Policy Research Working Paper 10105 Consumption Cities versus Production Cities New Considerations and Evidence Remi Jedwab Elena Ianchovichina Federico Haslop Latin America and the Caribbean Region Office of the Chief Economist June 2022 Policy Research Working Paper 10105 Abstract Cities dramatically vary in their sectoral composition across Compared to cities in industrialized countries, cities of the world, possibly lending credence to the theory that similar sizes in resource-rich and deindustrializing coun- some cities are production cities with high employment tries have lower shares of employment in manufacturing, shares of urban tradables while others are consumption tradable services, and the formal sector, and higher shares cities with high employment shares of urban non-tradables. of employment in non-tradables and the informal sector. A model of structural change highlights three paths lead- Results on the construction of “vanitous” tall buildings ing to the rise of consumption cities: resource rents from provide additional evidence on the relationship between exporting fuels and mining products, agricultural exports, resource exports and consumption cities. Finally, the evi- and premature deindustrialization. These findings appear dence suggests that having mostly consumption cities might to be corroborated using both country- and city-level data. have economic implications for a country. This paper is a product of the Office of the Chief Economist, Latin America and the Caribbean Region. 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 jedwab@email.gwu.edu, eianchovichina@worldbank.org, or fhaslop@gwmail.gwu.edu. 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 Consumption Cities versus Production Cities: New Considerations and Evidence Remi Jedwab, Elena Ianchovichina and Federico Haslop∗ JEL Codes: O11; E24; E26; O13; O14; O18; R1; R12 Keywords: Structural Change; Urbanization; Consumption Cities; Macro-Development Economics; Industrialization; Natural Resources; Deindustrialization; Construction ∗ Corresponding author: Remi Jedwab: Department of Economics, George Washington University. Elena Ianchovichina: Office of the Chief Economist for Latin America and the Caribbean, The World Bank. Federico Haslop: Department of Economics, George Washington University. We thank Martina Kirchberger, Somik Lall, Justin Yifu Lin, William Maloney, Tommaso Porzio, Luis Quintero, Martin Rama, Xiaodong Zhu and seminar audiences at George Washington, the International Annual Workshop of New Structural Economics (Beijing), the Urban Economics Association meetings, and the World Bank for helpful comments. We are grateful to Andrew Durrer for his research assistance. The findings, interpretations, and conclusions expressed in this paper are entirely ours. 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 or the governments they represent. 1 The period since the early 1960s has been one of rapid urbanization in developing countries (World Bank, 2009; United Nations, 2018). This process has been linked to a virtuous circle between economic development and urbanization (Henderson, 2010; Duranton, 2015). However, urbanization can also proceed without growth (Bairoch, 1988; Fay and Opal, 2000; Glaeser, 2014; Jedwab and Vollrath, 2015; Castells-Quintana, 2017). In the macro-development literature, urbanization is often a by-product of structural change. As countries develop, people move out of agriculture and engage in urban- based manufacturing and service activities (Herrendorf et al., 2014). Structural change occurs due to Green Revolutions, which increase food productivity and push labor out of agriculture (Matsuyama, 1992; Gollin, Parente and Rogerson, 2002; Restuccia, Yang and Zhu, 2008; Yang and Zhu, 2013; Lagakos and Waugh, 2013; Gollin, Hansen and Wingender, 2018). In addition, Industrial Revolutions or Service Revolutions increase manufacturing or service productivity and pull labor out of agriculture (Hansen and Prescott, 2002; Lucas, 2004; Alvarez-Cuadrado and Poschke, 2011).1 Gollin et al. (2016) (henceforth GJV16) show that only in some developing countries rapid urbanization has been accompanied by industrialization, following the historical patterns observed in richer nations. Manufacturing and tradable services agglomerated in production cities whose growth was driven by the countries’ increased production capacity. In others, the spending of resource rents on urban goods and services led to consumption cities whose growth was driven by increased consumption capacity. Since manufactured goods and tradable services – urban tradables – are often imported, non- tradable services – urban non-tradables – dominate their sectoral composition. This paper provides new evidence on the “origin” of consumption cities. Whether a country has production cities or consumption cities could then matter for economic development. Productivity in manufacturing and non-traditional services varies little across countries, and productivity catch-up in these sectors may explain some of the large gains in aggregate productivity observed across countries (Duarte and Restuccia, 2010, 2020). Hence, low productivity in urban non-tradables in developing economies, and thus the possible lack of catch-up in this sector, might contribute to explaining the lack of 1 Other mechanisms include quality-of-life amenities (Jedwab and Vollrath, 2019; Gollin, Kirchberger and Lagakos, 2021), demographic growth (Jedwab et al., 2017; Castells-Quintana and Wenban-Smith, 2020; Jedwab et al., 2021a), urban-biased policies (Lipton, 1977; Ades and Glaeser, 1995; Davis and Henderson, 2003; Castells-Quintana, 2017), trade (Glaeser, 2014; Venables, 2017), technology (Jedwab et al., 2021c), and disasters (Barrios et al., 2006; Henderson et al., 2017; Jedwab et al., 2021b; Castells-Quintana et al., 2021). 2 international convergence. In addition, returns to education and work experience, i.e. the rates of human capital accumulation at work, may vary across sectors (Islam et al., 2019). Likewise, agglomeration economies may also vary by sectors (Venables, 2017; Burger et al., 2022). That may also be the case for congestion externalities. Finally, international trade affects endogenous innovation and growth (Melitz and Redding, 2021). We first establish that, for a given population size and a given level of urban economic development, cities dramatically vary in their sectoral composition across the world. Using census data for about 65 countries, we obtain the sectoral composition of 7,000 agglomerations comprising three-fourths of the world’s urban population. We classify them as “production cities”, “consumption cities”, or “neutral cities”, depending on their employment share of manufacturing and tradable services. We also use our classification to highlight novel stylized facts related to the global distribution of urban employment and show that cities, not just countries, experience different patterns of structural change. We extend the theoretical analysis of GJV16, who show that countries can urbanize because of industrialization broadly defined (which includes tradable, or “industrialized”, services) or because they export natural resources. An increase in resource export earnings raises incomes. The resource export earnings are used to import food and other tradable goods whereas the higher demand for urban non-tradables is met by an increase in labor in that sector. Since the rural sector contracts, the country urbanizes. However, urbanization generated by resource rents differs from industrialization-led urbanization in that cities will have different employment shares of tradables and non-tradables. We consider two other paths leading to consumption cities. In GJV16, resources include fuels, mineral products, and a few high-rent cash crops, which require little labor. Yet, many countries export agricultural products, including food crops. Their production generates rents. However, it also requires rural labor, leading to ambiguous urbanization effects. We show that, under a reasonable set of assumptions, the urban share should increase. The agricultural export earnings can indeed be used to import additional food and other tradable goods. Much like what can be observed for non-agricultural resources, cities in these countries will have high employment shares of non-tradables.2 2 Any significant source of foreign income could also promote urbanization through the rise of consumption cities. For our sample countries, during the period 1960-2020, remittances and foreign aid averaged just 1.7% and 1.8% of GDP per year, respectively, compared to 7.8% earned from natural resource and agricultural exports (source: World Bank (2021)). 3 Next, many countries have experienced “premature” deindustrialization (Rodrik, 2016), due to the removal of import substitution industrialization (ISI) policies adopted by many nations up to the 1980s or increased trade competition due to trade liberalization or rising productivity related to advances in automation in some nations. We discuss why “premature” deindustrialization may not lead to de-urbanization. However, since the country experiences deindustrialization, its cities do so as well. Using GJV16’s sample of 116 developing countries (as of 1960) and long-difference and panel regressions for the period 1960-2020, we first show that: (i) higher urban shares are found in countries with higher GDP shares of manufacturing & services, a proxy for industrialization broadly defined (including tradable services); (ii) countries exporting natural resources, whether fuel & mining products or agricultural products, are also more urbanized; and (iii) urban shares are unchanged in deindustrializing countries. Second, we take advantage of newly available data, including IPUMS census microdata for about 60 countries over time and I2D2 household and labor force survey data for about 90 countries over time, to examine the correlations between the sectoral structure of urban areas and industrialization, resource exports, and deindustrialization.3 We study sectors not covered in GJV16 and informality, use panel regressions, and identify which parts of the city size distribution are affected by the structural change mechanisms mentioned above. Cities in industrialized countries have more employment in tradables and more wage employment, while cities in resource- rich and deindustrializing countries have higher shares of non-tradables and self- employment. Differences between industrialized, resource-rich, and deindustrializing countries are stable across city sizes. Thus, the origin of the urbanization process also impacts the largest cities, hence countries’ “engines of growth” (World Bank, 1999, 2009). Third, focusing on the mechanisms, we take advantage of novel data on urban construction across countries to shed light on the “quality” of spending in cities. For example, we explore whether the spending of resource rents in cities might have led to “white elephant” projects. We make use of a remarkable data set that inventories all the world’s tall buildings – whose height is above 80 meters – with information on their year of construction and height. Long-difference and panel regressions suggest that exporting 3 The International Income Distribution Database (I2D2) of the World Bank’s World Development Report unit consists of 1,500 individual-level household/labor force surveys. Details will be provided in Section 4.. 4 natural resources correlates with the construction of tall buildings whose economic rationale is questionable, for example very tall buildings that are more expensive to construct than other types of buildings. Since resource-rich countries do not have a larger construction sector, as measured by cement use or construction GDP, but use more cement in their tall construction sector than industrialized economies, the expansion of the sector might have occurred at the expense of non-tall construction.4 Fourth, an important theme in the macro-urban literature is the fact that urban-biased policies cause urban primacy (Ades and Glaeser, 1995). In particular, governments use resource rents to implement policies that disproportionately favor the largest city (Bates, 1981; World Bank, 2020). However, using both country- and city-level data, we do not find that resource-rich countries have relatively higher urban primacy rates. Indeed, resource- richness appears to have the same consequences for all population size categories of cities. Finally, we perform various analyses that suggest that having more consumption cities might have economic implications for a country. Using evidence on the number of years of schooling, education quality, and the returns to education in urban areas globally, we actually do not find that consumption cities have significantly less human capital than production cities. Yet, for a given level of human capital, consumption cities have more informality and higher employment in urban non-tradables than production cities. This implies that the human capital in consumption cities is employed in less productive sectors and jobs, at least in the short to medium run. Indeed, controlling for selection and sorting, we find large wage gaps between urban tradables and urban non- tradables. In addition, we find that returns to experience in urban areas are overall lower in countries with more consumption cities. Finally, we discuss various studies showing that agglomeration economies are possibly weaker in urban non-tradables. In addition to the structural change literature, we contribute to the macro- development literature on the determinants and characteristics of urbanization across countries (Gollin, Lagakos and Waugh, 2014; Gollin, Jedwab and Vollrath, 2016; Lagakos, 2016; Jedwab and Vollrath, 2019; Lagakos, Marshall, Mobarak, Vernot and Waugh, 2020; Lagakos, 2020; Porzio, Rossi and Santangelo, 2020; Gollin, Kirchberger and Lagakos, 2021; 4 ¨ These results on possibly “inefficient” construction echo the work of Collier, Kirchberger and Soderbom (2016) for roads and link this work to recent studies on the global role of market constraints (Kirchberger and Beirne, 2021) and geographical constraints (Jedwab et al., 2020b; Ahlfeldt and Jedwab, 2022) in construction. See Kirchberger (2020) for a survey of the literature on construction in developing countries. 5 Gai, Guo, Li, Shi and Zhu, 2021; Lagakos, Mobarak and Waugh, 2022). We study the links between changes in economic and export structures and the nature of the urbanization process, as well as urban structural change at the city level, not just the country level. Much like the macro-development literature, this paper offers a model of structural change and data to establish new stylized facts regarding the development process. We consider the potential effects of various factors on country- and city-level variables rather than focusing on identifying a clean effect for one of them, which would not be credible anyway since we compare countries. Our results are not causal and instead, our goal is to generate a number of theoretical predictions and empirical results that collectively contribute to corroborating, but not proving, our main messages. Consumption cities differ from the consumer cities in the urban literature. In our case, resource-rich countries experience the rise of consumption cities due to their increased consumption capacity. Deindustrializing countries see their consumption cities grow because their production cities lose their production capacity. Analyses on consumer cities rely on the Rosen-Roback model to show that, within a country, cities with better amenities attract residents that accept lower wages and/or higher rents to live there (Rosen, 1979; Roback, 1982; Glaeser et al., 2001; Glaeser and Gottlieb, 2009). Relatedly, Gollin et al. (2021) find that the quality of amenities is at least as high in cities as in rural areas in Africa, showing that the observed urban wage premia in developing economies (Gollin et al., 2014) do not represent a compensation for lower amenities. We do not claim that consumption cities are harmful for economic development. Urbanization through the rise of consumption cities may generate agglomeration economies and thus boost growth. Furthermore, even if consumption cities offer smaller economic benefits than production cities, they may still be economically significant. Lastly, consumption cities could evolve into production cities, especially considering that urban areas in resource-rich and urbanized economies have similar human capital levels.5 The paper is structured as follows. Section 1 discusses the data and methodology used to classify world cities into production or consumption cities. Section 2 presents the model and four propositions that guide our empirical analysis in Section 3 on 5 Porzio et al. (2020) show that human capital explains structural change out of agriculture. Countries with more human capital should thus be more urbanized. Yet, the study does not discuss how human capital impacts urban sectoral composition. Since we do not find large urban human capital differences between resource-rich and industrialized economies, urban sectoral structure does not appear to be related to urban human capital. However, we need better data to study this question and capture “omitted” factors. 