1 s2.0 S2590198222001397 Main

Transportation Research Interdisciplinary Perspectives 16 (2022) 100679
Contents lists available at ScienceDirect
Transportation Research Interdisciplinary Perspectives

journal homepage: www.sciencedirect.com/journal/transportation-
research-interdisciplinary-perspectives
The impact of access prices on train traffic: An econometric study for France
Carlos Augusto Olarte-Bacares a, *, Julien Brunel b, Damien Sigaud a
a
Economist, Marketing, Economy and Regulation Department, SNCF Réseau, 12 Rue Jean-Philippe Rameau, 93200 St Denis, France
b
Economist, Strategy and Corporate Affairs Department, SNCF Réseau, 18 Rue de Dunkerque, 75010 Paris, France
A R T I C L E I N F O A B S T R A C T
JEL Codes: To guarantee fair and nondiscriminatory access to rail networks, European Commission raised the importance of
H21 regulation by promoting short run marginal cost. Thus, to recover the full cost of infrastructure, Infrastructure
H54 Managers may use mark ups over the marginal cost, using Ramsey pricing for example. However, a Ramsey
L92
pricing implementation for rail infrastructure is strongly limited by the lack of information about the demand
R48
elasticity of train paths. The aim of this article is to try to fill this gap by conducting an empirical research on the
Keywords:
price sensitivity of the railway paths using a French dataset from 2003 to 2016. Preliminary results suggest that
Track Access Charges
Elasticity of rail traffic
track access charges influence negatively the train path demand for high-speed trains and freight trains but do
Railway infrastructure pricing not seem to influence the train path demand for regional trains under PSO. Nevertheless, if track access charges
Ramsey-Boiteux pricing are related to the track category (UIC category) or to the kind of line (high speed or conventional) results may
differ. Regarding regional and freight trains, the track category seems to determine the statistical significance of
track access charges elasticity while the kind of line seems to explain the impacts on the demand of rail paths for
high-speed trains.
Introduction charging but allows infrastructure managers (IM) to levy mark-ups to

