At the Transit Costs Project, we’re starting to build a database not just of infrastructure construction costs but also rolling stock acquisition costs. The database is extremely incomplete and not at all ready for public consumption, so please read the following preliminaries with the understanding that more data may reverse some conclusions. All costs below are in 2023 PPP dollars.

• The $100,000/linear meter cost for single-deck rolling stock, which I mentioned in previous blog posts, looks like it holds for European regional trains and Chinese metros. Smaller trains, such as trams, may just be more expensive per linear meter.
• The variation in costs is, as expected, much narrower than in the costs of infrastructure. $200,000/meter exists but is unusual and generally indicates something wrong happened, such as the Berlin S-Bahn stock, which suffered from a lawsuit by Alstom over losing the bid to Stadler.
• There is a rolling stock premium for rubber-tired metros in Paris, on the order of 40% over steel-wheeled metros.
• Berlin and London have expensive metro stock as well, and reading some about their production, I am left to wonder if there’s somewhat of a prima donna premium for cities of such size.
• To my surprise, in cases where the costs of a base order and options can be disaggregated, the options aren’t consistently cheaper than the base.
• There doesn’t seem to be any secular increase in real rolling stock acquisition stock going back to the early 2000s.
• The PPP conversion is dicey within Europe; there’s an integrated European supply chain, and then the question in non-euro countries is which PPP rate to delfate to, which makes a big difference where the PPP conversion differs substantially from the euro’s, in one direction in Switzerland (with its cross-border lines, to add a complication) and in the other in Eastern Europe. In contrast, it’s easy in China, with its domestic supply chain, and in the US, with its pretense of one.
• American costs are near the high end, and rising gently, but even the New York City Subway orders of the 1990s and 2000s had somewhat of a premium – I’m getting $148,000/meter for the base R142/R142A (source) and $120,000/meter for the base R160 (source) with higher costs for the options. Nonetheless, at worst the cost premium for the R211 ($160,000 base, $145,000 option) is a factor of 1.6, not the factor of around 10 we see for infrastructure.
• Some orders come with maintenance and some don’t, and we try to disaggregate this when we can, but sources are inconsistent on whether they mention maintenance bundles in the contract; in at least two cases, the maintenance contract drives up the cost premium, but in some others (like rubber-tired Parisian trains), the premium is not about maintenance.

After my last post detailing how high American subway construction costs cannot be attributed to high incomes, people in comments were talking about inequality instead. Matt was talking about lack of union power, calling high US and UK costs “social democracy by stealth,” and Michael James was talking about political elite elements of inequality including domination by Ivy League and Oxbridge graduates with their old boy networks. Just as the story that high US costs are about wealth does not stand up to closer scrutiny, the converse story, relating high US and UK costs to a negative exceptionalism about their inequality and class systems, is also false.

Inequality and costs

The Luxembourg Income Study has comparative data for inequality in most of the world, looking at inequality after taxes and transfers. The OECD has its own dataset, which mostly agrees with the LIS, except notably in the UK: on LIS data, the UK nowadays has similar inequality to Germany and France, after a long period of decline from post-Thatcher levels under Tony Blair and Gordon Brown, whereas on OECD data, the UK has higher inequality than nearly anywhere else in Europe, and is slightly more unequal than halfway between Germany and the US.

But regardless, among the countries and macro-regions that build a lot of subways, the highest inequality is in India and Latin America. India’s Gini index on LIS numbers is 0.5; Indian costs are in theory below-average, but only by virtue of high use of els, and correcting for that, its costs are high by global standards and low by Anglosphere ones. Chile, remarkable for its low costs (as explained by Eno), has a Gini index of 0.46. Much more expensive Brazil is rather similar. Next to Latin America there is China, which the LIS gives a Gini of 0.41, and which has rather average costs. Any attempt to do correlations between a country’s subway construction costs and its Gini will run into those outliers; I expect the correlation is still going to be positive because of high American inequality (0.39) and high American costs, but dropping the US it might not even be positive, let alone statistically significant.

Canada and Australia

The US unambiguously has high inequality and high costs. The UK has ambiguous inequality, and other features of high inequality, such as low income mobility (see for example an OECD review). However, Canada and Australia both have lower inequality and high income mobility; a more recent paper, by Corak, Lindquist, and Mazumder, finds that Canada has more mobility than Sweden and not just more than the US.

