High-Speed Train As a Feeder for Air Transport

1 ------Sources: Sources:

0:00 -0:18 -0:45 -0:48 -0:54 -0:54 -1:57 Difference 0:13 0:23 0:44 0:57 1:33 3:00 3:52 e-zine 49 edition After HSL After 0:13 0:41 1:29 1:45 2:27 3:54 5:49 TRANSPORT MARKET COMPETITION Before HSL Before Before and after (shortest) train travel times to AMS to times travel train (shortest) after and Before Amsterdam Rotterdam Breda Antwerp Brussels Paris London The traditional view on the relationship between markets. air range medium to short in transport compete they that is HST and More the recently, view of HST and air transport being seen as complementary modes has This appeared. paper extends the lit erature by looking at high-speed rail from the angle rail high-speed of the to connected airport well is airport an If competition. network, it increases its catchment area and, hence, its competi HST- the to connected well both are airports two If position. tive network, competition between them will increase. We focus on the link between Amsterdam and on Brussels, the and recently opened link between MadridThe Spanish andcase will Barcelona.serve as a reference for the Dutch-Bel gian case. We estimate a multinomial logit choice andmodel predict future marketof shares, airportbased on access time, 1: Table Own Calculations based on Deutsche Bahn timetable, 2009; NS high speed 2009 - - - tion, focusing on the new high-speed rail-link weenbet Amsterdam and Brussels. We conducted a detailed analysis regarding the airport passengers choice living in the of catchment areas of these airports and compared the resultschoice in with Spain, airportwhere a high-speed readytrains running. is al Air transport and high-speed trains are markets.competitorsrangemediumRe short toseen in as cently, the view of high-speed trains and air trans port being complementary modes has This emerged. paper extends the literaturethe impact of by high-speed lookingrail on atairport competi

delays, Fyra started operations Amsterdam, on Schiphol theand Rotterdam HSL-Southin September between2009 and Thalys started to make use of 2009, covering the the routeAmsterdam-Paris. HSL-South in December with the name Fyra and will cover the services between sterdamAm and Brussels, while the existing operator Thalys will continueto operate between Amsterdam and Paris. Aftermany ways (NS) (90%) and Royal DutchThis consortiumAirlines has the (KLM) right to (10%). operate high-speed transport on the entire route for 15 years.Trains of the are HSA branded For operating the high-speed line, the consortium High Speed Alliance (HSA) was founded, which comprises of Dutch Rail and London in four hours (HSL-Zuid, 2009). As travel by times train will be shortened what the impact significantly,will be on passenger air transport. the question arises trains(HST) run from these railway stations Antwerp,to Brus sels and Paris. Travelers have the opportunity to change trains in Brussels for trains to London. With the arrival of the HST, passengersareable toreach Paris inapproximately three hours network, the High-Speed Line South (HSL-South), runs from Amsterdamvia Schiphol Airport and Rotterdam to the Belgian border, with connections toThe Hague and Breda. High-speed Introduction Recently, the Netherlands became connected to the European network of High Speed Lines (HSL). The Dutch part of this by: Ilse Terpstra and Mark G. Lijesen G. Mark and by: Terpstra Ilse