recover the full cost of infrastructure but only “when the market can bear
Most of the European railway services were assured historically by it”. Indeed, track access charges (TACs) are one of the main tools for IM
vertically integrated State-owned companies until the nineties. In the to cover the costs that are not covered by direct subsidies or grants from
beginning of the nineties, European commission decided to adopt a governments.
policy of “open markets” incentivizing the introduction of competition Mark ups principles of most of European IMs are generally based
in the railway sector. The Directive 91/440/EC introduced the right of under three different approaches: Ramsey-Boiteux approach, cost
access to the infrastructure for limited categories to international freight causation approach or competitiveness of markets (segments) approach.
operators. In 2007, this right was extended to all freight, domestic and However, independently of the principle adopted, IMs must fulfill the
international services in the EU. Then, in 2009 the access to the railway requirements of European directives 2012/34/EU and 2015/909/EU.
infrastructure was extended to international passenger services. During The adoption and the definition of charging principle and TACs
this period, several EU countries have introduced competition in the calculation are one of the most recurrent subjects of discussion between
form of competitive tendering for the provision of passenger services regulatory bodies (RBs), States, IMs and railway undertakings (RU). The
under public services obligations (Sweden, Germany, UK, etc.) and/or in debate remains open as a continuation of the discussion in economics
the form of open access competition for the provision of commercial between the marginalist and the institutionalist schools.
passengers’ services (Sweden, UK, Czech Republic, Austria, etc.). However, regardless the charging principles employed by IMs, the
To guaranty fair and nondiscriminatory access to rail networks for calculation of the optimal TAC is highly dependent on the availability of
new entrants, EU commission raised the importance of regulation. In the market data. Those data are supposed to show how the final demand
this context, one of the main concerns of European regulation regarded (passengers) and the demand of rail paths (operators) may change with
the introduction of legislation on the issue of infrastructure charges that respect to key variables. In this regard, the calculation of the sensitivity
may prevent free entry to the market. Since the directive 2001/14/EC, of the demand to the price (for passengers or for operators) appears to be
the current European regulation is promoting short-run marginal cost a crucial information that is not always available or easily calculated.
* Corresponding author.
E-mail addresses: carlos.olarte-bacares@reseau.sncf.fr (C.A. Olarte-Bacares), j.brunel@reseau.sncf.fr (J. Brunel), damien.sigaud@reseau.sncf.fr (D. Sigaud).
https://doi.org/10.1016/j.trip.2022.100679
Received 7 February 2022; Received in revised form 7 June 2022; Accepted 8 September 2022
Available online 29 September 2022
2590-1982/© 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
C.A. Olarte-Bacares et al. Transportation Research Interdisciplinary Perspectives 16 (2022) 100679
Some IMs attempt to fill this lack of information by adopting some reduction of users.4 Indeed, affordability of railway service may be
assumption about the elasticity of train tickets prices on passengers’ significantly reduced (Hotelling 1938) which will lead to a reduction of
demand and the impact of this elasticity on the demand for train paths the aggregated demand.
by operators. This research considers this assumption to be very strong The marginalist school suggests that to offset a part of this reduction
because there is not necessarily a direct proportional relationship be of demand and assuming that the cost of raising public funds is negli
tween traffic (tr.km) and final demand (passenger.km). gible, a public intervention through subsidies may be required (Nash
The aim of this paper is to try to fill this gap, at least for France, by and Smith 2021). In this way, the number of users will be optimized and
conducting an empirical investigation of the price (track access charges) the amount of subsidies used to recover part of the full cost will be offset
sensitivity of the traffic through an econometric analysis using a French by the increase of the number of users thus approaching the social op
dataset covering the period 2003–2016.1 This approach represents the timum (Nash and Smith, 2021, Ait Ali et al., 2020, Börjesson et al.,
main novelty of this research making it the first study to attempt to 2020).
directly measure the price elasticity of demand for train paths. On the other hand, the “Institutional school” argued that the adop
After this brief introduction, the second part of this paper attempts to tion of these subsidies does not ensure the social optimum (Coase 1946,
identify the context in which this research could be positioned by Roth 2006, Winston 2010)). Indeed, the intervention of the state to fix
focusing on the controversy between Marginal and Institutional schools2 this market failure leads to a second-best scenario. Initial costs may then
and tries to summarize in big lines the charging principles used by IMs in be covered either by infrastructure users or by public funds.
almost all of the European countries. The third and the fourth parts of The first option assumes that users will have to pay significant
this research present the data and the models used to define the TAC charges that can limit the number of users and which may result in a
elasticity of the demand of French train paths. A fifth part reveals and reduction of the demand. The second option assumes that users cannot
discuss the results of the models. Conclusions are presented in the last cover full costs, which is not a social welfare optimum. Indeed, the
part. intervention of the state might adversely affect the equilibrium since the
subsidies needed to support the infrastructure result in a cost for the
Theoretical pricing controversy and charging principles applied society in the form of imperfections in the tax structure and the shortfall
in european countries of public resources.
Furthermore, Coase also suggests that by subsidizing an infrastruc
As previously stated, European TAC schemes are frequently subject ture, the Government may be encouraging its use, but this is not justified
to discussions between States, IMs and RBs. On one hand, two schools of by consumers’ willingness to pay (Coase 1960, 1970). Thus, users do not
thought go head-to-head: the theoretical pricing controversy between receive a signal on the full cost of the good, and producers do not receive
marginalist and institutionalist schools (Duffy 2004). On the other hand, information about the willingness to pay for the cost of new goods or
there is another controversy that tries to answer to the question if the improvements. Under these conditions, it is even more difficult for the
definition of TACs is more about a cost coverage problem or if it is more central planner to take decisions concerning planning and investments,
a question about the choice of a State to finance the rail infrastructure. which could be oversized.
This part tries to explain how these two kinds of debates take place and In addition, given the long-life asset aspect of railway infrastructure
how European states tries to face this question. and the uncertainties of policy decisions, maintenance, renewal and
improvement of railway networks may not be assured in the long term.
The funding of these activities along decades represents then a third
A theoretical pricing controversy negative consequence of subsidization and marginal cost pricing.
The main question remains if full cost should be covered by users or
From an economic point of view, an infrastructure is characterized by taxpayers. European railway market practices indicate that most of
by various attributes that generally make cost recovery very chal the networks using charging principles looking to recover the fixed cost
lenging. Rail infrastructure is capital intensive with high initial in of the infrastructure is framed in the form of mark-ups. In this regard, the
vestments for its construction or its renewal. It can be considered as a two most employed charging principles in Europe are Ramsey-Boiteux
long-life asset with very long recovering costs period, making it a oriented (Ramsey, 1927, Boiteux, 1956) or follow a competitiveness of
typical case of a natural monopoly (Bitzan, 2003, Pitman 2007). Given market segments approach (Dierker, 1991).
these characteristics, rail infrastructure can create market failures It is then supposed that the Ramsey-Boiteux pricing solve the opti
leading to inefficient results in terms of general welfare. mization problem where, in a natural monopoly scenario (as for most of
Regarding the cost recovery and thus, infrastructure pricing, two European railway IM) social welfare is maximized and where IMs’ costs
different “schools” of economic theory propose two different solutions. are covered by IMs’ revenues by applying an “optimal” mark-up. Ram
On one hand, the neoclassical literature based on the welfare theory sey-Boiteux pricing aims to maximize the social welfare under a budget
suggests that a marginal cost pricing model may allow the achievement constraint for the monopoly. Under balanced budget constraint, optimal
to the social optimum (Marshall 1890, Hotelling 1938). Given the mark-ups over the marginal cost of a natural monopoly are inversely
magnitude of the fixed portion3 of the full costs for the French infra proportional to the demand elasticity of each product and is given by a
structure manager and the fact that wear and tear cost is not the only classical “inverse elasticity rule”. Nevertheless, in the case of interde
element of marginal cost (e.g. capacity, scarcity costs), pricing above pendent demand, the mark-up should be larger for substitutes markets
these elements should be avoided because it may result in a significant and smaller for complement markets (Höffler, 2006).
Generally, the discussion between IMs, RBs, and States are centered
on the definition and the argumentation of the elasticity. Indeed, this
1
Due to a significant change in the data collection system, it was not possible study, as noticed before, focuses on the determination of this elasticity
to use the data in the same way after 2016. but does not discuss the level of the coefficient related to financing
2
Because Ramsey pricing principle is strongly recommended by European
Commission this sub-part gave a particular attention to this principle.
3
Around 85% of the total cost for the French railway infrastructure.
4
Even if the European (Directive 2012/34/EU) and French (decree No.
97–446 and decree No. 2003–194) legislations suggest that other elements like
environmental or congestion costs should be taken into consideration, the
marginal cost computed by the French infrastructure manager does not consider
these elements. Thus, French TACs do not take into account these elements.
2
Table 1 Most, if not all, European countries use the principle of marginal cost
European charging practices in 2018. pricing, combined with subsidies, mark ups or other resources.
Mark-ups “if the Market segments? Charge(s) reflecting As it can be noticed, the adoption of a charging model using mark-
market can bear this” (article 32,1 of direct costs (article ups depends firstly on the charging principle chosen by IMs (or RBs)
(article 32,1 of directive 2012/ 31,3 of directive but secondly on the availability of the data to calculate and to apply each
directive 2012/34/ 34/EU) 2012/34/EU)
charging principle. In this regard price elasticity of demand (for pas
EU)
sengers and for RUs) is a very important variable that must be calcu
Austria × ✓ ✓ lated. Nonetheless, the calculation of this variable is very limited
Belgium
because of the availability of data. Indeed, this kind of information is at
× ✓ ×
Bulgaria × × ✓
Croatia × × ✓ the base of the Ramsey-Boiteux charging model.
Denmark × × ✓ In Europe, there is a wide range of justification of mark ups over
Estonia × × × marginal costs for track access charges Ljungberg (2013), Crozet (2007).
Finland × × ✓
It is interesting to note that some of them have developed a methodology
France ✓ ✓ ✓
Germany ✓ ✓ ✓ inspired by the Ramsey method. This solution has been investigated by
Greece × × × several academic articles like Link and Nilsson (2005), Sauvant (2008)
Hungary ✓ ✓ ✓ or Sanchez-Borras et al. (2010), Broman and Eliasson (2019), Ait Ali and
Italy ✓ ✓ ✓ Eliasson (2021a) Ait Ali and Eliasson (2021b) or Börjesson et al. (2021)
Latvia
among others.
✓ ✓ ✓
Luxemburg × × ✓
Netherlands ✓ ✓ ✓ In addition, some European IMs are inspired by this methodology
Norway ✓ ✓ ✓ when calculating their TACs. Actually, DB Netz follows the Ramsey-
Poland × ✓ ✓ Boiteux pricing rule by considering the price elasticity of passenger
Portugal × × ×
demand of the market segments based on data from a stated preference
Romania × × ✓
Slovakia × × ✓ survey (DB Netz, 2019). Once this elasticity computed, DB Netz assumes
Slovenia × × ✓ that the demand of train paths evolves in the same proportions. There
Spain × ✓ × fore, Austrian, Slovenian and Dutch IMs use the same assumption as DB
Sweden × × ✓ Netz regarding the calculation of train path elasticity.
Switzerland
Regarding the German High-Speed segment, DB Netz considers that
✓ ✓ ✓
UK ✓ ✓ ✓
the share of access price in the total turnover of train operators is of 18 %
Source: Track Access Summit 2018. Presentation from Christiane Trampisch. and that the consumer price elasticity for the final consumer is − 0.5.
According to this approach, the price elasticity of train paths in Germany
constraint λ which is almost as important as εi as shown in equation (1). is equal to − 0.09 for high-speed trains.
The optimal TAC is then described by: Several other studies have been interested in this topic in particular
pi − cmi λ 1 concerning freight markets. Matthews et al. (2009) propose an inter
= − • (1) esting consideration on the role of track infrastructure charges on the
pi 1 + λ εi
use of infrastructure for freight transport. The British Office of Rail and
where, pi is the optimal TAC for service i, cmi is the marginal cost of the Road (ORR) suggests that the impact of access charges on rail freight
service i, λ represents a coefficient related to financing constraint and εi demand can be very diverse among commodity groups (MDS Trans
is the price elasticity of the demand of the service i. modal, 2012).
Despite these studies, the applicability of Ramsey-Boiteux pricing
rule is more complex for passenger trains under a public service con
European legislation and European charging practices tract. As IRG-Rail (2016) noticed, “this problem is all the more complex
when considering services within the framework of a public service.
Despite the theoretical and political controversies, European regu These public services are further subsidized and therefore the determi
lation on rail infrastructure charges suggests the adoption of a marginal nation of price elasticity for these market segments is more
cost pricing. Indeed, article 31 of Directive 2012/34/EU defines the complicated”.
basic charging principle for calculating charges for the minimum access As noted in the introduction, this research aims to enrich the
package. Article 31 states that “the charges for the minimum access reflection about the sensitivity of train paths demand by employing an
package and for access to infrastructure connecting service facilities empirical study to compute the direct price elasticity of train path de
shall be set at the cost that is directly incurred as a result of operating the mand in France. This approach represents thus the main novelty of this
train service.” In June 2015, the European Commission adopted the research by using an extend and very rich data panel for a period of 13
implementing act n◦ 2015/909/EU on the modalities for the calculation years.
of the cost that is directly incurred as a result of operating the train
service. In this document, the Commission defines the methodologies for Data
the calculation of this cost. In the recital n◦ 12 of this regulation, it can
also be read that “It is a well-established economic principle that user As noted above, even if some econometric works have estimated final
charges based on marginal costs ensure the optimum effective use of passenger railway market elasticity very few papers have investigated
available infrastructure capacity”. With this recital, European regulation the price elasticity of train path. An estimation of the sensitivity of the
inclined towards marginal cost pricing. demand of train paths to the TAC is then necessary.
However, given the critics of social welfare sub-optimality, it may be This research uses an internal dataset of French infrastructure
preferable for IMs to cover the fixed costs of infrastructure with higher manager SNCF Réseau, which compiles the traffic and the TAC of almost
access charges. As stated before, article 32(1) of directive 2012/34/EU all the sections of the French railway network between 2003 and 2016
sets out the exception allowing to get “full recovery of the costs for HSR and regional train services and for the period 2003–2009 for
incurred” by levying mark-ups on different market segments “if the
market can bear it”. The methodology therefore remains fairly open (see
Table 1). Although some countries use Ramsey pricing, it is not the only
way. Then, the method may vary depending on the country.
3
Table 2
Descriptive statistics, track sections, 2003–2016 (18 116 observations).
Variable Obs Mean Std.Dev. Min Max
Year 18116 2 010 4 2 003 2 016