The sort of elite that the UK and US have does not exist in Canada. The United States is a country of “where did you go to college?”; the UK is the same, except it’s called university and not college. Most of Continental Europe is not like this at all – what one studied matters much more (engineering is good, humanities are not) than where one studied. Canada is rather like Continental Europe in this – there just isn’t an equivalent of the Ivies or Oxbridge there, just a number of huge, very good research universities like Toronto, McGill, and UBC, none of which is anything like an elite American university. The current prime minister is literally the son of a previous prime minister, but his predecessor, Stephen Harper, worked his way up from the mail room of an oil company; the previous PM, Paul Martin, is the son of politicians; the PM before, Jean Chrétien, was born indigent, more reminiscent of classical Victorian literature more than modern first-world poverty. This is not the consistent middle-class or princeling upbringing of leaders in the US or UK.

And yet, Canada has horrifically high construction costs. This, again, is in a country without any of the elite hangups of the US or UK. It doesn’t have Nordic or even Franco-German inequality, but it’s very close to France and Germany and not at all close to the United States. It doesn’t have nearly so much core-periphery dynamics (every single province has received net equalization payments at least once in the history of the program). Its provinces have local core-periphery dynamics, but nothing by the standards of any region or small country that builds subways at reasonable costs. It doesn’t need to do stealth social democracy.

Italy and East Asia

While Canada (and Australia) has American costs without American inequality, in Italy we see low costs while inequality is, by European standards, high. On LIS numbers, Italy is the worst in Western Europe on this. It also has low income mobility, and very sharp core-periphery dynamics, with incomes in Lombardy clocking in at about twice the level of Campania and Sicily. If anywhere needs to do stealth social democracy, it’s Italy – the core-periphery dynamic does lead to social disaffection on both sides of the North-South divide.

And yet none of that happens. Construction costs in Central and Northern Italy are rather low. They’re higher in Naples, for geological reasons more than any stealth social democracy.

What’s more, Italian costs fell between the 1980s and 2000s, an era of much cleanup of the endemic corruption of the First Italian Republic. This did not involve any resolution of longstanding social tensions – the North-South divide remained as large as ever, economic inequality rose (slightly), the one-party rule of Democrazia Cristiana was replaced with left-right polarization and the growth of the both extremist and regionalist-to-secessionist Lega Nord as one component of the right coalition. Economic growth ground to a halt – Italy took until about last year to recover to 2007’s GDP per capita, which itself was not too far above 1990s levels.

And again, Italian costs fell.

Democratic East Asia has different tensions – very deep political polarization in Taiwan and South Korea, political corruption, distaste for a universal welfare state of exactly the kind that encourages stealth place-based replacements, brutal test meritocracies in which the leadership comes from the same narrow social milieus as the US and UK. Taiwanese costs are high, though not Anglosphere-high; Japanese ones are lower; Korean ones are quite low. None of this seems to matter.

The timing of Anglosphere cost growth

To be clear, there are elements of the Reaganite and Thatcherite package that I do think are responsible for high costs – namely, the delegitimization of the civil service. But they’re not usually core to what people perceive as Thatcherism, and much of the dismantling of state capacity happened under people trying to come up with a synthesis of postwar big government and the Reaganite and Thatcherite antithesis thereto, leading to public-private partnerships and outsourcing of government functions.

None of that has anything to do with inequality or stealth social democracy or elite theory, though. The same cost explosion happened in Canada, where on LIS numbers, inequality has been largely the same since the 1970s; Brian Mulroney, unlike Thatcher or Reagan, did not preside over a rise in inequality. It did not happen in Germany, where inequality did rise under Gerhard Schröder and early Angela Merkel; Schröder’s ideology was similar in broad outline to that of Bill Clinton and Tony Blair, except that under Schröder economic growth was weak and inequality rose, the opposite of Clinton and Blair, and yet Anglo-style privatization of state planning to consultancies did not happen.

Rather, the culprit for high costs must be a more specific set of project delivery mechanisms that have been popularized in 1990s Britain – the same globalized system I mentioned in the previous post, workshopped by British consultants in Hong Kong and Singapore, and now in the Gulf states. British neoliberalism had Singapore and Hong Kong to absorb consultants, whereas German neoliberalism had Russia, which led to different dynamics, such as gas dependence and geopolitical weakness, rather than Russia inviting Germans to teach it how to redo the way RZhD contracts infrastructure and operations.

