Category: Construction Costs

Public Transportation in Megacities

I’ve been talking so much lately about integrated timed transfer in the context of Boston that people started asking me if it’s also applicable to New York. The answer is that the basic principles are not scale-dependent, but the implementation is, so in very large cities, public transport planning should not look like in Switzerland, a country whose largest metro area is staring at 2 million people from the bottom.

The one caveat here is that most cities are not huge. The developed world has seven megacities: Tokyo, Seoul, New York, Los Angeles, Osaka, London, Paris. And Los Angeles doesn’t really have public transportation, so we’re down to six. The middle-income world has a bunch more for sanity checking – Mexico City, São Paulo, Rio de Janeiro, Buenos Aires, Johannesburg, Moscow, Istanbul, Tehran, Beijing, Shanghai, Guangzhou, Shenzhen, Bangkok – but all are either still in convergence mode building up their networks or (mostly in Latin America) have given up. So much of this comes down to the idiosyncrasies of six cities, of which the largest three networks are substantially in the same planning tradition.

Demand is huge

Big cities have big centers, which can’t really be served by any mode except rapid transit. Even in Los Angeles, what passes for a central business district has around a 50% public transport modal split. This means that the transport network has to deliver high throughput to a relatively small city center. Even in a low-kurtosis city like Paris, most Métro lines converge on a narrow area ranging from Les Halles to Saint-Lazare; in a high-kurtosis one like New York or Tokyo, there are a few square kilometers with 200,000 jobs per km^2, which require an exceptionally dense network of rapid transit lines.

Some other network design principles follow from the need to amply serve city center. Specifically, high frequency is rarely a worry, because there’s so much demand even off-peak that usually megacity subway systems do not venture into the frequency range where long waits deter traffic; New York’s 10-minute midday gaps are bad, but that’s unusual and it comes from a combination of the legacy of postwar fear of subway crime suppressing demand and excessive branching.

But other principles require careful planning still.

Electronics before concrete, megacity version

The driverless lines in Paris support peak throughput of 42 trains per hour – a train every 85 seconds. CBTC on Line 13 without driverless operation supports 38 tph, and London’s CBTC-equipped lines support 36 tph when the branching isn’t too complex. It is imperative for other cities to learn from this and do whatever they can to reach similar headways. The difference between 21 tph, as in Shanghai, and Paris’s 42, is equivalent to building a brand new subway line. And what’s more, in a city in the size class we’re talking about, the primary concern is capacity – coverage is already good, so there really is no reason to build two 21 tph lines instead of one 42 tph one.

The situation in Paris is in a context with self-contained lines. That said, extremely busy self-contained lines do exist in other megacities – London has a bunch with near-Parisian levels of throughput, New York has some, Tokyo has a few, Seoul and Osaka are both more self-contained than Tokyo is.

Throughput and organization

The primacy of throughput means that it’s worthwhile to build small infrastructure upgrades, even with concrete, if they help with capacity. Right now the Northern line reverse-branches with the branches to the north recombining with those in the center, and Transport for London would like to split the line in two, reducing branching complexity, which would increase capacity. But doing so requires improving pedestrian circulation in the corridors of the branch point, Camden Town, where TfL expects very large transfer volumes if there’s a split and already there are circulation problems today without a split. Hence the plan in the medium term is to upgrade Camden Town and then split.

If there are bumper tracks at the end of a line, as at 8th Avenue on the L or Flushing-Main Street on the 7, then it’s useful to dig up the street for another block just to add some tail tracks. That way, trains could enter the station at full speed. This increases throughput, because the terminal interlocking has trains heading in opposite directions crossing each other at-grade, which imposes schedule constraints; it’s best if trains can go through the interlocking as fast as possible to reduce the time they’re in a constrained environment, but that in turn requires short tail tracks so that an overrun of a few meters is not catastrophic. Ideally the tail tracks should even extend a full train length past the platform to place the interlocking on the other side of it, as is done in Paris and Moscow; in that case, trains cross the interlocking out of service, when it’s easier to control their exact timings.

Such projects are disruptive, but the disruption is very localized, to just one transfer station for a deinterlining project as in London or one terminal as in New York, and the impact on capacity is very large, if not quite as large as the full suite of signaling and track upgrades that make the difference between a train every 3 minutes and a train every 1.5 minutes.

Network design

The ideal metro network is radial. Megacities already support that just because so many lines have to serve city center. However, it’s important to make sure every pair of lines intersects, with a transfer. No large metro network in the world achieves this ideal – Mexico City’s network is the largest without missed connections, but it is not radial and its only three radial lines are overburdened while the other lines have light ridership. Paris has just a single missed connection on the Métro proper, not counting the RER, but it has many pairs of lines that do not intersect at all, such as M1 and M3. London is more or less a pure radial, but there are a handful of misses, including one without any transfer between the two lines anywhere, namely the Metropolian line (including Hammersmith and City) and the Charing Cross branch of the Northern line.

Big cities that plan out a metro network have to make sure they do better. Missed connections reduce passenger ridership and lead riders to overload the lines that do get connections; for example, in Tokyo one reason cited for the high ridership of the Tozai Line is that until Fukutoshin opened it was the only one with a transfer to every other subway line, and in Shanghai, Line 1 was extremely congested as long as the alternatives going north either had critical missed connections (like Line 8) or avoided city center (like Line 3).

The role of regional rail

Regional rail as a basic concept is mostly scale-invariant. However, the design principles for trains that come every half hour are not the same as those for trains that come every 5 minutes. If trains come every half hour, they had better connect cities in a roundtrip time equal to an integer number of half hours minus turnaround times, so that they don’t have to loiter 25 minutes at a terminal collecting dust and depreciating. If they come every 5 minutes, they’re not going to loiter 25 minutes anyway, and the difference between a 5-minute turnaround and a 7-minute turnaround is not really relevant.

The design principles are then mostly about throughput, again. The most important thing is to build independent trunk lines for trains to serve city center. Even in a huge city, the finances of building a purely greenfield subway deep into suburbia are poor; Tokyo has done it with the Tsukuba Express but it’s mostly above-ground, and for the most part regional lines there and elsewhere come from taking existing suburban lines and linking them with city center tunnels.

Tokyo’s insistence on making these city center tunnels also form a coherent metro network is important. Only one non-Tokyo example is worth mentioning to add to all of this: this is Berlin, which is not a megacity but has three independent S-Bahn trunk lines. Berlin, unlike London and Paris, painstakingly made sure the S-Bahn lines would have transfers with the U-Bahn; its network has only one U-Bahn/S-Bahn missed connection, which is better than the situation in Tokyo, Paris, or (with Thameslink and Crossrail) London.

The role of development

All first-world megacities, and I believe also all megacities elsewhere, have high housing demand by domestic standards. All are very wealthy by domestic standards except Los Angeles, and Los Angeles is still incredibly expensive, it just doesn’t have the high wages to compensate that London and New York and Paris have. In such an environment, there’s no need to try to be clever with steering development to transit-oriented sites. Anywhere development is legal, developers will build, and the public transport system has a role to play in opening more land for more intense development through fast trips to the center.

