In our study of the cost benefit assessment of measures reducing road trans-port emissions, we apply a capita selecta approach. The study consists of two main parts: behavioural analysis and policy simulation.
The first part analyses car driver preferences for alternative fuels and vehicle technologies. This part extensively draws on discrete choice theory for the analysis of survey data and the design of a simulation tool.
1,2 for the marginal cost of public funds and a share of labour income in total income of 70%.
The contribution of the MCPF component to welfare is then equal to 6,6% of the increase in tax income for the government. Other studies (for instance Kleven and Kreiner, 2003) present higher values for the marginal cost of public funds. In our simulations we will report the contribution of the MCPF cost component separately in order to provide an indication for the sensitivity of the net welfare result to the assumed allocation of the change in tax revenues and the valuation of the selected allocation. Note that these assumptions do not feed back in the model but affect the size of the MCPF social cost component only.
5In our simulations we use values for the marginal external emission cost coefficients from a draft version of TREMOVE 2 which are based mainly on Friedrich and Bickel (2001) and which are presented in appendix C. Other studies (for instance Maibach et al., 2008) as well as later versions of TREMOVE 2 (based on Holland, Pye, Watkiss, Droste-Franke, and Bickel, 2005) use values that are lower by a factor of up to three. In our simulations we will report the contribution of the external emission cost component separately in order to provide an indication for the sensitivity of the net welfare result to the valuation of external impact.
6As opposed to the other pollutants we use for CO2emissions an external cost factor that is based on a reference value of economy wide marginal abatement cost. The rationale is that total EU-wide CO2emissions are capped by the Kyoto agreements. An increase of CO2emissions in one sector needs hence to be compensated by a decrease elsewhere in the economy. The reference value we use for external abatement cost is provided by TREMOVE 2 and is based on Holland, Hunt, Hurley, Navrud, and Watkiss (2005). This value is of a similar order of magnitude as values provided in literature for marginal environmental impact by CO2emissions.
Methodological Approach & Reading Guide
A first chapter discusses the design of a conjoint choice experiment and presents the corresponding findings on car purchase behaviour.
The second chapter presents an approach to use the survey data set for the design of a simulation tool for car technology procurement. It is illustrated how a nested mixed logit model specification allows to account for the repeated choice setting of the survey in order to identify correlated preferences for technologies. In a subsequent step a computationally efficient nested logit specification that replicates the correlation pattern is presented.
Finally the alternative technology model is integrated in the larger trans-port modelling framework of TREMOVE which is a proven model for evalua-tion of emission policies. This integrated model will be used for simulaevalua-tion of transport scenarios in the second part of the study.7
The second part of the study presents the assessment of the effectiveness and efficiency of a series of transport scenarios in which technological innova-tion is used to reduce road transport emissions. A first chapter studies the contribution to be expected from alternative private car technologies and fuels.
A second chapter is dedicated to energy efficiency and the corresponding CO2
emissions from traditional private car technologies. A last chapter studies contributions to emission reductions from public transport activity in urban areas.
For each simulation chapter a customised version of the TREMOVE model is implemented. The role of this simulation tool is to provide a consistent framework for the comparison of the effectiveness and efficiency of the differ-ent technical and non-technical measures.
The model simulations each focus on the impact of a small change in external variables on transport activity and emissions over the modelling period. These simulations allow us to study the impact of isolated policy measures.
The time period over which to change the course of external variables in a scenario is entirely arbitrary in a simulation exercise. In the context of this study we should however account for the lifetime of technologies in the vehicle stock. In order to allow for an understanding of long term impact, simulated measures should cover a sufficiently long time period. At the same time the end of the modelling period was fixed to 2020 to fit the availability of sufficiently detailed (and consistent) external baseline forecasts for model
7The methodological approach for the simulations is to use the framework of the TREMOVE model as a starting point and extend it. This approach allows for the use of an existing and proven modelling framework (including a consistent reference scenario) while focussing on behavioural extensions of it. A drawback is that each extension needs to be designed consistent with the larger modelling framework dictated by TREMOVE (an issue which is much the focus of the second chapter). An alternative bottom-up approach would be to design a partial equilibrium model tailored for (and entirely consistent with) the behavioural topic of the study. That would be an enormous exercise, and while allowing for more theoretical flexibility it would still be confined by limitations in data availability.
calibration.8
It is key to understand that the strength of the modelling approach is in the consistent comparison of individual measures. By no means is the model to be regarded as a forecast tool. To forecast developments in transport activity one has to consider a wide range of variables, whereas in TREMOVE a baseline evolution of activity and prices over the modelling period is taken as given in the model calibration, hence any forecasting is external to the model.