And in the US, costs rose from an already high base. This is not the same situation as in the UK, where in the 1960s and 70s costs were the same as in Continental Europe. Rather, American costs were already high, New York costs having exploded in the 1920s. Things have gotten worse in the last 40 years, but the mechanisms aren’t always the same, and the atrophying of the civil service is not a matter of stealth social democracy, not when postwar New York was incapable of building.

The Baumol effect is a mechanism for how the real costs of goods and services can rise over time: wages rise due to economy-wide productivity growth, including in sectors with no productivity growth, and this raises their overall real costs of production. The original example for Baumol was classical concerts – they use the same number of musicians as in the 19th century, but wages have increased from 19th-century levels. More generally, it’s also used to explain higher real costs of services as service productivity growth lags manufacturing productivity growth.

Unfortunately, I’ve also seen people use Baumol as a way of explaining rising infrastructure construction costs, for which it is not at all a good explanation. In fact, even though growth in average infrastructure costs over time is documented, there is very little cross-national correlation between GDP per capita and per-km subway construction costs. Notably, the Anglosphere’s very high construction costs affect not just very rich countries like the US and Singapore but also ones that are poorer than the Western European average, like New Zealand, Ireland (which has high GDP per capita due to corporate profits but unimpressive local wages), or increasingly the United Kingdom. Conversely, Nordic and Swiss wealth has not at all led to high construction costs, and until recently the Nordic countries and Switzerland had some of the world’s lowest tunneling costs.

Metro construction costs and GDP per capita

In an earlier version of the construction cost database, there was some positive correlation between GDP per capita and construction cost per km (about 0.23), but nearly all of it came from the fact that poorer countries tend to build more els and fewer subways; correcting for that, the correlation fell to about 0.04, and turned negative if New York and Singapore were dropped. We made a scattergram at the national level:

While looking at the scattergram, bear in mind that the poorest country as of 2020 on our list, Pakistan (the small gray circle touching the much larger gray circle of India), built an all-elevated line, and in general, substantially-elevated or even all-elevated lines are common in developing Asia, including Vietnam, Bangladesh, Indonesia, the Philippines, Thailand, and India. The only all-underground Indian line in our database, Mumbai Metro Line 3, cost $535 million/km in 2023 PPP dollars; Mumbai Line 11 and Chennai’s first-phase program are the only other two items that are majority-underground, both a bit more than $300 million/km.

At this point, even if we restrict our attention to Europe, the correlation between GDP per capita and construction costs per km isn’t clear. For example, Railway Gazette has just reported on the groundbreaking of the first metro line of Cluj-Napoca, to cost 13.7 billion lei/21 km, which is $8 billion in PPP terms, or $380 million/km. But then rounding up the bottom of the EU’s GDP per capita table is Bulgaria, with fairly low costs.

This is not supposed to happen if the Baumol effect is what’s going on. Grocery prices in developing countries are lower than at first-world discounters like Walmart or Aldi, even in PPP terms. Even at tourist traps, the prices are usually lower than where the tourists came from, not because retail and food service are atypically efficient in developing countries, but because these are labor-intensive industries and labor wages are lower in Thailand or China, let alone in India or Pakistan, than in the US or Germany.

The issue of the Anglosphere

The Anglosphere has atypically high construction costs. A dummy variable that takes the value 1 in the US, Canada, Australia, New Zealand, the United Kingdom, Singapore, Hong Kong, and (when it starts building) Ireland, and 0 elsewhere, has a correlation of 0.41 with per-km construction costs. In contrast, the tunnel percentage only has 0.15 correlation, due to the aforementioned effect of high-cost developing countries building els. I’ve heard the high Anglosphere costs blamed on a kind of areal Baumol effect: high pay in professional services drags the costs up, on the theory that the United Kingdom may be poorer than Germany and no richer than France, but at least it has productive London finance that drags engineering wages up, so Britain can’t just Germanize or Francize, right?

Well, no. British engineering wages are not at all high by Continental Western European standards. London finance pays a lot, but also has been stagnating for a while, and a number of professional service firms and regional HQs have left the country in response to Brexit to locate in the rump-EU, for which Amsterdam is a popular destination. British costs remain high, and if anything, they’re exploding again. The Bakerloo line extension is now projected to cost £5-8 billion in 2023 prices for what looks like 8 km, which is around $1.2 billion/km in PPP terms, somewhat more than the much more complex Crossrail and about twice as expensive as the comparably complex Northern line extension to Battersea.