A laissez-faire approach to zoning is useful in such an environment. This contrasts with smaller cities’ reliance on finger plans, like the original one in Copenhagen or the growing one in and around Berlin. No limits on development anywhere are required. The state’s planning role remains strong through transportation planning, and the suburbs may well form natural finger plans if developers are permitted to replace single-family houses with apartment buildings anywhere, since the highest-value land is near train stations. But state planning of where housing goes is counterproductive – high transit ridership comes from the impossibility of serving a large central business district by cars, and the risk of politicization and policy capture by homeowners is too great.

The advantage of this approach is also that because in a high-demand city public transport can to some extent shape and not just serve development, it’s okay to build lines that are good from the perspective of network coherence, even if the areas they serve are a bit light. This principle does not extend indefinitely – subway and regional rail lines should still go where people are – but for example building key transfer points in near-center neighborhoods that are not in high demand is fine, because demand will follow, as is building lines whose main purpose is to close some gap in the network.

Construction costs

The larger the city, the more important cost control is. This may sound counterintuitive, since larger cities have more demand – only in Manhattan could a $1.7 billion/km extension like Second Avenue Subway pencil out – but larger cities also have a bigger risk of cost blowouts. Already Tokyo has stopped building new rapid transit in the core despite very high crowding levels on the existing network, and London builds next to nothing as well. New York’s poor cost control led Philip Plotch to entitle his book about Second Avenue Subway The Last Subway. Even Paris builds mostly in the suburbs. Extensive city center and near-center construction continues in Seoul, in the context of very low construction costs.

The flip side is that a New York (or even London) that can build subways at the cost of Paris, let alone Seoul, is one that can rapidly solve all of its transport problems. My Assume Nordic Costs map fixates on a region of the world with small cities, but the construction costs in South Korea are if anything lower than in the Nordic countries. And even that map, given free reins for developers, is underbuilt – some lines would look ridiculous at current costs and zoning but reasonable given low costs and liberal zoning, for example something meandering through currently industrial parts of New Jersey.

Small cities designed their public transportation philosophy around scarcity: Switzerland really can’t just draw crayon and build it, because housing and transport demand there are finite and limited. Cities like New York and London, in contrast, should think in terms of abundance of infrastructure and housing, provided their regulations are set up in a way that permits the state to build infrastructure at low costs and private homebuilders to redevelop large swaths as they become easily accessible to city center.

Low- and Medium-Cost Countries

I was asked a very good and very difficult question in comments yesterday:

What, specifically, are the best practices we all should be learning from the lower cost countries? I’m reading a lot of what not to do, but not on what do to.

I’ve gone a lot over bad industry practices leading to high costs in the Anglosphere, especially the United States, especially New York, and to some extent also on bad practices in developing countries like India. These I am contrasting with a set of good practices from a host of low-cost countries like Spain and Italy as well as medium-cost ones like Germany and France.

However, the question remains, what distinguishes low- and medium-cost countries? The differences between them are not small – underground rail extensions in German cities are averaging around 250 million euros per kilometer, and ones in French cities are around 200 million or just less than that, whereas Milan and Turin average around 100 million, and Spanish cities average even less. Germany is a higher-cost and higher-wage country than Italy and Spain, but Berlin wages are not higher than Madrid and Milan wages, and within these countries, richer cities don’t really have higher costs (Milan is cheaper to tunnel in than Naples, Madrid has lower costs than Barcelona and similar costs to the rest of Spain, the cheapest German tunneling seems to be in Hamburg of all places). No: this is almost certainly a real difference in institutions that enables Southern Europe, Scandinavia, South Korea, and Turkey to dig tunnels at one third to one half the cost of Germany.

This argues in favor of doing a deep dive case on a medium-cost example like Paris or Berlin, in addition to the work we’re doing on low-cost Milan and Istanbul. The problem is that it’s not clear, so far, what to even look at. I have a decent idea of what the difference between the high-cost world and the rest of the world is – but applying it to the low- and medium-cost world is dicey.

In-house engineering capacity

So much of the problem in the Anglosphere seems to come from the loss of in-house engineering capacity, and its replacement with private consultants. The latest iteration of this is the penchant for design-build contracts, in which the state contracts with one company to handle the entire process and doesn’t have much if any public-sector oversight. Design-build doesn’t exist in any of the countries in Continental Europe I have any familiarity with; France is looking into it as a reform in the future, but only under the aegis of a large public-sector planning team coordinating the private-sector design and construction. Moreover, Canada’s recent adoption of design-build, coming from an ideology imported from Britain and then falsely claimed to come from Madrid, preceded an explosion in costs.

However, this does not really explain the difference between France or Germany and Scandinavia or Southern Europe. Norway and Spain have separation of design and construction; so does France. Italy and Spain have in-house engineering teams responsible for a great deal of the design, but to some extent France does too, with a large in-house planning team overseeing the private-sector designs.

Procurement issues can’t really distinguish the low- from medium-cost world either. Madrid does not hand out lowest-bid contracts – at least in its big wave of expansion in the 1990s and 2000s, contracts were given 50% on the basis of a technical score marked by the in-house engineering team, 30% on that of cost, and 20% on that of how fast they could finish the project. Paris doesn’t have a 50-30-20 split but a 60-40 one; that may be significant, but I doubt it. Both systems contrast with the American system of lowest bid, or sometimes 30-70 in California. Moreover, Turkey is lowest-bid, but it’s a repeated game due to the country’s fast growth and construction – contractors who screw up do get penalized in future projects.

Citizen voice and NIMBYism

Germany has a huge problem with NIMBYism. Key segments of the national passenger rail network, for example Hamburg-Hanover, remain slow because local NIMBYs who don’t like fast trains have litigated high-speed rail to death. France has had anti-LGV NIMBYism in Provence as well, which NIMBYism is often extralegal (that is, aggrieved drivers blocking roads); this forced the state to change its plans for a high-speed railway to Nice from a mostly above-ground inland route to a tunnel-heavy coastal route through the Maritime Alps and, as the cost was prohibitive, eventually downgrade into a mixture of high- and low-speed rail line.

As I understand it, this is less of an issue in Southern Europe. I do not know to what extent it’s a problem in Scandinavia and Switzerland; Switzerland does have a lot of bucolic NIMBYism, where “bucolic” means “the city as it looked in 1957,” but I don’t think it’s had any that successfully scuttled infrastructure, and overall the political imperative there seems to reduce costs more than anything.

The NIMBY issue is also important in the US and UK. In the US, NIMBYs are not legally strong, but politicians prefer to avoid the appearance of controversy and therefore give local actors whatever they ask for, no matter the cost; many sources told Eric and me versions of this story regarding the high cost of stations on the Green Line Extension (which are, to be clear, maybe 20% of the cost of the whole project). Brooks-Liscow favor this explanation for the internal increase in the cost of highways in the US from the 1970s onward.

I do not know to what extent there’s an institutional explanation here. I do not even know if this is a real difference between on the one hand France, Germany, Austria, the Netherlands, most of the 2004-7 EU accession countries, and Japan, and on the other hand Southern Europe, Bulgaria, Turkey, and South Korea. It’s possible that this is a bigger problem in Northern Italy than I realize.

The most worrying possibility is that this is a real difference, and it comes not from something about institutions, but from surplus extraction. The European core and Japan are rich, and at $150 million/km, subways there would create immense social surplus and decent financial surplus. (The Japanese state is refusing to build at $500 million/km because it wants a 30-year financial payback). Southern Europe is less rich, so there is less social surplus to extract by local actors wanting to dip their beaks in state money; Switzerland and the Nordic countries are rich, but their cities are smaller and farther-apart, so there is less surplus there too.