Feeder for Air Transport High-speed Train as a

Aerlines Before HSL After HSL Difference is affected by the inauguration of a new high speed connection between them. Due to a decrease in travel time to the airports, Amsterdam 3:14* 2:06* -1:08 the catchment area of both airports will increase, as will the Rotterdam 2:13* 1:30* -0:43 overlap of their catchment area. The latter effect will lead to Breda 2:16* 1:19* -0:57 increased competition between the airports. Den Haag 2:30* 2:04* -0:31 Airport choice in the Netherlands, Belgium and surrounding Antwerp 1:02* 0:56* -0:06 countries forms a gap in the current literature, implying that Brussels 0:16 0:16 0:00 it is uncertain which consequences of the feeder effect will Paris - 1:59 - dominate (an enhanced catchment area or increased competi- London - 2:50 - tion with other airports) and to what extent the current market share will be affected in comparison with that of its competi- Table 2: Before and after (shortest) train travel times to BRU tors. Therefore, this research will mainly focus on the HST as * Including 20 minutes to travel by regular train from Brussels a landside feeder for the airport. Midi to Brussels AirportSources: Own Calculations based on Deutsche Bahn timetable, 2009; NS high speed 2009 We apply the framework of the feeder effect described above frequency, average fare and airport-specific constants. Direct to two cases, the link between Amsterdam and Brussels and the and cross-elasticities are calculated, and finally, future market recently opened link between Madrid and Barcelona. The Span- shares are predicted, incorporating changes in access time, due ish case will serve as a reference for the Dutch-Belgian case. to the introduction of the high-speed rail. High-Speed Train in Spain Possible effects due to the arrival of the HSL-South One of the most extensive high-speed rail networks in Europe Two possible effects on air transport can be distinguished, is in Spain. The first route came into operation in 1992 and con- namely the substitution effect and the feeder effect: nected the cities Madrid and Seville. The focus on this reference country will be on the route Madrid-Zaragoza-Barcelona, which Substitution Effect came fully into operation in February 2008. Already in 2003, a One of the expectations is that passengers will decide to travel HST started running from Madrid to Lleida, covering 519 kilo- the entire journey with the HST instead of traveling by air- meters. The second part of the route, from Lleida to Tarragona, plane, because of considerably reduced travel times. Since air came into operation in 2006. The last part of this railway, from travelers are faced with airport processes as checking in, se- Tarragona to Barcelona, was finally finished in 2008 and -re curity and boarding and airports are generally located at the duced travel times from Madrid to Barcelona by 77 minutes. outskirts of a city, total journey times from city center to city center by HST will be similar or even shorter than the total When comparing the HST train in Spain with the HSL-South, journey time by airplane. The HST is likely to be a substitute there are some similarities. First of all, the complete line was to traveling by plane on the following routes: inaugurated in 2008, but several parts of this line had operated Amsterdam – Paris since 2003. This is comparable to the HSL-South situation, as the HST from Brussels to Paris and Amsterdam – London Figure 1: Impact of high speed rail on the catch- London started operating several years Amsterdam – Brussels ment areas of nearby airports ago. Second, the total distance between Feeder Effect Barcelona and Madrid (around 600 ki- A HST can also be complementary to air lometers) is comparable to the total transport, as it can function as a supplier distance between Amsterdam and Paris of passengers to major airports. We can (around 500 kilometers). think of two ways how this may affect air transport. First, airlines may replace Factors Influencing Passengers Air- feeder flights by feeder train rides, as port Choice Air France has already done between Several empirical studies have studied Paris and Brussels. Second, the HST airport choice in multiple airport re- can serve as the landside feeder for the gions, with the majority focusing on the airport, and it can do so at shorter ac- San Francisco Bay Area (Harvey, 1987; cess times than regular landside feeders, Basar and Bhat, 2004; Pels et al., 1998; such as cars and regular trains. There- Suzuki, 2005) and the Greater London fore, an enlargement of the catchment area (Bencheman and Ashford, 1987; area of Amsterdam Airport Schiphol Hess and Polak, 2006). The general pic- (AMS), due to the arrival of the HSL- ture from those studies is that airport South, is to be expected. With HST sta- choice depends on airport access time, tions in Rotterdam, Breda, Antwerp and flight frequency, fare, flight time and fre- Brussels, AMS could benefit because quent flyer program membership. Fol- of this enhanced catchment area. If the lowing this general picture, we select catchment area (the area from which an access time, frequency and flight fare as airport attracts its passengers) of two the variables for our empirical analysis. airports overlaps, competition between Following Pels et al. (2003), we use the airports is also affected. Figure 1 shows logarithm of monthly frequency. how the catchment area of two airports

2 ventional train; by car then HS and by con- ventional train then HSL. We calculated these different types of access times using Google Maps Routeplanner and Deutsche Bahn timetable. Since we have no information on access mode choice, we assumed that passengers have a strict preference for short travel times and always choose the fastest connection available.