6 the respective roles of natural resources, industrialization, and de-industrialization in urbanization. Sections 4, 5, and 6 study the role of these factors for urban employment, construction, and urban primacy, respectively. Section 7 focuses on urban human capital. 1. The Global Sectoral Composition of Cities It is not obvious how cities of the same population size and located in countries with similar levels of economic development differ globally in their employment composition. GJV16 find examples of countries where urban areas have high shares of urban tradables, which include manufacturing (MFG) and FIRE (finance, insurance, real estate, and business services). They also find examples of countries where urban areas have high shares of urban non-tradables, which include the (non-tradable) wholesale and retail sector. Since the work of GVJ16, which is cross-sectional and only focuses on the urban sector as a whole, many countries have added censuses to the IPUMS repository (Minnesota Population Center, 2020) and GIS files corresponding to the second level administrative units in which a household is located.6 In addition, the Global Human Settlements Layer (GHSL) database provides geocoded polygons of urban extent boundaries for the world c. 2015 (Schiavina et al., 2019). More precisely, this database uses satellite data on built-up area to identify “Functional Urban Areas” (FUAs), i.e. commuting zones of at least 50,000 residents today. Combining the two data sets allows us to obtain the sectoral composition of most world cities. Using a simple methodology presented below, we then classify each city as being a production city – a city with a disproportionately high share of employment in tradables; a consumption city – a city with a disproportionately low share of employment in tradables; or a neutral city – a city in which the shares of employment in tradables is neither disproportionately low nor disproportionately high.7 Data. For 76 countries and 191 country-years (1960-2015), we have IPUMS census micro- data with information on the administrative unit in which the respondent lives, the respondent’s sector, and whether the respondent lives in an “urban” area. We process the data following several steps. First, we select “urban” observations.8 Second, we have information on the resident’s second level administrative unit. Such units are typically smaller than FUAs.9 Third, IPUMS classifies employment into 16 6 These units correspond to counties in the U.S. and municipalities in most Latin American countries. 7 We interchangeably use the words “FUAs”, “agglomerations” and “cities” in the rest of the analysis. 8 When the urban identifier is unavailable, we identify urban residents from information on the locality. 9 In some cases, information is only available for first level administrative units, which leads to coarser 7 groups. We focus on the employment share of urban tradables (MFG & FIRE). We could have alternatively focused on urban non-tradables but it is not yet clear which sectors should be included. Having obtained employment data for each administrative unit (1960-2015), the next step is to use these data to obtain the labor shares for each FUA. Censuses only take place every 10-15 years. To ensure that we have enough countries for our comparison, we select for each country-FUA the closest observation to the year 2000 (within the 1990-2015 period). Doing so, we are left with 6,865 FUAs in 74 countries. Lastly, we need data on their population size c. 2000. The GHSL database reports their population for the “epoch 2015”. However, population is almost always estimated for earlier years. In many cases, the last census indeed took place in the 2000s. Sampling. The 6,865 FUAs include 3 billion people and account for 75% of world’s urban population. Kolmogorov–Smirnov tests confirm that the distribution of city sizes is not significantly different between our sample and the world. However, data on a few large developed countries are not available in IPUMS (e.g., Japan). To increase our sample’s representativeness, we divide all the countries in the world into ten deciles based on their log per capita GDP c. 2000 (PPP; source: World Bank (2021)). We then obtain the share of each decile in the world’s urban population and compare these shares to the shares in our sample. Based on the differences, we create weights over-sampling developed countries. Methodology. Since we aim to identify which cities have high shares of MFG+FIRE relative to other cities of the same size and for a given level of urban economic development, we categorize the FUAs in 10 deciles. Based on a range of log population from 10.8 (50K) to 17.2 (30 million), the resulting thresholds are: 95K, 180K, 341K, 648K, 1,227K, 2,306K, 4.377K, 8,351K, and 15,570K. The top two bins end up having few FUAs (N = 22 and 13, respectively), so we aggregate them together (N = 35). Then, using the sample of 6,865 FUAs, we run a regression relating the FUA’s employment share in MFG+FIRE (%) c. 2000 to 8 size category (CAT) dummies (omitting the lowest size category, hence 50-95K) and their interactions with the 2000 urban share (URB) of the FUA’s country. We also add a dummy if the FUA is the capital city of the country in 2000 (CAP). For FUA a in country c and population category p, the model is: MFGFIREa,c,2000 = α+Σ9 9 p=2 βp 1(CATa = p)+Σp=2 γp 1(CATa = p)∗URBc +δ URBc +ζ CAPa +µa estimates of sectoral shares. All FUAs in the same first-level unit are then assigned the same composition. 8 Finally, we use as weights the populations of each FUA. However, to ensure sample representativeness, we modify the weights so as to oversample richer countries. While Web Appx. Table D.1 reports the estimated coefficients, Figure 2 shows the implied relative employment share in MFG+FIRE of each population size category for a mostly unurbanized country (urban share = 20%) and a highly urbanized country (85%). More urbanized countries have higher urban shares of MFG+FIRE. The share is higher for smaller cities, consistent with MFG moving away from larger cities as countries develop. The regression residual measures to what extent the FUA has a high, or low, MFG+FIRE share (%) given its size and its country’s level of economic development. The 5th, 10th, 25th, 75th, 90th and 95th percentile values are about -15, -10, -5, 5, 10 and 15. Classification. In our classification, a production city is any FUA with a residual value above 5, indicating a city with a disproportionately high share of employment in urban tradables. Our definition further distinguishes production cities with a “low” (5-10), “medium” (10-15) or “high” (15+) value. A consumption city is any FUA with a residual value below -5, also distinguishing consumption cities with a “low” (-5;-10), “medium” (-10;-15) or “high” (-15+) value. Cities in the [-5; 5] range are classified as neutral. Figure 1 shows the distribution of production (in blue), consumption (red), and neutral cities (grey) c. 2000. Paler shades of the blue and red colors indicate lower values for the extent to which a city can be classified as a production or consumption city. Production cities are located in China and parts of Europe, the United States, Mexico, Brazil and India, while consumption cities are located in parts of Africa, the MENA, South America, and Southeast Asia.10 The patterns for Asia, Africa, Europe and the Americas are shown in Web Appx. Fig. D.1-D.5. As discussed in Web Appx. Section A, they conform with our priors. Next, Table 1 shows the classification into production, consumption, or neutral cities for those FUAs in our sample whose population exceeds 10 million people (“Residual” is the value used to classify cities). Production cities include Ho Chi Minh, Bangalore, Istanbul and Paris as well as all Chinese megacities, while consumption cities include Kolkata and Chennai, two historically important Indian cities, as well as Lagos, Jakarta, Rio and Surabaya.11 Approximately half of the mega cities are classified as neutral. In 10 China’s cities are mostly production cities whereas India has a mix of production, consumption, and neutral cities (Appx. Fig. D.1). Different configurations within a same country are possible and frequent. 11 Web Appx. Fig. D.7 separately considers MFG and FIRE, classifying cities according to their “best” 9 these cities, the employment in tradables is neither too low nor too high, implying a relatively balanced distribution of employment between tradables and non-tradables. Unsurprisingly, 8 of the 13 neutral mega cities are capital cities, where the government sector generates substantial employment in urban non-tradables. Finally, we show how, for a given size, cities in Latin America and the Caribbean (LAC) have experienced sectoral employment changes over time.12 We use FUA data for 8 countries to estimate the mean population-weighted urban employment share of tradables for the whole region and the periods “pre-1980” (obs. from 1962 to 1982; mean = 1978), “early 1990s” (1990-1994; 1991) and “c. 2010” (2001-2012;2009). As seen in Figure 3, employment in tradables (MFG+FIRE) has been declining over time. The evolution of LAC’s urban system was driven by its largest cities, as the decline at the “top” is 15 percentage points. Large LAC cities have thus increasingly become consumption cities.13 Robustness. We obtain similar results if we (Web Appx. Section B): (i) include the square, cube, and perfect fourth of the urban share, and their interactions with the population dummies, in case there are non-linearities; (ii) use log per capita GDP instead of the urban share or control for the urban definition used by each country;14 (iii) compare the raw (i.e. non-residualized) employment shares; (iv) use other weights or ignore them; (v) consider other classifications for the population dummies; and (vi) study urban non-tradables or informal employment. Since our residuals are estimated relative to similar cities in the world, we may worry that adding more countries to the analysis could change the results. However, our sample captures close to three-fourths of the world’s urban population, which limits such possibility. Results are also very similar if we use the raw shares. To conclude, ceteris paribus cities dramatically vary in their sectoral composition across the world. Theil decompositions suggest that half of the variation comes from differences across countries, implying an important role for aggregate structural change. sector. While some cities are production cities because of their high MFG shares (e.g., Guangzhou and Ho Chi Minh), other production cities have high shares of FIRE (e.g., Bangalore and Paris). 12 Indeed, only the LAC sample has good coverage over several decades in IPUMS. 13 Additional analysis (not shown but available upon request) shows that this evolution was driven by declines in the relative urban employment shares of MFG, not FIRE. 14 In the appendix, we discuss why results should not depend on the use of different urban definitions. Relatedly, while the urban classification is not stable across countries and years, we do not use the population density of the administrative units to select urban observations. Some rural areas of the world have particularly high density, for example in Bangladesh, Burundi, Haiti and Rwanda, which are some of the densest and least urbanized countries in the world. Other areas of the world are undoubtedly urban despite their low density, in particular sprawling suburban cities (e.g., many million-plus U.S. cities). 10 Important differences are then observed for the largest cities. To better understand why consumption cities may grow, in the next section we discuss a model of structural change. 2. Theoretical Insights: Paths to Consumption Cities We consider four sectors. The urban economy has a tradable sector (e.g., MFG + FIRE) and a non-tradable sector (e.g., wholesale & retail trade + personal services). The rural economy has an agricultural sector, which produces a tradable agricultural good (mostly crops, but also livestock). A natural resource R is an endowment that is internationally traded and is a source of foreign exchange earnings. Natural resources include fuels and mining products but also, for the sake of simplicity, cash crops characterized by high rents. The model offers several paths to consumption cities. A commodity boom due to a resource discovery or a boost to commodity prices on account of strong external demand boosts resource revenues R and influences urbanization/cities through two channels: (i) an income effect, which through non-homotheticities in the domestic demand for food pulls workers into urban sectors; and (ii) export earnings increase domestic demand for non-tradable services and pull workers away from agriculture and urban tradables. We also consider faster productivity growth in agriculture, which has an income effect and a foreign earnings effect if the country exports agricultural products, which increases demand for, and employment in, urban non-tradables. However, if the level of agricultural productivity is not high enough, this increase may pull workers back to agriculture in order to meet the food sufficiency requirement. Some crops then “behave” as a pure natural resource, in which case their effects go through R. Regardless of whether the country urbanizes on account of fuels/mining or agricultural exports, its employment share of urban non-tradables increases and its cities become consumption cities. De-industrialization can occur due to the removal of ISI policies or due to increased trade competition from industrializing nations. We discuss why such cases may not lead to de-urbanization, but “consumption” cities in already somewhat urbanized nations. 2.1. Set-Up We assume a log-linear utility function over the consumption of rural agricultural products (cf ), urban tradables (cm ), and urban non-tradables (cn ): U = βf ln(cf − cf ) + βm ln(cm ) + βn ln(cn ) (1) where expenditure shares βf , βm , and βn are between 0 and 1 and sum up to 1, and cf is 11 the subsistence level of agricultural consumption. With income elasticity for agriculture less than one, any income increase drives up the budget shares of urban tradables and non-tradables. For the sake of simplicity, production in each sector only requires labor: 1−α Qj = Aj Lj . (2) Lj is the share of workers in each sector j ∈ {f, m, n}, and Aj is sector-specific productivity. Agricultural commodities produced mostly for export and mostly with land or capital and little labor are included in resource endowments R. Thus, the agriculture sector comprises other agricultural subsectors, including subsistence food crops. The prices of urban tradables and agricultural products are assumed to be exogenous (*) and the budget constraint of the individual is: z = p∗ ∗ f cf + pm cm + pn cn . Since the household first covers its agricultural subsistence requirement and urban non-tradables are produced only domestically, the total expenditure on urban non- tradables equals the value of their production: βn (z − p∗ f cf ) = pn Qn (3) Assuming balanced trade, the following accounting relationship must hold, where R is the revenue from exporting natural resources and both agricultural products and urban tradables can be produced domestically, imported from the rest of the world, or exported: (βf + βm )(z − p∗ ∗ ∗ ∗ f cf ) + pf cf = R + pf Qf + pm Qm (4) With perfect labor mobility, wages equalize across any two sectors j and k ∈ {f, m}: −α −α (1 − α)p∗ j Aj Lj = (1 − α)p∗ k Ak Lk (5) The above relationships are used to determine the implicit function for the allocation of labor in the non-tradable urban sector: (1 − Ln )α L n = βn 1+ R − p∗ f cf . (6) A 1 1 α A = ( p∗ α ∗ m A m ) + pf A f α is a composite measure of agricultural productivity and productivity in urban tradables. Given Ln , the rest of labor is allocated to the tradable sectors in proportion to the relative productivity in agriculture and tradable non- agriculture: 1 p∗ m Am α Lm = (1 − Ln ) (7a) A 1 p∗ f Af α Lf = (1 − Ln ) (7b) A 12 The urbanization rate, U , is then simply U = Lm + Ln . 