Then-Singaporean minister of transport Khaw Boon Wan excused the meteoric growth in Singapore’s MRT construction costs on the grounds that the Singaporean economy had grown rapidly as well. But we’re seeing the same cost explosion in the slower-growing United Kingdom, and conversely we’re not seeing high costs in fast-growing South Korea. New Zealand, which has had okay growth but from low levels for a Western country and remains poorer than Italy, has these extreme costs as well. It’s not that the Anglosphere is rich; it’s that it’s the Anglosphere and builds inefficiently.

So why have costs grown?

While there is no correlation between subway tunneling costs and GDP per capita, there is an evident secular growth in costs over time. It’s not uniform everywhere – German costs are barely up compared with the 1970s and Italian costs are slightly down – but it’s huge in the Anglosphere and also evident elsewhere (for examples, in France and in the Nordic countries). So what’s going on?

Well, we’ve divided the New York cost premium into three tranches: labor (mostly overstaffing, not high wages), station and system design, and procurement and soft costs. All three show evidence of having gotten worse, the first in the Northeastern United States and the other two throughout the Anglosphere and sometimes also elsewhere.

Ad labor, staffing levels in New York are just higher than elsewhere. More workers are required to service a tunnel-boring machine in New York than in Istanbul, let alone richer European cities. This is, in theory, an eexample of the Baumol effect: higher wages raise the real cost in an industry without productivity growth. But in fact, there has been plenty of productivity growth in this industry, the Northeast just refuses to make use of it. Stockholm has been able to keep up and keep its labor share of the hard costs to the same 20-something% as Turkey and Italy; New York and other Northeastern US cities are in the 40-60% range instead. Swedish construction productivity has grown at slower rates than the overall Swedish economy, but American construction productivity has fallen.

Ad station and system design, we have pointed out that stations for Second Avenue Subway Phase 1 dug a cavern twice as long as necessary for the train, for the benefit of extensive back-of-the-house spaces, where in non-UK Europe and in China, the digs are typically a single-digit percent longer than the train. This is a general North American problem, also evident in Los Angeles and Vancouver, and I believe also in London. It’s also a new problem: in the 1980s, the overage in the United States was small, comparable to contemporary European levels. Then there’s the issue of poor standardization of materials, systems, and designs; we are uncertain whether this is a growing or longstanding problem, but it is smaller in magnitude than that of excessive station size, and in general, standardization is more important in a richer economy than a poorer one since the richer economy will have more reliance on big businesses with division of labor, which is also one of the speculated causes of the falling construction productivity (it’s a less standardized sector).

Finally, ad procurement, the invention of the globalized system in which state planning is outsourced to private consultancies, and with time even the supervision of the consultants is outsourced to consultants, is an Anglosphere special, dating from the 1990s onward. This system comprises design-build procurement (confusingly called progressive design-build in New York), very large contracts sometimes growing to $1 billion apiece, lump-sum rather than itemized contracts, and privatization of risk. It’s turned entire countries, like the United Kingdom, incapable of building more than about one line per generation. The Nordic countries have been affected as well, leading to sharp cost growth from very low levels to rather average ones. Canada went from fairly normal costs in the early 2000s to building the most expensive subway outside New York with the new Ontario Line cost overruns, and this can be traced to Toronto officials visiting Madrid, a city that sticks to traditional design-bid-build procurement, and coming back convinced that to imitate Madrid’s low costs Toronto should adopt design-build.

None of this is the Baumol effect or some general cost disease. When agency officials lose interest in building things and instead want to outsource their own jobs to consultants, it’s not Baumol; it’s experimenting with a new way of project delivery and then refusing to admit that it’s a failure. The same is true when nobody bothers to say no as each operating department demands more back-of-the-house space until half the station dig is about providing high-cost underground break rooms and storage rather than about providing space for trains and passenger circulation.

It’s a comforting story for Americans, Brits, and Singaporeans to tell themselves that their infrastructure costs are so high as a byproduct of their wealth. It happens to be entirely false. It’s not even interestingly wrong; it’s just plain wrong, ignorant of the explosion in station size, of the failures of the globalized system of project delivery, and (in New York) of labor productivity innovations elsewhere. The Anglosphere is not expensive because it’s ahead in anything, but rather because it’s behind. And as we see in the United Kingdom, it doesn’t even require American or Singaporean wealth to be totally incurious of Continental European success.