Length of SELs 18116 28 21 1 112
(kms)
Traffic data
Freight train.kms 5776 195 061 325 640 0 3 476 757
Regional train.kms 12252 200 271 217 967 0,4 1 607 863
HSR train.kms 6750 272 986 873 594 0,9 11 500
000
Access price data
Freight path.km 7558 1,26 1,71 0,000 14,87
price
Regional path.km 12250 2,59 2,14 0,081 20,56
price
HSR path.km price 6861 5,01 3,50 0,002 28,57
Infrastructure data Fig. 1. Evolution of freight tr.km (Millions) vs access price (€/tr.km).
HSR - UIC 2-6 846 60 1 60 61
HSR - UIC 7-9 8 1 1 1
may have direct impacts on the demand of train paths. Finally, given
–
CRL - UIC 2-6 9 768 698 3,45 693 704
CRL - UIC 7-9 7 494 535 3,45 529 540 that the HSR infrastructure evolved during the period studied (from
Socio-economic data 1540 km in 2003 to 2166 km in 2016), the length of the HSR network
Petrol 18116 55,95 18,59 25,49 86,86 was also taken into consideration.
HSR length 18116 1 858,64 216,70 1 540,00 2 166,00
Another characteristic of the infrastructure used in the study is the
National GDP 18116 1 935 230 179 671 1 637 438 2 194 243
National 18116 24 300 282 722 23 700 24 800
pricing category (UIC) of the railway paths: UIC 2–6 are the SELs that
Employment 000 000 000 generally support higher volumes of traffic while UIC 7–9 SELs supports
National Ind. Empl. 18116 3 566 158 233 986 3 280 109 4 004 674 less. Moreover, most of TACs for UIC 2–6 SELs are generally higher than
Regional GDP 18116 191 900 153 746 58 794 680 717 those for UIC 7–9 SELs.
Regional 18116 5 794 653 2 651 2 484 288 12 100
For the sake of completeness, this research decided to look at the
Population 744 000
correlation level between the different variables as well as their evolu
Own elaboration, source SNCF Réseau and INSEE. tion. Regarding the correlation, there is no evidence that the demand is
significantly correlated with the other variables and more precisely with
freight.5 Indeed, there are almost 1300 pricing sections (1294) called TAC. However, it seems that HS trains traffic is positively correlated
SEL6 that count around 35.000 kms of the French railway network. The with the TAC (0.3586) and negatively correlated for regional and freight
result is a sample of 18,116 observations (each line represents a pricing (− 0.1094 and − 0.0864 respectively). However, correlation levels are
path section) with the traffic by year. not very significant (lower than 0.40 in absolute terms).8
In addition, as almost all of the European IMs, SNCF Réseau has Concerning the comparison between the HSR and CRL9 SELs in the
different type of pricing (or TACs) for each type of service which sig network, it can be seen that among 18,116 SELs studied in 14 years, 854
nifies that the TACs paid by RUs depend on the SEL and on the type of are HSR SELs and 17,262 are CRL SELs. In parallel, the total tr.km of
activity. Three types of services are differentiated: freight, regional HSR trains runing in the railway network achieve almost 1.842 millions
passenger and High-Speed passenger services. TACs for each section are of tr.km.
well defined as well as their length and the traffic they supported per Among this total for HSR trains:
year during the period 2003–2016.
Furthermore, the demand of rail paths does not depend only on TACs - 66 % of these trains run on HSL from the pricing category UIC 2–6.
but may also depend on other national and/or regional variables. As it - There are no trains runing on HSL from pricing category UIC 7–9
will be presented in the next section, these variables will be used as because there are no HSL from pricing category UIC 7–9.
control variable to better understand the estimation of the evolution of - 33 % run on CRL from pricing category UIC 2–6
the demand of train paths for each type of service. These variables come - 0,7% run on CRL from pricing category UIC 7–9.
from the French national statistics office, INSEE.
Table 2 shows the descriptive statistics of the main variables used on Among all the SELs used by HSR trains, almost 9,7% are HSL and all
the model that will be presented in the next section of this research. As it these lines are UIC 2–6. 85,6% of the SELs used by HSR trains are CRL
can be seen, the traffic, that represents the demand from railway oper lines and are from the UIC 2–6 pricing category. Finally, almost 4,7% of
ators to infrastructure manager is expressed in train-km and is identified the rail paths used by HSR trains are CRL from the pricing category UIC
by the type of service.7 7–9.
Since the demand of each rail service may be affected by different These two observations show that, even if the number of HSL SELs
variables, some other variables were also taken into consideration and are significantly fewer (9,7%) with respect to the total of SELs, the
have to play the role of control variables affecting (or not) the demand of length of these HSL paths is significantly higher than the length of CRL
train paths: the price of petrol, the GDP at a national but also at a SELs and this is reflected on the part of the traffic (66 % of HSR trains.km
regional level, the regional population and the national employment run on HSL SELs). This specific characteristic should then be taken into
consideration in the models that will be used in this study because it may
even be misleading the use of fixed or random effects as well as results of
5
the models.
It was not possible to collect data for a longer period for the freight service
Based just on descritive statistics, in can be noticed that the rela
because of a change in the method of accounting for TAC.
6 tionship between the freight trains traffic and the evolution of average
From the french « Section élémentaire de ligne ».
7
The number of observations resulting from the addition of the traffic of the
three services does not match the total of observations of the two first rows of
8
the table because each rail section (SEL) may or may not support the three kind Correlation matrix is available in annex 1.
9
of services which at their turn have to pay different access prices. Conventional railway lines.
4
on the effective rail traffic by estimating a panel data model using a fixed
and/or random11 effects for every section of the network given by:
Dit = αi + β • Pit + γ • Zt + εit (2)
where,
Dit represents the demand (traffic) of train-km in the section i during

year t.
Pit is the average access price of each path section (SEL).
Zt is a set of external factors that may influence rail demand such as
the macroeconomic variables like the GDP, the regional GDP (for
regional services), the population of each region (for regional ser
vice), the price of the petrol on the three models or the national
employment for HSR service.
αi the characteristics of each SEL that are invariant and that may or
Fig. 2. Evolution of regional tr.km (Millions) vs access price (€/tr.km).
not influence the predictor variables; in other words, it represents a
mutually exclusive intercept which captures the distinctive, time-
invariant features of each SEL.
εit is the error term.
Hence, β and γ are respectively the elasticities of track access prices

and of the other external factors that may influence the rail demand.
With the objective to choose the adequacy of a fixed effect model or a
random effect model, this research decides to use the Hausman test
(Hausman, 1978). Nonetheless, each SEL may have time-invariant
characteristics that may be fixed for one or more periods.
In addition, these characteristics playing the role of control variables
may marginally vary (when they vary). This should highlight the
attention for some results concerning the Hausman test. Indeed, if data
used in the model do not change across the periods or if the model uses
dummy variables that considers very few parts of the sample, it may be
Fig. 3. Evolution of HSR tr.km (Millions) vs access price (€/tr.km). more appropriate to use random effect than fixed effects (Caves et al
1985, Odolinski and Nilsson, 2017). However, this will depend on the
TACs is inversely proportional (Fig. 1). However, this relationship is not set of control variables used for each model.
very clear for regional and HSR trains traffic. On one hand the average Furthermore, with the purpose to establish the elasticity of the de
access prices for regional train passes from 1.5€/tr.km to 3.5€/tr.km but mand of train-km with respect to the average TAC of each section and to
the traffic remains almost the same (Fig. 2).10 On the other hand, the make comparable the results of this study with most of the studies about
average access prices for HSR service increase from 2,4€/tr.km in 2003 the elasticity of the demand of train-km this paper uses a log–log model.
to 6,6€/tr.km in 2016 while the demand of HSR train paths fluctuated Given that the traffic is characterized by type of services, this
between until + 11 % in this period (Fig. 3). research runs three main different models: one for the train path demand
These simple descriptive statistics give some hints about the possible of high-speed trains explained by control variables at national level,
relations, if not the possible incidences, of TAC on tr.km demand. another one for the demand of regional trains explained by control
However, this descriptive exercise shows just correlations but does not variables at regional level and a last one for freight train demand
mean that TACs is the variable having the most important impact on explained by control variables at national level.
traffic. Indeed, a most rigorous analysis should be done to establish the Hence, the main model for HSR path demands is defined by:
real elasticity between traffic and several other variables. It will then be
log(DHSRit ) = β0 + β1 • log(PHSRit ) + β2 • log(HSRt ) + β3 • log(GDPt ) + β4 log
possible to analyze whether some other variables have a greater impact
on rail traffic. • (Employmentt ) + β5 • log(Petrolt ) + β6 • HSR1 + β7
Thanks to the richness and the precision of the dataset which rep • UIC2− 6 + εit
resents one of the main strengths of this research, it is possible to make (3)
an econometric model which is presented in the next section of this
research. The main model for regional passenger service trains is then defined
by.
Model log(DTERit ) = β0 + β1 • log(PTERit ) + β2 • log(RegionalGDP t ) + β3
( )
• log RegionalPopulation t + β4 • log(Petrolt ) + εit (4)
As noted above, this research aims to establish the impact of the TAC
Because of the absence of data about the evolution of industrial
production for each year, this research takes as a proxy the national
employment on industrial sector as an explain variable that may directly
10
In addition, Fig. 2 shows an important evolution of regional and HSR TACs affect the demand of rail freight paths. Hence, the main model for freight
in 2010. Indeed, it can be noticed a manifest change of the slope of these TACs rail path demand is:
due to a change of pricing methodology conducted by the French IM. However,
even with this sudden increase of TAC, the tr.km did not change significantly
(or at all), because this change has mainly resulted in rebalancing between
11
different charges (some of them are not considered in this study) and has not Depending on the kind of service and on the intrinsically variability of in
changed too much the total amount of TAC paid by operators. dependent variables.
5
Table 3
HSR path demand modeling: fixed and random effects models and Hausman Test results.
Variable model HSR 1 model HSR 2 model HSR 3 model HSR 4 model HSR 5
fixed random fixed random fixed random fixed random fixed random
LN_HSR_TAC -0.129*** -0.090** -0.161*** -0.098*