2.2. Main Predictions We obtain four predictions (Web Theory Appendix C provides details and proofs): Proposition 1 (Urbanization through commodity rents and “consumption cities”) ∂U ∂Ln ∂Lm ∂Lf > 0, > 0, < 0, <0 ∂R ∂R ∂R ∂R Proposition 1 reiterates GJV16’s result that resource revenues R offer a path to urbanization U and the emergence of consumption cities. Indeed, employment in urban non-tradables Ln is increasing in R whereas employment in manufacturing and tradable services Lm is decreasing in R. In other words, a positive shock to R (a resource windfall) leads to the rise of consumption cities. The effect on urbanization U is also positive. Proposition 2 (Productivity growth in agriculture and “consumption cities”) So long as R < p∗ ∗ f cf , given y = pf Af , it follows that: ∂Ln ∂Lm > 0, <0 ∂y ∂y ∂U ∂Lf 1 ∗ 1 < 0, > 0, if α(p∗ f Af ) < (pm Am ) α α ∂y ∂y ∂U ∂Lf 1 ∗ 1 > 0, < 0, if α(p∗ f Af ) > (pm Am ) α α ∂y ∂y Faster productivity growth in agriculture has an income effect and a foreign earnings effect if the country exports agricultural products. Both result in a disproportionate increase of urban non-tradables, while the increase in foreign earnings enables the importing of urban tradables, whose share in employment decreases. If the level of agricultural productivity is high enough, the urban share increases as the urban non- tradable effect dominates the urban tradable effect. However, if the level of agricultural productivity is not high enough, an increase in agricultural productivity pulls resources back to agriculture in order to meet the subsistence requirement. Then the urban share decreases. However, we will show that urban shares almost never decrease. Proposition 3 (Urbanization through industrialization and “production cities”) ∂U ∂Ln ∂Lm ∂Lf ∗ > 0, ∗ > 0, ∗ > 0, <0 ∂pm Am ∂pm Am ∂pm Am ∂p∗ m Am 1 1 ∗ so long as R-p∗ ∗ f cf <0 and agricultural productivity is high enough: α(pm Am ) < (pf Af ) . α α Increasing MFG-FIRE productivity leads to an expansion of urban tradable 13 employment. Thus, a MFG or FIRE revolution causes urbanization and production cities. Proposition 4 (de-industrialization without de-urbanization and the transformation of existing production cities into consumption cities) When Lf is fixed, by definition U = 1 − Lf is also fixed, implying that a productivity shock that decreases (increases) employment in manufacturing would lead to a corresponding increase (decrease) in employment in non-tradables. Proposition 4 says that shocks to the manufacturing or FIRE sector can cause existing production cities to become consumption cities when de-urbanization is unlikely. In the next section, we show that urbanization rates almost never decrease and discuss why. Empirics. The urban share and the employment composition of urban areas should depend on the resource windfall R, (tradable) agricultural productivity (p∗ f Af ), and urban tradable productivity (p∗ m Am ). In our econometric analysis, we focus on the period 1960- 2020 and 116 countries that were still “developing” economies in 1960.15 We do not have reliable historical measures of Am . It is also not obvious which price levels should be used for p∗ m . FIRE GDP is only reported for some countries and recent years (previous ISIC classifications did not separate FIRE). MFG and FIRE employment is likewise only measured for some countries and years when there is a census or a labor force survey (surveys were rare before 1990). Productivity could then be high because employment is low and/or “selected”, for example if a country has a few MFG/FIRE firms and these belong to high-productivity subsectors or are politically connected. Given such issues, we use the GDP share of manufacturing and services (MFGSERV). Web Appx Fig. D.6 shows that countries with a high GDP share of MFG+FIRE are countries with a high GDP share of MFGSEFV today, thus validating this proxy (correlation of 0.78). It is also unclear how to distinguish agricultural products belonging to R (those generating high rents) from agricultural products belonging to p∗ f Af . In addition, productivity and yields are typically poorly measured for the latter and it is not obvious which price level to use. Since we aim to measure the fact that a country is urbanizing because it is exporting fuels and mining products or agricultural products, and given the difficulty in separating agricultural products in different categories, we aggregate them together and use as a proxy the ratio of natural resource exports (NRX) to GDP.16 15 The sample excludes a few ex-communist countries due to the lack of pre-1989 data. 16 NRX/GDP differs from the GDP share of natural resources, for which we have no historical data for 14 Finally, for deindustrialization, we use as a proxy the decline in the GDP share of MFG over time. Indeed, if MFG productivity decreases relative to the world, MFG employment should decrease. The GDP share combines information on productivity and employment. 3. Resources, (De-)Industrialization, and Urbanization Data. We study 116 relatively large countries, with a developing status as of 1960, and for which data on urbanization, resource exports, and the GDP shares of manufacturing (MFG) and services (SERV) are available every 5 years between 1960 and 2020. We obtain urban shares (%) from the United Nations (2018) and the share of fuel and mining exports in total exports for the period 1960-2010 from GJV16 who rely on data from World Bank (2021) and the Mineral Industry Surveys of USGS (2020). We extend their data up to 2020, modifying them when needed. The USGS data are important due to the fact that the World Bank data are incomplete or incorrect in many cases.17 Measurement error, if classical, would make us under-estimate the contribution of these resources to urbanization. The export shares of all agricultural products come from FAO (2020). Using these shares and knowing from World Bank (2021) the export-to-GDP ratio of each country in each year, we calculate NRX/GDP (%) as in Sachs and Warner (1995, 1997). Next, we obtain the time series of the GDP share of MFG+SERV by relying on World Bank (2021), Central Intelligence Agency (2021) and United Nations (1960-1980, 2020c) and the GDP share of FIRE c. 2020 from United Nations (2020a).18 Country-level results. For countries c, we estimate the following cross-sectional model: URBc,20 = α + β NRXGDPc,60−20 + γ MFGSERVc,20 + δ DEINDUc,80−20 + Xc π + µc (8) In this equation, URB is the urban share (%) in 2020, NRXGDP – the mean export-to-GDP ratio of natural resources (%) in 1960-2020, MFGSERV – the GDP share of MFG and SERV (%) in 2020 (capturing structural change due to manufacturing or tradable services), and DEINDU – the absolute decline in the GDP share of manufacturing (%) between 1980 and 2020 transformed so that a positive value indicates more deindustrialization (= 0 if no too many countries. Furthermore, NRX/GDP has the advantage of attributing to NRX the “value” of any input used in producing the resources, as such inputs also directly contribute to urbanization. 17 According to the WDI of World Bank (2021), fuel and mining exports accounted for 8% of Botswana’s exports in 2015. According to USGS: “Botswana’s exports were $6.33 billion, of which diamonds accounted for 82.9%; copper and nickel, 5.9%; and soda ash, 1.4%.” Likewise, according to the WDI, fuel and mining exports accounted for 6% of Burkina-Faso’s exports in 2013. According to USGS: “The International Monetary Fund estimated that gold production [...] accounted for about 71% of the country’s total exports.” 18 See Web Data Appx. Section D for details on the data sources. 15 decline is observed). Summary statistics are provided in the notes of Table 2.19 Xc is a set of controls that includes: (i) controls for country size: log area and log population in 2020 and their squares and a dummy if the country is a small island country because such countries are mechanically more urbanized; (ii) controls for the urban definition used by the country c. 2010;20 and (iii) controls for initial conditions, i.e. URB, NRXGDP and MFGSERV c. 1960. Our regression is thus a long-difference regression.21 Col. (1) of Table 2 Panel A shows similar coefficients for NRXGDP and MFGSERV. A one standard deviation in MFGSERV is associated with a 0.57 standard deviation in urbanization (Proposition 3) vs. a 0.49 standard deviation in NRXGDP (Propositions 1-2). No correlation is observed for DEINDU (Proposition 4). Indeed, cases of de-urbanization are very rare. For 116 countries x 13 years (1960, 1965...), 94% of country-years showing an economic decline do not show a decline in their urban share. In addition, while (annualized) economic declines can be large (mean = -2.6%; 10th percentile = -5.5%), urbanization declines are tiny (mean = -0.16 percentage points; 10th percentile = -0.24). Several factors could account for these facts. Glaeser and Gyourko (2005) first show that, due to housing being durable, city growth rates are skewed so that cities grow more quickly than they decline. Thus, negative shocks decrease housing prices more than they reduce population. In addition, workers born in urban areas may be particularly unproductive in agriculture, making it unattractive for them to move to rural areas even when faced with large negative shocks. They may also have strong urban preferences (Jedwab et al., 2017). Furthermore, urban incomes may still be high enough for workers to afford their subsistence requirement, implying they can do so without having to work in agriculture (Lagakos and Waugh, 2013). That should be the case for countries that are already somewhat urbanized, e.g. deindustrializing nations. Finally, migration costs that prevent workers from migrating from rural to urban areas (Lagakos et al., 2020; Gai et al., 2021) may also prevent workers from migrating from urban to rural areas. In cols. (2)-(4), we examine the correlations in a panel framework with country and 19 NRXGDP, MFGSERV and DEINDU are not correlated with each other. Countries with high NRXGDP values include Angola, Kuwait, Nigeria, Saudi Arabia, and the Republica ´ Bolivariana de Venezuela. Countries with high DEINDU values (conditional on MFGSERV) include Argentina, Brazil, Colombia, the Philippines, and South Africa. 20 We include dummies identifying whether the definition is based on a population threshold, another condition, an administrative function, or a combination of these, and the log of the threshold (U.N., 2011). 21 Since the world’s urbanization depends on the urbanization of countries with large populations, we use as regression weights the total population of each country in 2020. 16 year fixed effects (FE). More precisely, for countries c and years t, we estimate: URBc,t = α + β NRXGDPc,t−1 + γ MFGSERVc,t + δ DEINDUc,1980−t + κc + θt + Xc,t π + µc,t . (9) We use NRXGDP in the previous period t-1, as we expect rents from resource exports to have a lagged effect on urbanization; it takes time to build cities. MFGSERV is defined in t since the production of MFG and SERV mostly takes place in cities, where they have a contemporaneous effect on the urban share. The deindustrialization variable – DEINDU – measures the absolute decline in the GDP share of manufacturing between 1980 and t transformed so that a positive value indicates more deindustrialization, while 0 is assigned if no decline is observed or if t ≤ 1980. Finally, we include the time-varying controls (log population and its square) and standard errors (SEs) are clustered at the country level. In cols. (2), (3) and (4), we consider data every 20 years, 10 years, and 5 years, respectively. As seen in Panel A, the panel estimates of NRXGDP and MFGSERV are smaller than their corresponding long-difference estimates in col. (1). The point estimates of NRXGDP are lower for shorter time periods in the panel and in all cases, they are smaller in magnitude than the point estimates for MFGSERV. These results are expected given the delayed effect of increases in NRXGDP on urbanization. Lastly, DEINDU has a small positive impact on urbanization. DEINDU countries are countries that saw their MFGSERV share increase in the past, which might still generate urbanization in the shorter run as urbanization begets urbanization. Once we control for the urban share in t-1, DEINDU’s coefficient is close to 0 (0.01 with the 5-year panel; not shown). Timing. With country fixed effects, identification comes from within-country changes in NRXGDP, MFGSERV, and DEINDU. Web Appx. Table D.3 examines how urbanization correlates with the timing of such changes. We use the 10-year panel specification (9) and include leads and lags. The table shows that leads of the variables are small and not significant. In contrast, some lags are significant. For MFGSERV, all lags have significant coefficients. For NRXGDP, the second and third lags have larger and more significant coefficients, suggesting that urban shares increase 20-30 years after resource booms. Urbanization thus seems to follow industrialization/FIRE-ization and resource export booms, not the other way around. No significant correlation is observed for DEINDU. Additional results In Panel B of Table 2, we decompose NRXGDP into the export-to- GDP ratio of fuel & mining (FMXGDP) and agricultural products (AGXGDP). The long- 17 difference regression of Col. (1) where we control for initial conditions suggests that urbanization correlates with both. The Beta coefficient is twice as large for FMXGDP than for AGXGDP, likely due to agricultural exports requiring rural labor. In the panel specifications (Cols. (2)-(4)), only FMXGDP survives. Likewise, lags of AGXGDP have positive but insignificant coefficients (Web Appx. Table D.3). Indeed, not all agricultural exports may be associated with urbanization. Some crops require large amounts of labor and cities may develop too slowly for the correlations to appear with panel regressions. Web Appx. Table D.2 then shows stronger correlations for MFG than for SERV, and positive correlations for SERV, and FIRE or non-FIRE services. The results shown so far are aligned with Propositions 1-4. We further test these propositions in the next section, and in particular the fact that the mechanism of the urbanization process might also influence the employment composition of urban areas. 4. Resources, (De-)Industrialization, and Urban Employment Model. We rely on IPUMS census microdata for 62 sample countries accounting for almost two-thirds of the world’s total urban population, which is high considering that we ignore most developed economies. In total, we obtain 184 census samples during the period 1990-2015. For our cross-sectional country (c)-level analyses, we then select for each country the closest year to the year 2000 and estimate the following model: EMPc00 = α+β MFGSERVc00 +γ NRXGDPc60−00 +δ DEINDUc80−00 + URBc00 +Xc κ+µc (10) In this equation, EMP is a sector’s labor share in the urban areas of country c c. 2000. Since most censuses are centered around the year 2000, we use information from 2000 for MFGSERV in country c, the mean of NRXGDP in country c in the period from 1960 to 2000, and DEINDU in country c between 1980 and 2000 (a positive value indicates deindustrialization). To compare countries with similar levels of urban economic development, we control for the urban share (URB) in 2000. We can then ignore the initial conditions in 1960 but still add the controls (X ) for area (2000), population (2000), small islands, and urban definitions. We use country populations in 2000 as weights. Finally, we examine the differences between coefficients β and γ , and then between β and δ . Sectors. IPUMS has information on 16 industries from which we construct 10 sectors. As can be seen in Table 3, for a given urban share, urban areas in industrialized MFGSERV countries have more employment in tradables (MFG, FIRE and SUM = MFG + FIRE) and 18 less employment in non-tradables (NTR...) than the resource-rich NRXGDP countries or the DEINDU countries, which deindustrialized post-1980. NTR is a non-tradable sector corresponding in this analysis to domestic Wholesale and retail trade. NTR2 also includes Other services that sometimes correspond to domestic commerce-related activities not included for some countries in NTR. NTR3 additionally includes Household services. We find, if anything, more government workers (GOVT) in MFGSERV countries. However, differences are not significant when adding education and health (GOVT2). This result is not surprising since structural adjustment programs were implemented in NRXGDP and DEINDU countries in the 1990s. We see more natural resource (NRX) workers in both NRXGDP and DEINDU countries, but the differences are not significant. Yet, we find significantly more fuel & mining workers in NRXGDP countries (not shown). Finally, we find less construction (CONST) in NRXGDP and DEINDU countries. Informality. Using the same eq. (10), we test whether urban informal employment differs depending on the mechanism of urbanization. Our sample includes 55 sample countries for which we know whether the worker earns a wage, is self-employed, or works without pay. As shown in Cols. (1)-(3) of Table 4, the share of wage workers is lower and the share of self-employed workers higher in both NRXGDP and DEINDU countries (relative to MFGSERV countries). However, due to the sample size, differences are not significant.22 We expand on this further by testing whether tradables – proxied by MFG, the largest sector of MFG+FIRE – and non-tradables – proxied by NTR (domestic wholesale and retail trade) – have higher informality rates, as measured by self-employment. As seen in Cols. (4)-(6) and (7)-(9) of Table 4, MFG is more informal in the urban areas of NRXGDP and DEINDU countries than in the urban areas of MFGSERV countries, but these differences are not significant. NTR, however, is significantly more informal in the urban areas of NRXGDP countries than in the urban areas of MFGSERV countries. City-level results. We use the city-level residualized measure of Production / Consumption City-ness (PCC) obtained in Section 1 to examine if the patterns obtained at the country- urban sector level are driven by larger and/or smaller cities. For 6,211 FUAs, indexed with a, belonging to 59 sample countries, indexed with c, we regress PCC c. 2000 on the country-level variables of interest – MFGSERV (2000), NRXGDP (mean 1960-2000) and 22 Self-employed workers are almost always self-employed without employees. 