Part A 61 (2014) 100–120

103

Table 1 Comparison of the properties of freight transportation system with the conditions of applicability of ABM. Conditions of applicability

Properties of freight transportation systems

Problem has a natural representation consisting of interacting agents or entities (Weiss, 1999) Agents or entities have a limited range of action (The Economist, 2009) Agents have degrees of freedom to decide on their own reactions (The Economist, 2009) Agents are heterogeneous and it makes no sense to consider average (or aggregate) behaviours or properties (Bonabeau, 2002) Decisions and behaviours can be defined discretely, that is, with welldefined boundaries (Weiss, 1999) Interactions between the transport agents are identifiable, discrete, dynamic and temporary (North and Macal, 2007)

Table 2 identifies the elements of a freight transport system for each component of an agent-based model Each transport agent is bounded to perform certain roles (e.g.: forwarding) Transport agents are an private and independent companies Transport agents are highly heterogenous: perform different roles and actions, and have different levels of performance The actions and roles of each transport agent are well defined (e.g.: transport, handling) Transport agents interact on specific moments (e.g.: negotiation or transhipment) along the production of the freight transport service

increase in transit time or an increase in fuel consumption). The NODUS model, presented by Jourquin (1995), is an example of one of the few models that performs a multimodal assignment. 2.1.5. Limitations of the traditional modelling approaches Traditional freight transport modelling approaches do, however, suffer from some limitations (Baindur and Viegas, 2011; Holmgren et al., 2012b; Liedtke, 2009). They rely on statistical analysis and correlations between freight transport market parameters. Thus, disaggregated choices do not necessarily correspond to the actual decisions of transport agents. They are unable to consider the influence of the agents’ individual heterogeneity in the evolution of the freight transport system. Consequently, behavioural aspects of the transport agents (such as decision making, individual preferences on modes of transport or variations on individual performance) and respective interactions (e.g., negotiation, communication or handling operations) cannot be modelled (Holmgren et al., 2012b). Also, in traditional modelling approaches results are restricted to the options initially included in the distribution. Thus, the emergence of new phenomena (e.g., implementation of new network structures such as transport corridors) cannot be forecast (Liedtke, 2009). 2.2. Agent based modelling Agent-Based Modelling (ABM4) is a micro-simulation tool that recreates in a virtual context the interactions and behaviours of a set of autonomous agents and their environment. This tool considers that any system is made up of a set of entities – agents – that interact amongst themselves and with an environment that supports their very existence (Bonabeau, 2002). The concept of agent is the most important aspect in ABM. An agent is an independent entity with precise boundaries and specific goals that exhibits autonomous behaviour and has both sensorial and communicational capabilities. It may have incomplete information about its surroundings and limited capacity to influence others. Interaction refers to the ability of an agent to communicate with or contact others and gather information from its surroundings. The environment provides the physical support for the agents’ liveability and interactions (North and Macal, 2007).The system’s overall properties and behaviours emerge as result of the agents’ behaviours and dynamic interactions, which in turn are the consequence of agents either pursuing their own goals or reacting to certain external stimuli. ABM thus follows a bottom-up approach to understanding real world systems. The successful utilisation of agent-based technology depends on the phenomenon to exhibit certain properties. Table 1 lists the conditions of applicability of ABM and compares them with freight transportation systems. The ABM method offers a number of advantages in relation to traditional modelling approaches. Firstly, the mode choice process entails the ability to evaluate each carrier across a set of variables – mode choice variables. The ABM method enables the generation of an indefinite number of agents. Secondly, the performance of an intermodal transport service is a function of the performance of the carriers and the nature of the respective interactions (Reis, 2010). ABM allows explicit modelling of the behaviour and the interactions of each agent. Thirdly, an intermodal freight transport service reveals time dependency (e.g., trust is built over time). ABM is a dynamic modelling method in which both the system and agents have memory. The utilisation of agent technology for the study of freight transport systems is relatively recent, with the first publications dating back to the 1990s (Tavasszy, 2006). An early attempt to use ABM principles in solving transport-related problems was conducted by Fischer and his colleagues (1995). The purpose of the work was to explore the usefulness of ABM in the study of negotiation and cooperation dynamics in the road freight transport market. The authors addressed the problem of scheduling orders. They developed MARS – Modelling a Multi-Agent Scenario for Shipping Companies, which (i) simulates a freight transport market, (ii) considers two types of agents: shipping company and truck, and (iii) incorporates three 4 A caveat is that there are various terms (and respective acronyms) in use in the literature for what (for all intents and purposes) is essentially ABM, such as: Agent Based Computational Modelling (ABMC), Agent Based Social Simulation (ABSS), Agent Based Computation Simulation (ABCS), Agent Based Modelling and Simulation (ABMS) and Multi Agent Systems (MAS).In particular MAS has received a lot of attention over the past couple of decades, mainly owing to the advent of the internet. MAS originated within Artificial Intelligence, with the aim of exploring the advantage of ‘‘massive open distributed systems – such as the Internet’’ (Wooldridge, 2009 pp. xi).

104

V. Reis / Transportation Research Part A 61 (2014) 100–120 Table 2 Comparing ABM and freight transport systems. Components of ABM

Examples of properties of freight transport system

Agents

Customer Freight forwarder Transport providers (road or rail operators) Terminal agents (port, rail terminal)

Interactions

Requesting/sending information Negotiation

Environment

Technological properties of the vehicles (speed, capacity, etc.) Geographical locations Physical properties of the goods (volume, weight, etc.)

dimensions: cooperation, negotiation and task decomposition and allocation. The authors concluded that the ABM approach was suitable for overcoming the complexity of scheduling problems. Davidsson and his colleagues have been developing a micro-simulation model of intermodal transport chains, inspired by ABM principles (Davidsson et al., 2008; Holmgren et al., 2012a; Persson and Davidsson, 2005): the so-called TAPAS – Transportation And Production Agent-based Simulator. The agents in the model are: shipper, producer, transport agent, production planner, transport buyer and transport chain coordinator. The authors adopted a two-layer simulator5 to structure the intermodal freight transport system: one concerning the decision-making process and the other relating to the physical transport. The purpose is to investigate the transport agents’ reactions to the implementation of governmental policies, such as: fuel taxes, road fees or vehicle taxes. Public policies are tested as different scenarios. The model computes the transport service’s total transport costs (both direct and external costs, including environmental costs), society revenues and the satisfaction of the shippers (as the level of reliability). The ABM presented in Baindur and Viegas (2011) is aimed at supporting the design of public policies for the promotion of intermodal short sea shipping transport. The model shares some similarities with the TAPAS model, including the types of agents and interactions. In terms of structure, the model adopts a three-layer approach, the layers being: physical transport, market layer (decision-making) and regulatory layer. The latter represents the public authority, responsible for designing the policies (such as taxation or road tolls). Whilst this layer is a novel addition when compared with other models, it is applied in a relatively simple manner and, in the end effect, works as a scenario analysis (as in the TAPAS model). Each public policy determines a fixed regulatory layer that remains fixed for the duration of the experiment. Liedtke (2009) proposed the INTERLOG prototype model to simulate the commodity flows and logistical organisation within a region. INTERLOG follows the traditional 4-step model approach at a disaggregated level. Step 1 (generation and attraction) and Step 2 (distribution) are simulated through a set of companies (agents). Step 3 (modal split) is ignored since only road transport is taken into consideration. Step 4 (trip assignment) is carried out through an auction system of transport contracts, in which a set of freight forwarders (agents) choose from a set of transport companies (agents) considering a set of factors including feedback of lot-size and tour construction. The disaggregated results are then aggregated and compared with available statistics for validation purposes. The utilisation of agents to achieve the disaggregation of the various steps of the simulation process lends enhanced flexibility and scalability to the model and represents an important added-value. The models reviewed above evidence that ABM has been used to model freight transport systems. However, the models are mainly concerned with providing support for public policy design and none of them has been designed to assess the influence of transport distance on the mode choice process. This paper suggests an ABM for simulating the performance of freight transport services.

