Policy Research Working Paper 10365 Exciting, Boring, and Nonexistent Skylines Vertical Building Gaps in Global Perspective Jason Barr Remi Jedwab Urban, Disaster Risk Management, Resilience and Land Global Practice March 2023 Policy Research Working Paper 10365 Abstract Despite the widespread prevalence and economic impor- prominent than they really are once all the tall buildings tance of tall buildings, little is known about how their and core controls are included, which alters how regions patterns vary across space and time. This paper focuses on are ranked in terms of tall building stocks. Using results by vertical real estate, aiming to quantify differences across city size, centrality, height of buildings, and building func- major world regions over time (1950–2020). The paper tion, the paper classifies world regions into different groups, exploits a novel database on the location, height (above finding that international tall building stocks are mostly 55 meters), and year of construction of nearly all the tall driven by boring skylines of residential high-rises, and to a buildings in the world. It proposes a new methodology to lesser extent exciting skylines of skyscrapers and supertall estimate the extent to which some world regions build up office towers. Finally, land use regulations and preferences, more than others given similar economic and geographic not historical preservation nor dispersed ownership, likely conditions, city size distributions, and other features. The account for most of the observed differences. analyses reveal that many skylines may visually appear more This paper is a product of the Urban, Disaster Risk Management, Resilience and Land Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at rjedwab@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Exciting, Boring, and Nonexistent Skylines: Vertical Building Gaps in Global Perspective * Jason Barr (Rutgers-Newark) Remi Jedwab (GWU & NYU) JEL Codes: R14; R30; R38; R31; R33 Key words: Buildings Heights; Skyscrapers; Global Real Estate; Housing Supply * Corresponding author: Remi Jedwab: Department of Economics, George Washington University, and Fellow, NYU Marron Institute of Urban Management. Email: jedwab@gwu.edu and rj2513@nyu.edu. Jason Barr: Department of Economics, Rutgers University-Newark. Email: jmbarr@newark.rutgers.edu. We thank Oliver Harman, Megha Mukim, Mark Roberts, Tony Venables and seminar audiences at Oxford University (Blavatnik School of Government-IGC) and the World Bank. We gratefully acknowledge research support provided by Federico Haslop. Remi Jedwab gratefully acknowledges financial support from the International Growth Center (XXX-2018; “Gauging the Impact of Land-Use Regulation in Developing Countries”) and The World Bank (Name of unit: Office of the SD Chief Economist; Urban, DRM, Resilience and Land Global Practice – Office of Global Director Sustainable Development Practice Group. Project: Flagship Report on Making Cities Green, Resilient and Inclusive (P177249). TTLs: Mark Roberts and Megha Mukim). The findings, interpretations and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. 1 1. Introduction Humanity has experienced dramatic urbanization in the past century (World Bank, 2009; U.N., 2020). Today, more than half of the world’s population lives in cities. Alongside this urbanization has been the rise of tall buildings. Dramatic skylines in Chicago; New York; Hong Kong SAR, China; and Dubai, for example, are visible manifestations of each city’s growth. Despite the increasing role of tall buildings for both local and national economies (Ahlfeldt et al., 2022), little is known about how much tall building stocks vary globally and what is driving international differences (Ahlfeldt and Barr, 2022). This is true despite the fact that real estate is one the largest asset classes in the world. Given the durable nature of real estate, policies and preference regarding different types of structures can have long-run effects on the well-being of cities and their residents (Jedwab et al., 2020). The responsiveness of building stock growth to prices, employment, or population changes can have far-reaching impacts, affecting housing affordability, productivity, traffic congestion, and air quality, to name a few examples. While cities can physically grow via land expansion–by building “out”–they can also grow by becoming more vertical–by building “up.” If anything, we expect economic development to be associated with verticalization (Jedwab et al., 2020). This paper adds to our understanding of why there is so much global variation in tall building stocks. To do so, we analyze tall building completions using novel data from Emporis–a global provider of tall buildings data–on the location (city), year of construction, heights, and uses of nearly all the tall buildings in the world. For this study, we include all completed, occupiable structures above 55 meters (about 14 floors). Guided by predictions from the Standard Urban Model (SUM), we perform country- level econometric analyses (for 163 countries every five years from 1950 to 2020) and city- level econometric analyses (for 12,877 cities every five years from 1975 to 2020) in order to show that: (i) SUM-related variables explain a large fraction of the global variation in heights; but (ii) considerable differences across the world remain after controlling for city population sizes, income, and supply factors such as geographical constraints. We thus delve deeper into the nature of the regression residuals to better understand the patterns 2 of global tall building construction. To simplify the analysis of the residuals, we perform regional comparisons where we group countries by geographic areas with similar income levels and/or cultures. More precisely, we use the 19 United Nations (U.N.) subregions. Regions with positive residuals “over-build” relative to their demand and supply conditions; and the residuals give clues on the nature of frictions or preferences regarding tall buildings. We then conduct the same residuals analyses for large vs. small cities, central vs. peripheral city areas, high-rises between 55 and 100 meters vs. skyscrapers above 100 meters, and residential vs. commercial buildings. We classify the subregions into four groups–population-oriented central districts (CDs), capital-oriented CDs, population- oriented peripheral districts (PDs), and capital-oriented PDs–depending on whether the subregion’s skyline is specialized into central districts (large cities and/or central city areas) or peripheral districts (small cities and/or peripheral areas), and is serving people (residential high-rises) or capital (office towers and/or luxury residential skyscrapers). While cities like New York, Shanghai, and Tokyo have impressive skylines, they also have tremendous populations. Likewise, some skyline-oriented cities, such as Dubai; Hong Kong SAR, China; and Singapore, are some of the wealthiest in the world. Thus, on a per capita income basis, they might not be so impressive, hence the need for controls. Based on regression/residual analysis, the main takeaways are as follows. Compared to what the SUM would predict, (i) Latin America, Eastern Europe, and Western Europe are highly ranked, ceteris paribus. Asian regions and North America are not that well ranked; (ii) Within Asia, the leading region, ceteris paribus, is Southeast Asia, not East Asia. West Asia, which includes the Gulf nations, is not that well-ranked; and (iii) Some African regions–particularly East Africa and Southern Africa–are relatively well-ranked. Grouping regions based on skyline types, we find that, the capital-oriented CDs and population-oriented PDs contain most regions in the world. However, with a few exceptions, the leading regions belong to the latter group. Hence, subregions with the largest tall building stocks, ceteris paribus, have skylines specialized in residential high- rises, i.e., flatter skylines. Subregions with capital-oriented CDs which rank well in terms of skyscrapers and/or office towers are thus not the world’s leaders. Our results 3 demonstrate a key point: the most impressive skylines are not necessarily the ones with the densest tall-building stocks. For example, regions like South America and Eastern Europe make up in high-rise volume what they lose in supertall skyscrapers. Thus, focusing on exciting skylines and famous skyscrapers is misleading. About 70% and 80% of tall building heights in the world come from high-rises below 100 meters (not skyscrapers) and residential towers (not office supertall towers), respectively. Many skylines may thus visually appear as more prominent internationally than they really are once one includes all tall buildings and core controls. Hence, boring skylines of residential high-rises may actually be what is driving variations in heights globally. Finally, based on a comparison of regional differences in vertical development, the evidence suggests that land-use regulations and residential preferences, not historical preservation nor dispersed ownership, likely account for most observed differences. The rest of this paper proceeds as follows. The next section discusses the literature. Section 3. describes the data and provides the descriptive patterns. Section 4. details the conclusions of the SUM that guides the analysis. Section 5. provides regression results that guide our step-wise “controlling for” analysis. In Section 6., we provide the results of our residuals/“gaps” analysis. Section 7. provides further analysis on the types and locations of construction around the world. Section 8. concludes. 2. Literature Review The economic causes and consequences of tall building construction have only recently begun to be explored in detail in the real estate and urban economics literatures. In particular, this paper is one of the first global studies on the economics of tall buildings and skyscrapers (see Ahlfeldt and Barr (2020) for a recent survey). In terms of country or regional analysis, most research on tall buildings to date has focused on cities within the United States, though there is a long tradition of work on capital-land substitution elasticities in central cities (e.g., Clapp, 1980; McDonald, 1981; Yoshida, 2016).1 A key finding of this research is that the skyscraper heights and frequencies reflect the demand and supply fundamentals for tall buildings over time and space (Barr, 2010, 2012, 2016; Ahlfeldt and McMillen, 2015). 1 Related recent work explores the evolution of floor area ratios (McMillen, 2006; Barr and Cohen, 2014). 4 Other papers have explored the drivers of tall building construction in China. All land is owned by the respective municipalities which sell long-term leases to developers (Peng and Thibodeau, 2012). Because local officials have so much say over what gets developed, this generates incentives for them to “overbuild.” Evidence to date suggests this is the case (Zheng et al., 2017; Barr and Luo, 2020; Li and Wang, 2020).2 More broadly, a few papers explore the role of height competition in driving “too tall” skyscrapers (Helsley and Strange, 2008). Evidence on height competition, however, is mixed, and is, statistically speaking, rare (Barr, 2010, 2012, 2013; Ahlfeldt and Barr, 2022). Recent work on the benefits or revenues from tall buildings shows rental or price premiums for amenities (e.g., views) or firm productivity signaling (Liu et al., 2018a; Koster et al., 2013; Nase et al., 2019). Liu et al. (2020) find evidence for agglomeration benefits on higher floors of tall office buildings. Glaeser (2012) and Ahlfeldt et al. (2022) document that tall buildings reduce housing prices and sprawl, respectively. Because of the increased costs of adding more floors, after some threshold height, the marginal cost of adding another floor increases at an increasing rate. Thus, the optimal height–the one that sets the marginal cost equal to the marginal revenue–depends on the state of technology and on total factor productivity in tall building construction. Work on the supply side has attempted to measure the costs of construction and supply functions (Picken and Ilozor, 2015; Coulson et al., 2021; Eriksen and Orlando, 2022). Additionally, there is a considerable body of work on supply elasticities (e.g., Glaeser et al., 2006; Saks, 2008), and how land-use regulations affect these elasticities (e.g., Glaeser et al., 2005b; Gyourko and Molloy, 2015; Hilber and Vermeulen, 2016). Cities differ in the geographical constraints. A bedrock that is too deep, or too close to the surface, raises construction costs, as indicated by Barr et al. (2011) for Downtown Manhattan. Other studies have used bedrock depth as instruments for density (Rosenthal and Strange, 2008; Ahlfeldt et al., 2022). Ahlfeldt et al. (2022) examines if there is also an inverted-U relationship between tall building heights and bedrock depth globally. Ahlfeldt et al. (2022) also shows that a higher earthquake risk reduces tall building construction, either because they increase foundation costs, or because high-risk cities 2 Garza and Lizieri (2016) find that fundamentals are good predictors of heights in Latin America. 5 adopt land-use regulations that restrict vertical development.3 There are only a handful of papers that conduct international comparisons of skyscraper construction. Ahlfeldt and Barr (2022) provide a general equilibrium SUM to understand the causes and consequences of skyscraper construction. Their results show that skyscraper patterns around the world are largely consistent with model predictions. To our knowledge, there are almost no studies on how vertical development patterns vary globally. Jedwab et al. (2021) use tall building data from the Council on Tall Buildings and Urban Habitat (CTBUH) to show that cities are more vertical with economic development, whereas we quantify to what extent a region is vertically developed for a given level of economic development. One limitation of the CTBUH data is that it only includes tall buildings above 80 m (21 floors). Also related to this paper is Jedwab et al. (2020), who use the imperfect CTBUH data to obtain construction gaps at the country level. However, they miss a significant share of the tall residential sector (buildings in the 55-80 meters range are overwhelmingly residential). Their methodology also differs from ours. They select “laissez-faire” countries for which they obtain the relationships between changes in heights and changes in income and agricultural land rent. They then apply the estimated relationships to other countries and obtain their gap by comparing their actual and predicted height stocks. Our approach is more direct and easier to implement at the city level. Also, by construction their approach mis-estimates the gaps at low income levels, since a gap can only appear when income increases (and heights increase slower). Our method has the advantage of being simple. So far, only a few studies have implemented this type of indirect method to measure the stringency of regulation. However, while we believe that land-use regulation is a major determinant of the “gaps” that we estimate, other factors possibly contribute to the gaps, such as dispersed ownership and societal preferences for historical preservation and suburban living.4 Finally, there are a few studies on how urban expansion patterns vary across countries/regions over time. But most of these studies have focused on horizontal land 3 Other research, however, shows the “benefits” from earthquakes or other natural disasters in that the rebuilding can allow for more efficient land uses post disaster (Siodla, 2015; Fu and Shi, 2022). 4 See Bertaud and Malpezzi (2001), Glaeser et al. (2005a), Hilber and Robert-Nicoud (2013), Brueckner et al. (2017), and Brueckner and Singh (2018) for important works on land-use regulation. 6 expansion instead of vertical development, such as Angel et al. (2005), Angel et al. (2012), AUE (2016) and Mahtta et al. (2019). These studies all show how city land areas increase with income but do not characterize “gaps” across countries/regions over time.5 3. Data and Patterns 3.1. Data We perform the analysis at both the country and city levels. At the country level, our sample includes 183 countries x 15 years every five years from 1950 to 2020. At the city level, our sample includes 12,877 cities x 10 years every five years from 1975 to 2020. Sample of Cities. Using the Global Human Settlements-Urban Centre Database (GHS-UCDB) of Florczyk et al. (2019, version v1.2), we obtain the GIS boundaries of all (12,877) 50K+ agglomerations today, which they call “urban centres” (UCs). The UCs correspond to commuting zones (e.g., metropolitan statistical areas in the U.S.).6 GHS-UCDB reports population estimates for each city ca. 1975, 1990, 2000, and 2015. The 12,877 cities account for 90% of the world’s total urban population ca. 2015 (United Nations, 2018). Building Heights. Emporis (2022) (last accessed 02-07-2022) is a global provider of “international skyscraper and high-rise building data.”7 They rely on information provided by the industry for “thousands of cities worldwide. Emporis collects information about the full life-cycle of each building, from idea to demolition.” More precisely, the database contains data for 693,855 “existing [completed]” buildings.8 As can be seen in Web Appx. Fig. A.1 which plots the Kernel distribution of building heights in the data set, the mode of the distribution is 55 m (about 14 floors). Since cities are likely to have more buildings below 55 m than above 55 m, and since the distribution of buildings is relatively smooth after 55 m, this suggests that the data set mostly captures buildings above 55 m. Thus, the data set is likely unreliable for buildings below 55 m. Restricting the sample to buildings above 55 m, we obtain 266,336 completed buildings. 