There are a lot of objections that can be raised to the surplus extraction hypothesis: there is plenty of surplus in Seoul and not much in Vienna or Prague or Bucharest or Warsaw or Tel Aviv, Japan already reached $250 million/km in the 1970s when it was a lot poorer than Korea is today, the surplus hypothesis predicts that there should be higher costs in richer cities within the same country and yet this is not observed, local interference with Métro expansion in Paris unlike with LGVs doesn’t seem very significant.

Conclusion

There’s no good answer to what distinguishes low- from medium-cost countries. I wish more people here and in France were interested in this question – the activist sphere in Berlin seems far more interested in trams and bike lanes than in rapid transit. Nor do I imagine Germans and French are ready to hear that there’s something the Italians and Spaniards and Turks do better than they do. But it’s something Germany is going to need to learn to deal with if it wants better infrastructure; on the same budget, it can get 2-3 times as much as it’s getting now.

Hong Kong Construction Costs

I think we have found the #2 city in urban rail construction costs, behind only New York. This is Hong Kong, setting a world record for the most expensive urban el and encroaching on Singapore for most expensive non-New York subway.

As we look for more data to add to our transit costs website, I looked at Hong Kong to see what was going on. I remembered that its costs were high, but didn’t remember details – I think the project I was thinking of was the longest, the Sha Tin to Central link, but I looked at all recent, under construction, and planned MTR lines. I summarized the results on Twitter, but I’d like to cover this in more detail here.

The projects and their costs

Tung Chung Line extension: a planned line for construction in 2023-30, a total of 1.8 km underground, HK$18.7 billion, or around US$3.1 billion, or $1,730 million per km. Even giving it 1.8 km seems like I’m doing Hong Kong a favor – the extension is 1.3 km, and the other 500 meters are overrun tracks at the Hong Kong Island end, which I don’t ever count elsewhere since it is not in-service trackage. In addition to the tunneling (and single underground station), there is a single at-grade infill station, whose contribution to the budget is approximately zero.

West Island Line: 3 km underground, HK$18.5 billion. This is around US$3.4 billion, or $1,130 million per km. Only six lines globally are more expensive than this: phase 6 of the Circle line in Singapore, and the five New York lines, of which three are not even open yet. This is not even regional rail, but construction is entirely within the Hong Kong CBD, explaining why it is so expensive even by local standards.

Sha Tin-Central: 17 km, HK$87.3 billion, or around US$14.5 billion. This is $850 million per km. The line is not even fully underground, just 90%: the northernmost segment, totaling around 10%, is elevated. This line near-ties Crossrail and the Melbourne Metro Tunnel for most expensive line in the world longer than about 5 km – New York is building short lines, the longest (Gateway) around 5 km depending on source. The line is partially regional rail: it includes a 6 km extension of the East Rail Line under Victoria Harbour toward Admiralty, but the other 10 km is not regional rail.

Tuen Mon South Extension: 2.4 km, HK$11.4 billion, all elevated, in an outlying residential area. This is $790 million per km, making it the world’s most expensive el – New York’s most recent els, the JFK and Newark Airport connectors, were positively reasonable by this standard, only around $270 million/km adjusted for inflation (but don’t worry, the PATH estimates in the near future are a lot worse).

Kwun Tong Line extension: 2.6 km, HK$7.2 billion (same source as West Island). This is around US$1.3 billion, or $500 million per km. It’s the standard high cost of projects around the world, common for regional rail tunnels and CBD tunneling, except that this is strictly on the Kowloon side without as much older infrastructure to cross – it even misses a connection to the East Rail Line.

South Island Line (East): 7 km, HK$16.9 billion, around US$3 billion – see same source as West Island, or SCMP reporting. This is $430 million per km. This is not a fully underground line: as explained here, 2 km is on viaduct, serving Ap Lei Chau. Notice also that the original cost estimate was HK$7 billion, but by the time construction rose the budget had risen to $12.4 billion, and the final budget was $16.9 billion.

Is Hong Kong in the Anglosphere?

I’ve argued before that the single biggest predictor of an urban rail project’s cost is whether it is in the Anglosphere – the correlation of an Anglosphere dummy in our database is 0.54, more than even whether the project is underground or elevated. So it’s worth asking, is Hong Kong in the Anglosphere? There are arguments both ways, but I believe the preponderance of evidence points to yes.

  • Hong Kong was under British rule until 1997.
  • The legal system is traditionally based on English common law, even if there’s been a recent shift toward Chinese law.
  • There is extensive exchange of knowledge with the core (white) Anglosphere, with managers who’ve moved around like Jay Walder, political leaders who have second passports in the UK (like Carrie Lam) or sometimes Canada or Australia, and Anglo media that reprints MTR press releases about its property development model.
  • The design layout of the MTR has obvious British influences, including for example the use of cross-platform transfers between the core lines. Similarities with China are the result of convergent evolution (China is influenced by the USSR, which was influenced by Britain). There are some similarities with Japan, like the smartcard system’s use as electronic money, but they are smaller.
  • Hong Kong’s love of privatization and high inequality is very Thatcherite. Again, the similarities with Japan are smaller – Japan’s privatization is slower, and Japanese corporations rely on mutual obligations whereas Hong Kong (like Singapore) expects brutal working hours of employees without offering them lifetime employment in return.

The one non-British aspect of the MTR is its use of property development subsidies (and before anyone asks: no, the costs above are just infrastructure, not property development). MTR expansion is funded by a mixture of property development, for which the MTR receives land at below-market rates, and more direct subsidies.

However, this is still more an Anglo aspect than an Asian one. Democratic East Asia notably does not give corporations land for below-market prices, not in the 21st century. Moreover, the British fascination with the Hong Kong model, which fascination is not present in France or Germany or probably anywhere else with reasonable construction costs and democratic constraints on the state, suggests that the elites in Britain and the US would like to be governed this way, just as many would like to be governed by the Lee clan. There is, in contrast, almost no curiosity about democratic East Asian governance, even after that 200 million people region proved itself to deal with corona better than any other.

Corona is a little awkward to bring in because Hong Kong’s infection numbers look like those of an East Asian democracy (it has the civil service of one), whereas the most similar country to Hong Kong on most matters, Singapore, has those of a Gulf state full of indentured migrant workers who got infected at extremely high rates. But for engineering, it doesn’t seem terribly important what the immigration numbers are – for example, Sweden and Norway are extremely similar to Finland even though they have way more immigrants, and likewise Saudi Arabia is similar to the other, immigrant-heavier Gulf states. So overall, Hong Kong’s public transport situation can be seen as very similar to Singapore’s – and Singapore has very high costs as well.

What does this mean?

I don’t know. Singapore and Hong Kong’s costs are probably higher than those of the core Anglosphere, but I am uncertain – Singapore’s big projects are not unusually expensive by Canadian or British standards, and the Sha Tin-Central link is legitimately difficult, the kind that Sweden would build for $250 million/km rather than for $130 million/km. So it’s hard to tell whether there is something about Hong Kong that goes beyond standard Anglo dysfunction.