2.3. Mode choice variables Understanding the mode choice process has always been a major concern amongst transport researchers. McGinnis compared twelve studies on the mode choice process in the United States before and after the 1980 transport deregulation process (McGinnis, 1990). The author identified, but did not rank, six key attributes in the mode choice process: freight rate, reliability, transit time, safety, shipper market considerations and carrier considerations. Murphy and Hall (1995) updated McGinnis’s initial work and ranked the attributes as follows: reliability, freight rates, carrier considerations, transit time, shipper market considerations and safety. Jeffs and Hills (1990) conducted research to identify the mode choice variables in the industrial category of the paper, printing and publishing sector in the region of West Yorkshire (United Kingdom). Using the factor analysis technique, the authors identified the following list: reliability, monitoring, safety, security, transit time, flexibility, length of haul and size of shipment and loyalty towards either specific agents or mode of transport. 5

This division has already been used by some authors in the conceptual representation of intermodal transport services (e.g., Jensen, 1990; Woxenius, 1998).

V. Reis / Transportation Research Part A 61 (2014) 100–120

105

Matear and Gray (1993) conducted a survey of shippers and carriers operating within and between Ireland and the United Kingdom. The purpose of the survey was to identify the mode choice variables in sea and air transport services. The authors concluded that the relevant attributes for shippers were rapid response to problems, safety and on-time collection and delivery; whilst for carriers, they were availability of freight space, punctuality and high frequency of service, in the case of maritime services, and frequency, punctuality and availability of freight space, in the case of air services. A new study presented by Murphy et al. (1997) assessed the mode choice variables of US shippers and carriers. The results showed that both shippers and carriers considered the same two factors to be the most relevant: reliability and equipment availability. After these, shippers ranked transit time, pickup and delivery service, and the carriers’ financial stability; whilst the carriers ranked operating personnel, transit time, pickup and delivery service. In a survey of 75 other bibliographic references about route and mode choice process, Cullinane and Toy (2000) showed that price, transit time and reliability were consistently referenced and often considered as the most relevant factors. The EU co-funded project, Intermodal Quality (IQ), was carried out with the purpose of improving the quality of intermodal transport within the European Union (INRETS, 2000). IQ identified a core group of universal quality attributes, though the relative weight of each one may have differed for goods and shippers, respectively. The attributes are time indicators, reliability, flexibility, qualification, accessibility, monitoring, and safety and security. Another EU co-funded project, LOGIQ, was developed to identify the main actors in the decision-making process and to provide information on the underlying criteria and constraints in the use of intermodal transport (Gruppo CLAS, 2000). A set of interviews with a number of transport decision makers – namely, freight forwarders, shippers and carriers – was carried out. Two types of decision orientation were identified: cost oriented and quality oriented. The former has price as its main decision attribute. Any change in price has significant consequences for the mode choice process and quality is perceived as a by-product and therefore not fundamental to a transport decision. In the latter, qualitative factors and cost are evaluated equally in the mode choice process. The key quality factors are: reliability, flexibility, safety and frequency. Shinghal and Fowkes (2002) estimated the determinants of mode choice in the Delhi to Bombay (India) corridor for two transport services: road and intermodal service (road and rail). The authors considered a total of four variables: price, transit time, reliability and frequency. A total of 32 interviews were conducted in firms from different sectors (chemical, technological, industrial and food products). A logit model was used to analyse the data of the stated preference survey. The authors concluded that road transport shows a great advantage in all variables and for every sector and that the viability of intermodal services is dependent on reliable, fast and high-frequency services. García-Menéndez and his colleagues (2004) estimated the freight transport demand function using a conditional logit model. They considered two transport services: road and sea. A total of 157 interviews in exporting firms located in the Valencian Community (Spain) were conducted. Four industry sectors were considered: wood manufactures and furniture, ceramics, textiles and agro-industry. Taking into consideration eight explanatory variables, the authors concluded that price, transit time and frequency of shipment were the determinants of mode choice across the various sectors. Sea transport revealed higher sensitivity in all mode choice variables. Grue and Ludvigsen (2006) conducted an extensive survey consisting of 246 interviews with freight forwarders and shippers. These agents used road and rail transport services. The purpose of the research was to identify the determinants of mode choice in the intra-European freight transport market. Respondents were asked to rank by importance a total of 23 mode choice variables. The results show that reliability and price of transport were the two most relevant mode choice variables. Danielis and Marcucci (2007) investigated the cut-off levels in the mode choice attributes, using logit models. The cut-off is a threshold that leads to a change in the behaviour of the shipper decision-making process. The mode choice variables included: price, transit time, reliability and flexibility. The authors concluded with respect to the relevancy of these variables that price and flexibility are significant in all modes, while reliability and transit time are significant in most modes. The studies reviewed above reveal a wide diversity of variables influencing the mode choice process. Table 3 compiles the mode choice variables citing the above studies and others. The amount of literature concerning mode choice variables and respective influencing factors is substantial. The main conclusion that can be drawn from the literature review is that the distance of the transport service is seldom referenced (the few available sources include INRETS, 2000 and Jeffs and Hills, 1990). This absence either evidences that the distance of the transport service most likely plays a minor role in the mode choice process; or denotes a bias in research towards services of a similar distance (mostly long-distance services) in which case the influence of the distance variable becomes irrelevant. Both cases are at odds with the discussion in Section 1 on the importance of distance to the competitiveness of intermodality. We can thus justifiably assert that there is a gap in the literature concerning the influence of distance on the mode choice process. Taking the number of times that a variable is identified as a proxy to infer as to its scope of influence in the decision making process, we may conclude that only a limited set of variables is invariably relevant, while most others are influential only in specific situations. The limited set includes: reliability, transit time, flexibility and price. This duality raises a number questions as to the structure of the mode choice process and role of the variables. According to some authors (e.g., D’Este, 1996) mode choice process is stepwise. The decision maker chooses one mode from a reduced set of transport options, which were initially filtered from a survey of all available transport options. The filtering step serves to eliminate transport options that are either technically unfeasible or fail to meet service criteria. Variables used in this step are necessarily case specific. The wide number of variables utilised one or a few times likely results from this step. In the choice step, the purpose is to

106

V. Reis / Transportation Research Part A 61 (2014) 100–120

Table 3 Mode choice attributes referred in the reviewed literature. Attribute

Authors

Reliability

         

Flexibility

      

Frequency of saervice

     

Service level

Length of haul

Grue and Ludvigsen (2006) Norojono and Young (2003) De Maeyer and Pauwels (2003) Shinghal and Fowkes (2002) Cullinane and Toy (2000) Gruppo CLAS (2000) INRETS (2000) Murphy et al. (1997) Jeffs and Hills (1990) McGinnis (1990), updated by Murphy and Hall (1995)  McGinnis (1989)  Oum (1979)

Attribute

Authors

Transit Time

        

Danielis and Marcucci (2007) Grue and Ludvigsen (2006) García-Menéndez et al. (2004) De Maeyer and Pauwels (2003) Shinghal and Fowkes (2002) Cullinane and Toy (2000) Murphy et al. (1997) Jeffs and Hills (1990) McGinnis (1990), updated by Murphy and Hall (1995)  McGinnis (1989),  Oum (1979)

Danielis and Marcucci (2007) Grue and Ludvigsen (2006) Norojono and Young (2003) Gruppo CLAS (2000) INRETS (2000) Matear and Gray (1993) Jeffs and Hills (1990)

Price

     

Danielis and Marcucci (2007) Grue and Ludvigsen (2006) García-Menéndez et al. (2004) Shinghal and Fowkes (2002) Gruppo CLAS (2000) Matear and Gray (1993)

Monitoring

 INRETS (2000)  Jeffs and Hills (1990)

 Grue and Ludvigsen (2006)  Cullinane and Toy (2000)  De Maeyer and Pauwels (2003)

Shipper’s market considerations

 McGinnis (1990), updated by Murphy and Hall (1995)  McGinnis (1989)

 INRETS (2000)  Jeffs and Hills (1990)

Security

 Grue and Ludvigsen (2006)  INRETS (2000)  Jeffs and Hills (1990)

Danielis and Marcucci (2007) Grue and Ludvigsen (2006) García-Menéndez et al. (2004) De Maeyer and Pauwels (2003) Cullinane and Toy (2000) McGinnis (1990), updated by Murphy and Hall (1995)  McGinnis (1989)

select the solution that provides the highest benefits or utility from several that are equally feasible. Variables used in this step are most likely not case specific, but rather economic or transport operations-related. Thus, a stable set of variables may be consistently used by decision makers. We may then hypothesis the existence of a few universal and many case-specific variables. Further studies are required to support these assumptions.