Furthermore, TREMOVE is not an optimisation tool. In a static context the welfare optimal price levels under a series of constraints is an illuminating exercise. But the time dynamic nature of the technology stock makes opti-misation burdensome, and the results would be dependant on the applied discount rate. As noted by Arrow et al. (2004), the right value of the discount rate is a matter of much debate which we will not further consider here.
For clarity we want to draw the attention of the reader to the monetary unit used throughout all simulation exercises presented in this study. Unless mentioned differently, all costs and prices have been expressed in constant prices of the year 2000. For actualisation a social discount rate of 4% per year was used.9
The common base for the simulations in the second part is the TREMOVE model for Belgium. A hybrid version is used that draws mainly on version 1.3a with some relevant upgrades from version 2 included. A discussion of the base model including baseline and other assumptions is presented in appendix C.
In the first chapter of the second part the extended private car technology choice model developed in part one is added to the base model.
The second chapter uses a model where availability of private car technol-ogy is again limited to traditional diesel and gasoline technologies, but the model is now extended with an internal representation of fuel efficiency for all road transport vehicles.
The last chapter limits the model to the Brussels metropolitan area and uses an updated baseline for public transport modes. The model is here extended with an internal representation of optimal public transport user prices and service provision level.
8To provide an indication we mention here that in our simulation model the expected (technical) lifetime of a private car is 9,5 years (see appendix D). The modelling horizon sufficiently accommodates for the representation of the corresponding transient effects.
9Zhuang, Liang, Lin, and De Guzman (2007) provide an overview of the state of practise around the world in public discount rate policy. They draw the conclusion that developed countries apply rates of 3–7%. In our simulations we will report undiscounted costs for a selection of years over the modelling period, in order to provide an indication of the evolution of costs.
The discount rate is applied where these yearly costs are aggregated in a single net present value indicator.
Notation
Model design and estimation
The notation used in discrete choice model design and estimation is applied in chapters 1 and 2 and appendixes A and C.
Notation Meaning
j, k, m, n indexes used to indicate alternative j, nest k, choice set or choice situation m and consumer or respondent n
Ujmn random utility of choice alternative j as obtained by consumer n in choice situation m
Vjmn deterministic part of Ujmn, function of xjmn
α, β vector of model coefficients
ˆβ estimate of β
xjmn vector of variables relating to consumer n and alternative j in choice situation m
zjmn vector of variables relating to consumer n and alternative j in choice situation m (used in mixed logit utility specification) ejmn stochastic part of Ujmn
µjmn stochastic utility in a mixed logit model specification (expected value of µjmnis zero)
Pjmn choice probability of alternative j chosen by consumer n in choice situation m
E(γ) expected value of a stochastic variable γ Var(γ) variance of a stochastic variable γ
ηkmn stochastic utility in a nested logit model specification σ scale parameter of the Gumbel distribution
Ikmn inclusive value of nest k
λk inclusive value coefficient of nest k
p p-value, indication of significance of a coefficient estimate continued on next page
Notation
continued from previous page Notation Meaning
t t-statistic, indication of confidence interval of a coefficient esti-mate
K number of nests
N sample size (=number of respondents) M number of choice sets per respondent J number of alternatives per choice set
B amount the respondent indicated he or she would spend on a new car in case he or she had to buy one at the moment of the survey
δ dummy variable, can have values 0 or 1 only
∆ difference
r ratio of variance of two stochastic terms, used to compare two model estimations
f(γ) probability density function of a stochastic variable γ F(γ) cumulative distribution function of a stochastic variable γ Sk set of alternatives j that belong to nest k
a scale factor, used to scale utility Ujmn of a model s scale factor, used to scale the estimation data set (xjmn) dn expected annual mileage by respondent n
PCjmn value of the purchase cost variable for alternative j in choice set m faced by respondent n
ACjmn annual cost FCjmn fuel cost LFCjmn lifetime cost
i discount rate used to amortise medium run capital investment y expected vehicle lifetime
Vc,p,r,t speed in year t in period p on road type r for vehicle class c Fp,r,t flow (in passenger car units per hour) in period p on road type r Ac,r,t coefficient in congestion function
Bc,r,t coefficient in congestion function
c vehicle class: truck/bus or private car/motorcycle r road type: Brussels, other urban, motorway or other road p period: peak or off-peak
Policy simulation
The notation presented in the following table is applied in chapters 3, 4 and appendix D.