5 Studies on land expansion are constrained in their analysis by the fact that they rely on a few years of satellite data (they also mostly focus on the post-1990 period). We then do not simultaneously study heights and land area as we are interested in the present study in vertical development gaps instead of overall gaps. 6 We will interchangeably use “UCs”, “agglomerations” and “cities.” New York includes “New York; Islip; Newark; Jersey City; Yonkers; Huntington; Paterson; Stamford; Elizabeth; New Brunswick.” 7 See https://www.emporis.com/ for details. 8 We only consider “building with towers,” “high-rise building,” “low-rise building,” “multi-story building,” and “skyscraper,” thus excluding observation towers, telecommunications towers, etc. 7 For almost all buildings, we know the geographical coordinates or at least the locality in which it is located. This allows us to assign each building to a city in the GHS-UCDB data set. Since we know the year of construction (and demolition if demolished), we obtain the total sum of heights for each city-year and each country-year. Total Population and Total Urban Population. Knowing the total urban population of a country, we can calculate its total sum of tall building heights per urban capita, one of our main dependent variables. From United Nations (2018) we obtain the total population and the total urban population of each country every five years from 1950-2020 (for individual cities, population is only available in the years 1975, 1990, 2000, and 2015). National Income. Our main sources are Maddison (2008) and Bolt and van Zanden (2020), where we obtain per capita GDP annually from 1950 to 2018 (PPP and constant international 2011 $). In order to avoid short-term fluctuations in income, for each year t we consider a moving-average including the three years before and the three years after year t (“2020” is proxied by 2018, which avoids the undue influence from COVID-19).9 Night Lights. Historical income data at the city level is unavailable except in a handful of countries, hence the need to rely on national income data for some analyses. Another solution is to control for the log total sum of night lights within a city’s boundary in year t (based on the GHS-UCBD boundaries). We rely on two databases for this purpose. First, night lights data corresponding to the DMSP satellites are provided by NGDC (2015). We use the radiance calibrated version of this data, which is available for select years between 1996 and 2011, to avoid issues related to top-coding.10 The data are available at a fine spatial resolution (30 arc second, or ≈ 1km at the Equator) and we use GIS to obtain the total sum of DMSP-based lights for each city. For the years 1995, 2000, 2005, and 2010 in our data, we rely on the DMSP years 1996, 2000, 2005, and 2010. Night lights data corresponding to the VIIRS satellites and the period 2010-2020 are provided by Elvidge et al. (2021). The data is also not top-coded. The data are available at a fine spatial resolution (15 arc second, or ≈ 500m at the Equator) and we use GIS to 9 Likewise, construction started before the COVID-19 pandemic for tall buildings completed in 2020. We nonetheless verify that results hold if we only consider buildings completed before 2019 (not shown). 10 This data records levels of luminosity beyond the normal digital number upper bound of 63. The other version of the DMSP data set from 1992 to 2013 is top-coded, which is a major issue for cities. 8 obtain the total sum of VIIRS-based lights for each city. For the years 2010, 2015 and 2020 in our data, we use the VIIRS years 2010, 2015, and 2019 (due to COVID-19). 3.2. Tall Building Construction across Space and Time 3.2.1. Aggregate Patterns Figure 1 shows for the 12,877 world cities the sum of tall building heights per capita (1,352 cities have a tall building above 55 m). American, European, and Asian cities are more vertical than African ones. Tall buildings are not only found in developed regions. As seen in Figure 2, the total sum of tall building heights dramatically increased around the world during our main period of study (1950-2020). The left panel shows for the world and the United States separately the total sum of tall building heights (km) from 1885 to 2020 (note that we only consider completed buildings above 55 meters). The right panel shows for the world and the United States separately the total sum of tall building heights per urban capita (km per million residents) from 1900 to 2020. The left figure shows that while tall building construction can be observed until 1929 and the Great Depression, the period from 1929 to 1950 did not see much construction. Also, until 1970 the United States (U.S.) led in terms of tall building construction. Since then, the total stock of tall building heights in the U.S. has increased rather slowly. The right figure shows that, on a per capita basis, global heights have caught up, and now surpassed, the U.S., which had nearly a century head start in tall building construction. Just as importantly, global tall building stocks have outpaced urban growth. Next, Figure 3 shows for six United Nations (U.N.) “regions” the respective evolution of the total sum of heights per urban capita (km per million) and the urban share (%) every five years from 1950 to date. A large gap between the two lines suggests that the region is urbanizing with less vertical cities. North America, Europe, Oceania, and Latin America & the Caribbean (LAC) have similar urban shares today. But Oceania has more heights per urban capita, followed by North America, Europe, and LAC. For Europe, Oceania, and LAC, the “heights per urban capita” line has been converging towards the “urban share” line over time, suggesting verticalization. Asia and Africa have urbanized similarly post-1950 (with an urban share of about 10% in 1950 and 45% in 2020). However, 9 Asia did so by becoming more vertical whereas Africa has not built up nearly as much.11 3.2.2. “Horse Races” We now look at examples of how specific U.N. subregions–the level below the U.N. region–have grown their tall building stocks versus their per capita GDP levels. Figure 4 shows (in faint grey) the scatter plot of the log of each region’s height-GDP profile in five-year intervals from 1950 to 2020 (height is the log of total height per urban population and GDP is log per capita GDP in PPP terms). The scatter plot shows a tight relationship between the two (R2 = 0.72). Also presented in the figure are four subregions. By “connecting the dots” we can see their relative trajectories as compared to the global averages. The green line is Southeast Asia (diamonds), the pink line is East Asia (circles), the red line is South America (squares), and the blue line is Eastern Europe (triangles). See Web Appx. Table A.1 for the country-to-region mappings used throughout the analysis. Focusing on Southeast Asia (diamonds), we see that given its income, in 1950 it was far below the world average, but, particularly during the 1960s and 1970s, it added building height relative to its GDP growth. A similar pattern can be seen for East Asia (circles), but its building growth trajectory took place in the 1970s and 1980s. On average, Southeast Asia has been ten years behind East Asia in terms of income. But the former was ahead of the latter in terms of building heights until the 2000s. The two subregions are now similar in terms of heights despite incomes being almost half lower in Southeast Asia. One hypothesis why Southeast Asia started building up at low income levels is that Southeast Asian cities were influenced by the other “Asian Tigers.” These economies– Hong Kong SAR, China; the Republic of Korea; Singapore; and Taiwan, China–created special economic zones to develop their export sectors. International firms arrived to establish factories with their local corporate headquarters housed in tall buildings. Thus, Southeast Asian economies saw that tall buildings offered a path toward economic development. 11 Web Appx. Fig. A.2 shows the patterns focusing on developing economies and 6 World Bank “regions”. LAC is the most urbanized region followed by Europe & Central Asia (ECA), the Middle East & North Africa (MENA), East Asia & Pacific (EAP), Sub-Saharan Africa (SSA) and South Asia (SAR). For a given urban share, ECA is the most vertical region, due to Eastern Europe. LAC, EAP, and to a lesser extent SAR, have become more vertical since 2000. Developing MENA and SSA economies have few vertical cities. 10 Another hypothesis explored by Ahlfeldt et al. (2022) is the influence of Hong Kong SAR, China, in housing population in high-rises. After the People’s Republic of China was established in 1949, Hong Kong SAR, China, experienced a large influx of refugees. In a context of buildable land scarcity, authorities favored vertical development. From the 1960s to 1980s, Hong Kong SAR, China, became the world’s leader in terms of heights and the first Asian “model city,” inspiring cities with Chinese diasporas “connected” to Hong Kong SAR, China (Haila, 2000). The triangles in Figure 4 show that Eastern Europe’s height density remained stable until 1965. By the 1960s and following a period of rapid industrialization, urban housing supply became more limited. The Soviet Union responded by massively constructing “Brezhnevkas,” residential towers reaching 16 floors (Zadornin and Sheina, 2015). Between 1965 and 1985, Eastern Europe experienced one of the most impressive construction episodes in history. The economic crisis brought by the fall of communism only slowed down that process for a few years. Starting in 1995, one can see continued increases in height density, with this time private developers supplying the tall buildings.12 The trajectory for South America (squares) may, at first blush, appear surprising. In 1950, its height-GDP profile was along the global trend line, and by 2020, it was far above it. Though its height growth relative to GDP was not as rapid as those regions in Asia, its growth contradicts perceptions that South American cities have few tall buildings. We return to the causes of South America’s growth spurt in the sections below. The Web Appendix provides several more “horse race” graphs that show the relative height-GDP growth profiles for other regions. For example, South Asia has been 15-20 years behind Southeast Asia in terms of income (Web Appx. Fig. A.3). We also find similar height density patterns in South Asia as in Southeast Asia until 1975 for the former, where a plateau can be observed until 1995. The Indian government implemented liberalization reforms in the 1990s (Aghion et al., 2008). Since then, South India has been converging towards the trends observed in Southeast Asia. Indian cities have been building up in 12 Large companies, for example in the energy sector, also built towers for their headquarters, such as the Lakhta Center in Saint Petersburg, which is Gazprom’s headquarters and the tallest building in Europe. 11 recent times, especially Mumbai (Gechter and Tsivanidis, 2018). West Asia (i.e., the Middle East) has been 20-25 years ahead of Southeast Asia due to the wealth of its oil-exporting nations (Web Appx. Fig. A.4). But height density levels are higher in Southeast Asia. While West Asia’s “performance” has been spectacular these past decades (OEDC Development Center, 2020), it may be due to its high income levels. East Africa has been following similar patterns as in Southeast Asia (Web Appx. Fig. A.5). East Africa now has the same height density and incomes as Southeast Asia in the 1970s. However, it is too soon to tell if these patterns will be sustained.13 From 1960 to 1980 one sees vertical development in Southern Africa (Web Appx. Fig. A.6) (CNN, 2016). Post-1980, height density has remained similar. We then see divergence in terms of height density for the rest of Africa (Web Appx. Fig. A.7 for West Africa). Surprisingly, North America and Southeast Asia have similar urban height density levels today (Web Appx. Fig. A.8). If anything, North America has been converging towards the global trend line over time, implying slower vertical growth overall. The patterns for Western Europe are emblematic of the patterns for most of Europe (Web Appx. Fig. A.9). Southeast Asia has the same income today as in Western Europe in 1960. Yet, Southeast Asia has higher height density levels than Western Europe today. However, the cities of Western European countries were becoming more vertical until the early 1970s. Since then, one can observe slower growth for height density.14 The visual analyses ignore the fact that the technology of building up has improved over time (Ahlfeldt and Barr, 2022). One way to examine this is to study the relationship between log urban height density and log per capita income in both 1950 and 2020. Web Appx. Fig. A.10 shows that the relationship has shifted up over time, especially in richer nations. Indeed, the relative cost of accessing construction technologies (e.g., tower cranes and construction software) and workers (e.g., tower crane operators and construction 13 For example, The Citizen (2021) describes how the skylines of Nairobi (Kenya) and Dar es Salaam (Tanzania) have dramatically changed this past decade. They write: “According to MI, as a region, East Africa has the largest number of recorded projects” within the African construction industry. 14 What happened circa 1970 that led most European nations to slow down tall building construction? As discussed in Jedwab et al. (2020), in Paris the Tour Montparnasse completed in 1973 became one of the most hated landmarks in the world and led two years later to the ban of new buildings over seven stories high in central Paris. Likewise, Dublin’s Liberty Hall built in 1965 was immediately seen as one of Ireland’s ugliest buildings. The backlash led to the adoption of more stringent height restrictions a few years later. 12 engineers) may have disproportionately decreased in richer countries because such technologies originate from there (Barr, 2016). The cost of financing has also likely become smaller in more financially developed and thus richer economies (Jedwab et al., 2020). Overall, we find a rapid rise in the number of tall buildings across the world. Nonetheless, looking at regional patterns shows significant variation that is not explainable by relative incomes. Before we turn to a detailed treatment of these differences, we discuss the SUM, which guides the rest of the analysis. 4. Conceptual Framework We now discuss the conclusions of the monocentric Standard Urban Model (SUM) that guides our analysis (Brueckner, 1987; Duranton, 2015). Ahlfeldt and Barr (2022) provide a general equilibrium version of the SUM adopted for the purpose of studying tall buildings. Workers have preference over both vertical and horizontal space. Closer to the center is preferred to further away, ceteris paribus, and higher is preferred to lower, given the views and other vertical amenities. Households choose a quantity of (height-adjusted) residential floor area given the prices, exogenous incomes, and transportation costs. Firms have a production function that is determined by the agglomeration benefits (i.e., the number of firms close by), the amount of labor it hires, and the amount of office or floor space used for production (Liu et al., 2022, 2018b). Firms choose a quantity of floor area (and height) given the relative cost and productivity of each input. Profit maximizing developers consider construction costs and floor space rental prices when deciding how much floor area to provide at each location across the city. The equilibrium is defined by market clearing, and free mobility anchors utility to a reservation level. Assuming Cobb Douglas utility, production, and development functions, and typical parameter values, the model generates a series of predictions: • Firms typically occupy the center, where agglomeration benefits are greatest. • Building heights toward the center are taller than further away, since the price of space incentivizes developers to substitute building capital (i.e., heights) for land. • The greater the agglomeration benefits, the taller the tall buildings in central areas.15 15 The work from home phenomenon may cause office rent gradients to flatten (Rosenthal et al., 2022). 13 • The lower the cost of providing height, either from geographical constraints, technological improvements or building processes, the taller are the buildings. • A higher income city will have taller buildings than a lower-income city of the same population size. By extension, a more urbanized country will have taller buildings. More land-constrained cities and countries will then have taller buildings. The SUM typically assumes an exogenous center. However, a crop of models has studied the rise of poly-centric cities with multiple nodes of business and population density (Fujita and Ogawa, 1982; Anas et al., 1998; McMillen and McDonald, 1997). These models assume a trade-off between agglomeration benefits and congestion costs. When the costs of congestion relative to the agglomeration benefits hit a certain threshold, it will endogenously generate more than one urban center. As cities grow and “congestify,” they will face these trade-offs. But since different cities have different agglomeration- congestion ratios based on their respective policies and histories, we are likely to see variation in both tall building numbers and locations, based on the number of centers.16 Related to this line of work is that of Brueckner et al. (1999) who provide a framework for why in some countries central cities tend to have more low-income residents relative to high-income ones and vice versa. They show that if the wealthy prefer central city amenities to more housing space, they will cluster in the center, with the poor in the suburbs. However, if the wealthy have a relative preference for large houses, they will cluster outside the center. The point is that high-rise housing can be spread out throughout the city depending on the location of various income groups. Taken together, models predict that a regression of total tall building stocks on urban populations and income–demand factors–will generate positive elasticity estimates. Additionally, variables that impact the relative cost of height–supply factors, such as fault lines and other geographical and geologic disadvantages–will generate negative elasticities. Any variation in heights that is then not explained by these factors could then be due to land-use regulations, constraints on land assembly and/or preferences. The step-wise regressions will provide evidence consistent with the SUM. However, as 16 Poly-centricity does not alter the SUM’s main conclusions. In practice (and in theory), land values and heights from the exogenous center are universally found to decay exponentially (McDonald and McMillen, 2010). Poly-centricity, where it exists, tends to show up as “less-smooth” gradients (McMillen, 2001). 