I do not know what Hong Kong’s historic costs were. I expect them not to be so high – Singapore’s weren’t through phases 1-5 of the Circle line, and only exploded with the Downtown and Thomson lines, and Canada’s only exploded in the late 2000s and 2010s as it decided to privatize state planning and adopt design-build contracting.

However, in the present and near future, Hong Kong is a model to study purely for its failures, much like Singapore. The leaders of Hong Kong, in their rush to emulate Chinese repression tactics, should perhaps also learn something from Chinese construction techniques – or, ideally, Korean ones, Korea being the only Asian country among the world’s cheapest. People in other countries should aim to study Hong Kong’s infrastructure construction as an example to avoid, and not one to emulate.

The Costs of Subways and Els

I’m probably going to write this up more precisely with Eric and send this to a journal, but for now, I’d like to use our construction costs database to discuss the cost ratio of subways to elevated lines. The table I’m working from can be found here; we’re adding projects and will do a major update probably at the end of the month, but I don’t expect the new data to change the conclusion. Overall, the data is consistent with a subway : el cost ratio in the 2-2.5 range, but it’s not possible to get more precise estimates despite the breadth of the data.

Crude averages

Our database has 11,559 km of total length, but not all of that comes with cost estimates yet; subtracting lines for which we don’t have costs, we get 11,095 km. The total cost of all the lines in our database is, in PPP-adjusted but not inflation-adjusted dollars, $2.302 trillion, for an average of $207 million/km. Nearly all of the items are recent – the majority by length are still under construction, and only 10% opened by 2010. So inflation adjustment is minor, though nontrivial.

Moreover, looking only at 100% underground lines, we get 3955.3 km, for a cost of $945.3 billion, averaging $239 million/km. The other lines are mixed or elevated. The purely elevated lines total 2490.4 km, for a cost of $408.1 billion, or $164 million.

To be slightly fancier but use the same underlying data, the linear estimate of cost per km, treating the tunnel proportion as the independent variable, is 153.1406 + 117.5787*tunnel-proportion; this has a larger spread than just averaging pure subways and pure els, coming from both the inclusion of more data and from not weighting by line length.

However, even the larger spread has a subway : el cost ratio of 1.77, lower than found elsewhere in the literature. Why?

Els are disproportionately build in higher-cost countries

The most important quantitative fact coming out of the analysis of construction costs is that the most important independent variables are country-level dummies. The correlation between the tunnel proportion and cost per km is just 0.163; the correlation between cost per km and a dummy variable that takes the value 1 in the US, Canada, Britain, Australia, New Zealand, and Singapore and 0 elsewhere, is 0.543. If we instead set the dummy variable to take the value 1 in the countries I consider cheap – Spain, Portugal, Italy, Greece, Bulgaria, Switzerland, Sweden, Norway, Denmark, Finland, Turkey, South Korea – then the correlation with plain cost is -0.18, and since linear correlation is better at detecting high outliers than low-but-positive ones, we can take the reciprocal of cost and then the correlation is 0.258.

So it’s useful to figure out where the most els are being built. For example, China has 5,933 km in our database – that is, a slight majority – of which 3,851 are confirmed tunnel and another 1,046 are unconfirmed (Hangzhou in particular is bad about reporting tunnel proportions). Excluding lines with unconfirmed information, we have 9,842.6 km of which 6,436.4 are in tunnel, or 65% – but China is 3,851/4,887, or 79%.

In the lowest-cost countries, els are not common. In Spain, 205.7 km out of 253.8 in our database are underground, or 81%. The Korean lines in our database are 100% underground, and as we add more data, this will hardly change. Overall, the countries I consider cheap have 927 km of rapid transit in the database, which number will rise as we add more Korean data, and of those, 730.1 are underground, a total of 79%. What’s more, one third of the non-underground length in cheap countries consists of a single 63 km item, tagged CR3, consisting of surface improvements for Marmaray (the tunnel is costed separately, as BC1); 63 km is hefty, but as a single item, it is less visible to unweighted correlation estimates like the regression.

So if els are uncommon in China and in cheap countries, where are they common? The answer is high-cost developing countries and Gulf states. India has 1,046.7 km in the database, of which only 235.8 are underground, or 23%; when I continue my series of posts on rapid transit traditions and get to India, I will of course mention the predominance of els. Moreover, these Indian els are spread across many items – there are 29 Indian items, since individual lines in Mumbai and phases in other cities each get their own lines, which matters for unweighted correlation estimates. Similarly, Thailand is 20% underground, Vietnam 50%, Pakistan’s single line 6%, Bangladesh 48%, the Philippines 55%, Malaysia 22%, Indonesia’s single line 38%, Panama 12%, Saudi Arabia 14%, the UAE 22% – and all of these are high-cost. In the developed world, the el-happiest country is Taiwan, only 40% underground in our database, and it’s on the expensive side, its average cost at 40% underground still amounting to $240 million/km, and its three all-underground lines averaging $375 million/km.

It makes sense when you think about it. If construction costs in a country are higher, then it will look for ways to cut costs by building less visually desirable els (typically in developing countries) or slower light rail lines (as in the United States). If we included at-grade light rail lines, then our table would also have a wealth of high-cost American lines; as it is, we’re likely to add some at-grade heavy rail lines like the Silver Line in Washington and, if it actually begins construction, the planned PATH extension to Newark Airport.

Country-internal averages

So instead of averaging in the entire database, let’s look internally to countries, chosen to be big enough to have a mix of projects with different underground proportions. I’m also going to ignore some cases where I worry about comparability – for example, in France, above-ground lines are represented mostly by a metro extension in Toulouse and by the most outlying parts of Grand Paris Express, and I worry about comparing those with Parisian and inner-suburban tunnels. The worst exclusion has to be that of China: while there is a wealth of data there, China built more els 15-20 years ago than it does now, so comparing subways to els in (say) Shanghai is to some extent a comparison of costs in the 2010s to costs in the late 1990s and early 2000s. In that, China is hardly different from the United States – New York built many els from the 19th century until the mid-1920s, but subsequently built an almost 100% underground system.

Japan

In Japan, we go back to the 1990s, so using dollar amounts does have inflation artifacts. Thankfully, the yen has had no inflation, so we can just plug in raw yen numbers and convert at the 2020 rate of 100:1. The 100% underground lines in Japan have averaged $382 million per km, the elevated ones $123 million/km; the ratio is 3.1. The regression estimate, again using ¥100 = $1 throughout, is cost = 149.8978 + 255.9496*tunnel-proportion; the ratio using this method is 2.7.

India

India has a single 100% underground line in the database, Line 3 in Mumbai, built for $449 million/km. The pure els in India cost $158 million/km, for a subway : el ratio of 2.8. Looking only at els in Mumbai, the average inches up to $167 million/km, a ratio of 2.7. Inflation adjustment would have marginal impact as all of these lines are recent, the earliest priced in 2011 terms. The regression estimate (for all of India, not just Mumbai) is cost = 151.6146 + 222.2716*tunnel-proportion, which yields a ratio of 2.5.

Taiwan

As mentioned above, Taiwan’s three pure subways average $375 million/km. But as a note of caution, they are all regional rail tunnels, and we know from evidence in countries that build 100% underground metros and regional rail tunnels (Finland, Sweden, France, Britain, Germany…) that the latter are more expensive.