Notation Meaning
f fuel intensity in litre per kilometre p fuel price in euro per litre
e fuel price elasticity of fuel intensity
UC user cost (including taxes) in euro per vehicle kilometre
RC resource costs (including taxes on resource costs) in euro per vehicle kilometre
e emission factor in gram per vehicle kilometre Eap annual emissions of pollutant p at vehicle age a
Cp marginal external emission cost coefficient for pollutant p in€ per ton
C constant
δj dummy variable for technology j
βcat coefficient of choice model for technology class cat LFCj lifetime user cost of technology j
y expected vehicle lifetime
EC expected emission cost over the entire expected vehicle lifetime i discount rate used to amortise medium run capital investment
∆ difference
λk inclusive value coefficient of nest k of a nested logit model speci-fication
v CO2emissions per fuel unit
The notation presented in the following table is applied in chapter 5.
Notation Meaning
Co is the operating cost in euro per vehicle kilometre (vkm) D is the average occupancy rate (in travellers per vehicle) Vw is the value of time during waiting (in euro per hour)
f is the average frequency (in departures per hour per direction) Lt is the average trip length (in pkm)
Cw is the walking cost from/to the stop
Vt is the in-vehicle value of time (in euro per hour)
Ce is the marginal external emission cost (in euro per vkm) T is the commercial travel time (in hour per km)
q is the level of demand (in pkm per hour) Ln is the network length (in km)
B number of buses
S no-travellers speed
δ time (in hours) necessary to slow down, open and close the doors and to re-accelerate at a stop
continued on next page
Acronyms
continued from previous page Notation Meaning
e time (in hours) necessary to let a user embark/disembark d the average stop distance
a, b COPERT III technology specific parameters
Acronyms
Acronym Meaning
ACEA Association des Constructeurs d’Automobiles CAFE Corporate Average Fuel Economy
CATI computer assisted telephone interviewing CES constant elasticities of substitution CNG compressed natural gas
CH4 methane
C6H6 benzene
CO carbon monoxide
CO2 carbon dioxide
EU European Union
EV electric vehicle
GDP gross domestic product
H2 hydrogen
HDV heavy duty vehicle
HH household
IG integrated model
IIA independence from irrelevant alternatives iid independent and identical distributed ICE internal combustion engine
JAMA Japan Automobile Manufacturers Association KAMA Korea Automobile Manufacturers Association
LA Los Angeles
LDV light duty vehicle LFC lifetime cost LL log-likelihood
LPG liquefied petroleum gas MCPF marginal cost of public funds
continued on next page
continued from previous page Acronym Meaning
MIE main income earner
MIVB Maatschappij voor het Intercommunaal Vervoer te Brussel
ML mixed logit
MNL multinomial logit
NG natural gas
NL nested multinomial logit
NMBS Nationale Maatschappij der Belgische Spoorwegen NMVOC non methane volatile organic compounds
N2O laughing gas NOX nitrogen oxides NPV net present value pkm passenger-kilometre PM particulate matter RP revealed preference
RUM random utility maximisation SP stated preference
TEC Transport En Commun tkm ton-kilometre
TWC three way catalyst
UK United Kingdom
VAT value added tax vkm vehicle-kilometre
VOC volatile organic compound WTP willingness to pay