3. Simulation model 3.1. Assumptions of the agent-based model Based on the findings presented in the literature on intermodal freight transport modelling, ABM and mode choice, the following assumptions have been adopted:  Shippers convey a certain amount of loading units (containers) over a period of time between one origin and multiple destinations.  Lot size and timing of delivery is stochastically determined during the simulation but a range is fixed at the start of the simulation.  Key decision makers are: shippers, freight forwarders and carriers. Carriers include several individual road haulage companies and a single rail transport company.  Exchange of information occurs between shippers and freight forwarders and between these and carriers.  Road carriers base their price on the distance of the transport service, while the rail carrier has a fixed price. The agent-based model was developed using software named ANYLOGIC Version 6.4.0.6 6

More information at: http://www.xjtek.com (accessed on 1st June 2013).

107

V. Reis / Transportation Research Part A 61 (2014) 100–120

Shipper Agent

Shipping Order

Origin Terminal Agent

Containers generation

Information Storage

Rail Transport Agent

Transport Service Agent

Order

Train Agent Delays

Proposal / Acceptance / Delays

Inquiry / Order

Terminal Agent

Information Storage

Freight Forwarder Agent

Notification

Inquiry / Order Proposal / Acceptance

Road Transport Agent

Truck Agent Order

Destination Terminal Agent

Information Storage

Administrative Layer

Physical Layer

Legend: Agent

Physical Flow

Information Flow Fig. 1. Conceptual diagram of the agent interactions in freight transport service.

3.2. Structure of the model The model was structured in the form of two interrelated and connected layers: the administrative layer and the physical layer (Fig. 1). This conceptualisation of a freight transport service was initially proposed by Jensen (1990) and since adopted by other authors (e.g., Baindur and Viegas, 2011; Bergkvist et al., 2004; Reis, 2010). The administrative level encompasses the tasks not directly related to the physical transport of goods, such as inquiries, ordering or reservation; whilst the physical level includes the activities carried out during the physical transport of the freight, such as transport and handling. The agents in the administrative layer have cognitive functions, are goal-seeking and are independent decision makers; whilst the agents in the market layer are non-cognitive (e.g., vehicles), controlled by the agents located within the previous layer. The performance of the carriers and services is continuously monitored. 3.2.1. Agents The reviewed literature and the analysis of the case study provided information about the agents involved in the intermodal freight transport service. A total of ten agents were identified (Fig. 1): Shipper, Freight Forwarder, Rail Carrier, Road

108

V. Reis / Transportation Research Part A 61 (2014) 100–120

Carrier, Origin Terminal, Transhipment Terminal, Destination Terminal, Train, Truck and Transport Service. Here follows a brief description of each one. 3.2.1.1. Shipper Agent. The Shipper Agent generates the weekly demand of freight transport services – shipment orders. A shipment order contains information about the amount of goods, origin and destination and time windows for pick-up and delivery. The shipment order is conveyed to the Freight Forwarder. 3.2.1.2. Freight Forwarder Agent. The Freight Forwarder Agent organises (administrative layer) and manages (physical layer) the transport services. In the organisational phase, the Freight Forwarder receives the shipment orders and hires the (road or rail) carriers accordingly. Road Carriers can be used on a standalone basis or in combination with Rail Carriers; whereas Rail Carriers can only be used in combination with road carriers. The hiring stage consists in booking a transport service in a designated transport carrier. An inquiry order is generated for each shipment order, containing the following information: destination, scheduling (pick up time and expected delivery time) and carrier (or carriers, if an intermodal service). The road service schedule for an intermodal service is based on the transit time for the rail service plus the handling time at the transhipment terminal. The inquiry order is sent to the Carrier Agents that reply with a proposal order. The contracted carrier (or carriers in case of intermodal transport) is notified with a booking order. Transport capacity depends on the truck and train capacities. In high demand situations, congestion may occur and goods are temporarily stored whilst waiting for available transport capacity. In such situations, containers already being stored have priority over recent arrivals. The managerial phase begins with the physical movement of containers. The Freight Forwarder Agent’s main task is to ensure the arrival of the goods within the scheduled time. In the event of rail service delays, the subsequent road service is re-booked. 3.2.1.3. Rail and Road Carrier Agents. Rail Carrier and Road Carrier Agents convey the containers between two locations. Upon receiving an inquiry order, the Carrier Agent replies with a proposal order (if it has the available transport capacity). The proposal order contains the following information: price, schedule and transit timing, booking time and capacity. The booking time is the notice time that a carrier demands to either book, change or cancel a service. The contracted carrier generates a service order that contains the following information: arrival time at origin, amount of goods and destination. 3.2.1.4. Train and Truck Agents. Train and Truck Agents are non-cognitive agents controlled by the Carrier Agents and perform the sole task of moving cargo between origin and destination. They are generated at the beginning of a transport service and terminated upon delivery of goods. Train Agents are characterised by four variables: capacity, schedule, transit time and reliability (of transit time). Truck Agents are characterised by two variables: speed and reliability. 3.2.1.5. Origin, Destination and Transhipment Terminal Agents. Origin, Destination and Transhipment Terminal Agents are also non-cognitive agents, reacting to the presence of a Train or a Truck Agent. Their purpose is to handle containers (loading or unloading). Loading activities occur at the origin and transhipment terminal agents, whilst unloading activities occur at the transhipment and destination terminal agents. The key attribute for these agents is the handling productivity, i.e., the amount of containers moved per unit of time. 3.2.1.6. Transport Service Agent. The Transport Service Agent is another non-cognitive agent that collects and stores information on every transport service. It constitutes the memory of the system. A transport service agent is generated by the model for every shipment order. It is a vector containing the following information: unique identifier of the goods, time and date of generation of shipment order, unique identifier of the Shipper Agent, destination, expected departure schedule, actual departure schedule, expected arrival schedule, actual arrival schedule, unique identifier of the Carrier Agents and price of the transport service. 3.2.2. Interactions Two types of interactions may occur between the agents during the freight transport: physical interaction and logical interaction (Fig. 1). Physical interactions result from the physical movement of the containers from origin to destination. There are two stages of interaction in a road-only service: Origin Terminal Agent–Truck Agent–Destination Terminal Agent; and four stages of interaction in an intermodal service: Origin Terminal Agent–Rail Agent–Transhipment Agent–Truck Agent–Destination Terminal Agent (Fig. 1). Logical interactions are the exchanges of information that occur during the negotiation and transport

109

V. Reis / Transportation Research Part A 61 (2014) 100–120

stages. There are nine types of logical interaction: acceptance, container generation, delay, information storage, inquiry, order, notification, proposal, and shipping order (Fig. 1). The syntax of the communication system was based on the rules of the AnyLogic software (which is implemented on a Java Platform). A type of interaction was defined for every type of logical interaction plus one for the physical interaction, i.e., a total of ten. 3.2.3. Environment The environment recreates the conditions and properties of the context in which the agents operated. In the case of a freight transport system, the environment refers to the physical properties of the containers (such as size, weight or volume), the technological properties of the vehicles (such as capacity or speed) and the geographical characteristics of the region in question (such as distance). The units of measurement implemented in the model follow the metric system as defined by The International System of Units. 3.3. Validation of the model The difficulty in validating complex system models is recognised in the literature (Axtell, 2000; North and Macal, 2007; Sterman, 2004). The main problems include: complex feedback loops; the latent behaviour of the agents and random effects resulting in unexpected behaviour; and the lack of a formal equation structure that renders difficult to trace the chain of cause and effect. With a view to increasing confidence in the model, North and Macal (2007) proposed a set of validation steps to be performed during its development, as follows:  Requirement Validation: the model should respond to clear requirements and questions about the real world. The objectives of the model are presented and discussed in Section 1.  Data Validation: the data in the model should be valid. The data in the model were provided by the Freight Forwarder involved in the Case Study presented in Section 4.  Theory Validation: the assumptions of the model should be valid. The assumptions follow the practice of modelling as presented in Section 2 and the case study as presented in Section 4.  Process Validation: agents and the interaction structure and steps in the model have to be clear, meaningful and correspond to the real world process. The structure of the model is based on previous work as presented in Section 2 and replicates the process of production of a case study as presented in Section 4.  Agent Validation: agent behaviour, relationships or interactions have to correspond to real world actions. The agents were based on previous works as presented in Section 2 and on the case study as presented in Section 4.