With that caveat, the four pure above-ground lines in Taiwan average $170 million/km, a ratio of 2.2. The regression estimate is cost = 183.3252 + 163.0895*tunnel-proportion, a ratio of 1.9. This is a lower ratio than in India and Japan, despite the caveat; the reason could be that the underground lines in the dataset are in Kaohsiung, Taoyuan, and Tainan, whereas the lines in Taipei and New Taipei are elevated, as the database so far does not include the older Taipei MRT lines with their city-center tunnels.

Thailand

There are no pure subways in Thailand; even the underground MRT’s extension is only 20% underground. However, the under-construction Orange Line is 75% underground, and costs $531 million/km. Overall, the regression estimate is 155.9491 + 350.2821*tunnel-proportion, which includes a number of lines in Bangkok and a cheaper half-underground line in Chiang Mai. This is a ratio of 3.2; excluding the one Chiang Mai line, this rises to 3.9.

Conclusion

Our database is consistent with the observation in the literature that the subway : el cost ratio is about 2-2.5. But a crude averaging of global costs would lead to an underestimate, since higher-cost countries are more likely to be building els. This is partly coincidence – former colonies in the developing world tend to have high costs and also wide throughfares where els are more politically acceptable – and partly the use of els to reduce costs where the country’s ability to afford subways is limited.

This reinforces the need to look at other treatments for reducing costs more carefully. It’s plausible that some policy treatments are not found in low-cost countries because those treatments are undesirable for some reason but do reduce costs. Thus, it is critical to look at both the best industry practices and the variation in practices within the parts of the world one considers best.

Transit Costs Website

Go here to see the our construction costs website. The static dataset is here, but I encourage people to go to the site, which has some interesting mapping – in particular, because the coverage is close to comprehensive, it is easier to see where many subways are being built (China!) and where they are not.

There are still gaps in coverage, plus some numbers that I am not perfectly certain about because the projects are still under construction. Please email us if you have corrections or additional data, whether it’s current or historic. For example, I wish I had complete historical data for Paris, Berlin, and Tokyo – in all three cities I have current data, and in the first two I also have early 20th century costs, but I don’t know what the postwar costs were, or the 1930s costs in Berlin. (In London and New York I have better though still imperfect historical costs, they’re just not integrated into the site yet.)

And please thank everyone who has worked on this. The lines in the database that I added are not even a plurality of the database – the Chinese data comes from Yinan Yao, the Arab data comes from Anan Maalouf, we’re adding massive amounts of current and historic Korean data due to Abdirashid Dahir, Marco Chitti has added some Italian data, Eric has been invaluable in checking some of the Spanish-language numbers, and the Turkish data comes from Elif Ensari, who also built the website and is responsible for the data visualization and mapping.

Case Selection

Eric and I recently sent in a list of criteria for case selection. We’re currently funded for 6 detailed case studies, of which one is the Green Line Extension in Boston due to funding from a different grant. My guess is that we need about 15-20 different cities to have near-perfect information about the institutional and geographic factors that influence infrastructure construction costs. Because different subway lines in the same city tend to cost the same to build, and even in the same country, our 500 lines in the database are more like 50 independent observations, and there are even identifiable clusters of countries.

These clusters are important, because ideally we should have 2 cases per cluster. With 6 cases in total, we’d like to have a case for at least one per cluster, even though it’s unlikely, depending on where we can find the most detailed information and the most people who will talk to us.

Clusters

1. Very low-cost countries

The first cluster is the success cases. These really come in two flavors: one is Switzerland and the Nordic countries, and the other is everywhere else with costs lower than $150 million per km, that is Spain, Portugal, Italy, Greece, Bulgaria, Turkey, and South Korea. The difference between the two flavors is that the first one consists of very high-wage countries with populations that trust their institutions, and the second consistent of countries with wages at the bottom of the first world or top of the second with populations who don’t believe me when I tell them their infrastructure construction is cheaper than in Germany. Even then, there are some important differences – for example, contracts in Turkey are lowest-bid, using the country’s high rate of construction and multitude of firms (a contract must have a minimum of 3 bids) to discipline contractors into behaving, whereas Spain instead has technical scoring for bids and only assigns 30% weight to cost.

2. Middle-range countries

This is countries close to the global average, which is around $250 million per kilometer for underground construction. China has about the same average cost as the rest of the world, and since a slight majority of our current database is Chinese, it falls in this category. France and Germany are definitely in this category; Austria, Czechia, and Romania are also in this category but have fewer distinct metro tunnels; Japan may be in this category but it’s unclear, since the few tunnels it’s building nowadays are both more expensive and more uniquely complicated, rather like regional rail. Big parts of Latin America fall into this category too, though they bleed with the high-cost category too. There’s a good case for separating China, France, Germany, and Japan into four separate categories (Austria should probably be institutionally similar to Germany), each of which gets different things right and wrong.

3. Countries with recent cost growth

This cluster consists of places that have high costs but didn’t until recently. Canada and Singapore are both competing for worst construction costs outside the United States but were not until well into the 2000s. Australia may be in this category too – it’s unclear, since Melbourne is extremely expensive to tunnel in but Sydney isn’t. New Zealand’s regional rail costs suggest it might be too – initial electrification was cheap but the regional rail tunnel is expensive. All of these countries share the characteristic of extreme cultural cringe toward Britain and the US, adopting recent British and American ideas of privatization of the state, and it would be valuable to follow up and see if this is indeed what happened with all of their infrastructure programs.

4. Rich countries with very high costs

This cluster is dominated by the US and UK. Taiwan is there too but is much smaller and likely has completely different institutional reasons – one person told me of political corruption. Hungary and Russia might be in this category too – they have very high costs (Budapest is scratching $500 million per km), but their wages are at the first/second world boundary, rather like Bulgaria or Turkey.

5. Countries on the global periphery with very high costs

This cluster consists of the high-cost world that is too poor or peripheral to be in cluster 4. This includes ex-colonies like India, Pakistan, Indonesia, Egypt, and Vietnam, but also the never- or more-or-less-never-colonized Gulf states; these two categories, the Gulf and the rest, must form two distinct flavors, but I lump them together because both seem to have extreme levels of cultural cringe and to associate bringing in European and East Asian consultants with modernity and success. (Meanwhile, parts of Europe, at least in the less self-assured East, bring in Turkish contractors.) The higher-cost Latin American countries, like Brazil and possibly Colombia, belong here too, and may form a distinct flavor. Thailand is on the edge between this cluster and cluster 2, which may befit its liminal colonial status before and during World War 2.

Where we struggle

We’ve been sending feeler messages to people in a number of places. This is far from perfect coverage – so far none of these countries is poorer than Turkey. In general, we’ve had early success in the lower-income range in cluster 1 (Italy, Spain, Korea, Turkey) and in cluster 4. Cluster 3 seems reachable too, especially since Stephen Wickens did much of the legwork for Toronto’s cost growth; we may be able to look at Sydney as well, and Singapore and Auckland seem like it shouldn’t be too difficult to find sources, nor to get people to listen if our conclusion ends up being “your government reforms in the last 15 years are terrible and should be reversed.”