14 we will discuss below, while the SUM-suggested variables can account for a large fraction of global variation, there are regional differences that need to be better explored. For this reason, after the regression results, we will turn to the analysis of the regression residuals. Lastly, in the country-level analysis, since the building stock is measured for the entire country, not for individual cities, it is appropriate to divide the stock variable by the size of the country’s urban population. The dependent variable is thus the country’s total sum of tall building heights per urban capita. For the city-level analysis, we consider the total sum of tall building heights on the left-hand-side and control for city population size. 5. Econometric Frameworks and Selection of the Control Variables As shown graphically in Section 3 and summarized theoretically in Section 4, we need to control for demand and supply effects when examining vertical development differences (or “gaps”) across space over time. In this section, we describe the regressions that we use for our residual/gaps analysis. We first start with the country-level specifications because common discussions about skylines tend to begin with country comparisons (e.g., China vs. the U.S.). We then explain our city-level specifications. Lastly, we use a city-level regression to validate our use of selected demand/supply controls in these models. 5.1. Country-Level Specifications for the Residual/Gaps Analysis For this analysis, we consider 163 countries with available income data every five years during the 1950-2020 period.17 We first control for income using the following model: LHEIGHTSPUCct = α + β1 ∗ PCGDPct + β2 ∗ PCGDPSQct + µct (1) where LHEIGHTSPUCct is the log of (urban height density + 1) for country, c, and year, t, due to the zeros in heights in some country-years, and PCGDPct is per capita GDP (PPP and constant international dollars) and PCGDPSQct its square. Note that we use unlogged PCGDP as we do not want to compress the distribution of incomes, in order to estimate differences even among high-income countries (e.g., the U.S. vs. the UAE).18 The residuals µct capture to what extent the selected country-year has vertical cities 17 While the full sample used for the descriptive analysis comprised 183 countries, 20 small countries do not have consistent, available per capita GDP data for the whole period 1950-2020. Thus, N = 163. 18 Throughout the analysis when using the log of a variable + 1, we verify that we obtain similar results when relying on an inverse hyperbolic sine transformation or a negative binomial regression (not shown). 15 given its income level. By construction, since our aim is to examine “country effects” we cannot add country fixed effects. In this specification, we also do not add year fixed effects yet since we first aim to compare a country-year to other country-years with similar income levels (e.g., Chile in 2015 vs. Japan in 1980 vs. the U.S. in 1965). Next, we modify eq. (1) by adding year fixed effects (κt ): LHEIGHTSPUCct = α + β1 ∗ PCGDPct + β2 ∗ PCGDPSQct + κt + µct . (2) Thus, we control for income and across-the-board technological innovations. Thus, the comparison is now restricted to countries with similar income levels in the same year. Next, since technology may have disproportionately improved for wealthier nations, we modify eq. (2) by interacting the income variables with the year fixed effects. In other words, the coefficients of the income variables (β1,t , β2,t ) are now year-specific: LHEIGHTSPUCct = α + β1,t ∗ PCGDPct + β2,t ∗ PCGDPSQct + κt + µct . (3) We then modify eq. (2) by adding the country’s total land area and its square and the country’s total population in year, t, and its square, which we all interact with year effects to allow their effects to change over time.19 These controls are now included in Xc . LHEIGHTSPUCct = α + β1,t ∗ PCGDPct + β2,t ∗ PCGDPSQct + Xc Bt + κt + µct . (4) We then additionally include controls for geographical constraints. For example, buildings of 55 meters need foundations that are a few meters deep. As a result, a bedrock depth of a few meters may raise construction costs. Having said that, most of the variation in heights is explained by the negative effects of bedrock depth on heights, i.e., cases where the bedrock is too deep (Ahlfeldt et al., 2022). We use the bedrock depth data from Shangguan et al. (2017) to obtain the average (city population-weighted) bedrock depth in the cities of each country. Since city population estimates are only available in 1975, 1990, 2000 and 2015, we use the cities’ populations in 1975 as weights.20 We use peak ground acceleration data to obtain the average (city population- 19 The main sources for land area and total population are United Nations (2018) and World Bank (2022). 20 For the whole world, Shangguan et al. (2017) report depth to bedrock (meters) with a 30 seconds (≈ 1 km) resolution (for example, there are 8,118 such pixels in the New York UC). We then use the GIS boundaries of the UCs to obtain mean bedrock depth (m) for each city. According to the website of Shangguan et al. (2017) (last accessed on 03-13-2022), “this dataset is based on observations extracted from a global compilation of soil profile data (ca. 1,30,000 locations) and borehole data (ca. 1.6 million locations).” 16 weighted) earthquake risk in each country (using city populations in 1975 as weights).21 Finally, tall building construction could be more difficult in high-altitude and rugged areas. Conversely, countries with more mountainous land may experience less horizontal expansion of their cities, increasing the need for tall buildings. We use elevation data to obtain the (city population-weighted) mean as well as standard deviation of altitude in the cities of each country (using the cities’ populations in 1975 as weights).22 To summarize, the following controls are added to model (4) (within Xc ): mean bedrock depth, mean earthquake risk, and the mean and standard deviation of elevation, and their squares, which we then interact with the year fixed effects. 5.2. City-Level Specifications for the Residual/Gaps Analysis The country-level analysis does not account for how the size distribution of cities varies across countries over time. We thus redo the analysis at the city level, classifying the 12,877 cities, a, in our data into 10 categories, p, depending on their population size in time t: 0-100K, 100-250K, 250-500K, 500-750K, 750-1,000K, 1,000-2,500K, 2,500-5,000K, 5,000- 7,500K, 7,500-10,000K, 10000K+. Since city population is only available in 1975, 1990, 2000, and 2015 in GHS-UCDB, we restrict the analysis to the period 1975-2020. For each year t, we construct the population categories for the closest year in GHS-UCDB.23 For cities a in countries c and years t every five years in 1975-2020, the model is: LHEIGHTSact = α + Σ10 10 p=1 γp 1(CATact = p) + Σp=1 β1,p ∗ 1(CATact = p) ∗ PCGDPct (5) +Σ10 p=1 β2,p ∗ 1(CATact = p) ∗ PCGDPSQct + Xc B1t + Xac B2t + µact where LHEIGHTS is the log of (total sum of tall building heights + 1). By construction, since our aim is to estimate city effects, we cannot add city fixed effects. Also, since we do not have city population data every year, we do not consider heights per urban 21 For the whole world, Giardini et al. (1999) report peak ground acceleration (PGA; m per s2 ) at the 0.0833*0.0833 degree level (≈ 9 x 9 km). PGA takes into account the probability of strong earthquakes in each pixel as well as the probability of diffusion over space. Since land-use regulations related to earthquake risk tend to be adopted based on fuzzily defined local conditions, we consider buffers of 0.05 degree (5.55 km) around each city. We then obtain the mean PGA for each city/buffer. According to their website: “The GSHAP Global Seismic Hazard Map has been compiled by joining the regional maps produced for different GSHAP regions and test areas; it depicts the global seismic hazard as Peak Ground Acceleration (PGA) with a 10% chance of exceedance in 50 years, corresponding to a return period of 475 years.” 22 The altitude data comes from GMTED (2010) (resolution: 15 arc-seconds, or 463 m close to the equator). 23 For 2010-2020, 1995-2005, 1985-1990, and 1975-1980, we use 2015, 2000, 1990, and 1975, respectively. 17 capita on the left-hand-side. Instead, we include the 10 population size dummies on the right-hand-side. By including the 10 population size dummies and their interactions with the (national) income variables, the residuals then capture to what extent a city is more vertical for a given population size and given national economic conditions (remember that historical income data at the city level does not exist except in a handful of countries). In the second specification, we include year fixed effects. In the third specification, we interact the year fixed effects with the 10 city size dummies, the income variables, and all their interactions. In the fourth specification, we also include the country’s land area and time-varying population and their squares, and all their interactions with the city size dummies and the year fixed effects. Finally, the fifth specification also includes the city’s mean bedrock depth and earthquake risk and the mean and standard deviation of elevation in the city, and their squares, as well as all their interactions with the city size dummies and the year fixed effects. We thus flexibly control for the effects of country size and city-level geographical constraints across different city sizes and over time. Lastly, the country residuals in each year t are constructed using the city residuals in each year t (with the population of each city circa year t as weights). The subregional averages in each year t are then constructed using the country residuals in each year t (with the total urban population of each country in year t as weights). One issue is that, due to data limitations, we do not control for city income per se. Instead, we include city population size dummies interacted with national income variables. Thus, we will additionally control for city night lights, in particular the log of (total sum of DMSP lights + 1) in year t interacted with a dummy equal to 1 if the DMSP data is available for that year (1995, 2000, 2005 and 2010) and the log (total sum of VIIRS lights + 1) in year t interacted with the dummy equal to 1 if the VIIRS data is available for that year (2010, 2015 and 2020). Overall, for 12,877 cities, a, in countries c and years t every five years during the period 1995-2020, the most complete specification becomes: 18 LHEIGHTSact = α + LDMSP SUMact ∗ 1(t = DMSPYR) + LVIIRS SUMact ∗ 1(t = VIIRSYR) +Σ10 10 p=1 γp 1(CATa c = p) + Σp=1 β1,p ∗ 1(CATa c = p) ∗ PCGDPct +Σ10 10 p=1 β2,p ∗ 1(CATa c = p) ∗ PCGDPSQct + Σp=1 β1,p ∗ 1(CATa c = p) ∗ Xc Bp1t +Σ10 p=1 β1,p ∗ 1(CATa c = p) ∗ Xac Bp2t + µact . (6) 5.3. Validation of the Controls Used Given the 15 year dummies for the period 1950-2020 and their interactions with the other controls, the most complete country-level specification includes 187 controls (the R2 is 0.59). The full city-level specification without the night lights controls considers a smaller period (1975-2020). Yet, despite the lower number of year dummies (10), their interactions with the city population size categories and the other controls increase the number of controls to 1,001 (R2 = 0.63). Finally, the full city-level specification with the night lights controls considers an even smaller period (1995-2020). Nonetheless, the number of controls is still 650 (R2 = 0.64). As such, presentation of the full regression results here becomes impractical. Additionally, our key interest in analyzing the residuals and not the coefficients of the many controls. Nonetheless, we can demonstrate that the controls have the right signs for a simplified city-level regression for the year 2020. More precisely, for 12,877 cities, a, in countries c and the year 2020, we consider the following cross-sectional model: LHEIGHTSac20 = α + Σ10 10 p=1 γpH 1(CATa c = p) ∗ Hc20 + Σp=1 γpU M ∗ 1(CATa c = p) ∗ UMc20 +Σ10 10 p=1 γpLM ∗ 1(CATa c = p) ∗ LMc20 + Σp=1 γpL ∗ 1(CATa c = p) ∗ Lc20 +LVIIRS SUMac20 + Xc B1 + Xac B2 + µac20 (7) where H, UM, LM and L are four dummies equal to one if the city belongs a high- income country, an upper-middle-income country, a lower-middle-income country or a low-income country c. 2020, respectively (according to the classification of the World Bank).24 Equation (7) is similar to equation (6) except we only consider the year 2020, 24 datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending-groups. 19 we replace the PCGDP variables by four income group dummies, and we do not interact the other country- and city-level controls with the 10 city population size dummies. The resulting model has only 47 controls, whose coefficients are shown in Table 1.25 Heights increase with city population size and economic development. This can be seen more clearly in Web Appx. Fig. A.11 where we show the predicted level of (log) heights for each city population size category × income group. Across the four income groups, larger cities systematically have more height than smaller cities. As expected, wealthier income groups also have more height than poorer income groups. Consistent with the conceptual framework of Section 4., Table 1 shows the expected correlations for city lights (positive), the country’s total population (mostly positive) and land area (mostly negative), earthquake risk (mostly negative) and bedrock depth (mostly negative). For mean altitude and ruggedness, the relationship is more ambiguous. Web Appx. Fig. A.12 thus shows how heights vary with respect to country population (mostly positive, but negative for high values) and land area (negative), earthquake risk (negative), bedrock depth (negative), altitude (negative) and ruggedness (positive). One explanation for the negative relationship for altitude could be that it raises construction costs due to high transportation costs of conveying materials to high-altitude cities. Ruggedness could lead to more height due to buildable land scarcity. More generally, the R2 of this regression and the main regressions is below 0.65. Overall, there is still a lot of variation left that is captured in the tall building construction residuals. 6. Analysis of the Tall Building Construction Residuals The regression results in Section 5. suggest that income, technological change, and natural environmental features can account for a large fraction of building height stocks per urban capita around the world. In this section, we turn to an analysis of the residuals based on the step-wise country-level regressions discussed above. The key method is to look at the average subregional residuals after controlling for the variables suggested by the SUM, to see how the residual rankings (and variation) change. This analysis can give clues on the institutional, historical, and preference-based factors around the world. 25 Some of the controls are missing for a few cities, hence N decreases to 12,753. 20 6.1. Residuals Based on the Country-Level Analysis Col. (1) of Figure 5 displays the residuals in 2020 when including the income controls (eq. (1) in Section 5.1.). Asian, Africa, Latin American, and other subregions are shown in green (squares), pink (circles), red (triangles) and blue (diamonds), respectively.26 The results show that Asian countries, South America, and Eastern Europe are at the top of the rankings, while Anglo-speaking and other European regions are at the bottom. More generally, the average population-weighted residuals for the six U.N. regions are: Latin America = 1.9; Asia = 1.8; Europe = -0.3; Africa = -0.3; Oceania = -1.3; and North America = -2.1. Latin America and Asia appear to dominate the world, ceteris paribus. The second column of Figure 5 displays the residuals based on eq. (2), holding constant both income and year effects, which control for across-the-board technological innovation. The third column shows the residuals of eq. (3) whereby we also interact the income variables with year fixed effects, capturing how technological innovation has benefited wealthier nations. As seen, the residuals and rankings are barely modified. The fourth column shows the ranking when we additionally control for the country’s total land area and population and its square, which are all interacted with year effects (eq. (4)). More populated countries tend to have larger large cities. For example, the largest city of China or India is far larger than the largest city of El Salvador or Rwanda. Since the largest cities of China and India are larger, they will naturally have more tall buildings, so their residuals are better estimated by also controlling for a country’s population size. Likewise, countries with more available land tend to have less vertical cities. Indeed, if land is “cheap,” cities can expand more horizontally (Brueckner, 1987). With the addition of these controls. East Asia and South Asia are not as well- ranked. China and India as well as Bangladesh, Japan, and Pakistan have large populations/mega-cities. These countries are also densely populated. The newly estimated residuals take into account the fact that their large cities “need” even more tall 26 Note that we center the residuals around 0 throughout the analysis because: (i) The residuals are estimated relative to other years if year fixed effects are not included; and (ii) Aggregating the residuals at the subregion-year level may uncenter the residuals even if year fixed effects are included. Indeed, we rely on total urban population sizes (later, city population sizes) for the aggregation process but regressions are unweighted. Since what matters is the difference across regions, this is inconsequential for our analysis. 