For Spain, we calculated access times from the capital of the provinces of the point-of- sale to the airports, adding an additional twenty minutes for cities other than the prov- ince capital. The same has been done for provinces in Northern France, Luxembourg and Belgium. For Western Germany and the Netherlands, another approach is taken, due to the fact that the two digits postal codes areas are not equivalent to the provinces. For these two countries, we selected the main cit- Photo 1: The HSR-line near Amsterdam Airport Schiphol. Courtesy of Rail.One GmbH and graphically edited by Du Saar Photography ies per two digit postal code.

We use data from the Sabre Airport Data Intelligence, from the The dataset has some shortcomings. First of all, tickets bought Marketing Information Data Transfer (MIDT) data, consist- immediately at an airline are not included in the database. ing of bookings through eight computer reservation systems Flights performed by low-cost carriers are underrepresented in (CRS’s) used by travel agents, which cover approximately 60- the data, as they do not sell their tickets through travel agents. 65% of all bookings worldwide. The data contains the average Second, we had to make the rather strong assumption that the 1 (by booking) fare paid, the number of direct flights per month point-of-sale is the same as the place of residence, which might between airports and the location of the travel agent. We as- not be the actual case. Third data on bookings are aggregated sume that the location of the travel agent is a good proxy for per route for the selected periods. Therefore, only data on the the departure location of the passenger. Bookings made at the headquarters of large travel agents are unlikely to meet this as- Ln frequency Acces time Average fare sumption and are hence deleted from the data. For Barcelona BCN 1.623 -0.440 -0.247 For both cases (Spain and The Netherlands-Belgium), we fo- BIO -1.074 0.261 0.111 cus on those airports located near the beginning and end of the HST routes in question. For the Spain-case, these airports are MAD -0.822 0.226 0.136 BCN and MAD. For the Netherlands-Belgium-case, these air- SVQ -0.517 0.185 0.054 ports are AMS and BRU. The catchment areas of these airports VLC -1.683 0.389 0.164 are created by drawing a radius of 250 kilometers around them. For Madrid For Spain, the five airports with the most passengers buying their tickets within these areas are selected as competing air- BCN -1.438 0.443 0.273 ports. These are: Barcelona El Prat Airport (BCN), Bilbao Air- BIO -2.011 0.425 0.240 port (BIO), Madrid Barajas Airport (MAD), Seville San Pablo MAD 1.272 -0.358 -0.218 Airport (SVQ) and Valencia Airport (VLC). SVQ -2.645 0.458 0.317 For the route Amsterdam-Brussels, we selected the follow- VLC -1.264 0.377 0.184 ing five airports: Amsterdam Airport Schiphol (AMS), Brus- For Amsterdam sels Airport (BRU), Paris-Charles de Gaulle Airport (CDG), AMS 0.605 -0.536 -0.055 Frankfurt am Main Airport (FRA) and Düsseldorf International BRU -0.692 0.539 0.052 Airport (DUS).2 CDG -0.071 0.096 0.012 Our Spanish data refers to September 2007 and September DUS -0.721 0.626 0.057 2008, thus including data prior to and after the opening of the FRA -0.456 0.573 0.078 HSL connection. For the case of the Netherlands, we use data For Brussels from September 2008. To limit the amount of data, we only use data from one month. Furthermore, only O-D flights are AMS -0.299 0.203 0.023 selected, as data on the frequency for connecting flights was BRU 1.160 -0.844 -0.094 not available. CDG -0.290 0.210 0.026 DUS -0.310 0.231 0.024 The variable access time (in minutes) is constructed based on the point-of-sale of the ticket and five types of access times FRA -0.291 0.304 0.026 to the airport of choice are estimated by car; by HSL; by con- Table 3: Cross-elasticities for selected airports