Legend: Rail Link Road Link Terminal or Port Destination

Source: Google Earth (2012) Fig. 2. Case study geography.

110

V. Reis / Transportation Research Part A 61 (2014) 100–120

4. Case study 4.1. Description A freight forwarder launched some years ago an intermodal transport service of containers connecting the port of Sines, located in the southwest of Portugal, with seven destinations in the centre of the country (Fig. 2). The intermodal service has a rail service, linking the port of Sines with two terminals – Bobadela and Vale do Tejo – and road services linking these terminals with the final destinations. The Bobadela terminal serves five destinations (Alcobaça, Carregado, Sintra, Lisbon and Setúbal), while the Vale do Tejo terminal serves two destinations (Leiria and Montemor). The transport distances range between 130 km (to Setúbal) and 330 km (to Montemor). The current structure of the intermodal transport service is conceptualised in Fig. 3. The shipping companies (shippers) hire a freight forwarder to manage the intermodal transport service. In turn, the freight forwarder contracts the transport services from carriers. There are two modes of transport: road and rail transport. Trucking carriers provide transport services between the terminals (port of Sines, Bobadela and Vale do Tejo) and the final destinations. A Rail Carrier provides transport services between terminals. At each terminal there is one handling company. Transport capacity is determined by the capacity of the carriers and transhipment productivity of the handling companies. Transport demand is determined by the shippers’ orders and the containers already stored at the terminal of origin. Shippers send their orders to a freight forwarder at unknown times. Each order contains information about the number of containers and respective destinations. When the amount of containers is higher than the available capacity, congestion occurs and containers are temporarily stored at the terminal of origin. Transport choice decisions (scheduling, destinations, length of train, etc.) are taken by the freight forwarder. The current choices of the freight forwarder are constrained by shipper requirements (e.g., destinations or transit time), the carrier characteristics (e.g., speed or capacity), demand, and its own past choices (e.g., train schedules or capacity or stored containers). The fact that the past choices of the freight forwarder influence the current ones, which in turn will influence the future choices, creates complex and non-linear behaviour. The main categories of exported goods are extracted natural products (such as marble), construction materials and products for recycling (such as paper and plastic), while the main categories of the imports flow include construction machinery, clothing, toys, food products (such as alcoholic beverages) or raw materials (such as iron). Table 4 summarises the annual flows of containers for every destination. With the exception of Setúbal, for all destinations there is a significant imbalance in favour of inbound flows.

Receiver

Shipper

Freight Forwarder

Handling company

Rail Operator

Train

Terminal

Terminal Operator

Terminal

Road Operators

Truck

Legend: Information Flow

Agents

Physical Flow Fig. 3. Conceptualisation of the freight transport services.

Destinations

111

V. Reis / Transportation Research Part A 61 (2014) 100–120 Table 4 Annual demand of containers. Source: Data provided by the freight forwarder. Terminal

Vale do Tejo

Destination

Montemor Leiria Alcobaça

Bobadela

Carregado Sintra Lisbon Setúbal Total

Annual demand (# containers) Outbound

Inbound

Total

96 (23%) 91 (24%) 98 (24%) 206 (12%) 175 (19%) 676 (27%) 349 (56%)

329 (77%) 284 (76%) 314 (76%) 944 (82%) 767 (81%) 1834 (73%) 272 (44%)

425 (7%) 375 (5%) 412 (6%) 1150 (18%) 942 (15%) 2510 (39%) 621 (10%)

1691 (26%)

4744 (74%)

6435 (100%)

4.2. Calibration 4.2.1. Agents The agent-based model was calibrated to simulate the dynamics of the case study. The freight forwarder provided the data on the intermodal service and on the single-road services, for a full year (52 weeks) of operations. The agents’ behaviour is presented below. 4.2.1.1. Shipper Agent. The shipping company has three weekly services on Monday, Wednesday and Friday. The ship is scheduled to arrive in the early morning. The containers are ready for transport, after handling operations and customs clearance, at 16:00 (on the day of arrival). There is a time window of 72 h to deliver containers to receivers. The data reveals a fairly constant demand of containers, in both directions. A triangular distribution with the mean equal to the mean of containers (Fig. 4) and with an interval equal to 10% of the mean was implemented to simulate demand. The linear formulation of triangular distributions is easy to implement and, in cases of random variables with low variance, which is the case for the case study, offers a reliable alternative to normal distributions. The 10% interval encompasses more than 95% of all cases in the available data, which was considered enough to adequately represent the demand patterns. The shipping company sends the shipment order to the freight forwarder 48 h in advance of the expected arrival of the ship. 4.2.1.2. Freight Forwarder Agent. On a daily basis, this agent computes the expected number of containers to be transported on the same day and the following two days. The expected number is calculated based on the containers that are currently stored in the port (if any) and on the expected demand in the following 48 h. The reason for analysing the demand in the following 48 h is that a container may be stored up to 48 h in the port (with no extra costs) and still be delivered to destination within the 72 h, after customs clearance, as requested by the shipping company. Rail and road services are then hired or cancelled according to a set of rules presented below. Road services in the intermodal transport service are scheduled based on the rail service’s transit time and expected handling time at the transhipment terminal.

Fig. 4. Random demand function.

112

V. Reis / Transportation Research Part A 61 (2014) 100–120

The Freight Forwarder Agent selects Road Carriers Agents to award transport contracts either in the intermodal transport case or the road transport case, using a random utility model – multinomial logit model (Ben-Akiva and Lerman, 1985; Train, 2009). The Freight Forwarder Agent organises calls for bids for every transport service to a discrete set of Road Carriers Agents. The set consists of all Road Carriers Agents defined at the start of the simulation. Since the objective of the investigation was to compare the performance of intermodal against road transport services, in each simulation only one transport solution (intermodal or road) is available to the Freight Forwarder Agent. In this situation, the Freight Forwarder Decision Tree has a single level of decision (choice between Road Carriers). Therefore, the multinomial logit model was found adequate for simulating the Freight Forwarder decision-making process. During simulation, the Freight Forwarder applies the multinomial logit model to obtain a discrete choice probability for each Road Carrier Agent. Upon having the probabilities of all the Road Carriers Agents, a Monte Carlo simulation selects the specific carrier. In terms of management tasks, the Freight Forwarder receives notification on eventual delays of the ship’s arrival and on the rail and road services’ transit time. A ship delay means that the expected container will not be available for transport, which may give rise to the need to cancel the transport services or modify the list of containers to transport. A train delay implies a rebooking of the road service.

4.2.1.3. Rail Carrier Agent. The railway company offers a daily (Monday–Friday) block train service between the port of Sines and two terminals: Bobadela and Vale do Tejo. The default block train comprises 11 trailers and conveys up to 44 TEUs. Up to 3 more trailers can be requested, free of charge, resulting in a maximum capacity of 56 TEUs. During simulation, the size of the block train is determined by the following equation: max(11, min(N, 14)) where N is the number of containers indicated in the inquiry order of the Freight Forwarder. Situations of low demand lead to an increase in the price per container, due to the fact that railway company sells block trains with a minimum fixed capacity. Situations of high demand lead to longer transit times, as some containers may be delayed and are transported on the next train (the following day). Changes in schedule or train capacity can be made according to the following rules:  Cancellation: minimum of 24 h in advance (otherwise the train costs is charged).  Extra train: minimum of 48 h in advance (normal charge), up to a maximum of 12 trains per week (2 trains per day from Monday to Saturday).  Increased capacity: minimum of 48 h in advance, and up to 14 trailers (no extra charge). Table 5 presents the transit time and price per block train for each destination.

4.2.1.4. Train Agent. The Train Agent only exists for the duration of the transport service. During simulation, the train is generated 3 h prior to departure and terminated upon unloading of containers at the destination. Railway companies work according to fixed timetables. In the event of a delay, the train stops and waits for the following available slot (Table 6). Train reliability was obtained from the data of the Freight Forwarder and computed as the ratio of ontime services to all services (for both destinations).