19 DEINDU (change 1980-2000), their interactions with the 8 city size categories, p, and the controls X of model (10) (except urbanization, which was used as a control in the first-step residualization procedure). We use the FUAs’ populations in 2000 as weights: PCCa,c = α+β MFGSERVc +γ NRXGDPc +δ DEINDUc +Σ9 p=2 ζp MFGSERVc ∗1(CATa = p)+ Σ9 9 p=2 θp NRXGDPc ∗ 1(CATa = p) + Σp=2 λp DEINDUc ∗ 1(CATa = p) + Xc κ + µa (11) We consider two PCC measures capturing urban tradables – MFG+FIRE – and urban non-tradables – NTR2, which includes domestic wholesale & retail trade and other services. We plot the obtained correlations for each PCC measure-size category in Figures 4(a)- 4(b). Figure 4(a) shows that MFGSERV countries have higher shares of MFG+FIRE employment in their cities than NRXGDP and DEINDU countries for all population size categories of cities. The patterns for the employment share of non-tradables, NTR2, are different. MFGSERV countries have the lowest employment shares of NTR2 across all city sizes (Figure 4(b)). By contrast, DEINDUSTR countries have the highest employment shares of NTR2 across all city sizes. In NRXGDP countries, the employment share of NTR2 increases with city size and is largest in the case of the largest cities. This analysis suggests that the whole urban system in the deindustrializing countries has experienced a shift from jobs in manufacturing to jobs in nontradable services, while in resource-rich countries only larger cities have substantially higher shares of nontradable services. In both cases, larger cities have much higher shares of urban non- tradables than smaller cities, but for all city sizes these shares are much larger in the deindustrializing countries than in the resource-rich countries. Robustness: Panel. We obtain the employment shares for 184 country-years and 62 countries from IPUMS and run panel regressions with country and year fixed effects (FEs). We regress the selected sectoral labor share in urban areas in time period t on MFGSERV in t, NRXGDP in t-10, DEINDU between 1980 and t (= 0 before 1980), the urban share in t, and log population in t and its square. We include countries with at least three years of data and cluster the standard errors at the country level. Since the sample size reduces to 124 observations, we cannot include more lags. Most panel results are consistent with the long-difference results. We find significantly lower shares of employment in MFG and MFG+FIRE in the urban areas of DEINDU countries relative to MFGSERV countries (Web Appx. Table D.4). While these shares 20 are also lower in the NRXGDP countries, the difference with MFGSERV countries is not significant. Thus, any significant impact of resource exports on the employment composition in urban areas may take place over more than one decade. Next, the shares of non-tradables (e.g., NTR3) in the urban areas of NRXGDP and DEINDU countries are higher than those in MFGSERV countries. Informality is also higher in the urban areas of NRXGDP countries than in MFGSERV countries, but the difference is not significant. Robustness: I2D2. Another data source for employment information by country is the International Income Distribution Database (I2D2) of the World Bank. The full database consists of about 1,500 individual-level household and labor force surveys for more than 150 countries. Detailed information on sectoral employment and self-employment can be found in about 1,000 survey samples for almost 100 countries. The data was initially compiled by the World Bank’s World Development Report (WDR) unit between 2005 and 2011. The database has expanded since then allowing us to include more countries than IPUMS. We use its December 2017 vintage to replicate the IPUMS-based results with the expanded country sample. The surveys are nationally representative and sufficiently large for our purpose (i.e., they have more than 10,000 observations).23 We obtain for each country the mean labor shares of various sectors and types of employment in urban areas in 1990-2015.24 We then use the same cross-sectional model as for IPUMS (eq. (10)). Since 2005 is the mean population-weighted year in I2D2, the variables of interest are defined relative to 2000 (as in Panel A of Web Appx. Table D.5) or 2010 (as in Panel B of the same table). While we observe a strong correlation (about 0.9) between I2D2 and IPUMS for MFGFIRE, the correlation is weaker (0.55) for urban non-tradables NTR, which has to do with the way I2D2 was harmonized.25 NRXGDP and DEINDU countries have lower employment shares of MFG in their urban areas than MFGSERV countries (Web Appx. Table D.5). For MFG+FIRE, the differences are mostly not significant. No correlations are observed for FIRE and NTRI (the version of “NTR” in I2D2). However, DEINDU countries have significantly higher 23 See for example Jedwab et al. (2020a, 2022) for more details. 24 Most countries had several surveys during the period. Since the surveys vary in sample size across time periods, we use as weights the size of each survey, thus giving more weight to better measured years. 25 During the harmonization process, sectors had to be aggregated together due to the different classifications in different surveys. Wholesale & retail trade and Hotels & restaurants were merged together by the I2D2 team, while we ignored the latter in IPUMS due to the lack of correlations with our variables of interest. “NTR” in I2D2 (NTRI) might also include other sectors depending on the country. 21 shares of NTRI2, which includes NTRI as well as household services and personal services (using I2D2’s classification). Finally, the results indicate significantly higher self- /unpaid employment in DEINDU countries compared to MFGSERV countries. While I2D2 allows us to consider more countries, the database consists of surveys, not censuses, and classical measurement error in the dependent variable should affect precision. Notwithstanding, relying on various measures, specifications, and databases, the evidence presented here suggests that cities in NRXGDP and DEINDU countries may be disproportionately consumption cities relative to the cities in MFGSERV countries. 5. Mechanisms: Urban Construction and Consumption Cities We now explore some of the mechanisms behind the rise of consumption cities. In particular, cross-country differences in urban construction may stem from “white elephant” projects financed by resource rents. In the absence of a global historical data base on “white elephants”, we examine whether exports of natural resources correlate with the over-construction of very tall buildings, whose economic rationale might be questionable as discussed below. As shown by Ahlfeldt and Barr (2020), construction costs increase non-linearly with height. Very tall buildings are also more inefficient as they require more occupiable square feet to be dedicated to essential building functions (e.g., due to elevators). We use a novel data set that inventories all the world’s “tall buildings”, with information on their year of construction and height.26 The database is maintained by the Council on Tall Buildings and Urban Habitat (CTBUH) and is publicly available online.27 The database mostly captures buildings above 80 meters (Jedwab et al., 2020b). Since some countries have no such buildings, to avoid having their stock of heights equal to 0 when using logs, we consider for each country buildings above 80 meters as well as their 10 tallest buildings, even if some of them are below 80 meters. In the end, we use 16,369 tall buildings.28 We use as our dependent variable log urban height density, which is the 26 To our knowledge, the only other econometric study on white elephant projects and tall building construction is a non-economics study on autocracies and skyscrapers by Gjerlow and Knutsen (2019). The study finds that autocratic regimes are more likely to build super-tall and possibly “vanitous” buildings. 27 The full online database can be found here: http://www.skyscrapercenter.com/. As one example, the webpage for the Burj Khalifa in Dubai is: http://www.skyscrapercenter.com/building/burj-khalifa. According to their website, the data have been “collected by the Council for more than 40 years [...] The Council relies on its extensive member network [of academics, land developers, architectural firms, builders, city administrations, and banks] to maintain” the database with the help of “an Editorial Board”. 28 As discussed in Jedwab et al. (2020b, 2021c), measurement error is more likely for smaller buildings 22 sum of tall building heights + 1 divided by urban population.29 As previously shown by Jedwab et al. (2020b, 2021c), there is a strong relationship between log urban height density and log per capita GDP (PPP; see Figure 5). Next, we want to examine whether resource-rich countries have more tall buildings than equally developed non-resource-rich countries. And if they have more tall buildings, are these more likely to be “vanitous” in ways that are quantifiable in the data? Cross-sectional model. We first examine the long-difference correlations between log urban height density (LUHT) c. 2020 in country c, and the three mechanisms of urbanization: MFGSERV (2020), NRXGDP (mean in 1960-2020) and DEINDU (change in 1980-2020; positive values imply deindustrialization): LUHTc,20 = α + β MFGSERVc,20 + γ NRXGDPc,60−20 + δ DEINDUc,80−20 + Xc κ + µc . (12) In the above equation, Xc includes the controls for initial conditions (including log urban height density in 1960, which makes the regression a long-difference regression) and the controls for area and population in 2020, small islands and urban definitions. Col. (1) of Panel A of Table 5 shows a strong correlation between tall building construction per capita and MFGSERV. The point estimate for NRXGDP is not negligible either, but it is not significant. In cols. (2) and (3), we consider residential and non- residential buildings only, with the latter including office towers ((4)), hotels ((5)) and retail towers ((6)). Col. (7) shows the correlations for government buildings. We observe positive and significant correlations for most types of buildings and MFGSERV. We also see positive and significant correlations for residential and non-residential (office) towers and NRXGDP. Thus, MFGSERV or NRXGDP countries urbanize as they develop (Table 2) and build more and taller buildings to accommodate urban growth. In Panel B of Table 5, we aim to capture “excesses” in tall building construction per capita for a given level of urban economic development. We therefore add controls for urban economic development. These include the urban share and log per capita GDP (PPP; constant 2005 international dollars) in 2020 and the mean log per capita GDP in 1960-2020 (since past economic development also matters and since GDP can fluctuate). than tall buildings above 80 m. Classical measurement error in tall buildings then only affects precision. 29 The sum of heights is above 0 for almost all sample countries in 2020. For the full sample covering the period from 1960 to 2020, it is equal to 0 for half of the observations. It is 0 for most countries in 1960. Indeed, the U.S. contained almost all of the world’s tall buildings up to the 1980s (Jedwab et al., 2020b). Thus, few of the sample countries (with a developing status as of 1960) had tall buildings before the 1980s. 23 The results suggest that NRXGDP countries have higher stocks of non-residential towers than MFGSERV countries ((3)), with this correlation being driven by office ((4)) and retail towers ((6), not significant though). No differences are observed for government towers.30 Panel C examines the same correlations in a 10-year panel framework. The dependent variable is log urban height density in t and we include three extra lags of MFGSERV, NRXGDP and DEINDU.31 We control for the urban share, log per capita GDP and log population in t. Lastly, the panels show the total correlations estimated when summing up the coefficients across the lags. The correlations are not as strong as in the long- difference specification (panel B). Indeed, construction does not respond instantaneously to economic changes. However, by adding lags, we lose observations and degrees of freedom, making our standard errors larger. Abstracting from the lack of significance, one can see that the overall correlation for NRXGDP is always higher than for MFGSERV. We also use data on the characteristics of the buildings to explore their “vanity”. In Panel A of Table 6, we use the long-difference specification, control for urban economic development, and consider the dependent variable to be the log urban height density constructed using only non-residential buildings above a certain threshold. In our data, 125, 140, 160, 200 and 240 meters correspond to the median, the mean, the 75th percentile, the 90th percentile and the 95th percentile value in total height. The results indicate that NRXGDP countries have more very tall buildings, and the difference with respect to MFGSERV countries tends to be largest and significant for the tallest buildings in the data set. When studying the same relationships in our 10-year panel framework, considering four lags, and summing up the coefficients across the lags (Panel B), the point estimates are broadly consistent with the long-difference results for NRXGDP countries. We perform additional tests in Panel C. We only consider the long-difference specification and include the urban economic development controls. First, in col. (1), the dependent variable is an index of construction vanity, more precisely the sum of the differences (in meters) between height at the tip and the height of the top occupied floor. Buildings constructed to project power are more likely to include space at the top. However, the height of the top occupied floor is only available for enough buildings in 38 30 These results suggest that private companies or individuals directly or indirectly affiliated with the government and/or fuel, mining or agricultural firms might be using such buildings. 31 The lag structure is as follows: MFGSERV t, t-10, t-20, t-30; NRXGDP t-10, t-20, t-30, t-40; DEINDU t, t-10, t-20, t-30 (for DEINDU, all lags are defined relative to 1980). 24 countries. Nonetheless, we find stronger correlations for NRXGDP than for MFGSERV. We also know the main structural material used to construct the tall buildings in the data set. Steel is more durable, resistant and safe than concrete but also more expensive. Cols. (2)-(3) show the results with the dependent variable being log urban height density for buildings whose structural material is steel vs. concrete, respectively. Interestingly, we find strong correlations between steel use and MFGSERV (but steel could be cheaper in MFGSERV countries) and between concrete use and NRXGDP. Third, we examine if the correlations are stronger for larger cities. In col. (4), the dependent variable is log height density considering only buildings in the capital city. We find a stronger correlation with NRXGDP than MFGSERV, although the differences with respect to MFGSERV is not significant. If we consider both the capital city and the largest city, we find a weaker correlation ((5)). For other cities, we also find a possible gap, indicating higher height density in NRXGDP than MFGSERV countries ((6)). The excess of tall buildings in NRXGDP countries may be accompanied by lower occupancy rates. Since we do not have information on occupancy rates, we instead examine how the overall construction sector is potentially affected. For 80 countries, we know the annual cement production from 1970 to date.32 Cement – the main ingredient of concrete – is typically not traded (World Cement, 2013). Using UN COMTRADE data we find that the world trade of cement (tons) only accounted for 3.5% of world cement production (tons) during the period of study, and for just 2.5% today. For this reason, cement production should be a very good proxy for cement use. We rely on the same long-difference model as for tall buildings but use as our dependent variable the log sum of cement production over the period 1970-2020 while controlling for log urban population in both 1970 and 2020, thus capturing cement consumption per urban capita. The results in Cols. (1)-(2) of Table 7 show that cement production is positively correlated with MFGSERV and NRXGDP (significant for MFGSERV), consistent with the tall construction results. However, once we control for urban economic development, no differences are observed. If we use the COMTRADE data to reconstruct cement consumption, we find that the log sums of cement consumption and production over the period 1970-2020 are highly correlated (corr. = 0.98). Likewise, using the log sum of cement consumption produces the 32 We obtained the data from the Minerals Yearbooks of the U.S. Geological Survey (USGS). 