Vehicle stock modelling in long term projections - Survey of the literature [ Working Paper 08-17 - 26/07/2017]

  PDF & Download

• Publication (en - 903 Kb)

  Authors

Laurent Franckx (A)
 
A : Author, C : Contributor

  Publication type

Working Papers

The Working Paper presents a study or analysis conducted by the Federal Planning Bureau on its own initiative.

Our key conclusion is that the modelling approach taken should depend on the time perspective of the model and on the availability of data. In the short run (up to 5 years in the future), consumer preferences and vehicle technologies can be assumed to remain stable, and the evaluation of policies is best based on econometric models. In the long run (more than 15 years in the future), preferences and technologies are fundamentally uncertain, and a scenario approach is more appropriate than forecasting. For medium term projections (5 to 15 years in the future), we propose to enrich existing econometric models with models of social learning by households and of learning-by-doing by firms. The “synthetic utility” approach holds the most promise for rapid integration in PLANET of vehicle types that currently have low or zero market shares, but are expected to have an important potential in the long run, such as electric and automated vehicles.

Executive summary

It is expected that some externalities caused by the transport sector will get worse in the future, especially greenhouse gas emissions, congestion and local air pollution. However, the transport sector is also on the brink of a possibly transformative change caused by the simultaneous rise of shared, automated and electric mobility modes. The actual growth potential of these new technologies and business models and their impact are highly uncertain.

Depending on the policy context, they could solve the major externalities caused by transport, or exacerbate them to the point of disaster. On the one hand, the number of cars needed to satisfy mobility demand is likely to drop significantly if shared and automated cars become commonplace. On the other hand, the need to reposition shared and automated cars to get new passengers is likely to lead to a dramatic increase in vehicle kilometres, and to a much higher turnover of the vehicle stock. Transport models thus need to be adapted to reflect the impact of these developments. The current paper focuses on the issue of vehicle ownership modelling in long-term projections. Our objective is to derive lessons from the existing literature to find options to further improve the PLANET model, which has been developed by the Federal Planning Bureau to provide long-term projections of transport demand in Belgium.

Vehicle stock models are usually embedded in broader models, which also include projections of total mobility demand, modal choices, route choices etc. The representation of the vehicle stock is therefore related to other key choices in the design of the transport model, especially:

• The integration of the transport model in a model of the wider economy;
• The relation between overall travel demand and vehicle choice;
• The treatment of indirect emissions;
• The opportunity cost of time spent travelling;
• The relative weight given to explicit economic modelling versus expert judgement.

The PLANET model can be described as a sectoral model of the transport sector, which follows the “service demand approach”: it first models travel demand for all modes combined, and then allocates total demand to individual modes. The total number of cars is determined at the aggregate level as the number of cars that is needed to meet the expected mobility demand. Aggregate demand is further split over individual vehicle classes. No attempt is made to estimate the number of cars at the household level. PLANET considers only well-to-wheel emissions. For instance, it does not consider the environmental effects of the production and the scrapping of vehicles. The opportunity cost of time is included in the generalised cost of transport. Therefore, PLANET is well equipped to represent that automated cars reduce the opportunity cost of travel time.

In its representation of transport technologies, PLANET is a “hybrid” model: for passenger cars, the demand is explicitly modelled in detail while the technologies for other modes are described at a higher level of abstraction. It is thus neither a pure “top-down” nor “bottom-up” model. Several models represent non-technical parameters, such as the heterogeneity of the households, social influences and contextual conditions. PLANET, in contrast, does not account for individual household characteristics. Linking vehicle registration data with household surveys holds some potential to improve the representation of consumer behaviour in PLANET.

One increasingly important issue is the demand for vehicle types that have currently low or zero market shares, but that are expected to have an important potential in the long run, such as electric and automated vehicles. The most promising approach to deal with this issue consists in combining observations of ‘real’ markets with outcomes of ‘hypothetical’ markets. Alternatively, one can use “synthetic utility functions”, whose parameters are based on “first principles” or extensive literature surveys rather than on individual studies that use limited samples in a specific context. Taking into account data availability in Belgium, the “synthetic utility” approach holds the most promise for rapid integration in the existing version of PLANET.

Of all the approaches to vehicle stock modelling in long term projections, the one with the most solid foundations in economic theory consists in taking the annual sales of vehicles as “residual” variable:

• The existing stock of vehicles is retired according to a scrappage function.
• The total vehicle stock is estimated as the stock that is needed to meet travel demand (expressed in passenger kilometres), taking into account load factors and the average vehicle kilometres for existing vehicles.
• The annual vehicles sales are calculated as the desired car stock minus the actual car stock inherited from previous vintages.
• Total sales are split in classes (e.g. according to fuel or vehicle size) using a discrete choice function.

In order to represent the impact of shared mobility, it is important that parameters such as the load factors and the annual mileage of cars can be modified according to the scenario that is under consideration. The PLANET model can be readily modified to deal with this requirement.

Our key conclusion is that the modelling approach taken should depend on the time perspective of the model and on the availability of data. In the short run (up to 5 years in the future), consumer preferences and vehicle technologies can be assumed to remain stable, and the evaluation of policies is best based on econometric models. In the long run (more than 15 years in the future), preferences and technologies are fundamentally uncertain, and a scenario approach is appropriate: models should explore possible futures, and outline the implications of the most extreme scenarios that still seem plausible. For medium term projections (5 to 15 years in the future), we propose to start with existing econometric models, but enriched with models of social learning by households and of learning-by-doing by firms. Repeated simulations with random changes in the values of the key parameters will help us understand the robustness of our projections.