4.2.1.5. Road Carrier Agent. Road Carriers provide trucking services from ports or terminals to final destinations. They work with non-fixed schedules but require new services to be booked at least 4 h in advance. A service may be cancelled up to 4 h in advance without any penalty, otherwise the service cost is charged. Changes can be made without extra charges. Information provided by the Freight Forwarder speaks to the plentiful availability of Road Carriers, with no difficulty in booking trucking services ever registered, regardless of demand. Road Carrier Agents were simulated with an unlimited number of trucks. Prices and average transit times are listed in Table 7. Table 5 Transit time and price of rail services. Source: Data provided by the freight forwarder. Attribute

Service to Bobadela

Service to Vale do Tejo

Transit time (min) Price (€)

330 3000

540 4500

Table 6 Reliability and delay time of rail services. Source: Data provided by the freight forwarder. Service

Reliability (%)

Delay time (min)

To Bobadela To Vale do Tejo

85 70

240 360

113

V. Reis / Transportation Research Part A 61 (2014) 100–120 Table 7 Average transit time and price of road services. Source: Data provided by the freight forwarder. Destination

Montemor Leiria Alcobaça Carregado Sintra Lisbon Setúbal

Sines

Bobadela

Vale do Tejo

Transit time (min)

Price (€)

Transit time (min)

Price (€)

Transit time (min)

Price (€)

305 240 220 150 170 150 115

550 450 420 260 260 210 275

– – – 35 50 25 60

– – – 115 175 105 175

110 60 75 – – – –

275 175 185 – – – –

4.2.1.6. Truck Agent. The Truck Agent only exists for the duration of the transport service. It is generated 30 min prior to departure and terminated upon unloading of containers at destination. Truck capacity is 1 TEU. Truck transit time (TTT) was computed based on the following normal distribution function: let TTT be the truck transit time measured in minutes; let ATTT be the average truck transit time measured in minutes taken from Table 7; then u  N(ATTT, 0.1  ATTT) TTT = min(u, 80) The maximum legal speed is 80 km/h. 4.2.1.7. Origin Terminal. The Origin Terminal Agent simulates handling operations at the port of Sines. Containers are loaded onto either a train or a truck. Each specific movement has a different amount of productivity. Productivity is measured as the number of containers handled per hour. The values were obtained from interviews with employees at the port of Sines. All handling operations were simulated through a random function following a triangular distribution. Minimum, mean and maximum values of productivity are presented in Table 8. Truck handling operation productivity is higher because the rail tracks are located further away from the container terminal than the truck parking terminal. 4.2.1.8. Transhipment Terminal. The Transhipment Terminal Agent simulates handling operations between train and truck at the Bobadela and Vale do Tejo terminals. Based on information provided by local workers, productivity was concluded to be similar to the truck handling operations at the origin terminal (Table 8). 4.2.2. Mode choice variables The model computation for each variable is described in Appendix A. 4.2.2.1. Price. Price was defined as the total amount paid by the Shipper to the Freight Forwarder to organise and manage the transport of freight. Results are displayed in average price per container. 4.2.2.2. Transit time. Transit time was defined as the time elapsed between the unloading from the ship at the origin and the delivery at the destination. Results are displayed in average time per destination. 4.2.2.3. Reliability. Reliability was defined as the probability of freight being delivered at the destination within a given time window. In the absence of a consensual method to evaluate this variable (Danielis and Marcucci, 2007), in this research the Shannon (1948) concept of entropy was used. The author used entropy to assess the level of noise of a network, that is, the amount of interruptions or mishandling of information. The author proposed the following formula to measure the reliability: let X be a discrete random variable that can assume an unknown but finite set of states {x1, x2, . . . , xN}, and let p(Xi), i = 1, 2 , . . . , N be the probability of Xi, then

Table 8 Productivity of the terminals and port operators. Source: Data provided by the local handling workers.

Train handling Truck handling

Minimum (# containers/h)

Mean (# containers/h)

Maximum (# containers/h)

12 16

20 24

28 32

114

V. Reis / Transportation Research Part A 61 (2014) 100–120 N X EðXÞ ¼  ðpðX i Þ  InðpðX i ÞÞÞ i¼1

In a fully reliable network, the probability of a message arriving to the destination in the same state of departure is one. In this situation, entropy is minimum and equal to 0. In a non-reliable network, a given message may arrive at the destination in one of many different states. In this situation, entropy is greater than 0. Since the number of states is finite but not fixed, there is no upper level for the entropy value. Indeed, the higher the number of states then the higher the entropy value can be. The application of Shannon’s concept to a transport service is straightforward. The transport network represents the communication network, the movement of the freight corresponds to the transmission of information and the unforeseen events that will bring uncertainty correspond to the noise. Also, transit time is a discrete variable that can assume a finite but unknown set of values, which is a prerequisite for computing the entropy level. A low-reliability transport service reveals high uncertainty in the expected transit time – high entropy. An increase in entropy denotes a growth in the range of transit times. Results are displayed as the average entropy per container. 4.2.2.4. Flexibility. Flexibility was defined as the capability of a transport service to operate with no disruptions in situations of unexpected fluctuations in demand (typically, cases of sudden increase). Flexibility was measured as the time needed to recover from an unexpected fluctuation in demand. That is, the time needed to dispatch all extra containers. After consultation with the freight forwarder, recovery was considered when transport time after unexpected fluctuation was within 10% off the average transport time for the previous 8 weeks and fluctuation in demand was set to a value of 50% of average demand. This fluctuation would occur periodically but randomly with an interval ranging between 4 to 6 weeks. Results are given in average time per destination. 4.3. Simulations

ratio of prices of intermodal over road transport services

The agent-based model was run under differing demand conditions. Demand was set to range from 2/3 up to 4 times the base-case demand, which corresponds to the average value, in a total of eleven scenarios, as follows: Scenario 1 (67% of basecase demand), Scenario 2 (100% of base-case demand), Scenario 3 (133% of base-case demand), Scenario 4 (167% of base-case demand), Scenario 5 (200% of base-case demand), Scenario 6 (233% of base-case demand), Scenario 7 (267% of base-case demand), Scenario 8 (300% of base-case demand), Scenario 9 (333% of base-case demand), Scenario 10 (367% of base-case demand) and Scenario 11 (400% of base-case demand). The lower limit corresponds to the minimum amount ever received, while the upper limit represents the current maximum capacity of the intermodal transport service. Each scenario was run 200 times, in order to fully capture the randomness of the model, and for a 1 year operation (52 weeks). Appendix A provides details about the formulation used to calculate each model choice variable. Only results from week 40 to 52 were taken into consideration. Only inbound flows were considered. Firstly, they largely determine the performance of the transport services, since they account for around 75% of all container movements. Secondly, rail and road transport services are hired out, meaning that

1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00 1

2

3

4

5

6

7

8

9

Scenarios Lisbon Carregado Leiria Fig. 5. Simulation results: price variable.