Within the rich world, so far getting sources in Germany and Scandinavia has proved the hardest. I don’t know if it’s random or if it’s the fact that in countries that believe their standards of living are higher than those of the US and UK people are less likely to be forthcoming to someone who writes them in English. I’ve seen a decent amount of written material about rail capital construction projects in Germany, though not about the one I’m most interested in, that is the U5-U55 connection here in Berlin; but the rail advocates I’ve talked to are not quite in metro construction, though I have learned a lot about public transportation issues in Germany from them.

In Scandinavia things are even harder. Costs there seem pretty consistently low. A common explanation is that the rock in both Stockholm and Helsinki is gneiss, which forms a natural arch and makes tunnel boring easy, but a short tunnel in Oslo, the Løren Line, was even cheaper in softer rock. Moreover, the planned Helsinki-Turku high-speed rail is currently budgeted at €2 billion for 94 km of which 10 are in tunnel, so maybe equivalent to 140 km of at-grade line; this is noticeably below French costs, let alone German ones.

The low-income world is an entirely different situation. My suspicion is that the same cultural cringe that makes India build turnkey Shinkansen at something like 3 times its domestic cost (correcting for tunnel length) would make India eager to talk to us – if we were covered in the first-world discourse first. People in India, Nigeria, etc. know their countries are poor and are desperate to absorb the knowledge of richer places; they don’t understand the US as well as Americans do, but they understand it better than Americans understand the third world.

Cluster curiosities

The reasons I’d ideally like to have 20 case studies are that there are a lot of questions about internal differences, and that things that look like clusters from cost data may not actually be similar. There are a lot of questions that doing more cases might explain.

  • South Korea and Japan share many institutional similarities, and many of those are also shared with Taiwan. How come South Korea near-ties for lowest costs in the world, Taiwan near-ties for highest costs in the non-Anglophone first world, and Japan is somewhere in the middle?
  • What explains why different Eastern European countries with similar histories and institutions have such cost divergence?
  • Why does Italy have low metro construction costs (more in the North than in Rome and the South, but Rome is at worst average) and high costs of high-speed rail construction?
  • Why does Japan have high metro construction costs where it builds and low costs of Shinkansen tunneling?
  • Turkey seems similar in costs to Southern Europe, but it does things very differently – for one, it uses lowest-bid contracting. To what extent this is about Turkey’s very high rates of construction recently, and does this generalize elsewhere? Of note, there are extremely high construction rates all over middle-cost China, and also decently high rates in high-cost India, Singapore, and California.
  • The Netherlands is institutionally within the same range of what’s seen elsewhere in Northern Europe, and yet its construction costs are high. Is this just a matter of alluvial soil tunneling? If so, why did HSL Zuid cost so much?

The US Leads the Way in High Costs

Our current project timeline includes posting the dataset of urban rail lines and their construction costs in a month. This means looking at various spreadsheets and checking them item by item. Part of it is checking for mistakes, which do unfortunately occur for some items. Sometimes even the sources have mistakes – for example, most sources for the Sinbundang Line in Seoul say it cost 1.169 trillion won (e.g. here, a bit higher in PDF-p. 60 here, and my now-linkrotted original source), but one says 1.69 trillion, which I’m fairly certain is a typo. However, the biggest source of errors in my file is that the majority of lines I included were under construction as of 2018, so cost overruns and schedule slips remain possible. And unfortunately, while a number of projects have significantly higher costs, the US is especially rich in cost overruns.

The case of Los Angeles is the most infuriating. It is not the highest-cost American city, not even close – nothing is within a factor of 2 of dislodging Second Avenue Subway Phase 2 from its throne. However, it’s making a strong bid for the second highest. The third phase of the Purple Line extension in the Westside, connecting Century City (reached in the second phase) with UCLA and the VA Hospital in Westwood, is $3.6 billion for 4.2 km. Costs have been creeping up from what used to start with a 2, and now this is $857 million per kilometer. This is in year-of-expenditure dollars, so in 2020 money it’s more like $800 million per km.

The contrast to what LA looked like in the 2000s is huge. In 2010-11, it looked like the lowest-cost US city; it was still really expensive to tunnel in, but it seemed more like $300-400 million per km. But things keep getting worse. If Canada and Australia and Singapore and Britain today are like the US 10-15 years ago, the US is pulling ahead, eager to be #1 in everything.

Of note, this is an environment with high and stable funding levels. Transit funding in Los Angeles is bonded through 2060. Contracts in Los Angeles are let on a lowest-cost basis; sometimes there’s a technical score, but officials at LA Metro told Eric and me that unless the weight of the technical score is very high, around 70%, then in practice the contract will go to the lowest bidder. Now, it is not true that all low-cost countries have high technical score weights like Spain does; Turkey in particular uses lowest cost, and uses its high construction rates to discipline bidders into behaving, since shoddy work will risk their ability to get future contracts. Nonetheless, in Los Angeles the great extent of construction does not involve any such discipline. Metro prefers dealing with familiar contractors, even if their record is poor.

Americans, as a culture, would rather die than be more like another nation. Taiwan’s last domestic corona infection was on April 12th, the US averages 60,000 such infections a day. The sort of change required to make Americans forget about 2 generations of learned public-sector helplessness is immense, and will not come spontaneously (and no, your chosen revolutionary movement won’t do it – revolutionaries are selected for incompetence).

The upshot is that the share of current senior managers who have anything to contribute to improving public transportation in the US is very low. Not zero, but still very low. The process selects the least imaginative, least technically apt, and least curious people. Whether it’s best practices that do not look outside the Boston-Seattle-San Diego-Miami quadrilateral, or grants that have language that makes it clear foreign knowledge is unwelcome, or hiring practices that exclude immigrants on visas, everything about the process in the US screams it. It’s not a coincidence that the US has the world’s highest construction costs, and when other countries begin to catch up often thanks to adoption of American practices, the US keeps staying ahead.

Construction Costs in the Arab World

The construction costs of rail infrastructure in the Arab world are globally atypical. We looked at the entire region, because of the common use of the literary Arab language; nearly all of this work is due to a New School student named Anan Maalouf, who’s doing long-term work on urban planning as relevant to Nazareth. Here is a presentation he gave at NYU on the subject last monthUpdate 7/22: here is an updated version of the presentation.

There are identifiable clusters in the Arab world, which is not surprising – it’s similar to how there is a common Nordic cost (which is low), a common cost to the English-speaking world (which is high), and so on. Of course, these clusters are not perfectly predictable ex ante; in light of the most important global pattern with the coronavirus crisis, I keep stressing that there is no distinct Europe vs. East Asia cluster when it comes to costs, and instead both regions have similar averages and huge internal variations. The Arab world does not form an entire cluster itself, but its clusters are at least somewhat understandable based on internal divisions.

One cluster is the six states of the Gulf Cooperation Council: Kuwait, Saudi Arabia, Qatar, Bahrain, the UAE, and Oman. All are distinguished by high incomes, comparable to those of the developed world, but coming almost exclusively from oil extraction. All also have atypically large numbers of immigrants, who form large majorities of the populations of the UAE and Qatar, and who have few rights and earn very low wages by local standards. One might expect that in such an environment, construction costs should be low, since there is ample cheap labor but also money for imported capital. Instead, these states all have high costs; for example, the Doha metro project costs around $700 million per kilometer, and is not even 100% underground but only 90%.