3 average fare per travel agent per route is available and data on the cross-elasticities involving the other airports. AMS seems the actual fare for the ticket paid by each individual passen- to be more able to draw away market share from DUS and ger is lacking. In addition, for the alternative choices, we have FRA by improving its own services (because of the high cross- calculated the air fare as the average of all fares available on a elasticities), while BRU has an advantage in capturing market particular route. The choice set is therefore based on average share from CDG by improving its services. ticket fares paid by passengers, rather than actual prices available at the moment of purchase. New Market Shares Figure 2 provides the market shares of the airports in our sam- We estimate a multinomial logit (MNL) model to assess the ple before and after the introduction of the HST in Spain. The impact of access travel time on airport choice. In addition to majority of the passengers originating in Spain are flying from analyzing current demand, we used the model to predict new Madrid airport and Barcelona airport. A small shift from Bar- market shares. To do so, we calculate predicted access times celona airport towards Madrid airport is noticeable. In Septem- in the presence of a high-speed rail link between Amsterdam ber 2007, 34% of all O-D passengers traveled from Barcelona and Brussels, and predict airport choice after the introduction airport, while a year later the percentage decreased to 31,9%. of the link. In contrary, the percentage of O-D passengers traveling from Madrid airport increased from 58,1% in September 2007 to Estimated Elasticities 59,6% in September 2008. For all airports in the sample, frequency turned out to be the most important explanatory factor for airport choice, followed Figure 3 provides the old market shares and the predicted mar- by access time and average fare. When comparing the results to ket shares for the Dutch case. The future market shares are cal- previous research, some similar results are obtained for access culated by applying the parameters for the different variables time and average fare, but the Figure 2: Market shares before and after the introduction of we estimated in the MNL mod- direct elasticities with respect a HST in Spain el for the current situation to to frequency are substantially the dataset with the new access larger. These differences may times after the introduction of have several reasons. First of the HST. AMS is the only air- all, most of the previous stud- port that is expected to benefit ies have been done using data (in terms of market share) from from the San Francisco Bay the arrival of the HSL-South, Area or the Greater London with predicted market share in- Area, where airports are much creasing from 40,8% to 47,3%. closer to each other than in our BRU will lose considerably, as study areas. Because no earlier will DUS. This result is in line research on airport choice in with the outcomes of the cross- Spain and the Netherlands is elasticities, which already available, the results are hard pointed out the stronger market to compare. position of AMS in comparison with BRU. Looking at cross-elasticities of demand (table 3) for a change The pattern we see here is fairly in the level of the attributes of similar to the results we had for the airports, we focus on (MAD Spain. By improving the con- vs BCN and AMS vs BRU), we nection between two airports, find that BCN has high cross the larger airport gains mar- elasticities relative to VLC, in- ket share at the expense of the dicating that by improving its smaller one(s). The intuition Figure 3: Market shares before and after (predicted) the introduc- own services, BCN could cap- behind this result is that the ex- tion of the HSL-South ture market share from VLC pansion of the combined catch- more easily than from the other airports. A change in one of the ment area works in the favor of the airport with the higher variables of MAD has the biggest impact on SVQ and BIO, but level of service (frequency in this case). Passengers who were changes in the access time effect BCN, BIO and SVQ almost previously captives are now in a position to choose, and they equally. are more likely to choose the airport with the higher service level. Note that this is not only beneficial to the larger airport, A change in one of the variables of AMS will have just a small but also to the customers, as their freedom of choice increas- impact on CDG, which suggest that AMS and CDG are not re- es. The resulting welfare increase may lead to an increase in ally substitutes. This makes sense, as the geographical distance the total number of passengers using the airports, but this is between the airports is the largest of the five. A change in one beyond the scope of our model. In addition, people living in of the variables of BRU will influence the other airports nearly Belgium or the South of the Netherlands who currently travel equally. Stronger effects occur when changing one of the vari- to CDG, DUS or FRA may switch to Amsterdam, as it has ables of AMS than one of the variables of BRU, suggesting become easier to reach. Keep in mind that the HST between that the arrival of the HSL-South could be beneficial to AMS. Brussels and Paris is already running, and access times be- The stronger position of AMS is also visible when looking at tween those places remained the same.