21 buildings. North America also becomes more poorly ranked. The U.S. has a lot of land, which could have led to lower gaps. However, the U.S. has a large population/large mega-cities. Thus, it should have needed more buildings, and its ranking decreases. As seen in the fifth column, the rankings change when also controlling for bedrock depth, earthquake risk, elevation, and ruggedness (eq. (5)). Southern Africa, Central America, and East Africa are in the same leading group as South America, Southeast Asia, and Eastern Europe. East Asia decreases in the rankings. For example, it could be that Japan and Korea should have even more tall buildings given how mountainous they are. Northern Europe (which includes, for example, the United Kingdom, Ireland, and the four Scandinavian countries) and Western Europe (which includes, for example, Germany, France, the Netherlands and Belgium) see their rankings improve. More generally, the average population-weighted residuals for the six U.N. regions are: Latin America = 1.4; Europe = 0.3; Asia = 0.0; Oceania = -0.1; Africa = -0.3; and North America = -2.6. Latin America still dominates the world, ceteris paribus. Europe and Asia are also relatively well-ranked. Finally and as expected, Figure 5 shows that the residuals become less dispersed as more and more controls are added. Furthermore, Figure 6 shows the residuals for each U.N. subregion and the five specifications above when considering the mean of the (five-year) absolute changes in the tall building construction residuals during the 1950-2020 period. The rankings show how each subregion has improved or deteriorated on average over the past 70 years, which should better reflect current conditions (rather than pre-1950 conditions). Focusing on the last and most complete specification (fifth column), one can see that the rankings are dominated by Asian subregions (with the following order: C > SE > S > W > E). The other top subregions include East Africa and South America. Among the lowest-ranked subregions, we can see a few European and African regions as well as North America. More generally, the average population-weighted mean change in the residuals for the six U.N. regions are: Latin America = 0.05; Asia = 0.01; Africa = 0.01; Oceania = 0.0; Europe = -0.05; and North America = -0.1. Europe and North America thus rank at the bottom, showing how other regions have been “catching up” over time. We now examine how the patterns hold, or change, if we perform the analysis at the city level. 22 6.2. Residuals Based on the City-Level Analysis For the city residuals analysis, we perform a similar set of step-wise regressions (see Section 5.2.). The results for the five specifications are shown in Web Appx. Figs. A.13 and A.15 for the residuals in 2020 (without vs. with the lights controls) and Web Appx. Figs. A.14 and A.16 for the mean change in the residuals in the recent period (1975-2020 when not including the lights controls; 1995-2020 when including them). Since we are interested in how the rankings compare for the country- and city-level specifications, Figure 7 shows the 2020 rankings for four selected analyses: (1) “Cntry: Income:” Country-level analysis controlling for national income. (2) “Cntry: Geo:” Country-level analysis with the full specification with all country-level controls (incl. the geographical controls). (3) “City: Geo.:” City-level analysis with the full specification with all country- and city-level controls (eq. (6) above); and (4) “City: Geo+Lights:” City- level analysis with the full specification also including the lights controls (eq. (6)). The city-based results are thus reported in the last two columns of Figure 7. The 2020 rankings are not modified when controlling for night lights. Next, the five leading subregions are South America, Eastern Europe, Southeast Asia, East Africa and Southern Africa. The top half of the rankings then include four Asian regions (SE > E > C > S), as well as Western Europe (ranked sixth). North America and West Asia, which are often thought of tall building-oriented regions, are only ranked twelfth and thirteenth. With the lights controls, the average residuals for the U.N. regions are: Latin America = 0.9; Europe = 0.4; Asia = 0.1; North America = -0.1; Africa = -0.5; and Oceania = -1.5. Focusing on the more recent conditions, Figure 8 shows the rankings for the same selected analyses but considering the mean change in the residuals over time. Privileging the city-level analysis with the most complete specification and the lights controls (last column), we have South America, Southeast Asia, Eastern Europe, Central America, and East Africa as the leading regions, followed by West Asia, East Asia, South Asia, Central Asia, and Western Europe. North America is ranked thirteenth. Other European and African regions (including now Southern Africa) rank at the bottom. Also, the mean change in the residuals for the U.N. regions are: Latin America = 0.19; Europe = 0.06; Asia = 0.05; North America = -0.4; Africa = -0.04; and Oceania = -0.22. 23 The main takeaways are as follows: (i) Latin America and Europe are highly ranked, ceteris paribus, due to the construction patterns of South and Central America and of Eastern and Western Europe, respectively. Asia and North America are not that well ranked, respectively; (ii) Within Asia, the leading region, ceteris paribus, is Southeast ¨ Asia, not East Asia. West Asia, which includes not only the Gulf nations but also Turkiye, Iraq, the Syrian Arab Republic, and Jordan, is the lowest ranked Asian region; (iii) Among regions with large developed economies, Eastern Europe is ranked first, followed by Western Europe, and East Asia, and then Southern Europe, North America, Northern Europe, and Australia/New Zealand; and (iv) Some African regions–East Africa and Southern Africa–are relatively well-ranked, ceteris paribus. While some of these results may initially appear surprising, recall that we control for city population size and income. Whereas some cities “visually” appear to have impressive skylines (e.g., Tokyo, Shanghai, and New York), some of them are also heavily populated (33 million, 25 million, and 16 million, respectively, in the GHS-UCDB data). Likewise, some skyline-oriented cities also have some of the highest incomes in the world (e.g., Dubai; Hong Kong SAR, China; and Singapore). Finally, and as we will discuss below, focusing on the most well-known tall buildings in the world is misleading because about 70% and 80% of total heights in the world come from high-rises below 100 meters (not skyscrapers) and residential towers (not office super-towers), respectively. Thus, “boring” skylines made of shorter residential towers may actually be what is driving variations in heights globally. 6.3. Differences between the Country- and City-Based Residuals Why would conducting the analysis at the city level rather than at the country level change the rankings? Several mechanisms could explain the observed discrepancies. First, for the country-level analysis we divide the sum of tall building heights by the total urban population of the country. But for a given urban population, a country may have a more skewed city size distribution. If a country has a higher urban primacy rate, its largest city is relatively larger, so more tall buildings are “needed.” With the analysis at the city level, countries with more primate cities should thus be lower ranked.27 27 High primacy countries include, for example, Bangladesh, the Arab Republic of Egypt, and Japan 24 Second, countries with few small- and medium-sized cities for a given total urban population should be lower ranked as well. Indeed, if most of the urban population is concentrated in a few large cities, more tall buildings will be “needed” overall. Third, the city-level analysis could lower the ranking of countries whose spatial distribution of tall buildings does not match the spatial distribution of populations. While tall building supply should respond to tall building demand, it could be that the governments of some countries encourage tall building construction in some cities, hoping that populations follow. If that is not the case, a “spatial mismatch” occurs. Fourth, it could be that the country- and city-level controls have more impact for larger cities in a country, implying that the country-level analysis suffers from under-controlling. It is difficult to assign each change to one or several of the above-mentioned mechanisms. The patterns in North America and East Asia are likely dominated by the patterns in the U.S. and China given their population size. These countries as well as Western European countries have relatively low urban primacy rates (source: World Bank (2022)). Due to their long urban histories, Western European countries and East Asian countries are both densely populated and have many small- and medium-sized cities. That is also the case for Central Asia, due to its historically important “central” location between Europe and East Asia (Bairoch, 1988). These and other factors could contribute to these subregions’ improved ranks once one moves to a city-level analysis. Conversely, Australia/New Zealand, the Caribbean, Northern Europe, most African regions, and West Asia have relatively fewer small- and medium-sized cities, possibly due to their specific geographies (e.g., a lot of the land is uninhabited due to extremely cold or hot temperatures) and/or late colonization and/or urbanization processes (Bairoch, 1988). That could explain why they rank worse with the city-level analysis. 7. Measurement Issues, Heterogeneity, and Interpretation of the Gaps 7.1. Measurement Issues One could argue that taller buildings are better measured than shorter buildings, because they stand out more. Therefore, we verify that the results hold if we restrict our analysis to buildings above higher thresholds than 55 meters, in particular 60, 65, 70, and 75 meters. (World Bank, 2022). 25 The residuals that we then obtain for the year 2020 (with the most complete city-level specification including the night lights controls) are highly correlated with each other (correlation of 0.95-0.99). Likewise, the mean five-year changes in the residuals that we then obtain for the period 1995-2020 are also highly correlated with each other (0.97-1.00). Finally, the subregional rankings are barely changed (Web Appx. Figs. A.17-A.18). The residuals that we obtain, on which the rankings are based, are thus due to other factors, either an under-controlling bias–the fact that we miss some important controls and/or that the controls used are misestimated–or capture what we aim to capture, i.e., real gaps due to land-use regulations and other institutional or socio-cultural features (Titman and Twite, 2013). Due to data limitations, we do not believe that it is possible to accurately control for all the factors that we are supposed to control for. However, we now discuss what controls could be missing or mis-specified, and how this could affect the rankings. We then discuss what factors could drive the gaps if they are “real.” First, if our controls are imperfect, the gaps will be affected. However, the extent to which this affects the rankings is difficult to say. If all regions are similarly affected, the rankings are by construction not affected. Furthermore, if measurement error is classical, it should not dramatically affect the results. But if measurement error is non-classical, then some regions could be more impacted. The question is which ones and how much. The SUM suggests that building heights increase with the size of the urban population, incomes, and the land rent at the edge of the cities, and decrease with commuting costs (other factors may of course also matter). By controlling for urban/city population and national/city income, we likely control for the previous factors. Indeed, we expect wealthier nations to have higher urban per capita incomes and land rent, and better commuting technologies. But per capita GDP is an imperfect control that may not fully capture how each of the previous and other factors vary across countries over time.28 Likewise, we control for bedrock depth, earthquake risk, and mountainous-ness. These controls could be mis-estimated. Other geographical constraints could be missing. Overall, even if our controls are imperfect and some controls are missing, the extent to which the subregional rankings will be affected depends on: (i) to what extent our existing 28 Global historical data on how commuting costs vary across cities over time does not exist. 26 controls indirectly capture the missing controls; (ii) how imperfect our controls are and how important the missing controls are; and (iii) how the distributions of these controls and the related measurement errors vary across countries/subregions. It could well be that any mismeasurement/omission is non-consequential for our ranking analysis. 7.2. Heterogeneity by Type of Localities and Type of Buildings We now examine the residuals and rankings by city size and type of buildings. Large-City vs. Small-City Skylines. As of 2015, 221 cities among the 12,877 cities have more than 2 million residents and comprise 37% of total city population but 76% of total city heights in our data. We thus examine how the residuals vary for cities above 2 million residents (“large cities”) vs. below 2 million residents (“small cities”). To do so, we re-run the city-level regression with all the controls including night lights (specification: “City: Geo+Lights,” period = 1995-2020) but separately for each type of city. We then obtain the average population-weighted residuals for each subregion in 2020. Figure 9 shows how the residuals compare for large cities (x-axis) and small cities (y- axis). Subregions below the regression line have more vertical large cities–i.e., “large city skylines”–ceteris paribus relative to the rest of the world. Subregions above the line have relatively more vertical small cities–i.e., “small city skylines.” Larger cities have more tall buildings and small cities fewer such buildings, for example because land is more expensive there. But if an urban system is balanced, both its large cities and small cities should have similar residuals, i.e., the subregion should be close to the line. That is the case for most Asian subregions (squares) and North America (a diamond). Subregions with large city skylines include West Asia (a square well below the regression line), where numerous tall buildings might be “white elephants” located in the largest cities (Gjerløw and Knutsen, 2019; Jedwab et al., 2022; Jedwab, 2022). Northern Europe (a diamond), which includes the United Kingdom and thus London, also has large city skylines. That is also the case in all African subregions (circles below the line). Next, small city skylines as observed in other European regions (diamonds above the line) and South America (a triangle well-above the line) could suggest that large cities are height- constrained and that tall building supply has to increase in smaller cities as a result.29 29 Since large cities account for three-fourths of total heights, the baseline residuals should be driven by 27 Central vs. Peripheral Skylines. Emporis reports the locality of each building. For example, for the New York agglomeration, we know if the building is located in New York City or another locality such as Jersey City or Newark. For each city, we thus estimate the total sum of heights in the central areas vs. the peripheral areas (central areas account for 75% of heights globally). We then examine how the residuals vary for central vs. peripheral areas. To do so, we re-run the city-level regression with all the controls including night lights (specification: “City: Geo+Lights;” period = 1995-2020) but separately for each type of area. We then obtain the subregional residuals in 2020. Figure 10 shows how the residuals compare for central areas (x-axis) and peripheral areas (y-axis). Subregions below the regression line have more vertical central cities– “central skylines”–ceteris paribus relative to the rest of the world. Subregions above the line have relatively more vertical peripheral, i.e., “peripheral skylines.” The monocentric urban model suggests that central areas should be more vertical than peripheral areas (Brueckner, 1987). If employment is centrally located and central living reduces commuting costs, then land becomes more valuable in central areas, which leads developers to build up. However, with polycentric cities, the main central areas should not be as vertical, and other tall building centers may emerge. Nonetheless, since polycentric cities have a main center (e.g., Downtown LA for Los Angeles), this center should still be more vertical than other centers, and comprise most of the city’s skyline. West, East, and Central Asia (squares below the regression line) and South, East, West and North Africa (circles below the line) have central skylines. But Southeast Asia (a square well-above the line), most European regions and North America (diamonds above the line)30 as well as Central and South America (triangles above the line) have peripheral skylines, which suggests that their central areas might be height-constrained.31 large cities. Web Appx. Fig. A.19 shows the baseline rankings and the rankings by city size. The baseline rankings and the rankings for the large cities are broadly consistent (correlation between the residuals = 0.86). For small cities, European regions tend to be better ranked. But some of the high performers for large cities also perform well for small cities (e.g., South America, East Africa and most Asian subregions). 