25 same results as when relying on production ((3)-(4)). Lastly, for both cement production and consumption, if we use 10-year panel regressions with several lags of the variables of interest, we do not find significant differences between the types of countries.33 Therefore, although NRXGDP countries have more tall buildings, in particular concrete towers, they do not use more cement, which might imply eviction effects within the construction sector. Of course, with economic development and rising land values, tall buildings replace short structures (Jedwab et al., 2020b). However, we here compare NRXGDP and MFGSERV countries with similar levels of urban economic development, thus focusing on “excesses” that might lead to “gaps” in the rest of the urban sector. To study urban construction more generally, we turn to information on the value added of the construction sector (in PPP; constant 2005 international dollars) from 1970 to date.34 We use the long-difference model and consider as the dependent variable the log sum of construction GDP over the period 1970-2020 while simultaneously controlling for log urban population in both 1970 and 2020, thus capturing construction GDP per urban capita. Again, we observe no significant difference ((5)-(6)). If we use 10-year panel regressions with several lags of the variables of interest, we also do not find significant differences between the types of countries (not shown). However, one caveat with these analyses is that both cement use and construction include infrastructure. Finally, if residential construction is indeed crowded out in NRXGDP countries, one might see higher slum shares there. However, to project prestige NRXGDP may more willingly adopt slum clearance policies. Data on slum shares are only available for the most recent years (United Nations, 2020b). If we use the cross-sectional version of the long-difference model (thus not controlling for slums c. 1960), we find smaller slum shares in MFGSERV and NRXGDP countries when not controlling for urban economic development ((7)). Consistent with the results on tall building construction, when MFGSERV or NRXGDP countries develop and urbanize, whether due to industrialization broadly defined or resource windfalls, their urban sector also develops, which leads to some forms of formalization. But if we control for urban economic development ((8)), we find no correlations between the slum share and the variables of interest. Therefore, the slum share is not higher in NRXGDP countries. As a result, if any crowding takes places 33 These results are not shown, but are available upon request. 34 From (United Nations, 2020c) we know the GDP share of construction over time. We then obtain GDP using data from Bolt and van Zanden (2014), Gollin et al. (2016) and World Bank (2021). 26 in such countries due to the consumption of resources by the tall sector, the non-tall non- slum sector must be disproportionately affected. However, these results should be taken with caution given the lack of global data on supply and crowding in this sector. 6. Mechanisms: Urban-Biased Policies and Consumption Cities Is the “rise of consumption cities” in some countries driven by their capital/largest city? In resource-rich countries, there is evidence that governments disproportionately invest resources in their capital/largest city (Bates, 1981; World Bank, 2020). Yet, industrialization/FIRE-ization may also lead to the disproportionate growth of a country’s largest city if it requires resources found there (e.g., skilled labor or an international airport).35 In addition, resource exports cause the growth of small- and medium-sized mining towns or agro-towns (Jedwab, 2013), thus decreasing urban primacy. Therefore, a priori it is unclear whether urban primacy is more prevalent among MFGSERV or NRXGDP countries. Cross-Sectional Regressions: Country-Level. We use model (8) to study the correlations between the primacy rate – the share of the country’s urban population that lives in the largest city (source: World Bank (2021)) – in 2020 and the variables of interest: MFGSERV (2020), NRXGDP (mean 1960-2020) and DEINDU (change 1980-2020). We control for initial conditions, i.e. primacy, MFGSERV, and NRXGDP in 1960, thus capturing long- difference correlations. We control for the urban share in 2020, add the controls for area, population, small islands and urban definitions, and use populations in 2020 as weights. As seen in Col. (1) of Table 8, only slightly higher primacy rates are observed in NRXGDP and DEINDU countries relative to MFGSERV countries. The differences are not significant.36 In Col. (2), we decompose NRXGDP into the export-to-GDP ratios of fuel & mining (FMXGDP) and agriculture (AGXGDP). The results do not show particularly high primacy rates in FMXGDP countries, but they show less primacy in AGXGDP countries. Yet, the high standard errors suggest that it is not the case for all agricultural exports. Panel Regressions: Country-Level. We use the same 10-year panel regression as before 35 In the longer run, as the cost of space increases and more stringent environmental regulations are adopted in larger cities, industrial activities move to small and medium-sized cities, thus reducing primacy. 36 While some resource-rich countries such as Angola, Cote ˆ d’Ivoire, and Malaysia have high urban primacy rates (30%-40%), resource-poor countries such as Bangladesh, Japan and Thailand have similarly high rates. Argentina, the Arab Republic of Egypt and Peru – countries with intermediary levels of resource richness – also have high rates. 27 (eq. (9)) but now the dependent variable is the primacy rate in year t (116 x 7 = 812 obs.). The variables of interest are again MFGSERV (t), DEINDU (change 1980-t), and NRXGDP (t-1) (or FMXGDP and AGXGDP). In Cols. (3)-(6), we consider either three lags or four lags of the variables of interest (i.e., two or three extra lags in addition to the contemporaneous lag) and show the overall correlations across the various lags. We control for the urban share in t, include controls for populations in t, and cluster standard errors at the country level. Again, we find no significant correlation. City-Level Results. We use the Functional Urban Area (FUA)-level data of the GHSL database to study whether the largest cities, not just the largest city, grow differently than other cities depending on the country’s “type”. We have population estimates for 7,422 FUAs c. 1975, 1990, 2000 and 2015 in 115 sample countries. We regress the log growth of their population between 1975 and 2015 – log (pop. + 1) 2015 - log (pop. + 1) 1975 – on the (country-level) variables of interest and their interactions with a dummy for whether the FUA is the capital/largest city (2015; we call this dummy “top 1”). We also consider the capital city and the two (top 2) or five (top 5) largest cities, or the capital city only (top 0) as well as add the controls of the long-difference regressions at the country level. As seen in Web Appx. Table D.6, there are no significant correlations. However, point estimates suggest that the top cities in countries specialized in agricultural exports indeed grow slower relative top cities in industrialized countries. Yet, differences are not significant. We also do not find any significant correlation in a 10-year panel framework with one or two extra lags included (see Web Appx. Table D.7). The results presented in this section suggest that even if some resources might be associated with urban primacy in some countries, it is not the case in general. While we do not discuss this above, we also find no correlation with deindustrialization. Thus, in resource-rich and deindustrializing countries, consumption cities are found along the urban hierarchy and, as shown in Fig. 4(a), the shares of urban tradables in their cities are lower than those in MFGSERV countries for all population size categories of cities. 7. Concluding Discussion: Consumption Cities & Urban Human Capital Using census data for a large number of cities and countries, we established the following stylized fact: While some cities have high employment shares of urban tradables and can be characterized as production cities, a relatively high number of cities have high shares of urban non-tradables and can be characterized as consumption cities. This appears to be 28 the case not only for small cities, but also for many large cities, which are often seen as “engines of growth” in developing countries (World Bank, 1999, 2009). Furthermore, we can use our city-level measures of production/consumption city-ness (PCC; Section 1.) as well as population data for each city to obtain the population- weighted average level of PCC for each country c. 2000 (see Panel A of Web Appx. Table D.8 for the full list). While some developing economies have high aggregated PCC values, for example China, Costa Rica, Lesotho, Mauritius and Vietnam, other developing economies have low aggregated PCC values, especially Colombia, Indonesia, Nigeria, ´ Tanzania and the Republica Bolivariana de Venezuela. Some countries such as Brazil, India, Malaysia, Mexico and Turkey have both production cities and consumption cities, as indicated by a (city population-weighted) Gini index of PCC calculated for each country (Panel B for the full list).37 Either their urban system is chronically “dualistic” or it is transitioning, whether away from production cities as in Latin America or towards them as in Asia (Section 1.). Urban Human Capital. These facts pose the question of whether having an urban hierarchy dominated by consumption cities might have economic implications for a country, “beyond” the economic structure of its cities. If consumption cities are not as growth-enhancing as production cities, are they doomed to remain so or is it possible that they evolve into production cities in the future? An important factor that can shed light on these questions is the stock and quality of urban human capital. Indeed, cities with a bigger stock of high-quality human capital may be better able to adapt to future economic conditions and grow regardless of their current economic structure. We assess how urban human capital correlates with our structural variables of interest by obtaining from IPUMS the average number of years of education of adults (ages 25 to 65) living in the urban areas of 53 out of the 116 sample countries for the year closest to 2000. We use the same cross-sectional regression as in our study of sectoral or informal employment (eq. (10)) and include MFGSERV (2000), NRXGDP (average in 1960-2020) and DEINDU (change in 1980-2000), as well as the urban share (2000). The results, shown in col. (1) of Panel A in Table 9, suggest that NRXGDP and DEINDU countries might have 37 For each country one by one, and since we know the PCC and population of each city c. 2000, we can estimate the degree of statistical dispersion of the PCC measure – i.e., the Gini index – within the country. 29 more educated urban residents. However, the difference is not significant.38 Angrist et al. (2019) provide a measure of education quality that is built on globally harmonized test scores, obtained for all areas of a country, for 101 countries (c. 2020).39 If we use it as a proxy for urban education (human capital) quality, we find significantly lower values in NRXGDP countries ((2)). However, if we use the learning adjusted years of education,40 we again find no significant differences at the country level ((3)). Another imperfect proxy for education quality is a country’s return to education. Abstracting from human capital demand factors, a lower education quality should reduce the wage gains from higher educational attainment. We use the I2D2 data (1990-2017) and follow the methodology of Lagakos et al. (2018) and Jedwab et al. (2020a) to estimate the returns to education for urban areas, as well as the returns to experience for urban areas.41 More precisely, for individual i and country-year-sample t, we use OLS to estimate the following model for each country one by one (for 18-67 year-old “urban” workers only): 7 lnWit = α + βe expite + γeduit + θt + εit . (13) e=1 The dependent variable is the log of monthly earnings (lnWit ). As in Lagakos et al. (2018), experience is categorized into seven bins (expite ): [5-9 years] (which we call 5), [10-14] (10), [15-19] (15), [20-24] (20), [25-29] (25), [30-34] (30), and [35+] (35). The omitted bin is [0-4] (0). We include the number of years of education (eduit ) and country-year-sample fixed effects (θt ).42 Finally, we omit samples without at least 10 observations in each bin. We use γ as a measure of the returns to education. To estimate the returns to work experience we follow Jedwab et al. (2020a). For each bin one by one (5, 10, etc.), we 38 Likewise, if we use city-level data as we did in Section 1., we only find very weak correlations between the PCC residuals estimated using tradable employment and the residuals based on the number of years of education. These results are not shown but are available upon request. Thus, urban tradable employment and urban human capital seem unrelated for a given city size and the level of urban economic development. 39 From the notes of the Excel database containing the data, we learn that the “database harmonizes scores across major international student achievement testing programs measured in TIMMS-equivalent units, where 300 is minimal attainment and 625 is advanced attainment.” Url for the database: https: //www.worldbank.org/en/publication/human-capital$\sharp$Data, last accessed 03-29-2022. 40 From the notes of the Excel database containing the data (same Url as above), we learn that the “Learning-Adjusted Years of School are calculated by multiplying the [...] years of school by the ratio of most recent harmonized test scores to 625, where 625 corresponds to advanced attainment [...].” 41 We calculate potential work experience – which we call “experience” in the rest of the analysis – as follows: (i) For individuals with at least 12 years of education, we assume children start school at age 6 and calculate experience as age - years of education - 6; (ii) For individuals with less than 12 years of education, we assume that experience before age 18 is inconsequential and calculate experience as age - 18. 42 Most countries having several samples, we use individual weights divided by the size of the sample. 30 estimate an annualized return. We then take the mean of the annualized returns across the seven bins to obtain the mean annualized return throughout the experience distribution (remember that we use any available survey data during the 1990-2017 period).43 The results in Col. (4) of Panel A, Table 9 show that urban returns to education might be lower in NRXGDP and DEINDU countries, however not significantly so. If we use the product of the years of schooling and the returns, to obtain an imperfect measure of the value of education, we see a substantially lower value of schooling in DEINDU countries, but not in NRXGDP countries (col. (5)). In Panel B and the first column of Panel C of Table 9, we investigate the same correlations for the urban returns to education or their interactions with the urban number of years of education when also including country-year-sample-location fixed effects in eq. (13). Locations correspond to macro regional areas (“Reg 01”), first level administrative areas (“Reg 02”), or primary sampling units, hence survey clusters (“PSU”).44 Doing so allows us to compare individuals residing in the same regional or local urban labor market. While the number of available country estimates decreases, we tend to find lower returns to education in NRXGDP and DEINDU countries, and no significant differences for the interactions with the number of years (not shown). Finally, we tend to find lower returns to experience in NRXGDP and DEINDU countries, but not all the estimated differences are significant (see the last 4 columns, Panel C). We then use data from IPUMS to obtain the urban number of years of education for adults (ages 25 to 65) in 54 countries. We end up with 151 country-years. Since we select countries with at least two years of data, the sample size decreases to 104 observations and therefore we cannot include lags. We run panel regressions with country and year FEs, regressing the education variable in t on MFGSERV in t, NRXGDP in t-10, DEINDU between 1980 and t (= 0 before 1980), the urban share in t, and log population in t and its square. We again do not find significantly different correlations for MFGFIRE, NRXGDP, and DEINDU. The results are similar for younger cohorts (ages 25 to 45).45 43 More precisely, for individuals belonging to bin e, we obtain the bin-specific annualized return as rete = ((βe + 1)(1/e) − 1) ∗ 100, with βe being the estimated coefficient for bin e in eq. (13). For this subgroup of individuals, it tells us by how many percentage points wages increased on average for each extra year of experience. We then take the average of these seven bin-specific annualized returns so that each bin is 7 equally represented. Therefore, the overall return re is equal to ( e=1 rete ) ÷ 7. 