4 The gain in market share for AMS is however unlikely to be References large enough to compensate for the loss of passengers through Ashford, N. and M. Bencheman (1987), ‘Passengers’ Choice of air-rail substitution. Earlier research (Jorritsma, 2009) showed Airport: An Application of the Multionomial Logit Model.’ Trans- Aerlines that similar passenger behavior might be expected on, for ex- portation Research Record 1147 pp. 1-5. ample, routes as Amsterdam-London and Amsterdam-Paris, Basar, G., and C. Bhat (2004), ‘A Parameterized Consideration Set resulting in a decrease of air passengers between these cities. Model of Airport Choice: An Application to the San Francisco Bay Area.’ Transportation Research Part B Vol. 38 (10) pp. 889-904. End Notes Deutsche Bahn (2009), ‘Timetable’, Available via: http://www. 1 Aggregation of the fares by booking may lead to some infor- bahn.de/p/view/index.shtml (accessed on 19-05-2009). mation loss, as some of the heterogeneity in fares is not reflected, but Harvey, G. (1987), ‘Airport Choice in a Multiple Airport Region.’ it is by far the best measure available for actual fares in a European Transport Research A Vol. 21(6) pp. 439-449. context. Hess, S and J.W. Polak (2006), ‘Exploring the Potential for Cross- 2 Antwerp International Airport (ANR) and Rotterdam Airport (RTM) Nesting Structures in Airport-Choice Analysis: A Case Study of are also located near the HSL-South route, but they did not yield valid the Greater London Area.’ Transportation Research E Vol. 42 pp. results due to the low number of observations. 63-81. Jorritsma, P (2009), ‘Substitution Opportunities of High Speed About the Authors Train for Air Transport.’ Aerlines E-zine Edition 43. Mark Lijesen is an assistant professor at the department of spa- NS High speed (2009),http://www.nshispeed.nl/ tial economics of the VU university and has previously worked at Pels, E., P. Nijkamp and P. Rietveld (1998), ‘Airport Choice in a the Netherlands Bureau for Economic Policy Analysis (CPB) and Multiple Airport Region: An Empirical Analysis of the San Fran- Netherlands Institute for Transport Policy Analysis (KiM). His Phd research focused on market power and networks structures cisco Bay Area.’ Available at: http://www.tinbergen.nl/discussion- in civil aviation, which is still one of his main fields of research. papers/98041.pdf (accessed on 25-05-2009). Pels, E., P. Nijkamp and P. Rietveld (2003), ‘Access to and Com- Ilse Terpstra graduated in economic geography at the Univer- petition between Airports: A Case Ctudy for the San Francisco Bay sity of Groningen and in spatial, transport and environmen- Area.’ Transportation Research A Vol. 37, pp 71-83. tal economics at the VU University in 2009. She is now work- Sabre ADI (MIDT, 2009). ing as a market analyst at Amsterdam Airport Schiphol of the Suzuki, Y (2005), ‘Modeling and Testing the “Two-Step” Decision Schiphol Group. This article is written in a personal capacity. Ilse Terpstra is corresponding author for this article: ilseterpstra@ho- Process of Travelers in Airport and Airline Choices.’ Transporta- tmail.com tion Research Part E.

Photo 2: The Thalys train on a leg between Amsterdam and Brussels. Source: Wikimedia Commons

e-zine edition 49 AJ 5