Modelling Demand by Time of Day for High Speed Rail Services in France
E Bulman, A Hanif, NERA Economic Consulting, UK; A Sauvant, Réseau Ferré de France, FR
Réseau Ferré de France commissioned NERA to develop a time-of-day model to examine the potential for managing demand on its rail infrastructure. NERA devised a hybrid approach based on the logit model, diversion factors and fare elasticities.
Réseau Ferré de France (RFF), the rail infrastructure manager in France, is responsible for planning rail network enhancements, including extensions and expansions of capacity for high speed rail infrastructure. As such, RFF prepares projections in passenger demand over a long time horizon (50 years or so) and under a range of scenarios. It assesses both the benefits of new infrastructure, and the potential to manage demand (for train seats and hence from train paths) with reasonable investments in the existing infrastructure.
Within this context, RFF commissioned NERA Economic Consulting to develop a model of passenger demand differentiated by time of day; and to test, for a particular group of high speed services, how fares across the day, for different geographical groups and reasons to travel, would need to change so that demand could be accommodated within existing or slightly increased capacity.
NERA?s experience is that traditional discrete choice models, including standard logit models, may not suit highly disaggregate time-of-day modeling (where adjacent time periods can be close substitutes, but the relationship is not transitive). In a standard multinomial logit model, for example, an increase in fare in one time period would tend to result in passengers switching to all times of day; whereas the predominant response can b surmised to be a switch to adjacent hours.
An alternative approach, given cautious support by the UK rail Passenger Demand Forecasting Handbook (PDFH), is to forecast switching between ticket types (with time-of-day restrictions) using diversion factors derived from surveys, combined with evidence on fare elasticities. Such an approach is, however, vulnerable to misapplication by the uninitiated (it is easy to forecast an increase in rail demand in response to an increase in rail fares). More fundamentally, the factors used can only be applied to a narrow range of fares; and they are ill equipped to model multiple fare changes.
Instead, NERA developed a hybrid approach, using a more generalized form of the logit model. Under the approach, as with the ?PDFH method? of using diversion factors, switching time of travel was calculated separately for each time period. NERA?s method has the following properties:
- the proportion of passengers diverting from one time period to each of the other time periods is a function of the generalized cost of switching time of travel; the responses vary according to journey purpose, determined on the basis of empirical evidence;
- if fares in two time periods increase by the same percentage, NERA imposed the constraint that passengers do not switch between the periods;
- the conditional elasticity of rail demand (where all rail fares change by the same proportion) is pegged to external analytical evidence; and
- the proportion of passengers switching to and from rail (for example, to air travel) is a function (using a logit function formulation) of the conditional fares elasticity and the relative attractiveness of travel by rail at other times of day.
NERA tested the model on a variety of demand profiles, including longer and shorter high speed journeys to and from a major city, with different degrees of capacity constraint. The model differentiated between business and leisure passengers. NERA tested plausible fare structures, and determined the tariffs that result in demand satisfying the capacity constraints. The model showed the potential to differentiate fares by time of day in order to shift passengers to less busy times. As a consequence, the fall in passenger demand, and increase in fares, required to meet capacity constraints can be minimized. The model also takes into account different passengers? behaviour according to the reason to travel.
From RFF?s point of view, this model has been successfully implemented. It has highlighted the following strategic lessons:
- highly used high speed lines can be managed using innovating techniques for a while, combining limited investments and innovative demand management, using finely tuned infrastructure tarification.
- in the long term, however, major capacity investments, starting with the nodes, and then carrying on with a new line, cannot be avoided.
Further studies and research will be needed in order to implement this demand management strategy, but this study has led the way.
Association for European Transport