LN_LGV_INFRA 1.077*** 0.929*** 1.153*** 0.783** 1.323*** 0.953***
LN_GDP 0.878*** 0.833*** 0.893*** 0.807** 0.720*** 0.612***
LN_EMPL 2.464 2.400 2.413 2.484
LN_PETROL -0.058 -0.068 -0.058 -0.067
TAC_HSR_LINES -0.404** 0.468*** -0.409** 0.464*** -0.448** 0.545***
TAC_CRL_LINES -0.133*** -0.086** -0.137*** -0.091** -0.170*** -0.084*
YEAR_2004 -0.407*** -0.436*** -0.408*** -0.436***
YEAR_2005 -0.428*** -0.468*** -0.433*** -0.462***
YEAR_2006 0.011 -0.078 -0.001 -0.056
YEAR_2007 0.258** 0.157 0.245** 0.180*
YEAR_2008 0.154 0.020 0.136 0.055
YEAR_2009 0.317*** 0.182 0.298** 0.219*
YEAR_2010 0.577*** 0.359*** 0.548*** 0.414***
YEAR_2011 0.328** 0.113 0.296** 0.173
YEAR_2012 0.294** 0.090 0.261* 0.153
YEAR_2013 0.307** 0.084 0.271* 0.154
YEAR_2014 0.368*** 0.133 0.330** 0.207
YEAR_2015 0.309** 0.071 0.270* 0.148
YEAR_2016 0.238* 0.009 0.200 0.085
_cons -52.792 -50.868 -52.654 -50.946 -10.605*** -7.414** 9.639*** 8.565*** 9.590*** 8.663***
N 6656 6656 6656 6656 6656 6656 6656 6656 6656 6656
r2 0.015 0.015 0.015 0.036 0.036
r2_a -0.094 -0.093 -0.093 -0.071 -0.072
F 17.945 15.624 22.939 15.071 15.842
Hausman Test
chi2(5) 1837.10 chi2(6) 92.36 chi2(4) 95.06 chi2(15) 107.74 chi2(14) -377.41
Prob > chi2 0.0000 Prob > chi2 0.0000 Prob > chi2 0.000 Prob > chi2 0.0000 Prob > chi2 –
*p<0.05;**p<0.01;***p<0.001
Own elaboration.
Table 4
Regional train path modeling: fixed and random effects models and Hausman test results.
Variable model Regional 1 model Regional 2 model Regional 3 model Regional 4
fixed random fixed random fixed random fixed random
LN_REG_TAC 0.020 0.059*** -0.004 0.006

LN_REG_GDP 0.076 0.108 0.105 -0.083 0.033 -0.113
LN_REG_POP 0.679 -1.604*** 0.043 -1.472*** 0.290 -1.429***
LN_PETROL 0.020 0.060* 0.004 0.027 0.007 0.029
YEAR_D 0.060* 0.113*** 0.068** 0.123***
TAC_UIC_2_6 0.007 0.015 -0.008 -0.020
TAC_UIC_7_9 -0.085* -0.065 -0.110** -0.108**
YEAR_2004 0.022 0.019
YEAR_2005 0.043 0.043
YEAR_2006 0.011 0.010
YEAR_2007 0.025 0.017
YEAR_2008 0.048 0.046
YEAR_2009 0.055 0.056
YEAR_2010 0.029 0.032
YEAR_2011 0.098* 0.101*
YEAR_2012 0.134*** 0.138***
YEAR_2013 0.145*** 0.148***
YEAR_2014 0.183*** 0.185***
YEAR_2015 0.175*** 0.179***
YEAR_2016 0.139*** 0.141***
_cons -0.244 34.066*** 9.306 34.429*** 6.329 34.103*** 11.208*** 10.717***
N 12238 12238 12238 12238 12238 12238 12238 12238
r2 0.002 0.003 0.003 0.006
r2_a -0.090 -0.090 -0.089 -0.087
F 6.161 6.201 6.487 4.636
Hausman Test
chi2(4) 116.10 chi2(5) 83.42 chi2(6) 67.18 chi2(15) -127.28
Prob > chi2 0.0000 Prob > chi2 0.0000 Prob > chi2 0.0000 Prob > chi2 –
*p<0.05;**p<0.01;***p<0.001
Own elaboration.
6
Table 5
Freight train path modeling: fixed and random effects models and Hausman test results.
Variable model Freight 1 model Freight 2 model Freight 3
fixed random fixed random fixed random
LN_FREIGHT_TAC -0.051** -0.037*