30 efense Height restrictions in central Paris have led to the development of the peripherally located La D´ business district. It is similar in Washington DC where most office towers are located in neighboring Arlington and Fairfax County. The U.S. also tends to have more polycentric cities (e.g., in California). 31 Finally, Web Appx. Fig. A.19 shows the rankings for central and peripheral areas. The overall rankings should be driven by central areas. As expected, the baseline and central rankings are similar. Interestingly, European regions, North America and Australia/New Zealand are better-ranked when 28 Conic vs. Flat Skylines. Emporis defines “high-rises” as tall buildings below 100 meters and “skyscrapers” as tall buildings above 100 meters. We examine how the residuals vary for high-rises vs. skyscrapers. To do so, we re-run the city-level regression with all the controls including night lights (specification: “City: Geo+Lights;” period = 1995-2020) but separately for each type of heights. We then obtain the residuals for each subregion in 2020. High-rises and skyscrapers account for 68% and 32% of total heights in the world c. 2020, respectively. Thus, skylines are actually dominated by high-rises. Figure 11 shows how the residuals compare for high-rises (x-axis) and skyscrapers (y-axis). Subregions below the regression line have flatter vertical areas–“flat skylines.” Subregions above the line have relatively steeper vertical areas–“conic skylines”. Balanced urban systems should be close to the line. Most American regions including North America are close to the line. Other regions appear more specialized. Conic skylines can be observed in West Asia (i.e., in the Gulf nations), East Asia and Southeast Asia (squares above the regression line). It is also the case in Southern Africa and East Africa (circles above the lines). As we will discuss below, tall buildings in these two regions are overwhelmingly office towers, which tend to be taller. But that will not necessarily be the case in Asia, where tall buildings may be relatively more residential. In contrast, the four European regions have flat skylines (diamonds below the line), as well as South America (triangle below the line) and Central Asia (square below the line). In Eastern Europe and Central Asia, the focus during the Communist era was on building residential high-rises. Other European countries tend to discourage the construction of skyscrapers that may noticeably alter their skyline. In South America, height restrictions were imposed in several Brazilian cities for tall buildings above a certain height following the Joelma Building fire of 1971 and the Andraus Building fire of 1972 in Sao Paulo (Leahy, 2018). That could explain why Sao Paulo has a mostly flat skyline of high-rise buildings.32 considering peripheral areas. However, leading subregions in the central rankings are also among the top ranked regions for the peripheral rankings: Southeast Asia as well as South America and Central America. 32 Web Appx. Fig. A.20 shows the rankings for high-rises and skyscrapers. The rankings should be driven by high-rises. As expected, the baseline and high-rises rankings are similar. Considering skyscrapers, a few leading regions in terms of high-rises are leading regions in terms of skyscrapers: Southeast Asia, East Africa and South Africa. Exceptions include South America, Eastern Europe, Western Europe and Central Asia that see their rankings decrease when considering the latter only. Conversely, East Asia, West Asia and North America see their rankings improve when considering the latter. Their skylines may thus visually appear as more prominent internationally than they really are once one includes all tall buildings. 29 Residential vs. Office Skylines. Emporis reports each building’s function. Of all the buildings, 82% appear as residential vs. 12% as office towers (less than 1% of buildings appear as both residential and office towers). Other functions include hotels (3%), government buildings (0.5%), etc. We examine how the residuals vary for residential vs. office towers. To do so, we re-run the city-level regression with all the controls including night lights (specification: “City: Geo+Lights;” period = 1995-2020) but separately for each type of function. We then obtain the residuals for each subregion in 2020. Figure 12 shows how the residuals compare for residential buildings (x-axis) and office buildings (y-axis). Subregions below the regression line have more residential buildings– “residential skylines”. Subregions above the line have relatively more office buildings– “office skylines.” More balanced urban systems should be close to the line. It is the case for Southeast Asia (a square on the line). Other regions are more specialized. North America and most European regions (diamonds above the line) and African regions (circles above the line) privilege office towers. Some Asian subregions (squares below the line: W, S and C), South America (a triangle below the line) and Eastern Europe (a diamond below the line) privilege residential towers. Finally, as suspected, the residential and high-rises residuals are extremely correlated (correlation of 0.98). We just discussed why Eastern Europe and Central Asia privilege residential towers. For West Asia but also South Asia and South America, one explanation could be that wealthier residents see living in a high-rise as more glamorous as well as safer. For example, NY Times (2005) writes the following about Sao Paulo: “Twenty-four-hour security, closed-circuit television and an armored entrance create a very safe feeling in a city where all luxury buildings try to assuage fears of violent robbery and kidnapping.”33 33 Web Appx. Fig. A.20 shows the rankings for residential and office buildings. The overall rankings should be driven by residential buildings. As expected, the baseline and residential rankings are similar. Next, some leading regions in terms of residential buildings are also leading regions in terms of office buildings: South America, Eastern Europe, Southeast Asia, Southern Africa, Western Europe, East Africa and East Asia. Other European regions are better-ranked when considering office buildings. 30 7.3. Skyline Typologies We use these analyses to better understand the nature and characteristics of global skyline variations. Figure 13 provides a visual summary of the different types of skylines that appear around the world. To create the graph, we exploit the fact that the residuals’ correlations between residential/office and high-rise/skyscraper, and small/large city and periphery/core, are both high. The correlation between the residential and high-rise residuals is 0.97. It is 0.83 between the large city and central area residuals. To understand the difference in regional tall building outcomes, we perform the following steps. We take deviation of each residual from the respective trend lines shown in Figures 9 to 12. This gives for each gap how far they are from the respective world averages for each category. For example for Southern Africa, the small city skyline deviation is -0.40 vs. -0.22 for the peripheral skyline deviation. It is -0.25 for high-rises vs. -0.26 for residential buildings. Next, we take the maximum deviation from either the high-rise or residential deviations (i.e., -0.26 for Southern Africa). Then we take the maximum deviation of either the small city or peripheral area deviations (e.g., -0.40 for Southern Africa). The idea is to extract the “most extreme” deviations to see if they suggest something about the preferences and building behavior of each country. Figure 13 shows key features of each region. First, the regions closest to the (0, 0) point in the figure have less specialized skylines. This includes North America. Therefore, North America’s skyline, while relatively underdeveloped (remember that it is ranked overall twelfth in the world), does not seem to be “unbalanced.” The upper-right quadrant contains what we call Population-Oriented Peripheral Districts (PD) regions. They tend to build high-rise residential housing outside of the central cities (i.e., in smaller cities and/or peripheral areas in large cities). South America and Eastern Europe are two notable examples. Looking at their “behavior“ this way shows more broadly why they rank so well in the overall gaps analyses. Other regions included in this quadrant include Western Europe and Central Asia, which share many features with Eastern Europe. Southeast Asia and South Asia are also located there. We label the bottom left quadrant as Capital-Oriented Central Districts (CD) because these regions concentrate on building skyscrapers in their respective central cities (i.e., 31 in larger cities and/or central city areas). We label them as “capital-oriented” because most of these regions are specialized in (skyscraper) office buildings instead of (high-rise) residential buildings. They include almost all African regions (including Southern Africa and East Africa), East Asia, and Central America. West Asia (e.g., the Gulf Nations) also belongs to the quadrant and we also define it as “capital-oriented” despite the fact that it specializes in (skyscraper) residential buildings. Indeed, such buildings aim to capture some of the global luxury residential real estate sector, hence “capital” owners. The two previous quadrants contain most regions in the world. Hence, either skylines consist of population-oriented peripheral districts (PDs) or capital-oriented central districts (CDs). Three regions–North America, Australia/New Zealand, and Southern Europe–then belong to the upper left quadrant, which we call Capital-oriented PD. These tend to build skyscraper offices outside central districts, i.e. in smaller cities and/or peripheral city areas. Though places like New York and Chicago dominate our thinking about American skylines, most other cities across the U.S. are not so centralized. Likewise, since the U.S. remains a mostly suburbanites’ country, office parks and “Edge Cities” (Garreau, 2011) likely comprise an important part of how the U.S. builds upward. Finally, the lower right quadrant–Population-Oriented CDs–contains Northern Europe (which for example includes the UK, Ireland, and the Scandinavian countries). These countries tend to build high-rise housing closer to the urban centers (than say Western Europe and Eastern Europe, two regions with population-oriented PDs). 7.4. Interpretation of the Vertical Urban Development Gaps 7.4.1. Theories Assuming that the rankings that we obtain are “real,” what could be driving these? Answering this is important for policy, since the gaps may be due to policy differences across subregions. However, more socio-cultural, hence less policy-related, factors could also matter. We now turn to four key possibilities: (i) land-use regulations, (ii) historical preservation, (iii) land assembly, and (iv) residential living preferences. First, the gaps could be attributed to land-use regulations, in particular height restrictions in some, or all, areas of the cities in hopes of reducing shadows and congestion 32 (Glaeser et al., 2005b; Bertaud and Brueckner, 2005). If the height restrictions are binding given the local demand, then a gap may appear. Conversely, urban containment policies such as green belts and urban growth boundaries increase the need for taller buildings when urban demand is high. Additionally, land-use regulations may impact residential and office buildings similarly, except insofar as residential buildings tend to be less tall and centrally located than office buildings (Ahlfeldt and Barr, 2022). Land-use regulations should then disproportionately impact large cities (Saiz, 2010). Next, many countries have historical cities in them and their societies may demand land-use regulations that protect such cities against (modern) vertical re-development in their historical areas (as well as vertical development outside their historical areas if it affects the views of the sky from historical areas) (Jedwab et al., 2020). Since the central areas of cities tend to be more historical, historical preservation may disproportionately impact central areas, where tall buildings are more likely to be skyscrapers and office buildings (Brueckner, 1987; Ahlfeldt and Barr, 2022). In other cases, land-use regulations can arise because homeowners-as-home voters oppose development because they fear the effects of increasing housing supply on property prices (Fischel, 2005). One can thus separate regulations into history-related ones and non-history-related ones. However, even when land-use regulations are not binding, dispersed ownership–the fact that a candidate plot(s) of land for development/re-development as well as the structures on them may be owned by many owners–makes it intrinsically difficult to verticalize a city. Such constraints on land assembly could occur in more historical cities where plots may have been historically small and units subdivided (Lindenthal et al., 2017). That could also be the case in non-historical cities where, for various reasons, plot sizes are lower and crowding implies more subdivided units (Harari and Wong, 2018; Henderson et al., 2021). Relatedly, land assembly, particularly in regions where property rights are not strong, might hinder tall building construction, because ownership claims are vague and diffused. If anything, that should impact residential and office buildings similarly. Finally, due to various socio-historical factors, preferences for different types of urban living may vary across societies. There may be societies where people have an intrinsic preference for living in structures with few storeys or individual houses. There may then 33 be societies where people have an intrinsic preference for living in high-rise apartments, because of the centrality, views, and amenities they may offer (Yeh and Yuen, 2011). Why a society has specific preferences in favor or against high-rise apartments is difficult to say. Preferences and land-use regulations reinforce each other. In societies where high-rise apartments are preferred less, height restrictions are more likely to be imposed (or less likely to be removed in the face of high housing demand). But in contexts where stringent height restrictions exist, exceptions are often made for social housing. High-rise apartments may then be associated with poverty and crime, reinforcing the societies’ overall preference against high-rise apartments (Barr and Johnson, 2020). Lastly, that should mostly impact residential buildings as well as possibly central areas. Of course, with the right data, we could test which of the five identified factors could explain the differences in the residuals across regions. However, there does not exist global databases of land-use regulations and dispersed ownership rates in all cities across all countries. It is also, by construction, impossible to measure preferences. As such, we instead study the patterns established so far to try to make sense of the factors driving each subregion’s performance, mobilizing as much as possible the results by city size, centrality, building height, and building function as well as our skyline typology. 