Eurostar, the high-speed rail system connecting London and Paris, underperforms. Its ridership, 9.5 million in 2010, is very low relative to both ridership projections and the populations of the cities it connects. This is used by opponents of high-speed rail as a worst-case scenario, as evidence that high-speed rail is a lemon. In addition, Drunk Engineer has argued in comments that it comes from unique design problems such as security theater, not present on any other high-speed rail network.

I claim that it’s not Eurostar specifically that underperforms, but rather the entire London-Paris travel market, and that it’s probably due to its being an international link. Specifically, there should be fewer business ties between London and Paris than between two similarly-sized cities in the same country, or even in different countries speaking the same language. In addition, because London and Paris were traditionally separated by sea, there was never a large ground market between them for rail transportation to poach.

Exhibit #1: Eurostar’s mode share is quite normal by the standards of other HSR lines of comparable travel time. See for example figure 1 in this report on air/rail competition in Europe, with data a few years out of date, and figure 2-4 2-3 in this report on Brazilian HSR, which is more up to date.

Exhibit #2: before the Channel Tunnel opened, the total size of the London-Paris air market was 4 million per year. This is smaller than intranational links connecting smaller cities: for example, according to an EU report on the busiest single-airport pairs in Europe, Madrid-Barcelona was 4.6 million (largest in Europe) on the eve of the AVE’s opening, Paris-Toulouse and Paris-Nice were a total of 4.6 million between them, and Rome-Milan with 2.5 million. In Australia, Sydney-Melbourne has 6.8 million annual passengers, and is the fourth busiest air market in the world, after Tokyo-Sapporo (not served by HSR), Tokyo-Fukuoka (where HSR takes 5 hours), and Seoul-Jeju (Jeju is an island off of mainland South Korea).

Exhibit #3: international air links other than London-Paris underperform as well. The EU report cited above, a study by the Centre for Asia Pacific Aviation ranking all airport pairs in the world by seats flown, and a Brookings ranking of the top 100 city pairs (not airport pairs) in or into the US are all missing or underranking international city pairs with huge combined populations. Tokyo-Seoul has 21% the seats flown of Tokyo-Sapporo; Paris-Milan is three-fifths the size of Paris-Nice; the only air market from Zurich to the EU that makes the top 20 EU-to-not-EU pairs is Zurich-London, with 900,000; and New York-Toronto is not on Brookings’ list, and has only 1 million passengers per year, compared with 2.3 million for New York-Washington, which has to compete with rail, buses, and cars.

Note that this effect applies both to international pairs speaking the same language and international pairs not speaking the same language. However, language does have an effect: there are far fewer air travelers from the US to Montreal than to Toronto or Vancouver. London-Dublin is a large market (4 million) when one looks at all airport combinations. Paris-Brussels rail traffic (6 million) is  lower than Paris-Lyon (18 million in 1988, the last year the LGV Sud-Est was the only LGV) but almost as high as Paris-London despite a huge city size difference. And London-New York is the largest long-haul market, more than 4 million, though much of it must be connecting traffic – perhaps the finance links between the two cities contribute.

Indeed, while Eurostar flounders with its 9.5 million passengers, domestic HSR networks thrive – follow some of the links in the above paragraph to see numbers for the TGV, the KTX, and the premium-priced AVE.

Note that this pattern applies only to intercity passenger travel. Regional travel crosses borders frequently: the S-Bahn networks of Zurich and Basel both cross borders every day, with integrated tickets and fares, and the border crossings between San Diego and Tijuana and between Detroit and Windsor are infamous for their congestion. That the travel market between New York and Toronto is small says nothing about the travel market between Buffalo and Niagara Falls, Ontario.

There are two upshots to this pattern. First, proposals for high-speed rail within the US should be compared with higher-performing lines, such as the TGV network; they should avoid comparisons to the Eurostar flop. Second, US HSR plans should give much lower priority to international links, especially to Montreal; links to Toronto, Vancouver, and Tijuana may be justifiable on grounds of regional cross-border travel.

1. ant6n

True. But Montreal would probably benefit economically from a HSR connection to New York, and there have been proposals and plans to build a faster train for a really long time. In the end of the day the cost-benefit ratio is simply not that good (as you point out).

But I believe the conclusion should be that the anything to Montreal should be part of an integrated strategy to get the best cost-benefit ratio up by combining all different proposals into one – currently there are the ideas to upgrade the Adirondack, extend the Vermonter again, and build a Montreal-Boston connection through rail-averse New Hampshire. In Vermont, there are also plans to extend the Ethan Allan express to Burlington, and thus get a train to Montreal.

Out of all these proposals there should only be one line built to moderate hsr track standards (~200km/h), incrementally, with a track where the routes Boston-Montreal and Montreal-New York are shared as much as possible (i.e. by going through Burlington); and which connects to some of the tracks that will get upgraded anyway (Albany or Springfield). At the same time, there should be combination of regional/express trains (like on the NEC), to get as much passengers as possible out of the capital investments.

In a way, the cross-border services would only make sense if piggy-backed on top of each other, and of other lines that are getting upgraded in the States, anyway.