LN_PETROL -0.373*** -0.357** -0.358*** -0.345**
LN_NAT_GDP 1.969** 1.849** 1.865** 1.766*
LN_IND_EMP 7.288*** 7.610*** 7.246*** 7.588***
TAC_UIC_2_6 0.007 0.009 0.022 0.027
TAC_UIC_7_9 -0.156*** -0.118*** -0.137*** -0.092***
YEAR_2004 -0.229*** -0.265***
YEAR_2005 -0.482*** -0.521***
YEAR_2006 -0.516*** -0.553***
YEAR_2007 -0.586*** -0.642***
YEAR_2008 -0.646*** -0.696***
YEAR_2009 -0.963*** -1.040***
_cons -126.505*** -130.090*** -124.418*** -128.603*** 11.339*** 10.986***
N 4948 4948 4948 4948 4948 4948
r2 0.170 0.177 0.180
r2_a -0.035 -0.027 -0.023
F 203.243 170.542 109.055
Hausman Test chi2(4) 81.990 chi2(5) -2657.37 chi2(8) -337.25
Prob > chi2 0.0000 Prob > chi2 – Prob > chi2 –
*p<0.05;**p<0.01;***p<0.001
Own elaboration.
( ) ( )
log DFreightit = β0 + β1 • log PFreightit + β2 • log(GDPt ) + β3 of the control variables on model 2 do not change even if the models use
random effect.
• log(Industrial employmentlGDPt ) + β4 • log(Petrolt ) + εit
Once this short introduction about the choice between fixed effects
(5)
model or random effects model is done, the following paragraphs of this
Finally, concerning the explanatory variables, it would have been section focus on the impact of control variables for each model of each
desirable to include more precise variables about passenger or freight service.
demand like train tickets prices for passengers or the quality of infra
structure for freight and passengers’ operators. Unfortunately, it was not High speed rail (HSR) service
possible to access to this kind of information and the task remains even
more complicated at a charging section level (SEL). However, the in Concerning the demand for high-speed train paths, results from the
clusion of these kind of variables represents a major hint of improve model 1 show that the variables having a significant statistical impact on
ment of this research. the demand of paths for HSR trains are the TACs with an elasticity of
− 0.13, the length of the HSR network and the GDP with positive elas
Results and discussion ticities of around 1.1 and 0.9 respectively (Table 3).
Multicollinearity analysis of the model, and more precisely, collin
This section reveals the results of each model for each activity earity between GDP and the size of HSR network12 are not shown in
(Tables 3–5). Additionally, a treatment of a probable endogeneity be these results but even if the construction of HSR path is supposed to
tween track access charges and the error term of the main models is depend on the economic growth, the commissioning of HSR paths takes
evoked at the end of this part. place some years after the decision of the construction. Hence, the
impact of the size of the HSR network does not necessarily depends on
The pertinence between fixed and random effects models for each railway the economic cycle of the same period where the evaluation takes place.
service With the purpose to better understand the demand of HSR paths, this
research considered that a differentiation between high-speed lines and
Hausman test was done for each model with the purpose to establish conventional lines named in this research “nature” of the rail path,
the pertinence of the fixed effect or the random effect model as well as should be taken into consideration. In fact, elasticity of TACs on HSL is
their robustness. between − 0.404 and − 0.448 while elasticity on conventional lines
In the first instance, Hausman tests suggest that fixed effects model (CRL) is between − 0.13 and − 0.17.HSL. This may be explained because
should be used to estimate the three main models (models 1 from HSR, just 9,7% of the SELs used by HSR trains are HSL which in turn can
Regional and Freight). However, Hausman test results may differ for the explain that although the impact of TACs on HSL is almost three times
other models tested for each service. higher than the impact of TACs on CRL (models 2, 3 and 4) the overall
Indeed, concerning HSR rail paths demand models, Table 3 suggests impact (model 1 and 5) is close to the impact of TACs on CRL.
that fixed-effects models should be preferred for the models 2, 3 and 4
but not for model 5. However, it is important to note that results of the Regional railway service (TER)
model 5 do not diverge from the main results of previous models even if
random effects are chosen. Hausman test suggests that the most appropriate model is the one
Regarding regional rail paths demand models, Table 4 indicates that, with fixed effect. As suggested on Fig. 2, Table 4 does not allow to
as for the main model, fixed effects should be chosen for models 2 and 3
but not for model 4.
Finally, with respect to the demand of rail paths for freight trains, 12
Results of the correlation matrix of coefficients of xtreg model done in Stata
Table 5 shows that random effect models are more appropriated for show that there is no problem about the collinearity between these two
models 2 and 3. However, the magnitude and the statistical significance variables.
7
establish a clear relationship between the evolution of the demand of found some “unexpected” results that should be discussed. The first one
regional train paths and the control variables of model 1. Estimators concerns the elasticity to petrol prices and the other one concerns the
have not a statistically significant impact on regional tr.km. elasticity to GDP.
In this regard, results comfort in part the intuition that demand does Firstly, literature suggests that an increase of petrol prices should
not depend directly on the track access charges and that the model translate into a negative impact on road freight. Hence, this effect may
proposed in this research does not explain the determinants of the de be positive to rail freight activity because of the electrification of a part
mand of regional train paths. This may corroborate the idea advanced by of this activity. However, results suggest that even if a part of the rail
IRG Rail, that for public service trains, the problem of what the market freight activity is electrified, an increase of petrol prices has a negative
can bear is more complex than for high speed or freight trains. Indeed, impact on demand on freight tr.km (0.37). Indeed, even if it may be
because of its nature of Public Service Obligation (PSO), the traffic of expected that the rail freight market may enjoy from the negative impact
regional trains may be highly defined by other variables not integrated of the increase of the petrol prices, the rail freight market is so sensitive
in the model especially the willingness to pay of each Transport Orga to external shocks that this does not achieve to absorb the negative
nizer Authority (TOA). impact of road freight activity.
Nonetheless, as it was presented previously, a pricing reform took Secondly, it could be considered as unexpected to see a positive
place in 2010 which may have some impact on the demand of regional elasticity to GDP in rail freight tr.km and that in the same time, graph 1
trains paths. In addition, the pricing structure for regional train paths shows a notably decrease of the demand of freight rail paths. Actually, it
depends also on different characteristics (traffic, maintenance costs, could be explained by the fact that the trend of the demand of rail freight
etc.) of each railway paths (SEL). This research considers that the elas is so negative that even a virtuous economic cycle does not achieve to
ticity of demand of tr.km for regional service with respect to TACs may change this trend. Indeed, the highly competition of road freight due to
be different for SELs with pricing categories UIC 2–6 and UIC 7–9 and the rigidity of schedules observed in rail freight services makes that rail
this should be taken into consideration. freight activity does not achieve benefit from economic growth even if
On one hand, results suggest that the pricing reform may have a elasticity of GDP is positive (1.97).
positive impact. The dummy variable indicating periods before and after As for the regional service, this research takes into consideration the
the reform is positive and statistically significant in model 2. This differentiation of the elasticity of TACs on the demand of freight railway
finding is corroborated in models 3 and 4. Indeed, dummy variables for paths with respect to the pricing category of each SEL (model 2). As for
each period become statistically significant just one year after the regional service, results suggest a statistical significance of TACs on the
adoption of the pricing reform. It suggests then that the reform could demand of paths from pricing category UIC 7–9 (-0.118) but not for path
have a positive impact on the demand of regional train paths. The result from pricing category UIC 2–6. A preferential or privileged pricing for
of the reform could be interpreted as attractive for operators and TOA. freight trains using paths from UIC 7–9 may have sense in an effort to
On the other hand, elasticities of crossed variables between TACs and avoid a bigger decrease of freight activity using this kind of paths. The
the pricing category of each rail path (UIC) suggest that the impact of impact of the other control variables is the same than for model 1.
TACs on the demand of rail paths that are less used by regional trains Finally, model 3 presents an elasticity of TACs on the freight traffic
(UIC 7–9) is negative and that this impact may be statistically signifi on lines UIC 7–9 around − 0.092 confirming the findings of precedent
cant. This finding may suggest that to avoid a deeper deterioration of the models as an example of the robustness of the model proposed.
traffic on these tracks (UIC 7–9), a differentiation of the pricing scheme
with respect to the traffic they support and as it was mentioned before, Similarities and differences of these results with respect to other countries
by transitivity, frequently associated to their conditions in terms of
maintenance, is appropriate. This is corroborated by results for model 4 The way that this research computes the elasticity is based on traffic
confirming an absence of statistical significance of TACs for paths from data and infrastructure and represents a big difference with respect to
the category UIC 2–6 and the statistically significance of TACs for rail the methodology employed by most of European IMs. Indeed, because of
way path from the category UIC 7–9. the big difficulty to have data about traffic and TAC, the calculation of
TACs in other countries like Germany, Austria or Netherlands is based on
Freight railway service the end customer elasticity: the price elasticity of passengers.
DB Netz calculations are based on stated preferences surveys for each
Results from the main freight model (model 1) advise that freight rail service. Long distance passenger services as well as freight services
paths demand is highly sensitive to several variables which may be a compute their own end customers elasticities with respect to the market
sign of the fragility of the rail freight market (Table 5). Indeed, it can share of each kind of service, the income of each traveler (of each
explain the decrease of the freight tr.km of almost 50 % observed in company for freight services), the cost of transport services and the
Fig. 1 for the period studied (2003–2009). Concerning the TAC, as for purpose of each travel for customers. The customer elasticity calculated
HSR service, the elasticity is negative (-0.051). This may suggest that by DB Netz for long distance passengers was − 0.09 in 2015. The end
even if TAC elasticity is statistically significant for freight rail paths customer elasticity computed in 2015 by DB Netz for freight transport
demand, the demand is highly dependent of a pool of variables that was − 1.50.13
together may have big negative effects. Although the elasticities of HSR markets in France and Germany are
In the absence of information about industrial production by region, very similar, it is not the case for the freight elasticity. This may be due
this study decided to use as a proxy the regional employment in the to different reasons, such as the fact that it may hide other character
sector of industry. Results suggest that French industrial and manufac istics in each country that the model does not take into account, or it
tured production exerts an influence on freight path demand. This effect may also be due to the different methodologies used by the IMs. In any
is very high (the highest one) with an elasticity of 7.3. Beside the case, this research considers that it should be further analyzed.
amplitude of the elasticity, it is not surprising to have a high elasticity of Finally, as noted above, endogeneity between TACs and the traffic
industrial activity since the demand for transport can amplify the eco should be studied. To solve endogeneity problems one of the treatments
nomic cycles due to the stocks. Actually, economic slowdowns may advised by literature is based on instrumental variable (IV) regressions.
generate an increase of the stocks since the industrial production, rigid This research therefore followed the literature that suggests studying the
in the short run, cannot adjust to a drop of demand, and vice versa in
case of an economic recovery. However, results suggested by this model
seem to be very high which should be taken with caution. 13
For further specification of the calculation of HSR elasticity in the German
Concerning the elasticity to the other control variables, this study market, see annex 2.
8
relevance and the exogeneity of instruments and that will result in demand of railway paths is negative and statistically significant for high-
overidentified coefficients14.15 speed trains and freight trains services but without statistical signifi
Two different variables should, in theory, respect the conditions of cance for regional trains services.
relevance and of exogeneity and are also supposed to follow the logical Concerning the HSR rail paths demand, the magnitude of the elas
thought of railway system: the debt of the IM and its maintenance fixed ticity of TACs may be larger when the nature of the railway paths is
costs. taken into consideration. Indeed, TACs elasticity of rail paths on HSL is
The first instrumental variable is supposed to be covered by TACs but higher in absolute terms than the elasticity of TACs on CRL: between