The explanation, per Anan and an Israeli-British planner named Omer Raz, is that there is no interest in cost control in the Gulf region. The GCC states have money. They are buying prestige and the trappings of modernity; for all of their crowing about the superiority of their traditional values and Islamic law, they crave Western acceptance, in similar vein to Singapore. So they invite first-world consultancies to build their infrastructure to build what Aaron Renn would call “world-class in Doha” (or Dubai, or Riyadh, etc.), as opposed to “world-class Doha,” i.e. domestic production that is good enough that other people are attracted to it. On top of it all, Omer gave an example in which Saudi Arabia was not a reliable partner for these foreign consultancies; Anan, too, notes a plethora of postponements and cancellations of rail lines, sometimes because of changing economic conditions, sometimes because these lines are international and relationships between Saudi Arabia and Qatar deteriorated recently, etc.

Another cluster consists of Egypt and Iraq. Both have high costs, in line with other third-world ex-colonies, like India, Vietnam, Indonesia, and Nigeria. The Cairo Metro extensions are $600-700 million per km; the Baghdad Metro is, in PPP terms, $330 million per km for an all-elevated project. This is not surprising – these countries use first-world consultancies with background in high-wage, strong-currency, cheap-capital economies. Unlike in the other datasets, like mine or Yinan Yao’s, Anan included a crucial piece of information: who the lead contractor or consultant was. It’s often a foreign firm, from a much richer country – in Iraq’s case, it’s Alstom. On the other hand, Egypt is using a domestic contractor, Orascom, and costs there remain high.

Finally and most interestingly, there is the Maghreb. One would expect that Tunisia, Algeria, and Morocco should have high costs, just like the other ex-colonies. However, they do not. Anan pointed out that the Arab world inverts my theory about how ex-colonies have higher costs than never-colonized countries like Iran, Turkey, and China. He adds that these countries have much closer ties with France than other ex-colonies do, whether they used to be French outside Africa (i.e. Vietnam) or other European empires’ (e.g. India, Indonesia, Nigeria). Alstom has had continuous presence in the area for 20 years.

In a sense, France didn’t fully decolonize in the Maghreb and the Sahel. It still views these regions as its near abroad, with a forever war in Mali, currency pegs, and deep economic ties with the higher-functioning countries. One can even see the French way of building urban rail in the Maghreb, for example on the Sfax tramway. This isn’t quite every urban subway – the Algiers Metro is pretty expensive. But the Oran Metro has normal costs, and light rail systems like those of Sfax and Casablanca have reasonable costs, as does the TGV system running as Morocco’s high-speed rail system.

So perhaps the issue is that the French planners in the Maghreb are there for long enough that they know the local conditions, and build in accordance with them. In contrast, systems have higher costs if they try to imitate first-world methods either due to first-world consultants’ unfamiliarity with the local situation or due to local cultural cringe.

Construction Costs, Inflation, and Developing Countries

As our construction cost project moves forward, we are expanding our database to be as complete as possible. My original dataset is mostly in developed countries, but does have decent coverage in developing ones other than China. However, decent and good are two very different things, and expanding coverage showcases some problems. These are all resolvable, but they require some delicate care.

When I wrote about Yinan Yao’s work on construction costs in China, I mentioned we would expand to more parts of the world. We have a mostly complete table for the Arab world, thanks to the work of Anan Maalouf, and a growing table with exceptional detail thanks to the work of Elif Ensari. I’m going to give each a complete post fairly soon, later this month or in July, since they both have insights that have seriously challenged the way we have to think about costs. But for now, I want to focus on one cross-national issue: inflation, and generally currency conversion rates.

The best example of this is actually not in either the Arab world or Turkey, but in Iran. I have three Iranian projects in my dataset: the extension of Line 3, and Lines 6 and 7. They are fully underground and cluster around PPP$200 million per kilometer, which is slightly lower than the global median and roughly in line with the global median excluding the English-speaking world. The problem is figuring out what conversion rate to use. Line 3 cost 20 trillion rial and was built between 2012 and 2014. But what year should we deflate costs to? Iran had 30% annual inflation in that period, and after a brief lull of 10% inflation went up to 40% last year. A one-year error in the PPP conversion rate can lead to sizable errors in the final costs.

Moreover, high inflation leads to nominal cost overruns if it is higher than expected or if the project takes longer than expected. These nominal overruns can lead to real problems if there are contract disputes or a budget crisis, and usually if your inflation rate is 30% then your budget is in perpetual crisis mode. Check the source above for the costs of Lines 6 and 7: it mentions nominal overruns, disputes, and schedule slips.

The OECD has PPP conversion rates for different countries by year, going back to 2000. If the numbers increase over time, it indicates the country in question has more inflation than the United States; if they decrease, it indicates the opposite. For example, in Japan, the real value of the yen has strengthened from 154.718 to the dollar in 2000 to 101.474 in 2019, in line with Japan’s lack of inflation – in fact, it’s had slight net deflation. The eurozone has had positive inflation but less so than the United States, so the real value of the euro has increased from $1.159 to $1.416. These relative changes are significant in looking at the long-term evolution of costs, but they’re gradual, so over a 5-year construction period, they’re not too important.

In contrast, on the same table, we can look at Turkey. Between 2000 and 2019, the lira’s real value weakened from 0.282 to the dollar to 1.841. This is about 10% annual inflation above the US rate, so maybe 12% a year; moreover, this is an average of relatively moderate inflation in 2005-2015 and high inflation before and after. Getting the exact conversion correct is important, and evidently the data table I uploaded in November got one Turkish project wrong, making the spread in costs between different lines look larger than it really was.

But this is about more than just picking the correct year. The standard way to compare projects’ real costs is to deflate to the midpoint of construction. It’s an approximation that works when inflation rates are low – and for the purpose of this discussion, 5% over the course of a 6-year subway timeline is low; I am not making a macroeconomic claim about long-term price stability, but an econometric claim about measurement errors. However, when inflation is high, especially at the Iranian rates rather than the Turkish ones, we have to be more precise.

More precise here means having some idea when most of the money was spent. Was it spent evenly over the construction timeline? If so, then the midpoint is not a bad approximation at 10-15% inflation, and even at 20-30% inflation it is not terrible if the construction timeline is short, which it was for the Line 3 extension in Tehran. However, the money is not always spent at a uniform rate. Maybe there is a long preliminary engineering process followed by a quick construction period toward the end, or maybe most of the construction is done early and then thee timeline drags for the final elements like systems or testing or one particularly hard segment.

This introduces a new element – keeping track of how much money was spent in each year – that I didn’t do much before when I was only looking at first-world countries. About the only first-world projects for which I care much about the timeline are ones that have become legendary for how long they took, like Rome Metro Line C and Barcelona Metro Lines 9 and 10.

The broader point here is that it is often difficult to adapt knowledge from one context to another. The context in which I began looking at construction costs was that of New York during the construction of Second Avenue Subway, so I was focusing on fully underground lines in the centers of large first-world cities. I’ve since adapted it to a more global context, and in some cases it’s worked fine (e.g. smaller cities), but it’s critical to keep track of when new complications arise.

I wrote this thread a few days ago about third-world construction costs, and there I pointed out that it’s critical to analyze third-world issues in terms of what is relevant to the third world. Global consultancies (and here I’m including Japanese and European governmental organizations focusing on international development, and not just private consultancies) don’t often do this right – their money comes from the first world, so they think about how to be efficient in the first world. This is also relevant to us – our money comes from the US. But it’s critical to take developing-country factors into account nonetheless.