Sintra Montemor Alcobaça

10

11

115

V. Reis / Transportation Research Part A 61 (2014) 100–120

inbound and outbound flows are independently considered. Therefore, cross-influences between flows (in terms of funding or hiring carriers) do not occur. 5. Results 5.1. Price The price per container in road services is independent of demand and equal to the values listed in Table 7. Conversely in the case of intermodal transport, the price of each container depends on the train load factor, which in turn depends on demand. Fig. 5 presents the ratio of the price of each container in intermodal services to that for road service. The ratio decreases with the length of the route and with demand. Also, destinations served from the Vale do Tejo terminal, which have a longer rail leg, have lower ratios than those served from the Bobadela terminal. Demand is also favourable to intermodal services. An increase in demand results in a reduction of the ratio. On the longer routes there is a gain in competitiveness of around 15% when comparing the lowest with the highest demand, while on the shorter routes these gains are below 10%. Comparing both factors, distance seems to play a more important role than demand in the price competitiveness. Intermodal transport service is largely competitive on a price basis. With the exception of Scenario 1, in all other scenarios the weighted-average ratio (black line) is equal or below 1. The lower prices were obtained on the longest routes that account for 18% of total demand (Table 4). 5.2. Transit time

ratio of transit time of intermodal over road transport services

Fig. 6 presents the ratio of average transit time between intermodal and road services for each destination. Intermodal transport services have longer transit times than road transport for every destination and scenario. Longer transit times can be explained by the transhipment operations and the slow average speed of rail services. The impact of these factors reduces as the length of the route increases. The increase in transit time for road services ranges from 5% to 27% on the longest and shortest routes, respectively (excluding Scenario 11). Variations in road service transit time result only from the port’s reduction of productivity due to congestion, since the simulation assumption is the limitless availability of trucks. Intermodal transport services exhibit a higher sensitivity with mixed behaviour: an increase followed by a continuous decrease in transit time. This is essentially due to the finite and discrete capacity of rail services. In Scenarios 1 to 3, the Freight Forwarder Agent responds to the increase in demand by adding extra trailers to the scheduled rail services. In Scenarios 3 to 10, faced by the continuous increase in demand, the Freight Forwarder Agent responds by hiring extra rail services in order to keep transit times within the limits. As a result of the increase in capacity, the average storage time of the containers in the port reduces, leading to a progressive reduction in transit times and, consequently, in the ratio to road services. The reduction in transit time is higher on the shorter routes. However, simulation results reveal the existence of a limit to the gains in transit time that is similar for all destinations. The ratios converge towards the values for the longest routes.

30.00 25.00 20.00 15.00 10.00 5.00 0.00 1

2

3

4

5

6

7

8

Scenarios Lisbon Carregado Leiria

Sintra Montemor Alcobaça

Fig. 6. Simulation results: transit time.

9

10

11

V. Reis / Transportation Research Part A 61 (2014) 100–120

Entropy

116

Road Transport 7.0 6.0 5.0 4.0 3.0 2.0 1.0 0.0 1

2

3

4

5

6

7

8

9

10

11

10

11

Entropy

Scenario

Intermodal Transport 100.0 90.0 80.0 70.0 60.0 50.0 40.0 30.0 20.0 10.0 0.0 1

2

3

4

5

6

7

8

9

Scenario

ratio of recovery time btw intermodal and road transport

Fig. 7. Simulation results: reliability.

30.0 27.0 24.0 21.0 18.0 15.0 12.0 9.0 6.0 3.0 0.0 1

2

3

4

5

6

7

8

9

10

11

Scenarios Fig. 8. Simulation results: flexibility.

Finally, Scenario 11 corresponds to a situation of congestion in the intermodal service. The demand is higher than the rail services’ capacity. Excess containers are stored in the port leading to a continuous growth in transit time. 5.3. Reliability Fig. 7 presents the measurement of entropy levels in both road and intermodal transport services. Results show that road transport is inherently more reliable than intermodal transport, since it reveals lower entropy values in every scenario.

V. Reis / Transportation Research Part A 61 (2014) 100–120

117

Secondly, demand impacts reliability, since entropy continuously grows between scenarios. The sources of unreliability differ, however, between services. The reliability of road services is mainly determined by the port’s productivity, since a model works with the assumption of limitless availability of trucks. The reliability of intermodal transport services is mainly determined by the finite and discrete capacity of rail services. Thirdly, despite different sources of unreliability, the impact of demand on both transport solutions is similar as entropy grows at similar rates (the only exception being Scenario 11).

5.4. Flexibility Road services reveal almost constant flexibility due to the limitless availability of trucks; whereas intermodal services show decreasing flexibility owing to the limited availability of rail services. Fig. 8 presents the ratio of recovery time between intermodal and road services. In Scenarios 1 to 3 both transport services exhibit evolution of flexibility, as results show a constant ratio of around 1.3. From Scenario 3 onwards, intermodal transport reveals growing limitations in its flexibility: from Scenario 4 to Scenario 7 the ratio progressively grows to a value of around 5; from Scenario 8 to 10, it makes several jumps; and, finally, in Scenario 11 it grows indefinitely due to congestion of the intermodal transport service. The lower flexibility of the intermodal services can be explained by running rail services under fixed capacity and schedules. The capacity for adjustment is, therefore, limited and it decreases with an increase in demand.

6. Conclusions and directions for further research At the EU level, intermodality has been strongly advocated as a means of promoting the long-term sustainability of the freight transport sector. Many policies and incentives have been applied over the last two decades, all targeting medium to long distances (above 400 km). However, demand for freight transport services is evenly distributed between short and medium to long-distance services. Thus, current transport policies neglect approximately half of the freight transport market. As intermodal transport is currently revealing major difficulties in increasing its market share, opportunities in the short distance freight transport market must not be ignored. Despite the substantial amount of literature, the influence of distance in the mode choice process remains largely underresearched and, to some extent, ignored. Distance is seldom investigated as a potential decision variable and authors rarely discuss the impact of distance in investigation findings and their respective validity limitations. However, it is still to be demonstrated that distance does not influence mode choice process. Hence, the current practice may be incorrect, since findings obtained for a given transport distance are not necessarily transferable to other distances. Further research concerning the influence of distance on the mode choice process is required. Analysing the literature on mode choice variables, we may conclude that there is a wide set of variables that are occasionally referred to and a small set that is pinpointed in most sources. The latter includes the following variables: price, transit time, reliability and flexibility. The paper presents an investigation aimed at assessing whether mode choice variables used in medium to long-distance transport services can be used to explain the behaviour of agents in short-distance transport cases. A case study of a short distance intermodal transport service was used to test this hypothesis. The competitiveness of the intermodal transport services was compared against a hypothetical road transport service. Competitiveness was assessed by measuring the performance of each transport option in relation to the abovementioned four mode choice variables in different demand scenarios. A new simulation model based on agent technology was developed. The choice of agent-based modelling is related to the ability to model the behavioural aspects and interactions between transport agents and the influence of the individual heterogeneity of the agents in the evolution of the freight transport system. Traditional modelling techniques fail to model individuals. The agent-based model was applied to a running intermodal transport service and used to simulate a hypothetical road transport service. The development and deployment of an agent-based model is per se a contribution to the literature. Surveys conducted by Davidsson and his colleagues concluded that research based on ABM is still in a very early stage of maturity, with few (real world) experimental applications having been carried out (Bazzan et al., 2005; Davidsson et al., 2007). Future improvements to the model could include the addition of multiple freight forwarders and rail carriers and calculation of the reverse flows. The simulation results show a clear advantage for road transport. Intermodal transport only outperforms road transport in a few scenarios in the price variable. The results show that these variables (either individually or taken together) can hardly ever justify a Freight Forwarder’s choice for intermodality. Other variables or factors have certainly been more relevant. The hypothesis of the investigation could thus not be verified. We may therefore conclude that a gap in the body of literature concerning the relevant factors and mode choice variables in short distance transport services is likely to exist. Such a gap may be rendering it impossible to identify opportunities for short-distance intermodality and, consequently, develop tailored transport policies. Further research on the success factors and mode choice variables in short distance intermodal services is required.

118

V. Reis / Transportation Research Part A 61 (2014) 100–120

Appendix A. Formulation of model choice variables A.1. Price

Pn i

j¼1 ðPC ij Þ

Price Container i ¼

ni

where Price Containeri is the average price per container at destination i (i = 1, . . . , 7) paid by the Shipper to the Freight Forwarder, PCij is the price of container j (j = 1, . . . , ni) at destination i paid by the Shipper to the Freight Forwarder, and ni is the total number of containers dispatched to destination i. A.2. Transit time

Pni Transit Timei ¼

j¼1 ðTT ij Þ

ni

where Transit Timei is the average transit time to destination i (i = 1, . . . , 7), TTij is the Transit Time of container j (j = 1, . . . , ni) at destination i, calculated as:

TT ij ¼ TF ij  TIij ; where TFij is the time of unloading of container j at destination i, TIij is the time of unloading the container j from the ship at port of Sines with destination i, and ni is the total number of containers dispatched to destination i. A.3. Reliability

P7 Entropy ¼

i¼1 Ei

 ni

N

where Entropy is the value of entropy for a given transport service, ni is the total number of containers dispatched to destination i (i = 1, . . . , 7), and N is the total number of containers, calculated as:

N ¼ n1 þ n2 þ    þ n7 Ei is the entropy of destination i, calculated as:

