Modelling Production-consumption Flows of Goods in Europe: the Trade Model Within Transtools3

Modelling Production-consumption Flows of Goods in Europe: the Trade Model Within Transtools3


Gerard De Jong, ITS Leeds, Reto Tanner, ITS Leeds and TU Delft, Otto Anker Nielsen, DTU Transport


Estimation results and elasticities are presented for the trade model within the European transport model Transtools3. We also explain how the outcomes of this model are used in the overall freight model.


1. The general model
The Transtools3 model is a new forecasting model system for transport in Europe, developed by a consortium led by DTU Transport from Denmark for DG MOVE of the European Commission. It consists of three main blocks: the passenger transport model, the freight and logistics model and the network assignment model. This paper focuses on the trade model, a specific submodel within the freight and logistics model. This is the top-right part in Figure 1, which presents the overall model. The trade model produces growth between base year and future year in the goods flows between production and consumption zones (PC flows, in tonnes). Together with the base PC matrix, the trade model produces future year aggregate PC matrices.

2. The trade model
The trade model explains the transport flows, either between NUTS3 zones or between countries, by NST/R 1 commodity type, based on the (unconstrained) gravity formulation, from characteristics of the zones and their transport resistance. The dependent variable is the base PC matrix, as delivered by another project for DGMOVE, the ETIS+ project. The independent variables are GDP, GDP per capita (both obtained from the World Bank) and dummies for common trade zone (EU, EFTA), common currency zone (EURO), common language, zones being neighbours and zones being in the same country. The resistance variable in the current estimates is geographic distance, modelled using a spline function, but we plan to replace this later by transport costs to make the trade model sensitive to transport policies that alter transport costs.

To take into account the influence of relative trade resistance between countries instead of absolute resistance (in line with trade theory), we estimated not only least squares regression models but also fixed effects and random effects models.
A key problem of standard regression models is that they cannot handle the case where transport flows are zero, which is very relevant here since many countries do not trade (within a commodity group) with each other. As a result there can be a sample selection bias. In order to capture the decision whether or not to trade, we estimated Heckman sample selection models with a simultaneous trade selection equation and a positive demand equation to correct for this effect.

In the current model implementation, we only use the predicted changes over time (scenario-based) in GDP by zone and in the population to compute changes in the transport flows by commodity type and zone pair. All other variables are assumed to remain constant.

3. The combination of the future PC flows and the transport chain choice in the implemented model
The output of the trade model (after pivoting) consists of future year aggregate PC matrices. The logistics model is a disaggregate model for the choice of transport chain, estimated on the French ECHO survey and the Swedish CFS. In the application of the transport chain model, we use a prototypical sample of shipments. For each PC relation, we draw shipments from this prototypical sample until the sum of their shipment sizes is equal to the number of tonnes per year on the PC relation (by commodity type).
After this, we apply the logistics model for each PC relation on the shipments that were drawn for that PC relation. The output will then consist of choice probabilities that we sum over the shipments in the sample for the PC relation to get the predicted frequency for each transport chain for each PC relation and commodity type.

The legs of the transport chain by mode and commodity are summed over the PC relations to produce aggregate OD matrices by mode and commodity type (in tonnes), which are then (after pivoting) used in the network assignment. The model as a whole is a form of an aggregate-disaggregate-aggregate or ADA model (see Ben-Akiva and de Jong, 2010).

4. The paper
The paper will discuss the existing literature in gravity-based trade models. It will describe the data and model structures used and present the estimation results for various specifications and the limitations of the model. Elasticities for changes in GDP and population will be provided. The paper will also discuss the structure of the overall freight model and how PC matrices from the trade model are combined with the disaggreg


Association for European Transport