Predictions

I have a lot of readers who come from a rationalist or Effective Altruism background, and some more who come from an economics background, and both communities put a lot of stock in the idea of correct predictions about current events. The idea is that scientists have to make testable predictions about the results of their experiments, and therefore social scientists must equally make predictions about the state of the world. It’s become relevant in the corona crisis and is also relevant to my and Eric Goldwyn’s construction cost project in a specific way, so I’d like to talk about the complexities of what it exactly means to get things right.

Recession

Consider the following prediction: the economy is overheated and a recession will come soon. It’s a vague prediction. One can fill in details to make it strictly testable – “the German economy will have >6% unemployment in 2 years” – but what exactly is the point of one detail or another?

The real answer is that different classes of people have different uses for the prediction of recession, and therefore depend on different details. The investor wants to sell stocks near the peak. The Nasdaq went from 2,200 at the beginning of 1999 to a peak of 5,100. To the investor, knowing that there was a bubble at the beginning of 1999 would not have been useful – cashing out then would have meant missing on a stock market doubling over the year. It would take until about the onset of the 2001 recession for the Nasdaq to fall below January 1999 levels. To the successful investor, it is critical to know the exact timing of the peak to maximize income, and in pursuit of that goal it’s fine to miss some recessions, let alone to miss other important details like the length of the recession and the unemployment rate.

In contrast with the investor, the skilled worker has different concerns, like unemployment. In that environment, knowing that there’s going to be a recession is useful even if the timing is vague – such a worker can save more money, delay major purchases, avoid quitting a stable salaried job to start a small business, and maybe shift to a more recession-proof job even if it means taking a pay cut. Knowing how deep the recession will be is important as well, and remains important knowledge even as the recession takes place – the worker needs to know how stressed to be about savings running out if there is prolonged unemployment. All of this is equally valuable to the prospective immigrant who needs to make a decision on whether to emigrate.

The investor-worker duality is especially important for economists, and to some extent to rationalists who try to follow popular economists. They have money to invest, and often work as advisors to finance firms that pay them for investor-relevant information. But they are also researchers, who can respond to an impending recession by acquiring recession-relevant skills, like studying the history of depressions and conducting empirical research about unemployment and anti-poverty interventions. These are such big research programs that the exact timing of the recession doesn’t really matter, whereas its depth and length matter. An economist who can answer questions like “what is the impact of unemployment programs on long-term welfare?” is useful in a general period of economic weakness even if the papers appear a year into the beginning of the recession.

Predictions and construction costs

Before we started our current project, I had been writing about construction costs here, in comments, and on social media going back to 2009-10. I had some theories over the years, of which some would be confirmed by additional data and others wouldn’t:

  • The theory that common law leads to higher costs, based on high costs across the US, Singapore, the UK, Australia, Canada, India, and Bangladesh. I no longer believe this theory holds up; in the developed world, important edge cases disagree with the theory, including Quebec (expensive) and Israel (about average), and moreover Canadian and Singaporean costs only exploded in the last 15 years.
  • The theory that costs are consistent across projects in the same country, especially the same city; I’m pretty sure I brought it up even in the early 2010s, when I was saying Chinese costs seemed pretty average to me, but the starkest formulation is from 2019. This has subsequently been confirmed when thanks to Yinan Yao our knowledge of Chinese costs grew from two lines in Shanghai to more than 5,000 kilometers’ worth of lines across all major Chinese cities.
  • The theory that costs in developing countries are higher in ex-colonies than in never-colonized countries (like China and Iran) and distantly-colonized ones (like all of Latin America). As stated, there are counterexamples: I will report on our ongoing research into Arab construction costs, thanks to Anan Maalouf, but so far this is indicating that costs in never-colonized Saudi Arabia are pretty high. Call it half a correct prediction because Saudi Arabia is atypical enough I would not lump it a priori with China, Turkey, Mexico, or Iran.

With all that said, I am not too worried if my theories aren’t all confirmed by finding additional data. The reason is that this is not an experimental science but an observational one with a small, finite amount of data, so it’s much more important to have coherent mechanisms that can lead to actionable changes than to be able to predict every country’s construction costs from partial data.

In this case, the mechanisms posited in the 1.5 theories that do not stand up to additional data seem useful. The colonial theory is that high cultural cringe levels and weak state capacity lead ex-colonies to privatize planning to first-world (or Chinese) consultants, who use methods that are not appropriate for local conditions. On account of that explanation, I kept saying ex ante that I refused to make a prediction regarding Thailand, because it was never colonized but also has much more cultural cringe than China and uses first-world consultants; Thai costs are higher than Chinese ones but lower than ex-colonial ones. Saudi Arabia is similar – for all its bluster about rejecting Western governance norms, it craves first-world acceptance and the trappings of modernity, and extensively uses contractors from more developed countries. So the upshot regarding the importance of domestic state capacity and methods tailored for local urban geography and wages remains useful.

Likewise, the high costs across the Anglosphere remain a useful fact. Even more useful is the history of Singapore and Canada, which only aligned with British and American costs starting in the 2000s. The cost explosion in Singapore, Montreal, Toronto, and to some extent Calgary and Vancouver is a recent event, in accessible English-speaking cities; Stephen Wickens just wrote a long report about the Canadian cost explosion, which is of value in teasing out what happened. Even better, the persistent low costs in Scandinavia, Southern Europe, and South Korea provide ready-made sanity checks in knowing what to look for.

Timing issues

In one sense, I made a critical error that poses a serious threat to the project: I got the timing of the recession wrong. When applying for this grant throughout 2019, my assumption was that the American economy was overheated and would soon experience a demand-side recession, leading to stimulus – but that the contraction would be slow enough that the stimulus would come in 2021. With a jobs program announced in 2021, preliminary versions of our report would already be out, the full report with detailed case studies would be out later that year in time for agencies to request funding, and there would be enough time for agencies to implement our recommendations by the time of actual construction.

This may still happen, but the timeline is much less certain. People are talking about stimulus with infrastructure money now. I can promise a report with some actionable recommendations in 2021, but I can’t promise what costs I can promise, nor can I promise what investment to focus on. Our report centers on metro tunnels, but if there’s another push for high-speed rail then we’ll need to be able to adapt metro-based recommendations to a somewhat different context, in which high American costs may have different roots.

What’s more, based on what everyone knows in the United States, costs are so high there’s no point in planning for more. Maybe New York thinks it can finagle $40 billion in stimulus money; this can do a lot at Nordic costs, but unless New York thinks right now that this is possible, it won’t even try to plan more than a few lines like Second Avenue Subway Phase 2 and Gateway, each costing more than a full order of magnitude more than it would in Scandinavia or Southern Europe.

I am not that worried in the long run. There is ongoing investment in enough of the US for whatever we come up with to be relevant to at least some extent. And here too, a cost comparison with the cheaper parts of Europe would be instructive to many a German rail advocate or civil servant. I don’t expect to be in the situation of an investor who bet everything on a company that went bankrupt, just perhaps in that of one who missed a big stock market rally. Ultimately, don’t worry about me, worry about the virus this year and the unemployment rate of potentially the entire world in the next few years.