PStateDelayi

Eiu  Stateiu ni

u

Ei ¼

where StateDelayi is the amount of hourly delays at destination i, calculated as:

for every destination i ¼ 1 to 7 for every container j ¼ 1 to ni StateDelay½i ¼ max½Delay½i; jÞ þ 1 where TTij is the transit time of container j at destination i,

Delayij



intðTT ij  72Þ; if TT ij > 72 0;

otherwise

is the hourly delay of container j at destination i;

Stateiu is the matrix of the frequencies of hourly delays for destination i, such as

for every destination i ¼ 1 to 7 for every container j ¼ 1 to ni State½i; Delay½i; j þ 1 ¼ State½i; Delay½i; j þ 1 þ 1 Eiu is a vector of the entropy for StateDelayu at destination i, calculated as:

( Eiu

Stateiu ni

 In



Stateiu ni



; if u < 1ðdelayÞ

0;

u ¼ 1ðno delayÞ

A.4. Flexibility

RTi ¼ maxðTDij  TFDÞ;

j ¼ 1 . . . ni

; u ¼ 1 to StateDelayi :

PStateDelayi u

Stateiu ¼ ni , calculated as follows:

V. Reis / Transportation Research Part A 61 (2014) 100–120

119

where RTi is the recovery time for destination i (i = 1, . . . , 7), TFD is the timing of fluctuation in demand, ni is the total number of containers dispatched to destination i (i = 1, . . . , 7), TDij is the timing when the following condition is met: ð9w1wÞ

TTij 6 1:1  TTi

ð9w1wÞ

where TTij is the transport time of container j at destination i and TTi tween the T(f9w) and T(f1w) calculated as:

Pnðfi 1wÞ ð9w1wÞ

TT i

ðf 9wÞ

¼

j¼ni ðf 1wÞ

ni

is the average transport time at destination i be-

TT ij ðf 9wÞ

 ni

ðf 9wÞ

where T(f9w) is the TFD minus 9 weeks, T(f1w) is the TDF minus 1 week, ni is the number of containers transported at ðf 1wÞ timing T(f9w) at destination i and ni is the number of containers transported at timing T(f-1w) at destination i. Appendix B. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/ j.tra.2014.01.002. These data include Google maps of the most important areas described in this article. References Asariotis, R., 1999. The need for an integrated intermodal transport liability regime. Transp. Quart. 53, 45. Axtell, R., 2000. Why Agents? On the Varied Motivations for Agent Computing in the Social Sciences. Baindur, D., Viegas, J.M., 2011. An agent based model concept for assessing modal share in inter-regional freight transport markets. J. Transp. Geogr. 19, 1093–1105. Bazzan, A.L.C., Klügl, F., Ossowski, S., Davidsson, P., Henesey, L., Ramstedt, L., Törnquist, J., Wernstedt, F., 2005. An analysis of agent-based approaches to transport logistics. Transp. Res. Part C: Emerg. Technol. 13, 255–271. Ben-Akiva, M.E., De Jong, G., 2013. The aggregate–disaggregate–aggregate (ADA) freight model system. In: Ben-Akiva, M.E., Meersman, H., Voorde, E. Van de (Eds.), Freight Transport Modelling. Emerald Group Publishing Limited, Bingley, UK, pp. 69–90. Ben-Akiva, M.E., Lerman, S.R., 1985. Discrete Choice Analysis: Theory and Application to Travel Demand. Massachusetts Institute of Technology Press, Boston, MA, United States. Ben-Akiva, M.E., Meersman, H., Voorde, E. Van de (Eds.), 2013. Freight Transport Modelling. Emerald Group Publishing Limited, Bingley, UK. Bergkvist, M., Davidsson, P., Persson, J., Ramstedt, L., 2004. A hybrid micro-simulator for determining the effects of governmental control policies on transport chains. In: Davidsson, P., Logan, B., Takadama, K. (Eds.), Multi-Agent and Multi-Agent-Based Simulation: Joint Workshop MABS 2004. Springer, Berlin, Germany, pp. 236–247. Blauwens, G., Baere, P.De., Voorde, E., 2006a. Transport Economics. Duculot, Belgium. Blauwens, G., Vandaele, N., Van de Voorde, E., Vernimmen, B., Witlox, F., 2006b. Towards a modal shift in freight transport? A business logistics analysis of some policy measures. Transp. Rev. 26, 239–251. Blauwens, G., Voorde, E. Van de, 1988. The valuation of time savings in commodity transport. Int. J. Transp. Econ. CV, 77–87. Bonabeau, E., 2002. Agent-based modeling: methods and techniques for simulating human systems. Proc. Natl. Acad. Sci. USA 99 (Suppl. 3), 7280–7287. Button, K., 2010. Transport Economics. Edward Elgar Publishing Ltd, Cheltenham, UK. Cascetta, E., 1996. A multi-regional input–output model with elastic trade coefficients for the simulation of freight transport demand in Italy. In: European Transport Conference. Association for European Transport, Uxbridge, United Kingdom. Cascetta, E., 2009. Transportation Systems Analysis: Models and Applications. Springer, Berlin, Germany. Chow, J.Y.J., Yang, C.H., Regan, A.C., 2010. State-of-the art of freight forecast modeling: lessons learned and the road ahead. Transportation (Amst) 37, 1011– 1030. Cloodt, H., 2012. Modal Split of Freight Transport According to the Territoriality Principle (2008–2009). EUROSTAT-Stat. Focus 8. Cullinane, K., Toy, N., 2000. Identifying influential attributes in freight route/mode choice decisions: a content analysis. Transp. Res. Part E: Logist. Transp. Rev. 36, 41–53. D’Este, G., 1996. An event-based approach to modelling intermodal freight systems. In: Hensher, D., King, J., Oum, T. (Eds.), Proceedings of the 7th World Conference on Transport Research, Sydney. PERGAMON, Sydney, Australia, pp. 3–13. Danielis, R., Marcucci, E., 2007. Attribute cut-offs in freight service selection. Transp. Res. Part E: Logist. Transp. Rev. 43, 506–515. Davidsson, P., Holmgren, J., Kyhlbäck, H., Mengistu, D., Persson, M., 2007. Applications of agent based simulation. In: Antunes, L., Takadama, K. (Eds.), MultiAgent-Based Simulation VII. Springer, Berlin, Germany, pp. 15–27. Davidsson, P., Holmgren, J., Persson, J., Ramstedt, L., 2008. Multi agent based simulation of transport chains. In: Padgham, L., Parkes, D. (Eds.), 7th International Joint Conference on Autonomous Agents and Multi-Agent Systems. International Foundation for Autonomous Agents and Multiagent Systems, Estoril, Portugal, pp. 1153–1160. De Jong, G., Gunn, H., Walker, W., 2004. National and international freight transport models: an overview and ideas for future development. Transp. Rev. 24, 103–124. Erlander, S., Stewart, N.F., 1990. The Gravity Model in Transportation Analysis: Theory and Extensions. VSP. European Commission, 1995. COM (95) 691 – Green Paper – Towards Fair and Efficient Pricing in Transport Policy – Options for Internalising the External Cost of Transport in the European Union, Brussels, Belgium. European Commission, 2001a. COM (2001) 264 Final – A Sustainable Europe for a Better World: A European Union Strategy for Sustainable Development, Brussels, Belgium. European Commission, 2001b. COM (2001) 360 Final – White Paper – European Transport Policy for 2010: Time to Decide, Brussels, Belgium. European Commission, 2006. COM(2006) 314 Final – Keep Europe Moving – Sustainable Mobility for Our Continent Mid-term Review of the European Commission’s 2001 Transport White Paper, Brussels, Belgium. European Commission, 2007. COM (2007) 606 Final – The EU’s Freight Transport Agenda: Boosting the Efficiency, Integration and Sustainability of Freight Transport in Europe, Brussels, Belgium. European Commission, 2009. COM(2009) 279 Final – A Sustainable Future for Transport: Towards an Integrated, Technology-Led and User Friendly System, Brussels, Belgium. European Commission, 2011. COM(2011) 144 Final – White Paper: Roadmap to a Single European Transport Area – Towards a Competitive and Resource Efficient Transport System, Brussels, Belgium.