without a direct impact on the traffic which will erase its correlation − 0.404 and − 0.448 for HSL and between − 0.13 and − 0.17 for con
with the error term of the model.16 ventional lines. However, due to the results observed on the random
The second instrument proposed is the maintenance fixed cost which effects models, the elasticity for HSL should be considered with caution.
should not be correlated to traffic but should have a direct impact on Moreover, given that HSL do not represent 10 % of the paths used by
TACs.17 HSR trains, the elasticity of TACs on CRL absorbs almost all the impact of
Two different treatments of endogeneity suggested by the literature the TACs. Nevertheless, the impact is always negative and is close to the
were used in this study. The first one is the one suggested by Stock and elasticity of the passengers’ demand found by German infrastructure
Watson18 (chapter 12) by constructing data changes (evolutions) in the manager DB Netz (-0.09).
variables between two different time periods. The construction of data With respect to the elasticities of the demand for regional trains
changes concerns not only the evolution of dependent but also of in paths findings of the main model (model 1) confirm the hypothesis that
dependent variables. regional services are under PSO contracts then the demand of regional
The second methodology used in this research to treat endogeneity trains paths is not affected by explain variables used in this study. This
consists to use advanced packages on econometric software Stata19 that finding may not be completely surprising: regional authorities may not
makes possible the IV estimations across all the panel data using all the be influenced by the price signal, contrary to commercial market seg
sample. ments, since political reasons may also be considered by these author
However, after having treated endogeneity by these two different ities when they order trains paths. These other kinds of variables are not
techniques, results suggest that it may be more pertinent to use fixed included in the model proposed in this research; therefore, it cannot
effect models without instrument variables.20 explain the variability of regional tr.km. Hence, in French railway
network, when TAC are not evaluated with respect to their pricing
Conclusions category (UIC) the main model used in this research do not achieve to
explain significant impacts on regional rail path demand.
This article proposes an original empirical investigation under the Furthermore, when TACs are associated with the pricing category of
track access prices sensitivity of the traffic using a French dataset each rail path, results suggest that the elasticity of the demand of
covering a period from 2003 to 2016 for passenger services (regional regional trains paths with low level of traffic (UIC 7–9) is directly
service and High-Speed Rail service) and covering the period from 2003 impacted by TACs with a statistically significance (between − 0.09 and
to 2009 for freight service. This approach avoids passing by a calculation − 0.11). In addition, results suggest that the reform conducted by French
of the train tickets elasticity of passenger demand, which represents the IM in 2010 had positive impacts on the demand of regional rail paths.
main novelty of this paper. Indeed, elasticity showed by dummy variables for each period are al
This empirical analysis shows that when track access charges are ways positive and statistically significant (between 0.098 and 0.185).
taken as an isolated control variable (models 1), its elasticity on the Concerning the demand for freight railway paths, findings suggest
that, given the high sensitivity of the freight, this market has negative
relations with TACs when TACs are not related to the pricing category of
14
See Davidson and MacKinnon, 1993. Estimation and Inference in Econo paths. This “overall” negative impact of TACs on freight paths demand is
metrics. New York: Oxford University Press. principally induced by its impact on paths with low level of traffic, UIC
15
Judge et al., 1985. ¨The Theory and Practice of Econometrics. 2nd ed. New 7–9 (between − 0.092 and − 0.118).
York: Wiley. As noted at the beginning of this study, this research does not focus
16
Because of the nature of the HSR market that is supposed to be the only on the discussion of optimal TACs allowing French IM to recover its fixed
service to have a financial and economical sustainability, this research costs but aims to estimate the price elasticity of rail operators for each
considered that the debt may be pertinent as instrument for the HSR but also for French rail service. Indeed, as expected, each service and each market
regional and freight railway transport models. have different elasticities. However, in the case of the recovery of IM
17
This study would have wanted to take also into account the renewal fixed
costs, as the Ramsey-Boiteux rule suggests, it depends on the level of the
cost, but the data were unavailable.
18 elasticity and, in some cases, most importantly, on the level of the co
Stock and Watson (2003). Introduction to econometrics (Vol. 104). Boston:
Addison Wesley.
efficient related to financing constraint which is generally interpreted as
19
Mark E Schaffer, 2005. "XTIVREG2: Stata module to perform extended IV/ the social cost of public funds.
2SLS, GMM and AC/HAC, LIML and k-class regression for panel data models," Ramsey-Boiteux is then a powerful tool that enables all stakeholders
Statistical Software Components S456501, Boston College Department of Eco in the railway sector to determine a better TAC focusing on the calcu
nomics, revised 26 Jun 2020. lation of elasticity of each service but without leaving outside the
20
Results and the detailed description of the treatment of endogeneity are importance of the component associated to the cost of public funds
available in annex 2 under request.
14
See Davidson and MacKinnon, 1993. Estimation and Inference in Econometrics. New York: Oxford University Press.
15
Judge et al., 1985. ¨The Theory and Practice of Econometrics. 2nd ed. New York: Wiley.
16
Because of the nature of the HSR market that is supposed to be the only service to have a financial and economical sustainability, this research considered that the
debt may be pertinent as instrument for the HSR but also for regional and freight railway transport models.
17
This study would have wanted to take also into account the renewal fixed cost, but the data were unavailable.
18
Stock and Watson (2003). Introduction to econometrics (Vol. 104). Boston: Addison Wesley.
19
Mark E Schaffer, 2005. "XTIVREG2: Stata module to perform extended IV/2SLS, GMM and AC/HAC, LIML and k-class regression for panel data models," Sta
tistical Software Components S456501, Boston College Department of Economics, revised 26 Jun 2020.
20
Results and the detailed description of the treatment of endogeneity are available in annex 2 under request.
9
which, in the sense of this research, suggests the “willingness” of States the main difficulty to overcome.
to subsidy or not rail services (freight and regional services).
Finally, as noticed above, the inclusion of variables about train
tickets prices or the quality of railway infrastructure at a SEL level will Declaration of Competing Interest
certainly enrich considerably this research. However, the access to this
kind of information is very difficult if not impossible to have at the level The authors declare that they have no known competing financial
of a pricing section. Another hint to improve this research may pass by interests or personal relationships that could have appeared to influence
an analysis at the path level but the access to this information remains the work reported in this paper.
Annexes.
A.1
Table A1
Correlation between variables.
Own elaboration.
10
A.2
In the context of the subject of this research the specificities of the variables used in the models proposed may suppose that the relationship
between the railway paths demand and the track access charges (TACs) could be bidirectional. This relationship is translated by a simultaneous
causality that supposes that TACs may explain traffic, but traffic may also explain TACs. Indeed, this suggestion is based because TACs are supposed to
cover a part of IMs’ costs. Most of European IMs compute their TACs in accordance to the traffic of precedent years and hence in accordance to the
theoretical traffic which, as said before, may be traduced in a simultaneous causality that will biased results of proposed estimations. Econometrically
speaking, these biases will be traduced in a correlation between the error term and the regressor which is known as endogeneity between one or
several independent variables and the dependent variable. In this case, the railway demand (traffic) is then the dependent variable and the endog
enous variable is the TACs.
To solve endogeneity problems one of the treatments advised by literature are instrumental variable (IV) regressions. As their name evokes, this
kind of models use instrumental variables that explain endogenous variable(s) but that are not correlated to dependent variable (in this case the
traffic). Hence, instruments are variables that are supposed to explain the endogenous variables but that are not correlated with dependent variable.
The logic of the use of instruments is that one part of the endogenous variable is inexorably correlated to error term of the main econometric model
(with the dependent variable) while the other part of this variable is not correlated with error. Indeed, the use of instrumental variables are supposed
to capture and to isolate the part of the endogenous variable which is not correlated to the error term allowing to have consistent (not biased)
coefficients.
The model used for IV regressions generally takes a form of a Two Stage Least Squares estimation (2SLS). Actually, the first stage is an Ordinary
Least Square (OLS) estimation, the endogenous variable, in this case the TACs, is regressed with respect to the instrument(s) and to the other
exogenous variables of the main model. The predicted values of this First-stage estimation are then used on the second-stage regression regressing the
dependent variable, in this case the traffic, with respect to the predicted values of the First-stage estimation and with respect to the other exogenous
variables using an OLS estimation. Because instruments are supposed to be exogenous, predicted values of the First-stage estimation are supposed to
capture just the exogenous part of the endogenous variable. These predicted values are then used in the second-stage estimation assuring the exo
geneity of the endogenous variable and making estimators of the endogenous variable consistent.
The general IV regression model is:
Yi = β0 + β1 X1i + ⋯ + βk Xki + βk+1 W1i + ⋯ + βk+r Wri + ui
i = 1, ⋯, n,
Yi , is the dependent variable, in this case the traffic;
β0 , β1 , …, βk+r , are the estimators to find;
X1i , …, Xki are k endogenous regressors that are supposedly correlated to ui , in this case, there is one endogenous variable (TACs), so k = 1;
W1i , …,Wri are the exogenous regressors or control variables explaining the dependent variable and that are not correlated to ui ;
ui , is the error term that explain the measurement error of the model and/or the omitted variables in the model;
Z1i , …, Zmi are the instrumental variables used to control endogeneity problem.
However, the issue of IV regressions is to find valid instrumental variables that must satisfy-two conditions: relevance and exogeneity of instru
mental variables. The relevance of the instrument suggests that its variation is related to variation in endogenous variable. The exogeneity of an
instrument shows that, if there are more than one instrument, at least, one of them is not related to the error term and by consequence neither to the
dependent variable. Hence, this estimation of the exogenous variation of endogenous variable can be used to have consistent estimators.
The tests used to identify the validity of instruments take place after the first-stage of the estimation. Literature suggests that instruments may be
considered as relevant (not weak) if the first-stage F-statistic exceeds 10. Concerning the exogeneity of instruments, this can only be tested when
coefficients are overidentified. This implies that it must have more instruments than endogenous regressors (m > k): if there is one endogenous
variable the minimum number of instruments allowing to test the exogeneity of instruments must be two. Exogeneity test is based on the J-statistic
issued from the Sargan test where the null hypothesis is that all the instruments are exogenous. If J has a Chi distribution or if the p-value is inferior to
0,05 then the null hypothesis must be rejected which will indicates that at least, one of the instruments is not exogenous.
Finally, in order to simplify the endogeneity analysis and to better identify the better treatment of endogeneity, this research chosen to make the
analysis of endogeneity for the main models showed in precedent section (model 1) for each service.
Instrumental variables chosen for this research
This research decided to take into consideration two different variables that should, in theory, respect the conditions of relevance and exogeneity
and that are supposed to follow the logical thought of railway system.
The first instrumental variable considered by this research to be exogenous to the traffic but with an impact on the TACs level is the debt of the
Infrastructure Manager (IM). Indeed, as noted above, TACs are supposed to cover a (huge) part of IMs costs. This research considers that the level of the
debt of IMs is covered in a proportion by TACs but they do not have a direct impact on the traffic, so their estimators may not be correlated to the error
term of the model.
This instrumental variable is expressed by year which signifies that the model will take into account 14 different level of debt.
The second variable proposed as a pertinent instrument is the maintenance fixed cost which should not be correlated to the traffic but should has a
direct impact on TACs. This study would have wanted to take also into account the renewal fixed cost but due to the unavailability of data it was not