7.4.2. Suggestive Evidence for Developed Economies What could explain the patterns in the data? With the city-level analysis with all the controls including night lights, we obtain the following ranking among the subregions with large developed economies (Fig. 7): Eastern Europe, Western Europe, East Asia, Southern Europe, North America, West Asia, Northern Europe, and Oceania. Eastern Europe, ranked second among the 19 subregions, is ranked similarly for large/small cities (2), central areas (2, but for peripheral = 5), high-rises (2, but for skyscrapers = 10), and residential (2, but for offices = 6). Its best ranks are achieved for residential high-rises, which make up most of the tall building volume in the world. Eastern Europe provides an interesting case because high-rise apartment living was initially imposed by Soviet authorities in planned neighborhoods (Bertaud, 2018). However, these neighborhoods are classified as “central” in our analysis because they are still located within the main locality of each agglomeration (e.g., the city of Moscow 34 rather than urban satellites in the Moscow metro area). In major Eastern European cities today, whether Moscow, St. Petersburg or Warsaw, luxury high-rise apartments are now supplied by the private sector, and suburban living is possibly deemed less modern. More recently, there has also been a push towards skyscraper construction (Shuvalova, 2015). This could make one wonder to what extent (free market) preferences there were “shaped” by decades of heavily subsidized, or even imposed, high-rise apartment living. Western Europe and East Asia are then ranked 6th and 7th among the 19 subregions. In particular, Western Europe is ranked toward the top for small cities (3, but 13 for large cities), central and peripheral areas (6), high-rises (6, but 12 for skyscrapers), residential (6) and offices (5). Thus, Western European cities do not appear as vertical because they do not rank so well in terms of centrally located skyscrapers. But it is consistently ranked high for other dimensions, in particular residential high-rises and smaller cities. East Asia (rank = 7th) is ranked toward the top for large and small cities (7), central areas (7), offices (7), and skyscrapers (4), but lower for residential (9), high-rises (12) and peripheral (16). Overall, the largest cities in Japan; Korea; and Taiwan, China, as well as Hong Kong SAR, China, are undoubtedly vertical. This suggests that regulations are less stringent in such countries, even in more historical areas (e.g., Osaka has developed outside Kyoto). However, East Asian cities are not the most vertical in the world, ceteris paribus, especially given the fact that they include some of the largest cities in the world (Western European cities tend to be smaller). East Asian are highly ranked for skyscrapers, which are more visible, but lower ranked for residential high-rises, unlike Western Europe. Western Europe ranks better than East Asian cities for offices. Indeed, while some Western European cities are known for their stringent height restrictions in their central areas, for example Paris, these cities also have residential high-rises and business districts located in more peripheral areas. Many Western European cities are also pro-heights, for example in Germany and the Netherlands, where city sizes are not that large overall. In comparison, North America ranks 12th among the 19 subregions. It ranks higher for small cities (9) than large cities (12), peripheral areas (7) than central areas (12), skyscrapers (9) than high-rises (10), and higher for office buildings (11) than residential 35 buildings (12). Therefore, relative to the rest of the developed world, North America’s skylines appear to be disproportionately located in peripheral areas and small cities. In particular, regulations have become more stringent over time in the U.S. (Gyourko et al., 2008), as in New York following the adoption of the 1961 Zoning Resolution. The U.S. is also particularly suburban, and was so even before the White Flight of the 1950s- 1970s, due to the advent of the automobile from the 1920s (Jackson, 1987; Rappaport et al., 2003). The U.S. also has some of the largest and/or wealthiest cities in the world and many of these, such as Los Angeles and San Francisco, do not build tall (a significant share of tall buildings in New York and Chicago were also built before 1950). Except in a few cities, people still tend to associate high-rise apartment living with poverty and social ills (Gifford, 2007), and home voters are particularly “powerful” (Fischel, 2005). That is not the case in Asia where high-rise apartment living may be seen as more modern and luxurious than living in a suburban (i.e., more “rural”) house (Yeh and Yuen, 2011). Finally, West Asia is only ranked 13th among the 19 subregions. It is ranked similarly for large cities (10, but 17 for small cities), central areas (13, but 20 for peripheral areas) and residential buildings (11). It is ranked 7th for skyscrapers vs. 14th for high-rises and it performs more poorly for office buildings (15), suggesting that the Gulf nations have specialized in the luxury residential tall building sector, possibly at the expense of more peripheral urban locations and office buildings. This may have to do with their governments’ desire to project prestige and power in their central (and more politically important) city areas (Jedwab et al., 2022; Jedwab, 2022). Their skylines are highly visual but their cities appear to lack more “standard” tall buildings. A large share of their population also consists of migrants that tend to live in squalid conditions rather than modern buildings. As such, our conditional analysis reveals that West Asia may not be so pro-heights overall, just pro a certain type of heights. Indeed, the ruling dynasties in the wealthiest Gulf nations are also large developers (Jedwab, 2022). Their lack of land- use regulations may only be apparent if the same ruling families control what building is added to the skyline, thus blocking the construction by less politically connected firms of simpler (i.e., non-luxury) tall buildings for workers (including foreign workers). To summarize, land-use regulations and preferences likely account for most of the 36 observed differences. We do not believe that preservation and dispersed ownership play major roles, except maybe in some European subregions and North American cities. 7.4.3. Suggestive Evidence for Developing Economies With our best specification, eight subregions dominate the rankings (Figure 7): South America (1st ), Southest Asia (3rd ), East Africa (4rd ), Southern Africa (5th ), East Asia (7th ), Central America (8th ), Central Asia (9th ) and South Asia (10th ). The three other African regions (C, N and W) are nearly at the bottom. An important question is why cities in developing Asian and American nations are with a few exceptions better ranked than cities in African nations. Indeed, when computing for the six U.N. regions the average residuals with the best city-level specification (including the night lights controls), we obtained: Latin America = 0.9; Asia = 0.1; and Africa = -0.5. Relatedly, how much could we attribute the estimated differences to historical preservation? African cities are not as “historical” as Asian or American cities. Large African cities such as Abidjan, Accra, Dar es Salaam, Kinshasa, Lagos, Luanda, and Nairobi were founded, or only became true cities, in the late 19th century or at the turn of the 20th century. In comparison, Asian nations have existed for centuries and Latin America’s colonization process started at the turn of the 16th century. As such, historical preservation cannot explain why African regions tend to be less vertical. If anything, historical preservation should have led to larger gaps in Asia and Latin America. There are also no reasons to believe that dispersed ownership is behind the observed differences. Developing Asian, American, and African economies were likely not that different in 1950, with little manufacturing and a heavy economic reliance on natural resource exports. Their cities were also likely not that different in terms of population densities, levels of informality, and urban technologies and patterns more generally (AUE, 2016). Although we do not have the data to confirm this, we believe that plot sizes and the subdivision rate of existing units were unlikely to significantly differ then. This leaves two factors: land-use regulations unrelated to preservation and preferences. Do preferences differ between Asian, American, and African residents? It is difficult to say since preferences are not directly observable. South Africa has a lot in common with the U.S., with its history of segregation, urban decay in the central areas of cities, 37 and the White Flight, which led to the growth of suburban communities (The New York Times, 2019). East African cities were then historically less segregated than Southern African cities (Baruah et al., 2020). At the same time, Southern African cities experienced a massive construction boom in the 1960s-1970s (CNN, 2016) and East African cities as well as since 2010 (The Citizen, 2021). They did so at relatively low income levels and when their largest cities were still relatively small in comparison to the rest of the world. Among low-income economies, East Africa has the most vertical cities today, hence its high ranking, ceteris paribus. East African countries’ recent pro-heights stance has then been inspired by Chinese cities’ growth (Business Insider, 2015). In comparison, more stringent land-use regulations and the suburbanization of wealthier residents may be a feature of Northern, Central and Western African cities (e.g., Durand-Lasserve et al., 2015), but it is difficult to say without data (preferences are also not directly observable). By pro-vertical development-ness, we implicitly include the government’s will to clarify property rights, so that developers can develop or redevelop parcels without risk. Durand-Lasserve et al. (2015) suggest that competing claims for ownership that is unaddressed by government authorities could explain the lack of construction in West Africa. For example, Southern Africa, ranked 5th overall, is ranked similarly for large cities (5, but 12 for small cities), central areas (5, but 10 for peripheral areas), high-rises (6) and residential buildings (5). It is slightly better ranked for (skyscraper) office buildings (2), suggesting it has a surprisingly high number of 100m+ buildings given its income level and city population sizes. East Africa, ranked 4th overall, is ranked similarly for large cities (1), small cities (5), central areas (3, but 13 for peripheral areas), high-rises (3), skyscrapers (3) and office buildings (1). It is slightly worse ranked for residential buildings (7). Again, the top rank for large cities does not mean that East African cities are the most vertical cities in the world. It just means that for their low income level and relatively small large cities (in the order of 2-5 million vs. 2-40 million in developing Asia). In both cases, skylines comprise relatively more office buildings, suggesting an important part of the wealthier residential sector is suburban instead of urban. However, their relative lack of peripheral residential skyline is more than compensated by their central office skylines. In turn, it appears that the urban residents of developing Asian nations 38 do not mind living in tall buildings, for example in Southeast Asia (e.g., in Bangkok and Jakarta) and East Asia (e.g., in Shanghai and Chongqing). So residential preferences (combined with land-use regulations) could explain the observed differences. Southeast Asia (rank = 3 among the 19 subregions) is consistently ranked near the top for large cities (3), small cities (4), central areas (4), peripheral areas (1), high-rises (4), skyscrapers (1), and residential (3) and office buildings (4). We had found similar patterns for East Asian cities except that they were less specialized in residential high-rises. In comparison, South Asia (rank = 10 among the 19 subregions) is ranked in the middle for size (9 for large, but 11 for small cities), central areas (10), high-rises (8, but 11 for skyscrapers), and residential (8, but 13 for office buildings). In South Asia, skylines comprise relatively more residential high-rises in large cities, which could be consistent with the preferences mechanism whereby living in a house appears as more “rural.” Finally, land-use regulations likely vary between developing Asian, American, and Africa economies. Southeast Asian nations are particularly open to vertical development, possibly as a result of “Hong Kong SAR, China as a model city” effect highlighted by Ahlfeldt et al. (2022). This effect is being seen in East Asia as well (CNN, 2020). In Latin America, some cities appear to be particularly open to tall buildings, for example Panama ao Paulo. Even earthquake-prone cities such as Bogot´ City and S˜ a and Mexico City have business districts with numerous tall buildings (e.g., The New York Times, 2015). In particular, Central America, ranked 8th overall, is ranked similarly for large cities (6, but 13 for small cities), central areas (8, but 3 for peripheral areas), high-rises (9, but 6 for skyscrapers), residential buildings (10) and office buildings (8). Next, South America, ranked 1st overall, is similarly ranked for central areas (1), peripheral areas (2), high-rises (1) and residential buildings (1, with office buildings = 3). It is then higher ranked for small cities (1) than large cities (4) and high-rises (1) than skyscrapers (5). As discussed ao Paulo in particular have restrictions on very tall buildings, above, Brazilian cities and S˜ which may explain South America’s disproportionately “flat” skylines. In South and Central American cities, living in residential high-rises is seen as more modern and such cities also have tall business districts (NY Times, 2005; Sawyer et al., 2021). In contrast, most African cities outside Southern and East Africa have few tall 39 buildings, even high-rises (Emporis, 2022). Central, North and West Africa rank at the bottom in terms of high-rises. These differences likely reflect differences in land-use regulations and other constraints related to the permitting and construction processes.34 To summarize, land-use regulations and preferences, not historical preservation nor dispersed ownership, likely account for differences between developing subregions. 8. Conclusion In this paper, we aimed to quantify vertical urban development differences, or “gaps”, across world regions over time (1950-2020). To do so, we exploited a novel database on the location, height, and year of construction of all the tall buildings in the world. We used simple methodologies and various standard controls to estimate the gaps, i.e., the extent to which some regions of the world build “up” more, or less, than others given similar economic and geographic conditions, city size distributions, and other features. We found that many skylines may visually appear as more prominent internationally than they really are once one includes all tall buildings and core controls, which alters how regions are ranked in terms of tall building stocks. Latin America, Eastern Europe, and Western Europe are highly ranked, ceteris paribus. Asian regions and North America are not that well ranked. Within Asia, the leading region ceteris paribus is Southeast Asia, not East Asia. West Asia, which includes the Gulf nations, is not that well-ranked. Some African regions–East Africa and Southern Africa–are then relatively well-ranked. Using results by city size, centrality, height of buildings and building function, we then classified world regions into four different groups, finding that regions with capital- oriented central districts and population-oriented peripheral districts contain most regions in the world. However, with a few exceptions, the leading regions belong to the latter group. Therefore, world tall building stocks are driven by boring skylines of residential high-rises, and to a lesser extent exciting skylines of skyscrapers and office super-towers. Finally, land-use regulations and preferences likely account for most observed differences. 34 Another constraint could be the more unreliable supply of electricity in African nations. If tall buildings need electricity for their elevators and other machinery (e.g., HVAC units, water pumps, etc.), the lower reliability of electricity supply in Africa might impact vertical development there. The financial sector is also typically more developed in the Americas and Asia than in Africa (World Bank, 2020). 40 References Aghion, Philippe, Robin Burgess, Stephen J. Redding, and Fabrizio Zilibotti, “The Unequal Effects of Liberalization: Evidence from Dismantling the License Raj in India,” American Economic Review, September 2008, 98 (4), 1397–1412. Ahlfeldt, Gabriel M. and Daniel P. McMillen, “The vertical city: the price of land and the height of buildings in Chicago 1870-2010,” SERC discussion paper, 2015, 180. and Jason Barr, “The economics of skyscrapers: A synthesis,” CEPR Discussion Paper, 2020, 14987. and , “The economics of skyscrapers: A synthesis,” Journal of Urban Economics, 5 2022, 129, 103419. , Nathaniel Baum-Snow, and Remi Jedwab, “The Skyscraper Revolution: Global Economic Development and Land Savings,” 2022. Mimeo. Anas, Alex, Richard Arnott, and Kenneth A Small, “Urban spatial structure,” Journal of economic literature, 1998, 36 (3), 1426–1464. Angel, Shlomo, Jason Parent, and Daniel Civco, “The fragmentation of urban landscapes: global evidence of a key attribute of the spatial structure of cities, 1990-2000,” Environment and Urbanization, 2012, 24 (1), 249–283. , Stephen Sheppard, and Daniel Civco, The dynamics of global urban expansion 2005. AUE, Atlas of Urban Expansion. 2016. Bairoch, Paul, Cities and Economic Development: Frow the Dawn of History to the Present 1988. Barr, Jason, “Skyscrapers and the Skyline: Manhattan, 1895–2004,” Real Estate Economics, 2010, 38 (3), 567–597. , “Skyscraper Height,” Journal of Real Estate Finance and Economics, 2012, 45 (3). , “Skyscrapers and Skylines: New York and Chicago, 1885-2007,” Journal of Regional Science, 2013, 53 (3). , Building the Skyline: The Birth and Growth of Manhattan’s Skyscrapers, Oxford University Press, 2016. and Jeffrey P Cohen, “The floor area ratio gradient: New York City, 1890–2009,” Regional Science and Urban Economics, 2014, 48, 110–119. and Jennifer Johnson, “Skyscrapers and the happiness of cities,” Eastern Economic Journal, 2020, 46 (2). and Jingshu Luo, “Growing Skylines: The Economic Determinants of Skyscrapers in China,” The Journal of Real Estate Finance and Economics, 2020. , Troy Tassier, and Rossen Trendafilov, “Depth to Bedrock and the Formation of the Manhattan Skyline, 1890-1915,” The Journal of Economic History, 2011, 71 (4), 1060–1077. Baruah, Neeraj G, J Vernon Henderson, and Cong Peng, “Colonial legacies: Shaping African cities,” Journal of Economic Geography, 12 2020, 21 (1), 29–65. Bertaud, Alain, Order without design: How markets shape cities, Mit Press, 2018. and Jan Brueckner, “Analyzing building-height restrictions: predicted impacts and welfare costs,” Regional Science and Urban Economics, 2005, 35 (2), 109–125. and Stephen Malpezzi, “Measuring the costs and benefits of urban land use regulation: a simple model with an application to Malaysia,” Journal of Housing Economics, 2001, 10 (3), 393–418. Bolt, Jutta and Jan Luiten van Zanden, Maddison style estimates of the evolution of the world economy. A new 2020 update 2020. Brueckner, Jan K., “The structure of urban equilibria: A unified treatment of the muth-mills model,” Handbook of Regional and Urban Economics, 1987, 2, 821–845. and Ruchi Singh, “Stringency of Land-Use Regulation: Building Heights in US Cities,” Technical Report 2018. Brueckner, Jan K, Jacques-Francois Thisse, and Yves Zenou, “Why is central Paris rich and downtown Detroit poor?: An amenity-based theory,” European economic review, 1999, 43 (1), 91– 107. Brueckner, Jan K., Shihe Fu, Yizhen Gu, and Junfu Zhang, “Measuring the Stringency of Land Use Regulation: The Case of China’s Building Height Limits,” The Review of Economics and Statistics, 2017, 99 (4), 663–677. 