44 For example, “Reg 01” and “Reg 02” correspond to regions and states in the U.S., respectively. “PSUs” mostly likely correspond to neighborhoods or census blocks in the urban areas. 45 All these results are available upon request. 31 Wage Gaps. Overall, it does not appear that consumption cities have significantly less human capital than production cities. In other words, for a given level of human capital, the workers of consumption cities are more likely to work in urban non-tradables (and informal employment). Now, inspired by Gollin et al. (2014), we use I2D2 to investigate the productivity gaps between urban tradables (UT) and urban non-tradables (UNT). For individual i and country-year-sample t, we use OLS to estimate the following model for each country one by one, restricting the sample to 18-67 year-old urban workers belonging either to the UT sector or the UNT sector: lnWit = α + δTit + Xit λ + θt + εit (14) where the dependent variable is the log of monthly earnings (lnWit ) and Tit is a dummy equal to one if the worker belongs to the UT sector. δ measures the wage gap between the UT and UNT sectors. We first examine the gaps estimated without including controls for human capital (X ); then we re-estimate the gaps after adding human capital as a control and controlling for sorting. If we find large (conditional) wage gaps, that would suggest that the UT sector may provide “better” jobs than the UNT sector. Focusing on the sample countries (N = data available for 54 countries in I2D2), we find an unconditional wage gap of about 22%.46 If we control for gender, years of education, and the experience dummies, the gap reduces to 13%.47 If we additionally control for (spatial) sorting by including first-level administrative area or primary sampling units FEs, we get 15-17% (N = 28-41). In other words, we find significant gaps even when we compare workers with similar human capital levels and workers residing in the same urban regional or local labor market. Of course, these results should be taken with caution since there may still be unobservable factors that we cannot control for.48 Growth. The higher wages for jobs in UT relative to UNT imply that workers in UT are more productive than workers in UNT. If most urban areas in a country are consumption cities, this might have negative implications for aggregate labor productivity in the 46 Indeed, only 54 countries have large enough surveys with earnings data and detailed information on the respondent’s type of residence and subsector, allowing us to distinguish UT and UNT workers. 47 Like Gollin et al. (2014), if we use national accounts data (source: United Nations (2020c); ISIC Rev. 3) and sectoral employment data (ILO (2022); ISIC Rev. 3.1) to estimate GDP per worker, we find for the sample countries much higher gaps around 120%-205%, depending on which sectors are included in UNT. 48 If we use I2D2 data for the world instead, in order to increase sample size (thus adding developed economies), we find an unconditional gap of 24% (N = 91) and 16%-18% for the conditional gaps (N = 34-73). 32 country. But what is the implication of having consumption cities for growth? Comparing the returns to experience for the UT and UNT sectors might be one way to answer it. Indeed, if the returns to experience capture human accumulation at work (Lagakos et al., 2018; Jedwab et al., 2020a), and if there is more human capital accumulation at work in the UT sector, we should observe higher returns in UT than in UNT. However, we do not find that this is the case.49 Yet, in some specifications we found significantly lower urban returns to experience in the aggregate in NRXGDP and DEINDU countries, which may be due to more general factors in consumption cities.50 Another way to answer the question about the growth implications of consumption cities, especially large ones, would be to globally estimate separately the agglomeration economies (AEs) for the two sectors, UT and UNT. We cannot estimate AEs by sector using the I2D2 because we cannot identify cities in the database (only regions or the number of the survey cluster are given). Since this paper identifies production/consumption city-ness (PCC) for many urban areas in a large number of countries, we could obtain from the literature and compare the estimates of AEs for countries whose cities are mostly consumption cities and the AEs of countries whose cities are mostly production cities. One can draw on many such urban studies, which typically consider one country at a time and for a sample of cities regress log wages or log productivity on the log of population, population density or employment. Chauvin et al. (2017) estimated the AEs for China – a country which our analysis shows has mostly production cities – at more than 16%, for India and the United States – two countries that we show have a mix of production, consumption and neutral cities – at 8% and 5%, respectively, and for Brazil – a country with mostly neutral and consumption cities – at 3%. The fact that these estimated AEs decrease as consumption cities become more dominant in a country’s urban hierarchy suggests that urban economic growth might also weaken. Other studies corroborate the potential negative growth implications of countries’ having an urban hierarchy dominated by consumption cities. Venables (2017) show 49 If we estimate eq. (13) separately for UT and UNT and focus on the sample countries for the period 1990-2017, we find similar returns at around 2% (N = 59-67). This is also the case if we consider the whole world including developed economies for which the returns might be better estimated. 50 Human capital could be accumulated at work passively or actively. While Jedwab et al. (2020a) think of human capital accumulation at work as being a primarily passive process, Ma et al. (2021) show that the process can be active in developed economies. Note that our sample focuses on developing economies. 33 theoretically that the supply-side benefits of co-location should be small for UNT firms. Likewise, Glaeser and Resseger (2010) find for the United States that AEs are strong in cities with higher levels of skill and nonexistent in less skilled metropolitan areas. Lastly, Burger et al. (2022, Panel H of Table 2) use firm-level data for 649 metro areas to estimate AEs for 98 developing economies. They regress log labor productivity on log urban density, finding an average AE of 10%-15%. They then interact density with dummies proxying for whether the firm belongs to the tradable sector, finding AEs that are twice higher than in the non-tradable sector. Congestion forces might also be stronger for the UNT sector than for the UT sector.51 Finally, as shown by the trade literature, international trade affects endogenous innovation (Melitz and Redding, 2021). As such, having higher shares of firms and workers in UT rather than UNT may also promote growth. Just like production cities can evolve into consumption cities through de- industrialization, consumption cities could still evolve into production cities in the future. We showed that they do not necessarily lack human capital, but other constraints that we do not observe may constrain urban productivity growth; such constraints include institutions, control of corruption, and investment climate, among others. There are many examples of successful transitions. The United Arab Emirates’ largest cities – Dubai and Abu Dhabi – have uncontestably become production cities by expanding FIRE. Likewise, the growth of Malaysia’s manufacturing reshaped the economic structure of the Kuala Lumpur metro area. However, since understanding what distinguishes dynamic consumption cities, which evolve into production cities, from stagnant ones is beyond the scope of this paper, we leave this question for future research.52 REFERENCES Ades, Alberto F and Edward L Glaeser, “Trade and circuses: explaining urban giants,” The Quarterly Journal of Economics, 1995, 110 (1), 195–227. 51 The UT sector (MFG & FIRE) is more formal than the UNT sector (verified using both IPUMS and I2D2 data; not shown). As such, it may have smaller congestion externalities on other sectors and the city as a whole. Akbar et al. (2018) find that the roads of Indian cities are particularly congested in the middle of the day when few workers are commuting. Indeed, the (more informal) wholesale and retail trade sector (incl. roadside vendors) often causes traffic congestion. Burger et al. (2022, Table 4) find that the AEs of UT firms are reduced to a lower degree by congestion than the AEs of UNT firms, possibly because UNT firms are more dependent on local consumer markets, while UT firms rely less on local markets and employ workers who are better able to locate close to their workplaces and/or afford faster transport services. 52 Nonetheless, Venables (2017) shows theoretically that a city is unlikely to produce UTs if it faces high urban and hinterland demand for UNTs. That is especially the case with increasing returns in tradable production that make cities specialized in UNT globally uncompetitive for the production of UTs. 34 Ahlfeldt, Gabriel and Jason Barr, “The economics of skyscrapers: A synthesis,” CEPR Discussion Papers 14987, C.E.P.R. Discussion Papers July 2020. Ahlfeldt, Gabriel M. and Remi Jedwab, “Estimating the Global Economic and Environmental Effects of the Skyscraper Revolution,” Mimeo 2022. Akbar, Prottoy A., Victor Couture, Gilles Duranton, and Adam Storeygard, “Mobility and Congestion in Urban India,” NBER Working Papers 25218 November 2018. Alvarez-Cuadrado, Francisco and Markus Poschke, “Structural change out of agriculture: Labor push versus labor pull,” American Economic Journal: Macroeconomics, 2011, 3 (3), 127–58. Angrist, Noam, Simeon Djankov, Pinelopi Koujianou Goldberg, and Harry Anthony Patrinos, “Measuring Human Capital,” Technical Report 2019. Bairoch, Paul, Cities and economic development: from the dawn of history to the present, University of Chicago Press, 1988. Barrios, Salvador, Luisito Bertinelli, and Eric Strobl, “Climatic change and rural–urban migration: The case of sub-Saharan Africa,” Journal of Urban Economics, 2006, 60 (3), 357–371. Bates, Robert H, Markets and states in tropical Africa: the political basis of agricultural policies, Berkeley: University of California Press, Series on Social Choice and Political Economy, 1981. Bolt, Jutta and Jan Luiten van Zanden, “The Maddison Project: collaborative research on historical national accounts,” Economic History Review, 2014, 67 (3), 627–651. Burger, Martijn, Elena Ianchovichina, and Prottoy A. Akbar, “Heterogenous Agglomeration Economies in the Developing Countries: The Roles of Firm Characteristics, Sector Tradability, and Urban Mobility,” Policy Research Working Paper Series 9954, The World Bank, 2022. Castells-Quintana, David, “Malthus living in a slum: Urban concentration, infrastructure and economic growth,” Journal of Urban Economics, 2017, 98, 158–173. and Hugh Wenban-Smith, “Population Dynamics, Urbanisation without Growth, and the Rise of Megacities,” The Journal of Development Studies, 2020, 56 (9), 1663–1682. , Melanie Krause, and Thomas K J McDermott, “The urbanising force of global warming: the role of climate change in the spatial distribution of population,” Journal of Economic Geography, 2021, 21 (4), 531–556. Central Intelligence Agency, “The World Factbook,” 2021. Chauvin, Juan Pablo, Edward Glaeser, Yueran Ma, and Kristina Tobio, “What is different about urbanization in rich and poor countries? Cities in Brazil, China, India and the United States,” Journal of Urban Economics, 2017, 98, 17–49. Collier, Paul, Martina Kirchberger, and M˚ ¨ ans Soderbom, “The Cost of Road Infrastructure in Low- and Middle-Income Countries,” World Bank Economic Review, 2016, 30 (3), 522–548. Davis, James C and J Vernon Henderson, “Evidence on the political economy of the urbanization process,” Journal of urban economics, 2003, 53 (1), 98–125. Duarte, Margarida and Diego Restuccia, “The Role of the Structural Transformation in Aggregate Productivity,” The Quarterly Journal of Economics, 2010, 125 (1), 129–173. and , “Relative Prices and Sectoral Productivity,” Journal of the European Economic Association, 2020, 18 (3), 1400–1443. Duranton, Gilles, “Growing through cities in developing countries,” The World Bank Research Observer, 2015, 30 (1), 39–73. FAO, “FAOSTAT,” 2020. Fay, Marianne and Charlotte Opal, Urbanization without growth: A not so uncommon phenomenon, Vol. 2412, World Bank Publications, 2000. Gai, Qingen, Naijia Guo, Bingjing Li, Qinghua Shi, and Xiaodong Zhu, “Migration Costs, Sorting, and the Agricultural Productivity Gap,” Working Papers tecipa-693, University of Toronto, Department of Economics April 2021. Gjerlow, Haakon and Carl Henrik Knutsen, “Leaders, Private Interests, and Socially Wasteful Projects: Skyscrapers in Democracies and Autocracies,” Political Research Quarterly, 2019, 72 (2), 504–520. Glaeser, Edward L, “A world of cities: The causes and consequences of urbanization in poorer countries,” Journal of the European Economic Association, 2014, 12 (5), 1154–1199. Glaeser, Edward L. and Joseph Gyourko, “Urban Decline and Durable Housing,” Journal of Political 35 Economy, 2005, 113 (2), 345–375. Glaeser, Edward L and Joshua D Gottlieb, “The wealth of cities: Agglomeration economies and spatial equilibrium in the United States,” Journal of Economic Literature, 2009, 47 (4), 983–1028. Glaeser, Edward L. and Matthew G. Resseger, “THE COMPLEMENTARITY BETWEEN CITIES AND SKILLS,” Journal of Regional Science, 2010, 50 (1), 221–244. Glaeser, Edward L, Jed Kolko, and Albert Saiz, “Consumer city,” Journal of economic geography, 2001, 1 (1), 27–50. Gollin, Douglas, Casper Worm Hansen, and Asger Wingender, “Two blades of grass: The impact of the green revolution,” Technical Report, National Bureau of Economic Research 2018. , David Lagakos, and Michael E Waugh, “The agricultural productivity gap,” The Quarterly Journal of Economics, 2014, 129 (2), 939–993. , Martina Kirchberger, and David Lagakos, “Do urban wage premia reflect lower amenities? Evidence from Africa,” Journal of Urban Economics, 2021, 121, 103301. , Remi Jedwab, and Dietrich Vollrath, “Urbanization with and without industrialization,” Journal of Economic Growth, 2016, 21 (1), 35–70. , Stephen Parente, and Richard Rogerson, “The role of agriculture in development,” American economic review, 2002, 92 (2), 160–164. Hansen, Gary D and Edward C Prescott, “Malthus to solow,” American economic review, 2002, 92 (4), 1205–1217. Henderson, J Vernon, “Cities and development,” Journal of Regional Science, 2010, 50 (1), 515–540. , Adam Storeygard, and Uwe Deichmann, “Has climate change driven urbanization in Africa?,” Journal of Development Economics, 2017, 124, 60–82. Herrendorf, Berthold, Richard Rogerson, and Akos ´ Valentinyi, “Growth and Structural Transformation,” in Philippe Aghion and Steven Durlauf, eds., Handbook of Economic Growth, Vol. 2 of Handbook of Economic Growth, Elsevier, 2014, chapter 6, pp. 855–941. ILO, “Labor Force Statistics,” Geneva: International Labor Organization 2022. Islam, Asif, Remi Jedwab, Paul Romer, and Daniel Pereira, “Returns to Experience and the Sectoral and Spatial Allocation of Labor,” Mimeo 2019. Jedwab, Remi, “Urbanization without Industrialization: Evidence from Consumption Cities in Africa,” 2013. Unpublished manuscript, GWU. and Dietrich Vollrath, “Urbanization without growth in historical perspective,” Explorations in Economic History, 2015, 58, 1–21. and , “The urban mortality transition and poor-country urbanization,” American Economic Journal: Macroeconomics, 2019, 11 (1), 223–75. , Asif Islam, Paul Romer, and Robert Samaniego, “Human Capital Accumulation at Work: Estimates for the World and Implications for Development,” Working Papers 2020-3, The George Washington University, Institute for International Economic Policy March 2020. , Daniel Pereira, and Mark Roberts, “Cities of workers, children or seniors? Stylized facts and possible implications for growth in a global sample of cities,” Regional Science and Urban Economics, 2021, 87, 103610. , Federico Haslop, Roman David Zarate, Takaaki Masaki, and Carlos Rodriguez, “The Effects of Water Recessions: Evidence from the Lake Chad Post-1945,” Mimeo, GWU, 