possible. The calculation of the maintenance fixed costs was based on the last internal calculation of marginal costs for the year 2016 that is supposed
to be the most rigorous done by the French IM (SNCF Réseau). Once the marginal costs computed, a marginal cost with respect to the observed traffic
in each SEL is then computed. Finally, the result is subtracted to the full maintenance costs recorded by the IM giving as final result.
Different methodologies to treat the endogeneity
This research decided to make two different treatments of endogeneity suggested by literature. The first one is the one suggested by Stock and
11
Watson (chapter 12) by constructing data changes (evolutions) in the variables between two different time periods. The construction of data changes
concerns not only the evolution of dependent but also of independent variables. The second methodology used in this research to treat endogeneity
consists to use advanced packages on econometric software Stata that makes possible the IV estimations across all the panel data using all the sample.
Estimation based on the evolution of data between two different time periods
For the first method this research chose to treat the endogeneity of variables by estimating the demand of HSR and regional railway paths based on
the evolution of all the data between four different time periods: every year, every-two years, every-five years and every-seven years. Concerning the
demand of freight railway paths, because of the sample size (data for the period 2003–2009), the demand of freight railway paths is studied every year
and every-two years.
The purpose to use this methodology for five different time periods is to see if elasticities for short term (every year and every-two years) are
confirmed for the long term.
This methodology of endogeneity treatment results on the estimation of three different models for each kind of service (HSR, regional and freight
services). As results the models take the following form:
ln(Yit+n ) − ln(Yi t ) = β0 + β1 t [ln(X1it ) − ln(X1it+n ) ] + ⋯ + βk t [ln(Xkit ) − ln(Xkit+n ) ] + βk+1 t [ln(W1it ) − ln(W1it+n ) ] + ⋯ + βk+r t [ln(Writ ) − ln(Writ+n )] + uit
This model is then replicated three times for each service with three different combinations of instruments:
- A model with TAC as the endogenous variable explained by debt and fixed marginal costs as instruments
- A model with TAC as the endogenous variable explained by debt as the only instrument.
- A model with TAC as the endogenous variable explained by maintenance fixed costs as the only instrument.
By following this methodology, this study fit models of the evolutions of traffic between two time periods as a function of different control variables
where TACs are treated as endogenous.
In order to establish the endogeneity of TACs, this research follows the Durbin and Wu-Hausman test that uses as null hypothesis that TACs can be
treated as exogenous. A highly significance of test statistics leads to reject the null hypothesis so to continue to treat TACs as endogenous. When the
null hypothesis is not rejected (the variable seems to be exogenous), this suggests that it may be more pertinent to do usual fixed effect models without
IV variables because endogeneity seems not to be confirmed.21
Table A.2.1 shows the estimators for TACs of 30 main models: 12 for HSR traffic, 12 for regional trains traffic and 6 for freight trains traffic. F
statistics are shown to know the relevance of the instruments. Sargan test results are also shown to know the exogeneity of instruments. For endo
geneity test, this research decided to follow the Wu-Hausman test. Finally, it is relevant to clarify that, because of nature of data (changes between two
periods and not a real panel data) fixed effects are not taken in consideration in this methodology.
Concerning HSR trains traffic, endogeneity test appears to detect endogeneity of TACs in 9 (among 12) models. Based on these results, this study
proceeded to analyze the relevance and the exogeneity of instruments for the models suggesting that results for the estimators surrounded are
consistent: instruments are then relevant and exogenous. This suggests that results for the second and/or the third column (models with just one
instrument) can be also confirmed. TACs seem to be statistically significant for models using the variables “debt” as the only instrument but not
significant for models using the variable “fixed maintenance costs” (FMC) as only instrument. In addition, estimators are very different, and signs are
not the same. This leads this research to consider if a fixed effect model without IVs may be more appropriated for the analysis of HSR’ traffic.
With respect to regional traffic regressions it can be seen that the only cases where endogeneity of both instruments is confirmed is for the models
estimating changes every year. However, the Sargan test showed that, for this model, instruments are not exogenous. Hence, estimations with just
“debt” or “FMC” are not confirmed. In contrast, the other nine models for regional railway traffic show that instruments may be validated but
endogeneity is not ratified which leads this study to suggest that after considering the changes between two different periods, endogeneity is not
confirmed for the demand of regional railway paths; a fixed effects models without IV should then be more appropriated.
Finally, concerning the demand of railway paths for freight trains, none of the model verified endogeneity with TACs. This suggests that the
estimator for TACs variable are not correlated to the error term in the main model which would validate results from fixed effect model without
instrumental variables.
Instrumental variables estimations across all the panel data using all the sample
After following the methodology suggested by Stock and Watson,22 it can be seen that results are very heterogenous among markets but also among
time periods in different markets. Freight train demand paths seems not to be endogenous with TACs in none of the models proposed by this research.
The same seems to be noted for regional trains traffic with some exceptions. In opposition, endogeneity test for HSR train traffic suggests that the
models proposed corroborates an endogenous relationship between these variables but just two models seem to validate the use of the instruments.
Using this methodology has as for consequence the reduction of the sample which may affect the consistency of different tests and by consequence of
21
This research follows this suggestion from professor Jeff Wooldridge who posted his answer to a question of a forum member of the Stata forum “Statalist”. His
answer where posted on April 2014.
22
ibid.
12
the estimators.
As noted above, this research decided to treat the endogeneity issue also by using another econometrical methodology which is based on a Stata
package proposed by Shaffer (2010).23 This econometrical package permits to use the whole sample of the available panel data in a same analysis. This
avoids constructing data on changes (or evolutions) in the variables between two different periods. Indeed, the reduction of the sample using changes
in the variables in two different periods may increase bias on estimators. A treatment based on Shaffer (2010) package increases the sample and thus,
supposes to reduce possible bias of estimators; a higher sample may be traduced in lower bias of estimators.
Results for this methodology are presented in table A.2.2. As for the precedent methodology, the tests of relevance of instruments (F), exogeneity of
instruments (Sargan) and endogeneity of regressor (C test) based on results for the first stage of the model are presented in table A.2.2. However,
contrary to the precedent methodology, because of the model is run for all the sample of the panel data, these models take in consideration the fixed
effects since the first stage of the model.
Table A.2.2 shows that as for the precedent methodology, results for HSR path demand seem to confirm the endogeneity of TACs. In addition,
Sargan test shows that instruments are relevant and exogenous. Hence, both instruments may be use as single instruments. If the debt is used as the
single instrument, results suggest that there is a statistically significance of TACs on the HSR traffic (− 1.049). In contrary, if FMC is used as the only
instrument, endogeneity is not confirmed and in addition, the TACs estimator is not statistically significant. Finally, even if results for each model are
not the same, the sign remains negative which suggests that the impact of the TACs on HSR traffic is always negative once the endogeneity controlled.
Concerning results for regional traffic, table A.2.2 shows that, even if endogeneity and relevance are confirmed using both instruments, exogeneity
is not confirmed. In this case it is hard to validate the instruments, so results may be biased. It seems that using this methodology to control endo
geneity suggests that instruments are not adequate to control endogeneity between TACs and error term when estimating regional railway path
demand. The impact of TACs on the regional train traffic appears to be very complicated to establish which leads this research to use fixed effects
treatment without instrumental variables.
Finally, results of IV estimation for freight railway traffic are also very heterogenous. Column 1 shows that endogeneity of TACs is not confirmed on
freight traffic when both instruments are used as the exogeneity is not confirmed as well. With the treatment of endogeneity using IVs, TACs appear to
not have a clear impact in freight train path demand.
After this analysis of endogeneity, the forgoing discussion suggests the question whether the best way to study the impact of TACs on traffic is a
fixed effects treatment without IVs.
Hence, once adopted model with fixed effects without IVs, every main model of each service has some extensions in order to better understand the
way that TACs may affect traffic by considering some other characteristics of each traffic (nature of rail paths, kind of network between conventional
railway Lines and high-speed train lines, etc).
Indeed, extensions for HSR models take in consideration the kind/nature of rail paths between HS lines or Conventional Railway Lines (CRL) as
well as the kind of infrastructure that refers the state of the rail network or the traffic (UIC 2 to 6) are then included in the other models.24
Extensions for regional passenger service models take into consideration other kind of control variables such as the one that may denote the state of
the railway network (UIC category) and the period (before or after 2010). These variables may also explain in an important degree the demand of
Regional tr.km. The model proposed in this research tries to evaluate the impact of these variables on the demand of regional train paths.25
Considering the freight railway service, this research suggests that freight tr.km should also depend on the industrial activity and on the variable
suggesting the state of the railway network (UIC). This last variable is taken into consideration on the extended models for freight rail paths demand.26
After the estimation of these three main models, this research decided to cross the average access price of each path section (SEL) with HSR and UIC
variables to evaluate the pertinence of its use. Once this exercise done and in order to confirm the robustness of the model, the demand (tr.km) of rail
paths is then estimated with respect to the variable resulting from the cross between the price of each path section, the HSR and the UIC.
The other independent variables are 16 dummy variables (9 for the freight model) that are supposed to capture the effects of the other explain
variables used in the three main models as well as the non-observable effects of each year in each of the three main models.
Hence, the supplementary variables included in the three main models and that are used to test the robustness of the model are:
- Dummy variable for regional model indicating before and after the pricing reform between 2009 and 2010 (YEAR_D).
- Dummy variable for each year to capture the missing information and the fixed effects of each year. (YEAR_i).
- Dummy variable to separate the HSR traffic that uses the HS lines and the one that uses the conventional lines. (HSR).
- Crossed variables between prices of SELs and pricing category of railway lines (UIC) which are differentiated into two group. Those groups are
defined by the traffic that SELs support but also by pricing schemes where UIC 2 to 6 are SELs that support higher traffic than SELs from UIC 7 to 9.
In addition, SELs from group UIC 2–6 have TACs that are generally higher than for SELs from UIC 7–9 group. The first group referred to the lines on
which the traffic is generally higher and that are in “better” conditions in terms of maintenance. The second group of UIC supports less traffic and
are generally in worse conditions in terms of maintenance.
23
Mark E Schaffer, 2005. "XTIVREG2: Stata module to perform extended IV/2SLS, GMM and AC/HAC, LIML and k-class regression for panel data models," Sta
tistical Software Components S456501, Boston College Department of Economics, revised 26 Jun 2020.
24
( )
The model 2 takes then the following form:.log(DHSRit ) = β0 + β1 • log(PHSRit ) + β2 • log(HSRt ) + β3 • log(GDPt ) + β4 log • Employment t + β5 • log(Petrolt ) +
β6 • HSR1 + β7 • UIC2− 6 + εit
25
Model 2 for regional tr.km takes then in consideration other explain variables. It takes the following form:.log(DTERit ) = β0 + β1 • log(PTERit ) +
( ) ( )
β2 • log RegionalGDPt + β3 • log RegionalPopulation t + β4 • log(Petrolt ) + β5 • UIC2− 6 + β6 • YEAR2010 + εit
26
( ) ( )
Model 2 in table 6 takes then in consideration the state of the railway network. Hence the model is defined by:.log DFreight it = β0 + β1 • log PFreightit +
β2 • log(GDPt ) + β3 • log(Industrial employmentlGDPt ) + β4 • log(Petrolt ) + β5 • UIC7− 9 + εit
13
Table A.2.1
.
Table A.2.2
Regressors for TACs by service across all the panel data using all the sample.
Endogenous variable: TACs Instruments: debt, service debt and fixed maintenance costs
debt-MC debt FMC
TAC estimator TGV -1.121*** -1.049*** -0.416