41 Business Insider, “Africa’s carbon-copy cities show how much it wants to be the new China,” 2015. New York: Business Insider. Clapp, John M, “The elasticity of substitution for land: The effects of measurement errors,” Journal of Urban Economics, 1980, 8 (2), 255–263. CNN, “Johannesburg: City of skyscrapers,” 2016. Atlanta: CNN. CNN, The Shenzhen effect: Why China’s original ’model’ city matters more than ever, Published 22nd May 2020. Url: https://www.cnn.com/style/article/shenzhen-effect-china-model-city-intl-hnk/ index.html. Accessed 08-07-2022., 2020. Coulson, N Edward, Zhi Dong, and Tien Foo Sing, “Estimating supply functions for residential real estate attributes,” Real Estate Economics, 2021, 49 (2), 397–432. Durand-Lasserve, Alain, Maylis ¨ Durand-Lasserve, and Harris Selod, Land delivery systems in West African cities: The example of Bamako, Mali, World Bank Publications, 2015. Duranton, Gilles, “Determinants of City Growth in Colombia,” 2015. Working paper. Elvidge, Christopher D., Mikhail Zhizhin, Tilottama Ghosh, Feng-Chi Hsu, and Jay Taneja, “Annual Time Series of Global VIIRS Nighttime Lights Derived from Monthly Averages: 2012 to 2019,” Remote Sensing, 2021, 13 (5). Emporis, Emporis 2022. Eriksen, Michael D and Anthony W Orlando, “Returns to scale in residential construction: the marginal impact of building height,” Real Estate Economics, 2022, 50 (2), 534–564. Fischel, William A., The Homevoter Hypothesis: How Home Values Influence Local Government Taxation, School Finance, and Land-Use Policies, Cambridge, MA: Harvard University Press, 2005. Florczyk, A., M. Melchiorri, C. Corban, M. Schiavina, L. Maffenini, M. Pesaresi, P. Politis, F. Sabo, S. Carneiro Freire, D. Ehrlich, T. Kemper, P. Tommasi, D. Airaghi, and L. Zanchetta, “Description of the GHS Urban Centre Database 2015,” Publications Office of the European Union, 2019. Fu, Yuming and Song Shi, “Barriers to urban spatial development: Evidence from the 2010–2011 Christchurch earthquakes,” Journal of Regional Science, 2022. Fujita, Masahisa and Hideaki Ogawa, “Multiple equilibria and structural transition of non- monocentric urban configurations,” Regional Science and Urban Economics, 1982, 12 (2), 161–196. Garreau, Joel, Edge city: Life on the new frontier, Anchor, 2011. Garza, Nestor and Colin Lizieri, “Skyscrapers and the economy in Latin America,” Journal of Property Research, 2016, 33 (4), 269–292. Gechter, Michael and Nick Tsivanidis, “The welfare consequences of formalizing developing country cities: evidence from the Mumbai Mills redevelopment,” Work. Pap., Pa. State Univ./Dartmouth Coll. Google Scholar Article Locations: Article Location Article Location Article Location, 2018. Giardini, D, G Grunthal, KM Shedlock, and Zhang P, “The GSHAP Global Seismic Hazard Map,” Ann. Geophys., 1999, 42 (6). Gifford, Robert, “The consequences of living in high-rise buildings,” Architectural science review, 2007, 50 (1), 2–17. Gjerløw, Haakon and Carl Henrik Knutsen, “TRENDS: Leaders, Private Interests, and Socially Wasteful Projects: Skyscrapers in Democracies and Autocracies,” Political Research Quarterly, 2019, 72 (2), 504–520. Glaeser, Edward, Triumph of the City: How Our Greatest Invention Makes Us Richer, Smarter, Greener, Healthier, and Happier 2012. Glaeser, Edward L, Joseph Gyourko, and Raven E Saks, “Why have housing prices gone up?,” American Economic Review, 2005, 95 (2), 329–333. , , and , “Urban growth and housing supply,” Journal of economic geography, 2006, 6 (1), 71–89. Glaeser, Edward L., Joseph Gyourko, and Raven Saks, “Why Is Manhattan So Expensive? Regulation and the Rise in Housing Prices,” The Journal of Law and Economics, 2005, 48 (October), 331–369. GMTED, GMTED2010 Global Grids 2010. Gyourko, Joseph, Albert Saiz, and Anita Summers, “A new measure of the local regulatory environment for housing markets: The Wharton Residential Land Use Regulatory Index,” Urban Studies, 2008, 45 (3), 693–729. 42 and Raven Molloy, “Regulation and housing supply,” in “Handbook of regional and urban economics,” Vol. 5, Elsevier, 2015, pp. 1289–1337. Haila, Anne, “Real estate in global cities: Singapore and Hong Kong as property states,” Urban studies, 2000, 37 (12), 2241–2256. Harari, Mariaflavia and Maisy Wong, “Slum Upgrading and Long-run Urban Development: Evidence from Indonesia,” Technical Report 2018. Helsley, Robert W. and William C. Strange, “A game-theoretic analysis of skyscrapers,” Journal of Urban Economics, 7 2008, 64 (1), 49–64. Henderson, J Vernon, Tanner Regan, and Anthony Venables, “Building the city: from slums to a modern metropolis,” The Review of Economic Studies, 2021, 88 (3), 1157–1192. Hilber, Christian AL and Fr´ eric Robert-Nicoud, “On the origins of land use regulations: Theory ed´ and evidence from US metro areas,” Journal of Urban Economics, 2013, 75, 29–43. and Wouter Vermeulen, “The impact of supply constraints on house prices in England,” The Economic Journal, 2016, 126 (591), 358–405. Jackson, Kenneth T, Crabgrass frontier: The suburbanization of the United States, Oxford University Press, 1987. Jedwab, Remi, “Future Climate Change and Durable Real Estate: Is the World Over-Building in “Bad” Locations?,” 2022. Mimeo. , Elena Ianchovichina, and Federico Haslop, “Consumption Cities vs. Production Cities: New Considerations and Evidence,” 2022. Mimeo. , Jason Barr, and Jan Brueckner, “Cities without Skylines: Worldwide Building-height Gaps and their Implications,” 2020. , Prakash Loungani, and Anthony Yezer, “Comparing cities in developed and developing countries: Population, land area, building height and crowding,” Regional Science and Urban Economics, 1 2021. Koster, Hans R A, Jos van Ommeren, and Piet Rietveld, “Is the sky the limit? High-rise buildings and office rents,” Journal of Economic Geography, 5 2013, 14 (1), 125–153. Leahy, Joe, “S˜ao Paulo zoning laws spur a building boom,” 2018. Li, Qiang and Linlin Wang, “Is the Chinese skyscraper boom excessive?,” Journal of Urban Affairs, 2020, pp. 1–19. Lindenthal, Thies, Piet Eichholtz, and David Geltner, “Land assembly in Amsterdam, 1832–2015,” Regional Science and Urban Economics, 2017, 64, 57–67. Liu, Crocker H., Stuart S. Rosenthal, and William C. Strange, “The vertical city: Rent gradients, spatial structure, and agglomeration economies,” Journal of Urban Economics, 2018, 106, 101 – 122. Liu, Crocker H, Stuart S Rosenthal, and William C Strange, “The vertical city: Rent gradients, spatial structure, and agglomeration economies,” Journal of Urban Economics, 2018, 106, 101–122. , , and , “Employment density and agglomeration economies in tall buildings,” Regional Science and Urban Economics, 2020, 84, 103555. , , and , “Agglomeration Economies and the Built Environment: Evidence from Specialized Buildings and Anchor Tenants,” 2022. Maddison, Angus, Statistics on World Population, GDP and Per Capita GDP, 1-2008 AD 2008. Mahtta, Richa, Anjali Mahendra, and Karen C Seto, “Building up or spreading out? Typologies of urban growth across 478 cities of 1 million,” Environmental Research Letters, dec 2019, 14 (12), 124077. McDonald, John F, “Capital-land substitution in urban housing: A survey of empirical estimates,” Journal of urban Economics, 1981, 9 (2), 190–211. and Daniel P McMillen, Urban economics and real estate: theory and policy, John Wiley & Sons, 2010. McMillen, Daniel P, “Nonparametric Employment Subcenter Identification,” Journal of Urban Economics, 2001, 50 (3), 448–473. , “Testing for monocentricity,” A Companion to Urban Economics. Oxford: Blackwell, 2006, pp. 128– 140. and John F McDonald, “A nonparametric analysis of employment density in a polycentric city,” Journal of Regional Science, 1997, 37 (4), 591–612. Nase, Ilir, Nick van Assendelft, and Hilde Remøy, “Rent premiums and vertical sorting in 43 Amsterdam’s multi-tenant office buildings,” The Journal of Real Estate Finance and Economics, 2019, 59 (3), 419–460. NGDC, Global Radiance Calibrated Nighttime Lights 2015. NY Times, Living in style, and security, in Sao Paulo 2005. OEDC Development Center, “Economic Outlook for Southeast Asia, China and India 2020,” 2020. Peng, Liang and Thomas Thibodeau, “Government Interference and the Efficiency of the Land Market in China,” The Journal of Real Estate Finance and Economics, November 2012, 45 (4), 919– 938. Picken, David and Benedict Ilozor, “The Relationship between Building Height and Construction Costs,” in “Design Economics for the Built Environment” Wiley Online Books 3 2015, chapter 4, pp. 47–60. Rappaport, Jordan et al., “US urban decline and growth, 1950 to 2000,” Economic Review-Federal Reserve Bank of Kansas City, 2003, 88 (3), 15–44. Rosenthal, Stuart S and William C Strange, “The attenuation of human capital spillovers,” Journal of Urban Economics, 2008, 64 (2), 373–389. , , and Joaquin A Urrego, “JUE insight: Are city centers losing their appeal? Commercial real estate, urban spatial structure, and COVID-19,” Journal of Urban Economics, 2022, 127, 103381. Saiz, Albert, “The Geographic Determinants of Housing Supply,” The Quarterly Journal of Economics, 2010, 125 (3), 1253–1296. Saks, Raven E, “Job creation and housing construction: Constraints on metropolitan area employment growth,” Journal of Urban Economics, 2008, 64 (1), 178–195. Sawyer, Lindsay, Christian Schmid, Monika Streule, and Pascal Kallenberger, “Bypass urbanism: Re-ordering center-periphery relations in Kolkata, Lagos and Mexico City,” Environment and Planning A: Economy and Space, 2021, 53 (4), 675–703. Shangguan, Wei, Tomislav Hengl, Jorge Mendes de Jesus, Hua Yuan, and Yongjiu Dai, “Mapping the global depth to bedrock for land surface modeling,” Journal of Advances in Modeling Earth Systems, 2017, 9 (1), 65–88. Shuvalova, Elena, High-rise construction in Russia: Asian way vs. Middle Way 2015. Siodla, James, “Razing San Francisco: The 1906 disaster as a natural experiment in urban redevelopment,” Journal of Urban Economics, 2015, 89, 48–61. The Citizen, “Africa’s skyline undergoes major face-lift worth billions,” 2021. Johannesburg. The New York Times, “Modern Tower Shelters a Piece of Mexico City’s Past to Counter Critics,” 2015. New York: The New York Times. , “In South Africa, a Sense of Security Draws Many to Golf Communities,” 2019. The New York Times. Titman, Sheridan and Garry Twite, “Urban density, law and the duration of real estate leases,” Journal of Urban Economics, 2013, 74, 99–112. U.N., World Cities Report: The Value of Sustainable Urbanization 2020. United Nations, Patterns of Urban and Rural Population Growth 1980. , World Urbanization Prospects: The 2018 Revision 2018. World Bank, World Development Report: Reshaping Economic Geography 2009. , Doing Business Comparing Business Regulation in 190 Economies 2020. , World Development Indicators 2022. Yeh, Anthony GO and Belinda Yuen, “Introduction: High-rise living in Asian cities,” in “High-rise living in Asian cities,” Springer, 2011, pp. 1–8. Yoshida, Jiro, “Structure depreciation and the production of real estate services,” Available at SSRN 2734555, 2016. Zadornin, D and K Sheina, “Towards a Typology of Soviet Mass Housing Prefabrication in the USSR 1955–1991,” 2015. Zheng, Siqi, Weizeng Sun, Jianfeng Wu, and Matthew E. Kahn, “The birth of edge cities in China: Measuring the effects of industrial parks policy,” Journal of Urban Economics, 2017, 100 (C), 80– 103. 44 Figure 1: Total Sum of Tall Building Heights Per Capita, 12,877 World Cities, c. 2015 Notes: This figure shows for 12,877 urban agglomeration of at least 50,000 residents c. 2015 their total stock of tall building heights per capita (meters per thousand residents) c. 2015. To obtain total building heights, note that we only consider tall buildings above 55 meters in the Emporis database. Map cleared by the map clearance department of the World Bank on Dec 6, 2022 (Job 87949). Figure 2: Evolution of the Total Sum of Tall Building Heights, World vs. USA, 1890-2020 (a) Total Heights, 1885-2020 (b) Total Heights Per Urban Capita, 1900-2020 Notes: The left panel shows for the world and the United States separately the total sum of tall building heights (km) from 1885 to 2020 (note that we only consider completed buildings above 55 meters). The right panel shows for the world and the United States separately the total sum of tall building heights per urban capita (km per million residents) from 1900 to 2020. Sources for the total urban population data: United Nations (1980, 2018) as well as the Wikipedia webpage: “Urbanization in the United States .” 45 Figure 3: Heights per Urban Capita, United Nations Regions, 1950-2020 Notes: This figure shows the evolution of the urban share (%) and the total sum of heights per urban capita (km per million) for the six United Nations “regions” every 5 years from 1950 to 2020. Figure 4: Heights and Income, Four Selected Subregions, 1950-2020 Notes: This figure shows for Southeast Asia (green hollow diamonds), East Asia (pink hollow circles), South America (red hollow squares) and Eastern Europe (blue hollow triangles) the relationship between tall building heights per urban capita (meters per million) and log national per capita GDP (PPP and 2011 cst international dollars) from 1950 to 2020. The observations for all subregion-years are shown in grey. 46 Figure 5: Tall Building Construction Residuals, c. 2020, Country-Level Analysis Notes: This figure shows the tall building construction residuals estimated using the country-level regressions and with various country-level controls. To center the variable around 0 in 2020, we remove the mean of the variable across the 19 subregions in 2020. See text for details on the controls. Figure 6: Residuals, Mean Five-Year Change 1950-2020, Country-Level Analysis Notes: This figure shows the mean five-year change in the residuals during the 1950-2020 period. To center the variable around 0 in 2020, we remove the mean of the variable across the 19 subregions in 2020. 47 Figure 7: Rank of Residuals c. 2020, Country vs. City-Level Analysis Notes: “Cntry: Income:” Country-level analysis controlling for national per capita GDP in year t and its square. “Cntry: Geo:” Country-level analysis with the full specification with all country-level controls. “City: Geo.:” City-level analysis with the full specification with all country- and city-level controls. “City: Geo+Lights:” City-level analysis with the full specification also including the night lights controls. Figure 8: Rank of Change in the Residuals, Country vs. City-Level Analysis Notes: Period for: (i) the country analysis (first three columns): 1950-2020; (ii) the city-level analysis without night lights (fourth column; “City: Geo”): 1975-2020; and (iii) the city-level analysis with night lights (fifth column; “City: Geo+Lights”): 1995-2020. See the notes of Figure 7 for details on each specification. 48 Figure 9: Residuals for Cities Above vs. Below 2 Million Residents, c. 2020 Notes: This figure shows for each subregion and the year 2020 the (centered) residuals estimated for cities above 2 million residents vs. cities below 2 million residents (as of 2015). The residuals are based on the city-level analysis and the most complete specification with the night light controls (“City: Geo+Lights”). Figure 10: Residuals for Central vs. Peripheral City Areas, c. 2020 Notes: This figure shows for each subregion and the year 2020 the (centered) residuals estimated for the central areas vs. peripheral areas of urban agglomerations. The residuals are based on the city-level analysis and the most complete specification with the night light controls (“City: Geo+Lights”). 49 Figure 11: Residuals for High-Rises vs. Skyscrapers, c. 2020 Notes: This figure shows for each subregion and the year 2020 the (centered) residuals estimated for high- rises (between 55 and 100 meters) vs. skyscrapers (above 100 meters). The residuals are based on the city- level analysis and the most complete specification with the night light controls (“City: Geo+Lights”). Figure 12: Residuals for Residential vs. Office Towers, c. 2020 Notes: This figure shows for each subregion and the year 2020 the (centered) residuals estimated for residential buildings vs. (private) office buildings. The residuals are based on the city-level analysis and the most complete specification with the night light controls (“City: Geo+Lights”). 