120

V. Reis / Transportation Research Part A 61 (2014) 100–120

Eurostat, 2012. Road Freight Transport Statistics. (WWW Document, accessed 07.01.13). Fiorello, D., Fermi, F., Bielanska, D., 2010. The ASTRA model for strategic assessment of transport policies. Syst. Dyn. Rev. 26, 283–290. Fischer, K., Kuhn, N., Müller, H.J., Müller, J.P., Pischel, M., 1995. Sophisticated and distributed: The transportation domain – exploring emergent functionality in a real-word application. In: Castelfranchi, C., Müller, J.-P. (Eds.), From Reaction to Cognition. Springer, Berlin, Germany, pp. 122–138. Frémont, A., Franc, P., 2010. Hinterland transportation in Europe: combined transport versus road transport. J. Transp. Geogr. 18, 548–556. García-Menéndez, L., Martínez-Zarzoso, I., Miguel, D.P.De., 2004. Determinants of mode choice between road and shipping for freight transport: evidence for four Spanish exporting sectors. J. Transp. Econ. Policy 38, 447–466. Garrido, R.A., 2000. Spatial interaction between the truck flows through the Mexico-Texas border. Transp. Res. Part A: Policy Pract. 34, 23–33. Grue, B., Ludvigsen, J., 2006. Decision factors underlying transport mode choice in international European freight transport. In: European Transport Conference. Association for European Transport, Strasbourg, France. Gruppo CLAS, 2000. Final Report for Publication. LOGIQ Project, 4th Framework Programme. Holmgren, J., Davidsson, P., Persson, J.A., Ramstedt, L., 2012a. TAPAS: a multi-agent-based model for simulation of transport chains. Simul. Model. Pract. Theory 23, 1–18. Holmgren, J., Ramstedt, L., Davidsson, P., 2012b. Multi-agent-based simulation for analysis of transport policy and infrastructure measures. In: Cranefield, S., Song, I. (Eds.), Agent Based Simulation for a Sustainable Society and Multi-Agent Smart Computing. Strasbourg, France, pp. 1–15. INRETS, 2000. Final Report for Publication. IQ – Intermodal Quality, 4th Framework Programme. Janic, M., 2007. Modelling the full costs of an intermodal and road freight transport network. Transp. Res. Part D: Transp. Environ. 12, 33–44. Jeffs, V.P., Hills, P.J., 1990. Determinants of modal choice in freight transport. Transportation (Amst) 17, 29–47. Jensen, A., 1990. Combined Transport: Systems, Economics and Strategies, Stockholm, Sweden. Jourquin, B., 1995. Un outil d’analyse économique des transports de marchandises sur des réseaux multi-modaux et multi-produits: Le réseau virtuel, concepts, méthodes et applications. Facultés Universitaires Catholiques de Mons, Belgium. Konings, R., Priemus, H., Nijkamp, P., 2008. The Future of Intermodal Freight Transport. Edward Elgar Publishing, Cheltenham, United Kingdom. Leinbach, C., Capineri, T.R., 2006. Freight transport, seamlessness, and competitive advantage in the global economy. Eur. J. Transp. Infrastruct. Res. 6, 23–37. Liedtke, G., 2009. Principles of micro-behavior commodity transport modeling. Transp. Res. Part E: Logist. Transp. Rev. 45, 795–809. Macharis, C., Van Hoeck, E., Pekin, E., van Lier, T., 2010. A decision analysis framework for intermodal transport: comparing fuel price increases and the internalisation of external costs. Transp. Res. Part A: Policy Pract. 44, 550–561. Maeyer, J. De, Pauwels, T., 2003. Mode Choice Modelling – A Literature Review on the Role of Quality of Service Attributes and Their Monetary Valuation in Freight Demand Models. Matear, S., Gray, R., 1993. Factors influencing freight service choice for shippers and freight suppliers. Int. J. Phys. Distrib. Logist. Manage. 23, 25–35. McGinnis, M., 1989. A comparative evaluation of freight transportation choice models. Transp. J. 29, 36–46. McGinnis, M., 1990. The relative importance of cost and service in freight transportation choice: before and after deregulation. Transp. J. 30, 12–19. Murphy, P.R., Daley, J.M., Hall, P.K., 1997. Carrier selection: Do shippers and carriers agree, or not? Transp. Res. Part E: Logist. Transp. Rev. 33, 67–72. Murphy, P.R., Hall, P.K., 1995. The relative importance of cost and service in freight transportation choice before and after deregulation: an update. Transp. J. 35, 30–38. Norojono, O., Young, W., 2003. A stated preference freight mode choice model. Transp. Plan. Technol. 26, 195–212. North, M.J., Macal, C.M., 2007. Managing Business Complexity: Discovering Strategic Solutions with Agent-Based Modeling and Simulation. Oxford University Press, Oxford, England. Oum, T.H., 1979. Derived demand for freight transport and inter-modal competition in Canada. J. Transp. Econ. Policy 13, 149–168. Oum, T.H., 1989. Alternative demand models and their elasticity estimates. J. Transp. Econ. Policy 23, 163–187. Persson, J.A., Davidsson, P., 2005. Integrated optimization and multi-agent technology for combined production and transportation planning. In: Proceedings of the 38th Annual Hawaii International Conference on System Sciences. IEEE, Waikoloa Village, HI, United States, p. 72b–72b. Reis, V., 2010. Development of Cargo Business in Combination Airlines: Strategy and Instrument. Instituto Superior Técnico, University of Lisbon. Rich, J., Kveiborg, O., Hansen, C.O., 2011. On structural inelasticity of modal substitution in freight transport. J. Transp. Geogr. 19, 134–146. Savy, M., 2009. Freight Transport Modes: Competition, Cooperation or Areas of Advantage? Brussels, Belgium. Shannon, C.E., 1948. A mathematical theory of communication. Bell Syst. Tech. J. 27, 379–423. Shinghal, N., Fowkes, T., 2002. Freight mode choice and adaptive stated preferences. Transp. Res. Part E: Logist. Transp. Rev. 38, 367–378. Slack, B., 2001. Intermodal transportation. In: Brewer, A.M., Button, K.J., Hensher, D.A. (Eds.), Handbook of Logistics and Supply-Chain Management. Emerald Group Publishing Limited, pp. 141–154. Sterman, J.D., 2004. Business Dynamics: Systems Thinking and Modeling for a Complex World. Emerald Group Publishing Limited. Tavasszy, L., 2006. Freight modelling – an overview of international experiences. In: TRB Conference on Freight Demand Modelling: Tools for Public Sector Decision Making. Transportation Research Board, Washington, DC, United States, p. 12. Tavasszy, L.A., Ruijgrok, K., Davydenko, I., 2012. Incorporating logistics in freight transport demand models: state-of-the-art and research opportunities. Transp. Rev. 32, 203–219. Tavasszy, L.A., Smeenk, B., Ruijgrok, C.J., 1998. A DSS for modelling logistic chains in freight transport policy analysis. Int. Trans. Oper. Res. 5, 447–459. The Economist, 2009. Crowd Modelling – Model Behaviour. Econ. Train, K., 2009. Discrete Choice Methods with Simulation. Cambridge University Press. Tsamboulas, D., 2008. Development strategies for intermodal transport in Europe. In: Konings, R., Priemus, H., Nijkamp, P. (Eds.), The Future of Intermodal Freight Transport – Operations, Design and Policy. Edward Elgar Publishing Ltd, Cheltenham, United Kingdom, p. 360. Tsamboulas, D., Vrenken, H., Lekka, A.-M., 2007. Assessment of a transport policy potential for intermodal mode shift on a European scale. Transp. Res. Part A: Policy Pract. 41, 715–733. United Nations, 2001. Terminology on Combined Transport, Geneva, Switzerland. Weiss, G. (Ed.), 1999. Multiagent Systems: A Modern Approach to Distributed Artificial Intelligence. Massachusetts Institute of Technology Press, London, United Kingdom. Wooldridge, M., 2009. An Introduction to Multi Agent Systems. John Wiley & Sons Ltd., West Sussex, United Kingdom. Woxenius, J., 1998. Development of Small Scale Intermodal Freight Transportation in a Systems Context. Chalmers University of Technology.