Specification of a Logistics Model for Norway and Sweden
G de Jong, RAND Europe, NL and ITS, University of Leeds, UK; M Ben-Akiva, RAND Europe, NL and MIT, US; M Florian, INRO Consultants, CN
A logistics model (including shipment size and mode choice and use of consolidation and distribution centres) is specified. This was developed as a new component of the national freight transport forecasting systems of Norway and Sweden.
Specification of a logistics model for Norway and Sweden
Gerard de Jong, RAND Europe and ITS Leeds
Moshe Ben-Akiva, RAND Europe and Massachusetts Institute of Technology
Mike Florian, INRO Consultants
The national model systems for freight transport in Norway and Sweden are lacking logistic elements (such as the use of distribution centres). The same can be said about practically every other national or regional freight transport model system (an exception is the SMILE model in The Netherlands) in the world. This paper contains the outcomes of a project to specify how a new logistics module for such model systems could look like. The project was carried out for the Work Group for transport analysis in the Norwegian national transport plan and the Samgods group in Sweden by RAND Europe, together with Solving International, Solving Bohlin & Strömberg and Michael Florian of INRO Canada.
The paper will discuss the scope and structure of a logistics model, as well as ideas on the implementation.
The scope of the logistics model concerns the boundary lines with other parts of the national freight model systems, notably the base matrices and the network model. We recommend that both for Norway and Sweden the base matrices will (initially) be PWC (production-wholesale-consumption) matrices, using the current zoning systems and at least the current 12/13 commodity groups. This means that not only flows from P (production) to C (consumption by intermediate producers and retail) are included in the base matrix, but that flows from P to W (wholesale) and from W to C are included as well (treating W just as C and P respectively). The use of consolidation centres and distribution centres between P, W and C by shippers and carriers is covered in the logistics model, which will produce transport flows at the origin-destination (OD) level.
For the new freight model systems, an ?aggregate-disaggregate-aggregate? model is proposed. The base matrix (initially PWC flows and later P/C flows) and the network model need to be specified at an aggregate level (with zones as unit of observation), for reasons of data availability and ease of interpretation. For the logistics model, that is positioned between these two models, we proposed a disaggregate model, at the level of the firm, which is the actual decision unit in freight transport.
This logistics model itself consists of three steps:
A. disaggregation to allocate the flows to individual firms at the P and C end
B. models for the logistics decisions by the firms
C. aggregation of the information per shipment to origin-destination (OD) flows for assignment.
The allocation of flows in tonnes between zones (step A) to individual firms at the P and C end can to some degree be based on observed proportions of firms in local production and consumption data and business register data.
The logistics decisions in step B are derived from minimisation of the full logistics and transport costs (modelled as discrete choice models). The different logistics decisions that might be included in step B are:
? Which agent (=a firm) controls the relevant supply chain?
Choice set: the manufacturer at the P end, the retailer or manufacturer at the C end.
? Lead time.
The choice set could contain as alternatives: within 24 hours, within 48 hours, within a week, more than a week.
? Frequency/shipment size (so inventory decisions are endogenous).
The choice set for shipment size could be based on a categorisation in tonnes. Alternatively a functional classification (e.g. less-than-truckload, more-than- truckload) could be used. The latter will probably provide more insight.
? Whether the shipping firm carries out the transport and logistics itself or it contracts out to carriers/logistic service providers (one could go even further and distinguish several types of outsourcing).
Choice set: do it yourself versus contracting out.
? Choice of loading unit.
Choice set: especially the distinction between containerised versus non-containerised
? Use (and location) of distribution centres, freight terminals, ports and airports and the related consolidation and distribution of shipments and formation of tours (batching shipments at consolidation centres, multi-stop deliveries).
Choice set: chains of zones, with a specific activity (e.g. origin, consolidation, distribution and destination) at each zone. For the decision of the optimal location of consolidation and distribution centres only a limited number of candidate sites might be available.
? Mode used for each tour leg.
Choice set: air transport, road transport (possibly several vehicle types), rail transport (possibly with different train types, such as regular trains, block trains and intermodal rail transport), and maritime transport (possibly with different vessel types).
However, a logistics model which would include all these choices would be very complicated and data-demanding. We developed different options for the logistics model by simplifying the above full model and compared these options. In our recommended option, mode choice is handled in the logistics model, where observed and unobserved variation in choices made can be treated within the random utility maximisation framework. Serious difficulties can be expected when trying to make lead time and supply chain dominance endogenous and having the segmentation (partly) result from the mode estimation, and this is not part of the recommended option. Also the cutting of the logistics choices in two parts (shipment size choice and transport chain and mode choice) will make the model more tractable and easier to develop and apply. For short term applications of the model, it is also not needed that the location choice for consolidation and distribution centres is included. The existing locations can be used to define the choice alternatives for the transport chain selection. However, for long-run applications, it would be of considerable value to have the locations of consolidation and distribution centres in the private sector produced by the model (the locations of ports and airports remain exogenous).
The paper will further work out the structure and components of the model sketched above, and also discuss the data requirements, which will be confronted with the data that are available in Norway and Sweden to identify data gaps. Estimation and validation issues for the logistics model will be discussed as well.
Association for European Transport