50 Figure 13: Classification of Subregions Based on their Skyline Characteristics, c. 2020 Notes: This figure shows how we classify the 19 U.N. subregions in four groups depending on the characteristics of their skyline c. 2020. CD vs. PD = “central” districts vs. “peripheral” districts, i.e., whether it is larger cities and/or central city areas OR smaller cities and/or peripheral city areas that disproportionately have tall buildings within the subregion. To obtain this measure, we use for each region the max of the deviation from the regression line in the figures for small cities vs. large cities (Figure 9) and for peripheral vs. central areas (Figure 10). Capital-Oriented vs. Population-Oriented = whether it is skyscrapers and/or office towers OR high-rises and/or residential towers that are disproportionately tall in the subregion. To obtain this measure, we use for each region the max of the deviation from the regression line in the figures for high-rises vs. skyscrapers (Figure 11) and for office towers vs. residential towers (Figure 12). Table 1: Determinants of City Tall Building Stocks c. 2020, Global Analysis Dependent Variable: Log of (Sum of Tall Building Heights + 1) c. 2020 (N = 12,753; R2 = 0.60) 100-250*H 2.14*** 100-250*UM 0.09* 100-250*LM -0.04 100-250*L -0.04 Land Area (Mil. Sq Km) -0.23*** (0.15) (0.05) (0.02) (0.02) (0.03) 250-500*H 4.25*** 250-500*UM 1.29*** 250-500*LM 0.05 250-500*L -0.13* Land Area Sq 0.02*** (0.22) (0.13) (0.06) (0.08) (0.00) 500-750*H 6.09*** 500-750*UM 2.41*** 500-750*LM 0.05 500-750*L -0.47*** Earthquake Risk (m/s2 ) -0.23*** (0.27) (0.27) (0.12) (0.09) (0.03) 750-1000*H 6.50*** 750-1000*UM 3.49*** 750-1000*LM 0.41 750-1000*L -0.44*** Earthquake Risk Sq 0.02*** (0.45) (0.43) (0.26) (0.12) (0.01) 1000-2500*H 7.50*** 1000-2500*UM 5.51*** 1000-2500*LM 1.58*** 1000-2500*L 2.67*** Bedrock Depth (meters) -0.01*** (0.23) (0.31) (0.29) (0.80) (0.00) 2500-5000*H 8.38*** 2500-5000*UM 8.47*** 2500-5000*LM 3.49*** 2500-5000*L 2.81*** Bedrock Depth Sq 0.00* (0.29) (0.22) (0.65) (0.98) (0.00) 5000-7500*H 10.03*** 5000-7500*UM 8.96*** 5000-7500*LM 5.14*** 5000-7500*L 5.84*** Altitude (meters) -0.00*** (0.46) (0.64) (1.00) (0.74) (0.00) 7500-10000*H 9.76*** 7500-10000*UM 9.43*** 7500-10000*LM 8.00*** Log Sum Lights 0.13*** Altitude Sq 0.00** (0.44) (0.16) (0.15) (0.01) (0.00) 10000+*H 11.06*** 10000+*UM 9.98*** 10000+*LM 8.88*** Nat. Pop (Mil.) 0.00*** Ruggedness (meters) 0.00 (0.62) (0.47) (0.55) (0.00) (0.00) Nat. Pop Sq -0.00*** Ruggedness Sq 0.00 (0.00) (0.00) Notes: Obs. = 12,753 urban agglomerations of at least 50,000 inhabitants c. 2015 (163 countries). We classify the cities into 10 categories based on their population size c. 2015. We then interact the 10 population size category dummies with four dummies if the city belongs to a low-income 51 country (L), a lower-middle income country (LM), an upper-middle income country (UM) or a high-income country (H) c. 2020 (based on the classification of the World Bank). We control for the log of total VIIRS-based lights in the city in 2019, the total pop. (millions) and land area (million sq km) of the city’s country in 2020, and their squares, and city-level measures of earthquake risk (PGA; m/s2), mean bedrock depth (meters), mean altitude (meters), and ruggedness proxied by the standard deviation of altitude (meters) in the city, as well as their squares. 52 Web Appendix Figures and Table (NOT FOR PUBLICATION) Figure A.1: Distribution of Building Heights (m) in Emporis, World, c. 2022 Notes: This figure shows the Kernel distribution of heights (meters) for all 693,855 “existing [completed]” buildings in Emporis (accessed 02-07-2022). We only consider buildings with the following type: “building with towers”, “high-rise building”, “low-rise building”, “multi-story building”, and “skyscraper”. Restricting the sample to buildings above 55 m, we obtain 266,336 buildings. Figure A.2: Heights per Urban Capita, World Bank Regions, Developing Economies Only Notes: This figure shows the evolution of the urban share (%) and the total sum of heights per urban capita (km per million) for the six World Bank “regions” every 5 years from 1950 to 2020. We restrict the descriptive analysis to developing economies according to the classification of the World Bank in 2020. Developing economies are economies that the World Bank does not classify as “high-income economies”. 53 Figure A.3: Heights and Income, South Asia vs. South East-Asia, 1950-2020 Notes: This figure shows for Southeast Asia (blue hollow squares) and another UN subregion (red hollow circles) the relationship between tall building heights per urban capita (meters per million) and log national per capita GDP (PPP and 2011 constant international dollars) for 20 UN subregions from 1950 to 2020. Figure A.4: Heights and Income, West Asia vs. South East-Asia, 1950-2020 Notes: This figure shows for Southeast Asia (blue hollow squares) and another UN subregion (red hollow circles) the relationship between tall building heights per urban capita (meters per million) and log national per capita GDP (PPP and 2011 constant international dollars) for 20 UN subregions from 1950 to 2020. 54 Figure A.5: Heights and Income, East Africa vs. South East-Asia, 1950-2020 Notes: This figure shows for Southeast Asia (blue hollow squares) and another UN subregion (red hollow circles) the relationship between tall building heights per urban capita (meters per million) and log national per capita GDP (PPP and 2011 constant international dollars) for 20 UN subregions from 1950 to 2020. Figure A.6: Heights and Income, Southern Africa vs. South East-Asia, 1950-2020 Notes: This figure shows for Southeast Asia (blue hollow squares) and another UN subregion (red hollow circles) the relationship between tall building heights per urban capita (meters per million) and log national per capita GDP (PPP and 2011 constant international dollars) for 20 UN subregions from 1950 to 2020. 55 Figure A.7: Heights and Income, West Africa vs. South East-Asia, 1950-2020 Notes: This figure shows for Southeast Asia (blue hollow squares) and another UN subregion (red hollow circles) the relationship between tall building heights per urban capita (meters per million) and log national per capita GDP (PPP and 2011 constant international dollars) for 20 UN subregions from 1950 to 2020. Figure A.8: Heights and Income, North America vs. South East-Asia, 1950-2020 Notes: This figure shows for Southeast Asia (blue hollow squares) and another UN subregion (red hollow circles) the relationship between tall building heights per urban capita (meters per million) and log national per capita GDP (PPP and 2011 constant international dollars) for 20 UN subregions from 1950 to 2020. 56 Figure A.9: Heights and Income, Western Europe vs. South East-Asia, 1950-2020 Notes: This figure shows for Southeast Asia (blue hollow squares) and another UN subregion (red hollow circles) the relationship between tall building heights per urban capita (meters per million) and log national per capita GDP (PPP and 2011 constant international dollars) for 20 UN subregions from 1950 to 2020. Figure A.10: Heights per Urban Capita and Income, 2020 vs. 1950 Notes: This figure shows for 183 countries in both 2020 and 1950 the relationship between the total sum of tall building heights per urban capita (meters per million) and log national per capita GDP (PPP and 2011 constant international dollars). Note that we only show the observations for the year 2020. 57 Figure A.11: City Population Size, National Income, and Heights c. 2020 Notes: This figure shows for four income groups c. 2020–low, lower-middle, upper-middle and high-income groups (based on the classification of the World Bank)–the conditional relationship between the log sum of tall building heights c. 2020 and city population size categorized into 10 groups c. 2015. Figure A.12: Relations between Heights and Six Country/City Variables (a) National Population (b) National Land Area (c) Earthquake Risk (d) Bedrock Depth (e) Mean Altitude (f) Ruggedness (SD of Altitude) Notes: These figures show the relations between the log sum of tall building heights c. 2020 and six selected country- or city-level variables (based on the 25th, 50th and 75th percentile values of each variable). 58 Figure A.13: Tall Building Construction Residuals, City-Level Analysis, c. 2020 Notes: This figure shows the tall building construction residuals estimated using the city-level regressions and with various country- and city-level controls. To center the variable around 0 in 2020, we remove the mean of the variable across the 19 subregions in 2020. See text for details on the controls. Figure A.14: Residuals, City-Level Analysis, Mean Five-Year Change 1975-2015 Notes: This figure shows for each UN subregion the mean fifteen-year change in the residuals during the 1975-2015 period (data for 12,877 agglomerations in 183 countries). To center the variable around 0 in 2020, we remove the mean of the variable across the 19 subregions in 2020. See text for details on the controls. 59 Figure A.15: Residuals, City-Level Analysis, Night Lights Controls, c. 2020 Notes: This figure shows the tall building construction residuals estimated using the city-level regressions and with various country- and city-level controls incl. night lights. To center the variable around 0 in 2020, we remove the mean of the variable across the 19 subregions in 2020. See text for details on the controls. Figure A.16: Residuals, City-Level Analysis, Mean Five-Year Change 1995-2020 Notes: This figure shows for each UN subregion the mean fifteen-year change in the residuals during the 1995-2015 period (data for 12,877 agglomerations in 183 countries). To center the variable around 0 in 2020, we remove the mean of the variable across the 19 subregions in 2020. See text for details on the controls. 60 Figure A.17: Rankings c. 2020, Different Building Height Thresholds Notes: This figure shows the c. 2020 rankings when the log total sum of tall building heights in the city, and the residuals, are based on different tall building thresholds. The rankings are based on the city-level analysis and the most complete specification with the night light controls (“City: Geo+Lights”). Figure A.18: Rankings, Mean Five-Year Change 1995-2020, Different Thresholds Notes: This figure shows mean five-year change in the residuals when the log total sum of tall building heights in the city is based on different tall building thresholds. The rankings are based on the city-level analysis and the most complete specification with the night light controls (“City: Geo+Lights”). 61 Figure A.19: Rankings by City Size and Within-City Centrality, c. 2020 Notes: This figure shows the c. 2020 rankings by city size (above vs. below 2 million residents as of 2015) and by centrality (for central areas vs. peripheral areas). The rankings are based on the city-level analysis and the most complete specification with the night light controls (“City: Geo+Lights”). Figure A.20: Rankings by Building Height and Building Function, c. 2020 Notes: This figure shows the c. 2020 rankings by building height (high rises: 55-100 m vs. skyscrapers: above 100 m) and by building function (residential vs. private office). The rankings are based on the city- level analysis and the most complete specification with the night light controls (“City: Geo+Lights”). 62 Table A.1: List of Countries by U.N. Region and Subregion U.N. Subregion U.N. Region Country Pop. 2020 (Mil.) Income Group 2020 Eastern Africa Africa Ethiopia 112.8 L Eastern Africa Africa Tanzania 62.8 L Eastern Africa Africa Kenya 53.5 LM Eastern Africa Africa Uganda 47.2 L Eastern Africa Africa Mozambique 32.3 L Eastern Africa Africa Madagascar 27.7 L Eastern Africa Africa Malawi 20.3 L Eastern Africa Africa Zambia 18.7 LM Eastern Africa Africa Zimbabwe 17.7 LM Eastern Africa Africa Somalia 16.1 L Eastern Africa Africa South Sudan 13.6 L Eastern Africa Africa Rwanda 13.1 L Eastern Africa Africa Burundi 11.9 L Eastern Africa Africa Eritrea 5.4 L Eastern Africa Africa Mauritius 1.3 UM Eastern Africa Africa Djibouti 1 LM Eastern Africa Africa Comoros 0.9 LM Middle Africa Africa Congo, Dem. Rep. 89.5 L Middle Africa Africa Angola 32.8 LM Middle Africa Africa Cameroon 26 LM Middle Africa Africa Chad 16.3 L Middle Africa Africa Congo, Rep. 5.7 LM Middle Africa Africa Central African Republic 4.9 L Middle Africa Africa Gabon 2.2 UM Middle Africa Africa Equatorial Guinea 1.4 UM Middle Africa Africa ao Tom´ S˜ ıncipe e and Pr´ 0.2 LM Northern Africa Africa Egypt, Arab Rep. 102.9 LM Northern Africa Africa Sudan 43.5 LM Northern Africa Africa Algeria 43.3 UM Northern Africa Africa Morocco 37.1 LM Northern Africa Africa Tunisia 11.9 LM Northern Africa Africa Libya 6.7 UM Northern Africa Africa Western Sahara 0.6 LM Southern Africa Africa South Africa 58.7 UM Southern Africa Africa Namibia 2.7 UM Southern Africa Africa Botswana 2.4 UM Southern Africa Africa Lesotho 2.3 LM Southern Africa Africa Eswatini 1.4 LM Western Africa Africa Nigeria 206.2 LM Western Africa Africa Ghana 30.7 LM Western Africa Africa ˆ d’Ivoire Cote 26.2 LM Western Africa Africa Niger 24.1 L Western Africa Africa Burkina Faso 20.9 L Western Africa Africa Mali 20.3 L Western Africa Africa Senegal 17.2 LM Western Africa Africa Guinea 13.8 L Western Africa Africa Benin 12.1 L Western Africa Africa Togo 8.4 L Western Africa Africa Sierra Leone 8 L Western Africa Africa Liberia 5.1 L Western Africa Africa Mauritania 4.8 LM Western Africa Africa Gambia, The 2.3 L Western Africa Africa Guinea-Bissau 2 L Western Africa Africa Cabo Verde 0.6 LM Central Asia Asia Uzbekistan 33.2 LM Central Asia Asia Kazakhstan 18.8 UM Central Asia Asia Tajikistan 9.5 L Central Asia Asia Kyrgyz Republic 6.3 LM Central Asia Asia Turkmenistan 6 UM Notes: Income Group 2020 (income group circa 2020 according to the classification of the World Bank): L = low-income; LM = lower-middle income; UM = upper-middle income; and H = high-income. 63 U.N. Subregion U.N. Region Country Pop. 2020 (Mil.) Income Group 2020 Eastern Asia Asia China 1424.5 UM Eastern Asia Asia Japan 126.5 H Eastern Asia Asia Korea, Rep. 51.5 H Eastern Asia Asia Korea, Dem. People’s Rep. 25.8 L Eastern Asia Asia Taiwan, China 23.8 H Eastern Asia Asia Hong Kong SAR, China 7.5 H Eastern Asia Asia Mongolia 3.2 LM South-Eastern Asia Asia Indonesia 272.2 LM South-Eastern Asia Asia Philippines 109.7 LM South-Eastern Asia Asia Vietnam 98.4 LM South-Eastern Asia Asia Thailand 69.4 UM South-Eastern Asia Asia Myanmar 54.8 LM South-Eastern Asia Asia Malaysia 32.9 UM South-Eastern Asia Asia Cambodia 16.7 LM South-Eastern Asia Asia Lao PDR 7.2 LM South-Eastern Asia Asia Singapore 5.9 H South-Eastern Asia Asia Timor-Leste 1.4 LM South-Eastern Asia Asia Brunei Darussalam 0.4 H Southern Asia Asia India 1383.2 LM Southern Asia Asia Pakistan 208.4 LM Southern Asia Asia Bangladesh 169.8 LM Southern Asia Asia Iran, Islamic Rep. 83.6 UM Southern Asia Asia Afghanistan 38.1 L Southern Asia Asia Nepal 30.3 L Southern Asia Asia Sri Lanka 21.1 UM Southern Asia Asia Bhutan 0.8 LM Southern Asia Asia Maldives 0.5 UM Western Asia Asia ¨ Turkiye 83.8 UM Western Asia Asia Iraq 41.5 UM Western Asia Asia Saudi Arabia 34.7 H Western Asia Asia Yemen, Rep. 30.2 L Western Asia Asia Syrian Arab Republic 18.9 L Western Asia Asia Jordan 10.2 UM Western Asia Asia Azerbaijan 10.1 UM Western Asia Asia United Arab Emirates 9.8 H Western Asia Asia Israel 8.7 H Western Asia Asia Lebanon 6 UM Western Asia Asia West Bank and Gaza 5.3 LM Western Asia Asia Oman 5.2 H Western Asia Asia Kuwait 4.3 H Western Asia Asia Georgia 3.9 UM Western Asia Asia Armenia 2.9 UM Western Asia Asia Qatar 2.8 H Western Asia Asia Bahrain 1.7 H Western Asia Asia Cyprus 1.2 H Eastern Europe Europe Russian Federation 143.8 UM Eastern Europe Europe Ukraine 43.6 LM Eastern Europe Europe Poland 37.9 H Eastern Europe Europe Romania 19.4 UM Eastern Europe Europe Czech Republic 10.6 H Eastern Europe Europe Hungary 9.6 H Eastern Europe Europe Belarus 9.4 UM Eastern Europe Europe Bulgaria 6.9 UM Eastern Europe Europe Slovak Republic 5.5 H Eastern Europe Europe Moldova 4 LM Northern Europe Europe United Kingdom 67.3 H Northern Europe Europe Sweden 10.1 H Northern Europe Europe Denmark 5.8 H Northern Europe Europe Finland 5.6 H Northern Europe Europe Norway 5.4 H Northern Europe Europe Ireland 4.9 H Northern Europe Europe Lithuania 2.9 H Northern Europe Europe Latvia 1.9 H Northern Europe Europe Estonia 1.3 H Northern Europe Europe Iceland 0.3 H 64 U.N. Subregion U.N. Region Country Pop. 2020 (Mil.) Income Group 2020 Southern Europe Europe Italy 59.1 H Southern Europe Europe Spain 46.5 H Southern Europe Europe Greece 11.1 H Southern Europe Europe Portugal 10.2 H Southern Europe Europe Serbia 8.7 UM Southern Europe Europe Croatia 4.1 H Southern Europe Europe Bosnia & Herzegovina 3.5 UM Southern Europe Europe Albania 2.9 UM Southern Europe Europe North Macedonia 2.1 UM Southern Europe Europe Slovenia 2.1 H Southern Europe Europe Montenegro 0.6 UM Southern Europe Europe Malta 0.4 H Southern Europe Europe Kosovo 1.9 UM Western Europe Europe Germany 82.5 H Western Europe Europe France 65.7 H Western Europe Europe Netherlands 17.2 H Western Europe Europe Belgium 11.6 H Western Europe Europe Austria 8.8 H Western Europe Europe Switzerland 8.7 H Western Europe Europe Luxembourg 0.6 H Western Europe Europe Jersey 0.1 H Caribbean LAC Cuba 11.5 UM Caribbean LAC Haiti 11.4 L Caribbean LAC Dominican Republic 11.1 UM Caribbean LAC Puerto Rico 3.7 H Caribbean LAC Jamaica 2.9 UM Caribbean LAC Trinidad and Tobago 1.4 H Caribbean LAC Bahamas, The 0.4 H Caribbean LAC Barbados 0.3 H Caribbean LAC Curacao 0.2 H Central America LAC Mexico 133.9 UM Central America LAC Guatemala 17.9 UM Central America LAC Honduras 9.7 LM Central America LAC El Salvador 6.5 LM Central America LAC Nicaragua 6.4 LM Central America LAC Costa Rica 5 UM Central America LAC Panama 4.3 H Central America LAC Belize 0.4 UM South America LAC Brazil 213.9 UM South America LAC Colombia 50.2 UM South America LAC Argentina 45.5 UM South America LAC Peru 33.3 UM South America LAC Venezuela, RB 33.2 UM South America LAC Chile 18.5 H South America LAC Ecuador 17.3 UM South America LAC Bolivia 11.5 LM South America LAC Paraguay 7.1 UM South America LAC Uruguay 3.5 H South America LAC Guyana 0.8 UM South America LAC Suriname 0.6 UM North America North Am. United States 331.4 H North America North Am. Canada 37.6 H Australia/New Zealand Oceania Australia 25.4 H Australia/New Zealand Oceania New Zealand 4.8 H Melanesia Oceania Papua New Guinea 8.8 LM Melanesia Oceania Fiji 0.9 UM Melanesia Oceania Solomon Islands 0.6 LM Melanesia Oceania French Polynesia 0.3 H Melanesia Oceania New Caledonia 0.3 H