• emdx

200 km/h accross the Appalachians/Adirondacks is not a very practical proposition. It is unfortunate to say that the likeliest high-speed rail connection to Montréal would have to be through Toronto…

• ant6n

Well, The Montreal-Burlington-Albany connection is pretty flat. That would’ve been on the (rail Vermont) track between Rutland, VT and Burlington, which VT applied to have upgraded to passenger rail, to extend the Ethan Alan Express from Rutland to Burlington. Unfortunately that didn’t get funded, but it does represent a ROW with fewer topological difficulties.

Even the current Adirondack doesn’t go through the Appalachians – it winds along lake Champlain. I doubt that that track could be brought up to much speed – maybe 160km/h?

• Alon Levy

The current route is ridden with 250-meter curves, including a few parts with so many reverse curves that superelevation would have to be limited. Squeezing 100 km/h from this line would be a challenge; a few short segments have gentle enough curves to allow 130 km/h, but that’s about it.

2. Stephen Smith

Is it possible that the demand for Paris-London traffic has growth a lot since 1994 when the Chunnel began operations? As I understand it, London has really grown as a financial and tech hub since then, and the French are heavily represented in those high-wage sectors. And given France’s centralization in Paris, it seems like a good number of these people would be from Paris. I also remember hearing that London is a necessary stop for every French presidential candidate.

Anyway, maybe London’s growth as a financial center since 1994 hasn’t been enough to increase Paris-London travel, but I just thought I’d throw that out there.

3. Tom West

Hang on, Eurostar carries twice as many passenegers as planes did just before it opened and iis a failure?

(I never count performacne against predictions as a success or failure. What mattrs is mode share and number of trips).

• Alon Levy

Yeah, exactly. For other HSR lines, which have a large preexisting ground market to poach, the factor is much higher than 2. And the air market from Paris to London was not that large to begin with relative to city size.

4. Danny

It isn’t just international links. It is large city pairs as well. And this phenomenon has been observable going all the way back to Railway Mania.

The companies that fared best in in the Railway Mania bubble were the ones that did not overestimate their revenues from long distance city-pair trains, and accommodated the less-attractive but far more numerous short span travel.

From a utilitarian perspective, it makes sense. There are some things that big cities have that small cities don’t. If you want an Hermes scarf, you can buy one in Los Angeles and you can buy one in San Francisco, but you can’t buy one in Fresno. The same thing goes for international embassies, major airports, niche and specialty stores, media centers, sociocultural communities (eg…I used to travel every year to Los Angeles from Sacramento for an annual Portuguese festival), and about a million other things.

With these things, it isn’t always the allure of the specific city that drives ridership, but rather the fact that the city contains something that is demanded. And when you think about it from that perspective, the utilitarian demand for Los Angeles will be much higher for someone from Fresno than someone from San Francisco. Even San Diego-Los Angeles is projected to have higher ridership than SF-LA.

• Alon Levy

Danny, are you sure about this? I’ve read about the first third of Andrew Odlyzko’s manuscript, and it doesn’t indicate anything about length or city size. (On the contrary, the London and Birmingham was successful.)

More to the point, many relatively long HSR lines can succeed – for a ready source of examples, consider the Tokaido and Sanyo Shinkansen, and the first few LGVs. Length obviously affects ridership, but in California’s specific case, San Francisco’s larger population and its higher density and better connecting transit should overwhelm the fact that it is much farther from Los Angeles than Fresno.

At the risk of being anecdotal, let me say that my own travel is almost exclusively personal or professional, both things that scale perfectly with city size. Of course, my professional travel is usually to college towns, but averaged out with similar professional travel among the rest of the population, it should be a perfect proxy for a city’s economy’s size. Personal travel is more inscrutable and depends on social and ethnic networks and this could throw simple ridership models off-track, but to a first-order approximation it still should correspond to city size; more people in San Francisco should have personal reasons to go to Los Angeles than in Fresno. Perhaps I’m weird in that I don’t use the singular cultural attractions of New York and after moving to Providence am going to travel back to see people I know or perhaps my advisor rather than the Metropolitan Opera (and no, it’s not because I’m going to see its equivalent in Boston, which I’m told is not as good).

• Danny

It is in Section 16. Sorry, I read it a while ago and I had forgotten how long it was. The entire section details the failure of direct lines from city center to city center, explaining that the shorter distance fares added in the middle of the routes were often what made the difference between a failing railroad and a successful one.

There are more ways to look at it than just a utilitarian perspective, like for example the market positioning perspective. From SF to LA, you have several daily direct flights. From Madera to SF? Zero. Sure, HSR within the right range can be competitive, but generally speaking businesses are more successful when they position themselves so they don’t have to compete. This positioning is how Target can survive in the age of WalMart.

If you observe vehicle traffic patterns, this is also implicitly visible. Traffic congestion grows within proximity to the cities, and is lowest in the midpoint between the cities. If you travel from SF to LA through the central valley, you will notice that traffic congestion drops off like a rock after you reach Tracy or Salinas (depending on which pass you choose). With the exception of the interspersed cities along the route, you should still notice a growth in traffic in the Fresno/Bakersfield area that grows exponentially as you get closer to the city center.