F statistic 109.063 270.839 7.769
Sargan test 0.570 – –
Wu-Hausman C 34.299*** 34.867*** 0.094
TAC estimator TER 0.241*** 0.291*** 0.052
F statistic 1021.490 1522.177 412.138
Sargan test 5.732** – –
Wu-Hausman C 33.380*** 37.799*** 0.192
TAC estimator FRET -0.254 -0.800* 0.761
F statistic 12.070 13.350 11.364
Sargan test 17.489*** – –
Wu-Hausman C 0.611 6.309** 6.476**
legend:* p<0.05; **p<0.01; ***p<0.001
References Caves, D.W., Christensen, L.R., Tretheway, M.W., Windle, R.J., 1985. Network Effects
and the Measurement of Returns to Scale and Density for U.S. Railroads. In:
Daugherty, A.F. (Ed.), Analytical Studies in Transport Economics. University Press,
Ait Ali, A., Eliasson, J., 2021a. European railway deregulation: an overview of market
Cambridge, pp. 97–120.
organization and capacity allocation. Transp. A: Transp. Sci. 1–25.
Coase, R., 1946. The Marginal Cost Controversy. Economica 13 (51), 169–182.
Ait Ali, A., Eliasson, J., 2021b. The value of additional data for public transport origin-
Coase, R.H., 1960. The Problem of Social Cost. J. Law Econ. 3, 1–44.
destination matrix estimation. Public Transport. https://doi.org/10.1007/s12469-
Coase, R., 1970. The Theory of Public Utility Pricing and Its Application. Bell J. Econ.
021-00282-0.
Manage. Sci. 1 (1), 113–128.
Ait Ali, A., Warg, J., Eliasson, J., 2020. Pricing commercial train path requests based on
Davidson, R., MacKinnon, J.G., 1993. Estimation and Inference in Econometrics. Oxford
societal costs. Transp. Res. Policy Pract. 132, 452–464.
University Press, New York.
Bitzan, J.D., 2003. Railroad Coasts and Competition: The implication of introducing
Dierker, E., 1991. The Optimality of Boiteux-Ramsey Pricing. Econometrica 59 (1),
competition to Railroad Networks. J. Transp. Econ. Policy. 37 (2), 201–225.
99–121.
Boiteux, M., 1956. Sur la question des monopoles publics astreints à l’équilibre
Duffy, J.F., 2004. The Marginal Cost Controversy in Intellectual Property. Univ. Chicago
budgétaire. Econometrica 24, 22–40.
Law Rev. 71 (1), 37–56.
Börjesson, M., Eliasson, J., Rubensson, I., 2020. Distributional effects of public transport
Hausman, J.A., 1978. Specification Tests in Econometrics. Econometrica 46 (6),
subsidies. J. Transp. Geogr. 84, 102674.
1251–1271.
Börjesson, M., Rushid, A.R., Liu, C., 2021. The impact of optimal rail access charges on
Hotelling, H., 1938. The General Welfare in Relation to Problems of Taxation and of
frequencies and fares. Econ. Transp. 26-27, 100217.
Railway and Utility Rates”. Econometrica 6 (3), 242–269.
Broman, E., Eliasson, J., 2019. Welfare effects of open access competition on railway
IRG-Rail, 2016. Initial approach to market segment definition and criteria for an
markets. Transp. Res. Part A: Policy Pract. 129, 72–91. https://doi.org/10.1016/j.
assessment of mark-ups in consideration of Directive 2012-34, Nov. 2016.
tra.2019.07.005.
Judge, G.G., Griffiths, W.E., Hill, R.C., Lutkepohl, H., Lee, T.-C., 1985. The Theory and
Practice of Econometrics, 2nd ed. Wiley, New York.
14
Link, H. and J. Nilsson, 2005. ’Infrastructure’, in Nash, C. and B. Matthews (eds.), Odolinski, K., Nilsson, J.-E., 2017. Estimating the marginal maintenance cost of rail
Measuring the Marginal Social Cost of Transport, Research in Transportation infrastructure usage in Sweden; does more data make a difference? Econ. Transp.,
Economics, Volume 14 Elsevier, Amsterdam. Elsevier 10 (C), 8–17.
Ljungberg, A., 2013. Impacts of increased rail infrastructure charges in Sweden. In: Pitman, R., 2007. Options for restructuring the Stata-Owned Monopoly Railway. Res.
Research in Transportation Economics, THREDBO 12: Recent Developments in the Transp. Econ.. 20, 179–198.
Reform of Land Passenger Transport, vol. 39, pp. 90–103. doi:10.1016/ j. Ramsey, F.P., 1927. A Contribution to the Theory of Taxation. Econ. J. 37 (145), 47–61.
retrec.2012.05.027. Roth, G. (Ed.), 2006. Street Smart, Competition, Entrepreneurship and the Future of
Marshall, Alfred. 1890. Principles of Economics. Available online at the Archive for the Roads. The Independent Institute. Transaction Publishers.
History of Economic Thought: http://www.marxists.org/reference/subject/econom Sanchez-Borras, M., Nash, C., Abrantes, P., Andres Lopez-Pita, A., 2010. Rail access
ics/marshall/. charges and the competitiveness of high-speed trains. Transp. Policy 17 (2),
Matthews, B., Evangelinos, C., Johnson, D., Meunier, D., 2009. Impacts and incentives of 102–109.
differentiated rail infrastructure charges in Europe - focus on freight. Eur. Transp. Schaffer, Mark E 2005. “XTIVREG2: Stata module to perform extended IV/2SLS, GMM
43, 83–112. and AC/HAC, LIML and k-class regression for panel data models,” Statistical
Nash, C., Smith, A., 2021. Regulation and competition in railways. In: Vickerman, R. Software Components S456501, Boston College Department of Economics, revised
(Ed.), International Encyclopedia of Transportation. Elsevier Ltd, United Kingdom. 26 Jun 2020.
DB Netz, 2019. Description of how the market segmentation, the costs directly Stock, J.H., Watson, M.W., 2003. Introduction to econometrics. Addison Wesley, Boston.
attributable to train operation and the full cost mark-ups were derived. Annex 6.1 of MDS Transmodal, 2012. Impact of changes in track access charges on rail freight traffic,
the Network Statement. Available online at https://fahrweg.dbnetze.com/resourc Feb. 2012.
e/blob/1363812/409a105680d8beff3c231e3cd4dfefcb/snb_2019_annex_6-1-data. Winston, C. 2010. Last Exit – Privatization and Deregulation of the U.S. Transportation
pdf. System. Brookings Institution Press. Washington D.C.
15

1 s2.0 S2590198222001397 Main

Uploaded by

Copyright:

Available Formats

1 s2.0 S2590198222001397 Main

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 s2.0 S2590198222001397 Main

Uploaded by

Copyright:

Available Formats

Transportation Research Interdisciplinary Perspectives 16 (2022) 100679

Contents lists available at ScienceDirect

Transportation Research Interdisciplinary Perspectives

Introduction charging but allows infrastructure managers (IM) to levy mark-ups to

Year 18116 2 010 4 2 003 2 016

Dit represents the demand (traffic) of train-km in the section i during

Hence, β and γ are respectively the elasticities of track access prices

LN_HSR_TAC -0.129* -0.090 -0.161*** -0.098*

fixed random fixed random fixed random fixed random

LN_REG_TAC 0.020 0.059*** -0.004 0.006

fixed random fixed random fixed random

LN_FREIGHT_TAC -0.051** -0.037*

Instrumental variables chosen for this research

Different methodologies to treat the endogeneity

debt-MC debt FMC

TAC estimator TGV -1.121* -1.049* -0.416

You might also like