2021. , Jason Barr, and Jan Brueckner, “Cities without Skylines: Cities Without Skylines: Worldwide Building-Height Gaps, their Determinants and their Implications,” 2020. , Luc Christiaensen, and Marina Gindelsky, “Demography, urbanization and development: Rural push, urban pull and. . . urban push?,” Journal of Urban Economics, 2017, 98, 6–16. , Prakash Loungani, and Anthony Yezer, “Comparing cities in developed and developing countries: Population, land area, building height and crowding,” Regional Science and Urban Economics, 2021, 86, 103609. , Roberto Samaniego, Paul Romer, and Asif Islam, “Scars of Pandemics from Lost Schooling and Experience: Aggregate Implications and Gender Differences Through the Lens of COVID- 19,” Working Papers 2022-02, The George Washington University, Institute for International Economic Policy February 2022. Kirchberger, Martina, “The Role of the Construction Sector in Developing Countries: Some Key 36 Issues,” Page, J & Tarp, F. (Eds.). Mining for Change: Natural Resources and Industry in Africa, Oxford University Press 2020. and Keelan Beirne, “Concrete Thinking About Development,” Trinity Economics Papers tep0621, Trinity College Dublin, Department of Economics June 2021. Lagakos, David, “Explaining Cross-Country Productivity Differences in Retail Trade,” Journal of Political Economy, 2016, 124 (2), 579–620. , “Urban-Rural Gaps in the Developing World: Does Internal Migration Offer Opportunities?,” Journal of Economic Perspectives, Summer 2020, 34 (3), 174–192. and Michael E. Waugh, “Selection, Agriculture, and Cross-Country Productivity Differences,” American Economic Review, April 2013, 103 (2), 948–980. , Benjamin Moll, Tommaso Porzio, Nancy Qian, and Todd Schoellman, “Life Cycle Wage Growth across Countries,” Journal of Political Economy, 2018, 126 (2), 797–849. , Mushfiq Mobarak, and Michael E. Waugh, “The Welfare Effects of Encouraging Rural-Urban Migration,” Staff Report 635, Federal Reserve Bank of Minneapolis January 2022. , Samuel Marshall, Ahmed Mushfiq Mobarak, Corey Vernot, and Michael E. Waugh, “Migration costs and observational returns to migration in the developing world,” Journal of Monetary Economics, 2020, 113 (C), 138–154. Lipton, Michael, “Why poor people stay poor: urban bias in world development,” 1977. Lucas, Jr. Robert E, “Life earnings and rural-urban migration,” Journal of political economy, 2004, 112 (S1), S29–S59. Ma, Xiao, Alejandro Nakab, and Daniela Vidart, “Human Capital Investment and Development: The Role of On-the-job Training,” Working papers 2021-10, University of Connecticut, Department of Economics May 2021. Matsuyama, Kiminori, “Agricultural productivity, comparative advantage, and economic growth,” Journal of Economic Theory, 1992, 58 (2), 317–334. Melitz, Marc J. and Stephen J. Redding, “Trade and Innovation,” NBER Working Papers 28945, National Bureau of Economic Research, Inc June 2021. Minnesota Population Center, “Integrated Public Use Microdata Series, International: Version 7.3 [dataset],” 2020. Porzio, Tommaso, Federico Rossi, and Gabriella Santangelo, “The Human Side of Structural Transformation,” Technical Report 2020. Restuccia, Diego, Dennis Tao Yang, and Xiaodong Zhu, “Agriculture and aggregate productivity: A quantitative cross-country analysis,” Journal of Monetary Economics, 2008, 55 (2), 234–250. Roback, Jennifer, “Wages, rents, and the quality of life,” Journal of Political Economy, 1982, 90 (6), 1257–1278. Rodrik, Dani, “Premature deindustrialization,” Journal of Economic Growth, 2016, 21 (1), 1–33. Rosen, Sherwin, “Wages-Based Indexes of Urban Quality of Life,” Peter Miezkowski and Mahlon Straszheim, eds., Current Issues in Urban Economics, Johns Hopkins, 1979. Sachs, Jeffrey D and Andrew M Warner, “Natural resource abundance and economic growth,” Technical Report, National Bureau of Economic Research 1995. and , “Fundamental sources of long-run growth,” The American economic review, 1997, 87 (2), 184–188. Schiavina, M., A. Moreno-Monroy, L. Maffenini, and P. Veneri, “GHSL-OECD Functional Urban Areas 2019,” EUR 30001 EN, Publications Office of the European Union, 2019. U.N., “World Urbanization Prospects: The 2011 Revision, Department of Economic and Social Affairs, Population Division,” 2011. United Nations, “Statistical Yearbooks,” 1960-1980. , “World Urbanization Prospects 2018, Department of Economic and Social Affairs,” 2018. , “National Accounts Official Country Data database,” 2020. , Proportion of the Urban Population Living in Slums, United Nations, SDG Global Database., 2020. , “System of National Accounts (SNA) - Analysis of Main Aggregates (AMA) Database,” 2020. USGS, “Mineral Industry Surveys 1960-2020,” 2020. Venables, Anthony J., “Breaking into tradables: Urban form and urban function in a developing city,” Journal of Urban Economics, 2017, 98 (C), 88–97. 37 World Bank, World development report 1999/2000: Entering the 21st Century, The World Bank, 1999. , World development report 2009: Reshaping economic geography, The World Bank, 2009. , “Convergence: Five Critical Steps toward Integrating Lagging and Leading Areas in the Middle East and North Africa,” 2020. , “World Development Indicators,” 2021. World Cement, Global Trading Patterns in Cement, World Cement, Monday, 29 April 2013., 2013. Yang, Dennis Tao and Xiaodong Zhu, “Modernization of agriculture and long-term growth,” Journal of Monetary Economics, 2013, 60 (3), 367–382. Figure 1: World Map of Production Cities, Neutral Cities, and Consumption Cities, Data for 74 Countries, c. 2000 38 Notes: This figure shows for 74 countries in the world the location of production cities (Prod.; in blue), neutral cities (grey), and consumption cities (Cons.; yellow, orange or red), based on our analysis and our definitions. High, Mid and Low correspond to high, medium and low values. 39 Figure 2: Employment Share of Urban Tradables by City Size, Cross-Section, c. 2000 Notes: This figure shows the implied relative employment share of MFG+FIRE for the various population size categories in mostly unurbanized countries (urbanization rate = 20%) and highly urbanized countries (85%). In both countries, the shares are estimated relative to the omitted category (50K-95K). Figure 3: MFG+FIRE Employment in Cities, Latin America & the Caribbean 1960s-2010s Notes: The figure was created using FUA-specific data for 8 countries: Argentina (’80 ’91 ’01), Bolivia (’76 ’92 ’12), Brazil (’80 ’91 ’10), Chile (’82 ’92 ’02), Colombia (73 ’93 ’05), Ecuador (’62 ’90 ’06), Guatemala (’81 ’94 ’02), Mexico (’70 ’90 ’10), Panama (’80 ’10), Paraguay (’82 ’92 ’02), Peru (’93 ’17), Venezuela (’81 ’90 ’01). Figure 4: City Size & Urban Sectoral Shares for Each Group, Cross-Section, c. 2000 (a) Urban Tradables = MFG+FIRE (b) Urban Non-Tradables = NTR2 Notes: Mean pop. (000s): 1 = 76; 2 = 134; 3 = 250; 4 = 479; 5 = 921; 6 = 1,737; 7 = 3,268; 8 = 6,140; 9 = 18,184. 40 Figure 5: Tall Building Stock Per Capita & Economic Development, 2020 Notes: This figure shows for the 116 countries of our main sample the relationship between the log sum of tall building heights per urban capita (m per inh.) and log per capita GDP (PPP, cst 1990 intl $) c. 2020. Table 1: Production/Consumption City Classification for the Largest World Cities, c. 2000 Rank Name Category Residual (Pct) Pop. 2000s (Mil.) Country 1 Delhi Neutral -1.7 30.1 India 2 Jakarta Cons-Low -5.3 29.8 Indonesia 3 Shanghai Prod-Mid 10.2 26.9 China 4 Manila Neutral 1.6 25 Philippines 5 Cairo Neutral -0.8 23.5 Egypt 6 Kolkata Cons-Low -5.8 23.1 India 7 Mumbai Neutral -2.1 22.3 India 8 Sao Paulo Neutral 0.5 21.7 Brazil 9 Mexico City Neutral 1.6 21.4 Mexico 10 Beijing Neutral 3 21.3 China 11 New York Neutral -4.4 19.5 USA 12 Guangzhou Prod-High 15.7 16.7 China 13 Bangkok Neutral 1.3 16.3 Thailand 14 Los Angeles Neutral -2.5 15.7 USA 15 Buenos Aires Neutral 0.3 15 Argentina 16 Istanbul Prod-Low 6.3 14.8 Turkey 17 Tehran Neutral 1.2 13.4 Iran 18 Ho Chi Minh Prod-Low 6.6 12.8 Vietnam 19 Jieyang Neutral -3.1 12.7 China 20 Lagos Cons-High -18.1 12.3 Nigeria 21 Bangalore Prod-Low 5.3 11.9 India 22 Chengdu Cons-Low -6.7 11.7 China 23 Suzhou Prod-Low 9.9 11.4 China 24 Paris Prod-Low 7.7 11.2 France 25 Rio de Janeiro Cons-Mid -10.2 10.8 Brazil 26 Surabaya Cons-Mid -11.7 10.8 Indonesia 27 Chennai Cons-Low -8.8 10.6 India Notes: This table classifies 27 ten million plus cities into production cities, consumption, or neutral cities. 41 Table 2: Resources, Industrialization, Deindustrialization & Urbanization, 1960-2020 Specification: Long-Diff. Panel Analysis (Country FE & Year FE) Dep. Var. : Urban Share URB (%) in ... 2020 Year t Year t Year t Timing for the Panel: Every ... 20 Years 10 years 5 years Panel A: Baseline (1) (2) (3) (4) NRXGDP (%) (1): 2020; (2)-(4): t-1 1.02*** 0.28** 0.19** 0.14* [0.239] [0.106] [0.089] [0.072] MFGSERV (%) (1): 2020; (2)-(4): t 1.09*** 0.44** 0.43*** 0.38*** [0.195] [0.179] [0.141] [0.121] DEINDU (%) (1): 1980-2020; (2)-(4): 1980-t 0.04 0.53 0.42 0.41* [0.343] [0.402] [0.266] [0.238] Beta Coef. NRXGDP 0.49 0.16 0.10 0.07 Beta Coef. MFGSERV 0.57 0.26 0.26 0.24 Beta Coef. DEINDU 0.01 0.10 0.07 0.07 Panel B: Decomposing NRXGDP (1) (2) (3) (4) FMXGDP (%) (1): 2020; (2)-(4): t-1 0.91*** 0.29*** 0.22** 0.17** [0.235] [0.101] [0.090] [0.076] AGXGDP (%) (1): 2020; (2)-(4): t-1 1.22** 0.20 0.02 -0.03 [0.610] [0.254] [0.175] [0.070] Beta Coef. FMXGDP 0.40 0.15 0.11 0.08 Beta Coef. AGXGDP 0.22 0.05 0.00 0.00 Obs.; Controls; Country FE, Year FE 115; Y; N 347; Y; Y 693; Y; Y 1,387; Y; Y Notes: Robust SE (clust. at the country level in cols. (2)-(4)) in parentheses. The six variables have the following summary statistics in col. (1): URB: mean = 52.1; SD = 19.0; min = 13.3; max = 100.0; NRXGDP: mean = 7.9; SD = 9.0; min = 0.4; max = 63.9; MFGSERV: mean = 70.5; SD = 9.9; min = 34.2; max = 93.6; DEINDU: mean = 4.6; SD = 5.0; min = 0.0; max = 21.3; FMXGDP: mean = 4.4; SD = 7.5; min = 0.0; max = 59.8; and AGXGDP: mean = 2.2; SD = 2.7; min = 0.0; max = 37.1. * p<0.10, ** p<0.05, *** p<0.01. 42 Table 3: Resources, (De-)Industrialization & Urban Employment, Cross-Section, c. 2000 Dep.Var. = (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Empl. Sh. of: MFG FIRE SUM NTR NTR2 NTR3 GOVT GOVT2 NRX CONST MFGSERV 0.18** 0.10** 0.28** -0.21* -0.29* -0.23 -0.02 -0.01 0.05 -0.02 [0.09] [0.05] [0.11] [0.12] [0.14] [0.15] [0.03] [0.06] [0.10] [0.04] NRXGDP -0.02 0.03 0.01 0.09 0.16 0.22 -0.13** -0.09 0.11 -0.17** [0.11] [0.07] [0.15] [0.15] [0.20] [0.26] [0.06] [0.13] [0.16] [0.07] DEINDU -0.73* 0.09 -0.64* 0.38 0.88* 1.15* -0.23 -0.20 0.20 -0.40* [0.37] [0.11] [0.36] [0.36] [0.50] [0.63] [0.15] [0.29] [0.29] [0.21] NRXGDP -0.20** -0.07* -0.27** 0.29* 0.45*** 0.44** -0.11** -0.08 0.06 -0.15** - MFGSERV [0.10] [0.04] [0.11] [0.15] [0.15] [0.20] [0.05] [0.10] [0.14] [0.06] DEINDU -0.91** -0.01 -0.92** 0.58 1.17** 1.38** -0.22 -0.18 0.15 -0.37* - MFGSERV [0.35] [0.10] [0.35] [0.38] [0.48] [0.61] [0.14] [0.28] [0.28] [0.21] Mean; Ctrls 20.7; Y 5.0; Y 25.7; Y 20.7; Y 25.3; Y 25.4; Y 6.5; Y 13.9; Y 13.0; Y 6.8; Y Notes: Observations = 61 countries. This table shows the correlation between the employment share of each sector in urban areas c. 2000 and measures of natural resource exports, industrialization/FIRE-ization, and deindustrialization, also defined with respect to 2000. (3) SUM = MFG + FIRE. Robust SE. Table 4: Resources, (De-)Industrialization & Urban Informality, Cross-Section, c. 2000 Dep.Var. = Empl. Sh. in Urban Empl. Empl. Sh. in Urban MFG Empl. Sh. in Urban NTR Type of Wage Self Unpaid Wage Self Unpaid Wage Self Unpaid Employment: Work Empl. Empl. Work Empl. Empl. Work Empl. Empl. (1) (2) (3) (4) (5) (6) (7) (8) (9) MFGSERV 0.54* -0.62** 0.08 1.00*** -0.91*** -0.09 0.63* -0.84*** 0.21 [0.28] [0.24] [0.07] [0.30] [0.25] [0.07] [0.31] [0.26] [0.14] NRXGDP 0.06 -0.22 0.16 0.56 -0.52 -0.05 -0.14 0.00 0.14 [0.37] [0.28] [0.12] [0.37] [0.32] [0.09] [0.43] [0.40] [0.13] DEINDU -0.33 0.19 0.15 -0.47 0.31 0.16 0.52 -0.88 0.36* [0.69] [0.57] [0.17] [1.08] [0.90] [0.21] [0.88] [0.80] [0.21] NRXGDP -0.48 0.40 0.08 -0.44 0.39 0.05 -0.77** 0.84** -0.07 - MFGSERV [0.31] [0.24] [0.09] [0.35] [0.29] [0.08] [0.33] [0.32] [0.11] DEINDU -0.88 0.80 0.07 -1.47 1.22 0.25 -0.10 -0.05 0.15 - MFGSERV [0.72] [0.62] [0.15] [1.07] [0.90] [0.20] [0.83] [0.79] [0.15] Mean 58.0 34.8 7.2 61.2 33.9 5.0 35.8 58.4 5.8 Obs.; Ctrls 55; Y 55; Y 55; Y 54; Y 54; Y 54; Y 54; Y 54; Y 54; Y Notes: This table shows the correlation between the employment share of each type of employment in urban areas or specific sectors in urban areas c. 2000 and measures of natural resource exports, industrialization/FIRE-ization, and deindustrialization, also defined with respect to 2000. NTR = non- tradables (domestic wholesale and retail trade). Robust SE. * p<0.10, ** p<0.05, *** p<0.01. 43 Table 5: Natural Resources, Structural Change & Tall Building Construction, 1960-2020 Type of Buildings: All Resid. Non-Res. Office Hotel Retail Gvt (1) (2) (3) (4) (5) (6) (7) Panel A: Core Controls Long-Difference: Dep. Var.: Log Urban Height Density c. 2020 MFGSERV2020 0.08*** 0.09** 0.10*** 0.11*** 0.07** 0.07* 0.02 [0.024] [0.040] [0.029] [0.029] [0.034] [0.043] [0.029] NRXGDP1960−2020 0.05 0.10* 0.06** 0.06* 0.02 0.05 0.04 [0.031] [0.057] [0.030] [0.031] [0.051] [0.052] [0.038] DEINDU1980−2020 -0.01 -0.04 -0.00 -0.05 0.04 -0.06 -0.03 [0.028] [0.050] [0.035] [0.036] [0.039] [0.054] [0.047] NRXGDP - MFGSERV -0.03 0.01 -0.04 -0.05 -0.04 -0.02 0.01 [0.03] [0.06] [0.03] [0.03] [0.05] [0.06] [0.04] DEINDU - MFGSERV -0.08*** -0.13** -0.10** -0.16*** -0.03 -0.13** -0.06 [0.03] [0.06] [0.04] [0.04] [0.05] [0.06] [0.05] Obs; Controls 115; Y 115; Y 115; Y 115; Y 115; Y 115; Y 115; Y Panel B: + Urban Dvt Ctrls Long-Difference: Dep. Var.: Log Urban Height Density c. 2020 MFGSERV2020 -0.01 -0.02 0.01 0.02 -0.05 0.01 -0.02 [0.023] [0.043] [0.022] [0.024] [0.029] [0.040] [0.035] NRXGDP1960−2020 0.02 0.03 0.05** 0.07** -0.02 0.07 0.01 [0.027] [0.050] [0.023] [0.026] [0.046] [0.056] [0.051] DEINDU1980−2020 -0.02 -0.04 -0.03 -0.06** 0.03 -0.09 -0.03 [0.026] [0.042] [0.024] [0.025] [0.031] [0.056] [0.048] NRXGDP - MFGSERV 0.03 0.05 0.05** 0.04* 0.02 0.06 0.03 [0.02] [0.05] [0.02] [0.02] [0.04] [0.07] [0.06] DEINDU - MFGSERV -0.01 -0.01 -0.04 -0.09*** 0.07* -0.10 -0.00 [0.04] [0.06] [0.03] [0.03] [0.04] [0.06] [0.06] Panel C: + Urban Dvt Ctrls 10-Yr Panel (w/ 4 Lags): Dep. Var.: Log Urban Height Density in t Sum for MFGSERV Lags 0.05 0.08 -0.01 0.01 -0.03 -0.11 0.05 [0.06] [0.09] [0.06] [0.06] [0.08] [0.12] [0.12] Sum for NRXGDP Lags 0.10** 0.09 0.04 0.04 0.01 0.10 0.07 [0.04] [0.06] [0.04] [0.05] [0.05] [0.07] [0.05] Sum for DEINDU Lags -0.08 0.09 -0.06 -0.05 -0.03 0.40* -0.79*** [0.07] [0.14] [0.07] [0.07] [0.15] [0.23] [0.30] Obs; Cntry & Yr FE, Ctrls 346; Y 346; Y 346; Y 346; Y 346; Y 346; Y 346; Y Notes: Robust SE in parentheses (clust. at the country level in Panel C). Summary statistics for LUHT in Panels A-B: All: Mean = 4.6; SE = 1.7; Min = -2.0; Max = 9.6. Resid: Mean = 3.3; SE = 2.7; Min = -3.7; Max = 9.3. Non-Res.: Mean = 3.9; SE = 1.8; Min = -2.4; Max = 8.6. * p<0.10, ** p<0.05, *** p<0.01. 44 Table 6: Natural Resources, (De-)Industrialization & Vanitous Tall Buildings, 1960-2020 (1) (2) (3) (4) (5) (6) Panel A: + Urban Dvt Ctrls Long-Difference: Dep. Var.: Log Non-Residential Height Density c. 2020 Buildings ≥ ... Meters: All 125 140 160 200 240 MFGSERV2020 0.01 -0.00 -0.03 0.06 0.03 0.01 [0.022] [0.039] [0.037] [0.045] [0.051] [0.049] NRXGDP1960−2020 0.05** 0.04 0.08** 0.09* 0.13** 0.14** [0.023] [0.035] [0.036] [0.043] [0.063] [0.062] DEINDU1980−2020 -0.03 -0.06 -0.07* -0.04 0.02 0.01 [0.024] [0.044] [0.038] [0.047] [0.065] [0.061] NRXGDP - MFGSERV 0.05** 0.05 0.11*** 0.03 0.11 0.12* [0.02] [0.03] [0.03] [0.06] [0.07] [0.07] DEINDU - MFGSERV -0.04 -0.05 -0.04 -0.10 -0.01 -0.00 [0.03] [0.06] [0.05] [0.06] [0.08] [0.08] Obs; Controls 115; Y 115; Y 115; Y 115; Y 115; Y 115; Y Panel B: + Urban Dvt Ctrls 10-Yr Panel (w/ 4 Lags): Dep. Var.: Log Non-Residential Height Density t Buildings ≥ ... Meters: All 125 140 160 200 240 Sum for MFGSERV Lags -0.01 0.01 -0.03 0.01 -0.05 -0.15 [0.06] [0.09] [0.09] [0.09] [0.11] [0.11] Sum for NRXGDP Lags 0.04 -0.01 0.00 0.13* 0.13* 0.10 [0.04] [0.07] [0.07] [0.08] [0.08] [0.08] Sum for DEINDU Lags -0.06 0.10 0.13 0.58** 0.46* 0.39 [0.07] [0.13] [0.12] [0.26] [0.27] [0.25] Obs; Cntry & Yr FE, Ctrls 346; Y 346; Y 346; Y 346; Y 346; Y 346; Y Panel C: Long.Diff; Dep.Var.: Vanity Log Non-Residential Height Density c. 2020 Based on ... + Urban Dvt Ctrls (Tip-Occup.) Steel Concrete Capital Cap+Largest Other MFGSERV2020 -0.06 0.13*** 0.01 0.01 0.01 0.05 [0.080] [0.040] [0.038] [0.024] [0.041] [0.038] NRXGDP1960−2020 0.09 0.03 0.08** 0.05* 0.03 0.09** [0.065] [0.052] [0.033] [0.026] [0.054] [0.037] DEINDU1980−2020 0.00 0.02 -0.04 -0.03 -0.03 0.09 [0.035] [0.054] [0.033] [0.029] [0.044] [0.059] NRXGDP - MFGSERV 0.15*** -0.10* 0.07** 0.03 0.03 0.04 [0.05] [0.05] [0.03] [0.03] [0.06] [0.04] DEINDU - MFGSERV 0.06 -0.11 -0.05 -0.04 -0.04 0.04 [0.07] [0.07] [0.06] [0.03] [0.06] [0.07] Obs; Controls 38; Y 115; Y 115; Y 115; Y 115; Y 115; Y Notes: Robust SE in parentheses (clust. at the country level in Panel B). 