The long distance lines you mention don’t necessarily refute the point. The Tokaido line hits five major metropolitan areas in its path. I don’t have actual data, but I would be willing to bet that less than 10% of the total ridership is from terminus to terminus.

And you are right, there should be more people from SF that would travel to LA than from Fresno. But what about Fresno, Bakersfield, Modesto, Stockton, Turlock, Sacramento, etc? My guess is that absolute numbers you are right, but proportionally speaking, the percentage of those that demand travel to any specific city center will grow relative to their distance from any city center.

• Alon Levy

I’ll look at section 16 later, but let me just note that most Tokaido Shinkansen ridership (and a huge portion of Tokaido Shinkansen passenger-km) is between just the top three metro areas, since a small majority of trains are Nozomi and don’t stop in the smaller cities. Travel from Tokyo to Osaka is 128,000 people per day – I can’t tell whether it’s just Shinkansen or also air, in which case Shinkansen would be just 105,000 – whereas total ridership is 410,000 per day. Obviously the intermediate markets are very important, both in Japan and in California, but it doesn’t really look like longer-distance links underperform any model.

SNCF uses a gravity model to estimate ridership: the size of the market should be proportional to $(p_{a}p_{b})^{\alpha}c^{-\beta}$, where $p_{a}, p_{b}$ are the metro area populations, $\alpha$ is an experimental constant measured to be between 0.8 and 0.9 (i.e. less than 1, which accords with what you say), $c$ is a generalized cost function including time cost, and $\beta$ is another experimental constant measured to be between 1.9 and 2.2 (i.e. around 2, making this similar to gravitational attraction).

• ant6n

Mmmmh. Maybe this idea is similar to the idea that short subway lines suck.
With every station added on a line, you add n number of possible trips – so even if the number of passengers on each of those trips is small, it can have a big effect since the #trips grow as #stations^2.

5. emdx

Could it also be that France and England had been some of the bitterest ennemies (either outright at war with each other, or under some kind of cold war) until roughly a century ago?

Another thing that does not help is that Britain is not in the Schengen zone, so the extra customs procedures could be a deterrent…

• Alon Levy

Yeah, that’s possible, historically.

I’m somewhat skeptical about the Schengen bit. All flights, domestic or international, have security theater, and customs and immigration inspections take a very small number of minutes in Europe. Therefore, such checks should be a deterrent to train travel but not to flying, and this would be seen in lower HSR mode share rather than in smaller pre-HSR air market size. Although Eurostar’s mode share vs. air performs slightly worse than the trend line, the difference is small, and can be attributed to the half an hour people are told to show up before the train leaves (vs. 10 minutes for domestic TGV).

The worst security problem is the Brussels-London trains that stop in Lille (and only that direction).

Because of Schengen, passengers using a Brussels-London train to travel Brussels-Lille are allowed to skip border control in Brussels. UKBA have found a few passengers (about a hundred a year) who buy a Brussels-Lille ticket and a Lille-London ticket for the same seat, only show the Brussels-Lille ticket when boarding and so skip passport checks in Brussels. As a result, all passengers disembarking from Brussels-Lille-London trains are now required to go through passport control in St Pancras station, adding a typical 20-30 minutes queueing time. This is in spite of the fact that 99+% of these passengers went through passport control in Brussels or Lille.

This doesn’t arise for Paris-London where the French government requires French citizens to show their passport to British authorities for an internal journey within France (ie Paris-Lille on a Eurostar). They can do that, because there’s no international treaty prohibiting restrictions on internal travel.

• po8crg

No, all passport checks on Eurostars are done on departure – there’s a treaty from 2005 when that rule was changed.

6. David Alexander

New York-Toronto is not on Brookings’ list

I suspect it’s because of the high cost of the flight between the two cities. I’ve seen \$300 flights to Toronto and Montreal for what is basically a turboprop and a 1:30 flight. It doesn’t help that the bus and Amtrak are basically perpetual trips to each respective city. At least for a roadgeek, the drive features some of the better looking roads in the US.

US HSR plans should give much lower priority to international links, especially to Montreal

There’s a part of me that would kill for HSR to Montreal, but admittedly, as you noted, the ridership is questionable at best, and I would argue for limited improvements so the track doesn’t suck or at least an overnight sleeper service.

• Alon Levy

I was meaning to mention less competition on international links leading to higher fares and lower ridership. It’s part of it, but can’t be all, because the same issue is observed in Europe, where there’s plenty of competition and low-cost service on international routes.

7. Pingback: Do Not Compare NEC with HSR Ridership | Pedestrian Observations
8. Pingback: Carolyn Maloney’s International HSR Proposal | Pedestrian Observations
9. Pingback: Sometimes, Half a Line is as Good as No Line | Pedestrian Observations
10. Pingback: Difficult Transit | Pedestrian Observations
11. Pingback: High-Speed Rail for the Eastern United States | Pedestrian Observations
12. Pingback: International Links: a Revision | Pedestrian Observations

This site uses Akismet to reduce spam. Learn how your comment data is processed.