45 Table 7: Natural Resources, (De-)Industrialization & Construction, Long-Diff., 1970-2020 Dependent Variable: Log Sum Cement (Tons; 70-20) Log Sum Constr. Slum Share Long-Difference Production Consumption GDP (70-20; PPP) (%) c. 2010 (1) (2) (3) (4) (5) (6) (7) (8) MFGSERV 2020/2010 0.05*** 0.01 0.05*** 0.01 0.01 -0.01* -0.55* 0.18 [0.018] [0.014] [0.019] [0.014] [0.02] [0.01] [0.32] [0.31] NRXGDP 1970-2020/2010 0.02 0.02 0.01 0.01 0.00 -0.01 -0.65* 0.14 [0.021] [0.017] [0.020] [0.016] [0.02] [0.01] [0.36] [0.37] DEINDU 1980-2020/2010 -0.01 0.01 -0.01 0.01 -0.01 -0.01 -0.16 0.06 [0.034] [0.021] [0.034] [0.019] [0.02] [0.01] [0.74] [0.57] NRXGDP - MFGSERV -0.03 0.01 -0.03 0.01 -0.01 -0.02 -0.10 -0.04 [0.02] [0.02] [0.02] [0.02] [0.02] [0.03] [0.36] [0.33] DEINDU - MFGSERV -0.06* -0.00 -0.06 0.01 0.01 0.00 0.39 -0.12 [0.04] [0.02] [0.04] [0.02] [0.01] [0.02] [0.70] [0.61] NRXGDP - DEINDU 0.03 0.00 0.03 0.00 0.01 0.01 -0.49 0.09 [0.04] [0.02] [0.04] [0.02] [0.03] [0.03] [0.62] [0.46] Obs.; Core Controls 80; Y 80; Y 80; Y 80; Y 115; Y 115; Y 91; Y 91; Y Ctrls Urban Econ Dvt. N Y N Y N Y N Y Notes: Robust SE in parentheses: * p<0.10, ** p<0.05, *** p<0.01. 46 Table 8: Natural Resources, (De-)Industrialization & Urban Primacy, Country-Level Dependent Variable: Urban Primacy (%) Urban Primacy (%) in t (Panel) in 2020 (Long-Diff.) 3 Lags Incl. 4 Lags Incl. (1) (2) (3) (4) (5) (6) MFGSERV ((3)-(6): sum of lags) -0.19 -0.2 -0.10 -0.05 -0.09 -0.05 [0.13] [0.14] [0.10] [0.13] [0.10] [0.13] NRXGDP ((3)-(6): sum of lags) -0.11 0.02 0.05 [0.15] [0.07] [0.09] DEINDU ((3)-(6): sum of lags) -0.08 -0.11 -0.09 -0.05 -0.1 -0.05 [0.17] [0.17] [0.18] [0.19] [0.18] [0.18] FMXGDP ((3)-(6): sum of lags) -0.09 0.06 0.08 [0.16] [0.08] [0.11] AGXGDP ((3)-(6): sum of lags) -0.46 -0.11 -0.11 [0.36] [0.18] [0.20] NRXGDP - MFGSERV 0.08 0.12 0.10 [0.18] [0.09] [0.14] DEINDU - MFGSERV 0.11 0.09 0.01 -0.01 0.00 0.00 [0.20] [0.18] [0.24] [0.23] [0.28] [0.27] FMXGDP - MFGSERV 0.11 0.15 0.13 [0.20] [0.10] [0.16] AGXGDP - MFGSERV -0.27 -0.01 -0.06 [0.36] [0.19] [0.21] Country, Year FE N N Y Y Y Y Observations 115 115 462 346 462 346 Notes: Robust SE (clust. at the country level in (3)-(6)) in parentheses. * p<0.10, ** p<0.05, *** p<0.01. 47 Table 9: Resources, (De-)Industrialization, & Urban Human Capital, Cross-Section, c. 2000 (1) (2) (3) (4) (5) Panel A: Dep. Var.: Avg Number Harmonized Yrs Educ. x Returns Yrs Educ. x (urban; c. 2000-2020) Years Educ. Test Score Test Score to Educ. Returns MFGSERV 0.04* 0.93 0.03 0.15*** 2.06*** [0.02] [0.62] [0.02] [0.04] [0.49] NRXGDP 0.07** -0.30 0.04* 0.11** 2.01*** [0.03] [0.93] [0.03] [0.05] [0.55] DEINDU 0.16** 1.82 0.14 -0.03 0.05 [0.07] [3.14] [0.12] [0.13] [0.93] NRXGDP - MFGSERV 0.03 -1.23** 0.01 -0.05 -0.05 [0.03] [0.54] [0.03] [0.03] [0.55] DEINDU - MFGSERV 0.11 0.89 0.11 -0.18 -2.00* [0.08] [3.00] [0.12] [0.13] [1.01] Observations 53 101 51 84 48 Panel B: Dep. Var.: Ret. Educ. Ret. Educ. Ret. Educ. Ret x Yrs Edu Ret x Yrs Edu (urban; c. 2000-2020) Reg 01 FE Reg 02 FE PSU FE Reg 01 FE Reg 02 FE MFGSERV 0.25*** 0.18*** 0.14*** 2.00*** 2.05*** [0.07] [0.05] [0.03] [0.50] [0.48] NRXGDP 0.12 0.13** 0.17** 0.65 2.07*** [0.09] [0.06] [0.08] [0.66] [0.62] DEINDU 0.16 0.09 -0.05 0.11 0.91 [0.14] [0.15] [0.37] [1.40] [0.86] NRXGDP - MFGSERV -0.14* -0.04 0.03 -1.35 0.02 [0.07] [0.02] [0.07] [0.99] [0.61] DEINDU - MFGSERV -0.09 -0.08 -0.18 -1.89 -1.14 [0.13] [0.14] [0.37] [1.15] [0.89] Observations 46 66 51 28 40 Panel C: Dep. Var.: Ret x Yrs Edu Returns to Ret. Exp. Ret. Exp. Ret. Exp. (urban; c. 2000-2020) PSU FE Experience Reg 01 FE Reg 02 FE PSU FE MFGSERV 1.35*** 0.01 0.02 0.01 0.01 [0.30] [0.01] [0.02] [0.01] [0.01] NRXGDP 1.67** -0.02 0.02 -0.02 0.01 [0.75] [0.01] [0.01] [0.01] [0.02] DEINDU 3.57 -0.02 -0.01 -0.03 -0.21*** [2.08] [0.03] [0.03] [0.03] [0.07] NRXGDP - MFGSERV 0.32 -0.03*** 0.00 -0.03*** -0.00 [0.69] [0.01] [0.03] [0.01] [0.01] DEINDU - MFGSERV 2.22 -0.03 -0.02 -0.04 -0.22*** [2.22] [0.03] [0.03] [0.03] [0.07] Observations 29 84 46 66 51 Notes: Cross-sectional regressions c. 2000 (2020 for the test scores). See text for details. Robust SE. 48 WEB APPENDIX NOT FOR PUBLICATION A Regional Patterns of Production & Consumption Cities Asia. Web Appx. Fig. D.1 shows the location of production cities and consumption cities in Asia. Unlike China, India has a mix of specialized production cities (e.g., Bangalore and Kanpur) and consumption cities (e.g., Kolkata). Apart from some large production cities in Malaysia and Vietnam, other cities in Asia (e.g, in Indonesia, the Islamic Republic of Iran, Iraq, and the Philippines) are either consumption or neutral cities. Africa. Data are sparser in Africa (Web Appx. Fig. D.2). Nevertheless, one can see the dearth of production cities. With the exception of South Africa, most African cities are consumption cities, with more extreme consumption cities (in red) in Nigeria, Sierra Leone, Sudan or Mozambique. This is expected given the reliance of many African countries on resource exports, which fuels consumption in urban areas. Europe. In Europe (Web Appx. Fig. D.3), countries have production and neutral cities. Productive urbanization increases as one moves away from the “edge” of Europe, i.e. Southern Spain or Eastern Turkey, in line with the division of labor and production sharing within the EU bloc and the importance of intra- and inter-regional trade. North America. North America (Web Appx. Fig. D.4) has many production and neutral cities located in the Northeast or the East Coast of the United States and the North and Center of Mexico (so either where there are maquiladoras or close to Mexico City). Production cities can be seen in Central America, where production sharing brought in manufacturing activities from the United States (or Mexico) following reforms in the 1990s. Californian cities are neither consumption nor production centers, while Southern U.S. cities (e.g., Miami and Houston), coastal cities in Mexico (e.g., Acapulco and Cancun), and cities in the Dominican Republic or Haiti are (non-extreme) consumption cities (in yellow or orange). South America. In South America (Web Appx. Fig. D.5), only the Southeastern areas ˜o Paulo is, however, a neutral city). The rest of of Brazil have production cities (Sa Brazil has consumption cities (Rio is a clear consumption city). In Argentina, Bolivia, Chile and Paraguay, cities are either neutral (e.g., Buenos Aires and Santiago de Chile) or consumption cities. Ecuador and Peru have slightly more consumption cities and ´ Colombia and the Republica Bolivariana de Venezuela have many large and small 49 a is a “medium” consumption city but Barranquilla is an consumption cities (Bogot´ “extreme” consumption city). B Robustness for the Mapping Analysis Non-Linearities. We obtain similar residuals to those presented in the main text if we also include the square, cube, and perfect fourth of the urban share in 2000, and their interactions with the population dummies (coefficient of correlation = 0.99), in case there are non-linearities in the relationship between MFG+FIRE employment, urban economic development, and city size. Per Capita GDP. We obtain a coefficient of correlation of 0.97 if we use log per capita GDP in 2000 (PPP and constant international dollars) instead of urbanization (including the square, cube, and perfect fourth of log per capita GDP, and their interactions with the population dummies). This is not surprising since urbanization rates and log per capita GDP are highly correlated cross-sectionally (correlation of 0.91 in 2000; N = 178). Omitting Urbanization. The correlation is 0.99 if we do not control for the urban share and do not interact the population dummies with it. We then do not allow the relationship between city employment and size to change with urban economic development. Using the Raw Employment Shares. We obtain a coefficient of correlation of 0.89 if we simply consider the raw, i.e. non-residualized, city-specific employment shares. While the residualization was a priori important to ensure we compare apples with apples, the very high correlation indicates the residualization is not entirely necessary. Weights. We obtain a coefficient of correlation of 1.00 if we do not modify the weights so as to over-weigh developed countries (which are under-represented in IPUMS). In that case, the weights are only based on the FUAs’ population levels c. 2000. We obtain a coefficient of correlation of 1.00 if we do not use weights at all. Alternative City Categorizations. If we do not combine the top two population categories into one category, we obtain a coefficient of correlation of 1.00. If we use 5 population categories instead of 10 categories, we still get a correlation of 0.99. Urban Definition. We can control for the urban definition used by the country c. 2010. We include dummies identifying whether the definition is based on a population threshold, another condition, an administrative function, or a combination of these, and the log of the threshold (U.N., 2011). We then interact these variables with the population dummies. The correlation of the residuals remains very strong (about 0.9). Another related question 50 is whether we could instead of focusing on urban observations in FUAs consider all observations in administrative units with a population density above a certain threshold. However, this would include rural workers. In addition, population densities in urban areas are much higher in developing countries than in developed countries (Jedwab et al., 2021c). It is likely similar in rural areas. A high threshold would exclude rich country cities. A low threshold would then include rural areas/workers in poorer countries. Urban Non-Tradables. The correlation with the residuals when the dependent variable is the employment share of the non-tradable domestic “wholesale and retail trade” sector (DWRT) is -0.51 (we include “other services” as it appears that IPUMS mistakenly reclassified some DWRT activities for a few countries). Adding “household services”, it becomes -0.53. It is lower than -1.00 as other sectors see their share increase when MFG+FIRE decreases. Also including “public administration”, it becomes -0.60. Informality. The correlation with the residuals when the dependent variable is the self- employment share (estimated including unpaid workers) is about -0.45. It is lower than -1.00 because self-employment is an imperfect proxy for informality. Indeed, high-skilled workers of the MFG+FIRE sector could be self-employed despite belonging to the formal sector. Nonetheless, the fact that the correlation is almost equal to -0.5 is reassuring. C Theory Appendix for Propositions 1-4 In this appendix we explain how we obtain propositions 1-4. C1. Resource Revenues and Consumption Cities Our Proposition 1 reiterates GJV16’s result that resource revenues R offer a path to urbanization U and the emergence of “consumption cities”. Indeed, employment in urban non-tradables Ln is increasing in R whereas employment in manufacturing and tradable services Lm is decreasing in R. In other words, a positive shock to R leads to the emergence of “consumption cities”. The overall effect on urbanization is also positive. Proposition 1 (Urbanization through commodity rents and “consumption cities”) From (C.23), (C.17), (C.19) and (C.21) below, we have the following: ∂U ∂Ln ∂Lm ∂Lf > 0, > 0, < 0, <0 ∂R ∂R ∂R ∂R Proof: From (6) we have the following implicit function for Ln : 51 (1 − Ln )α F = Ln − βn 1 + R − p∗ f cf =0 (C.15) A From the implicit function theorem: ∂Ln FR =− (C.16) ∂R FLn The partial derivatives of F with respect to R and Ln are respectively: (1 − Ln )α FR = −βn (C.16a) A (1 − Ln )(α−1) FLn = 1 + βn α (R − p∗ f cf ) (C.16b) A From (C.16) we obtain that: α ∂Ln βn (1−Ln ) A = (1 −L )α−1 (C.17) ∂R 1 + βn α n (R − p∗ f cf ) A Both the numerator and the denominator are positive. The denominator is positive not only when the country is resource rich and R − p∗ ∗ f cf > 0, but also when R − pf cf < 0 because in this case we can show that the following inequality holds using the fact that since both α < 1, βn < 1 then αβn < 1. Replacing αβ with 1 results in a smaller expression because R − p∗ f cf < 0. Then from (6) and because both Ln < 1 and βn < 1 we have: (1 − Ln )α−1 ∗ (1 − Ln )α−1 Ln (1 − βn ) 1 + αβ n R − p f cf > 1 + R − p∗ f cf = >0 A A (1 − Ln ) βn After substituting (6) in (7a), we obtain: 1 (1 − Ln )α p∗ m Am α Lm = 1 − βn 1+ R − p∗ f cf (C.18) A A Differentiating with respect to R, from (C.18) we obtain: 1 ∂Lm (1 − Ln )α p∗ m Am α = −βn <0 (C.19) ∂R A A Eq. (C.19) shows the (urban) Dutch Disease effect of an increase in resource revenue. The effect is larger for countries with small non-tradable sectors (these are mostly low- income countries) and for countries with relatively productive tradable urban activities. Substituting (6) in (7b) we obtain: 52 1 (1 − Ln )α p∗ f Af α Lf = 1 − βn 1+ R − p∗ f cf (C.20) A A Differentiating (C.20) with respect to R we get: 1 ∂Lf (1 − Ln )α p∗ f Af α = −βn <0 (C.21) ∂R A A Resource windfalls shift resources away from agriculture. The shift is stronger the smaller the non-tradable sector and the higher agricultural productivity is relative to the average in the country. Now we turn to the urbanization rate. From (6) and (C.18) we have: (1 − Ln )α U = Ln + Lm = βn 1 + R − p∗ f cf A 1 (1 − Ln )α p∗ m Am α + 1 − βn 1+ R − p∗ f cf (C.22) A A Differentiating (C.22) with respect to R we obtain: 1 ∂U (1 − Ln )α p∗ m Am α = βn 1− >0 (C.23) ∂R A A End of Proof: Resource windfalls cause de-industrialization, but enable urbanization and the shift from rural and urban tradables to urban non-tradables in “consumption cities”. C2. Agricultural Growth and Consumption Cities Faster productivity growth in agriculture has an income effect and a foreign earnings effect if the country exports agricultural products. Both result in a disproportionate increase of urban non-tradables, while the increase in foreign earnings enables the importing of urban tradables, whose share in employment decreases. Lastly, if the level of agricultural productivity is high enough, the urbanization rate increases as the urban non- tradable effect dominates the urban tradable effect. However, if the level of agricultural productivity is not high enough (especially relative to urban tradables), an increase in agricultural productivity may have an effect of pulling resources back to agriculture in order to meet the agricultural sufficiency requirement (in which case more of the urban tradable consumption is provided internally). Then the urbanization rate decreases. While this is possible, we discuss below why we think that de-urbanization is unlikely. For brevity in exposition and for the sake of simplicity, we define x and y as follows: 53 x = p∗ ∗ m Am for urban tradables and y = pf Af for urban non-tradables. This allows us to explore not only the effect of productivity changes on urbanization and employment, but also the effect of price shocks affecting the agricultural and manufacturing sectors. For example, agricultural exports could increase because of productivity, Af , increases and/or because of increases in world demand, and therefore the world price (p∗ f ) for the country’s agricultural product. It is important to clarify that, in our mind, the world agricultural price, p∗ f , in the agricultural subsistence constraint, differs from the world price for the agricultural commodities exported by the country. The price in the subsistence constraint represents the price for the basket of goods consumed locally which can be assumed to be fixed or to change less in response to global demand changes than the prices of the country’s main agricultural exports. Proposition 2 (Productivity growth in agriculture and “consumption cities”) So long as R < p∗ f cf , from (C.26), (C.27), (C.29) and (C.30) below, it follows that: ∂Ln ∂Lm > 0, <0 ∂y ∂y ∂U ∂Lf 1 ∗ 1 < 0, > 0, if α(p∗ f Af ) < (pm Am ) α α ∂y ∂y ∂U ∂Lf 1 ∗ 1 > 0, < 0, if α(p∗ f Af ) > (pm Am ) α α ∂y ∂y Proof: From (C.15) and the implicit function theorem and noticing that y = p∗ f Af , we have: ∂Ln Fy =− (C.24) ∂y FLn From (C.15) the partial derivatives of F with respect to y is respectively: (1 − Ln )α 1 −1 Fy = βn α+1 y α R − p∗ f cf (C.25) 1 1 x +y α α From (C.16b), (C.24) and (C.25), we have: (1−Ln )α 1 −β n 1 1 α+1 y α −1 R − p∗ f cf ∂Ln x α +y α = α− 1 (C.26) ∂y 1+ βn α (1−LA n) R− p∗ f cf As long as R < p∗ f cf , i.e. the country is not particularly resource rich, the increase 54 in agricultural productivity shifts resources into non-tradables ( ∂L ∂y n > 0). In this case, the numerator is positive and, as shown above, the denominator is positive too. Differentiating (C.18) with respect to y , we get: 1 1 ∂Lm 1 x α y α −1 (1 − Ln )α = 2 (βn − 1) + βn R − p∗ f cf (1 + α) (C.27) ∂y α 1 1 A xα + y α The first term in the brackets in (C.27) is negative and so is the second one in resource poor countries as R − p∗ f cf < 0. Therefore, ∂Ln ∂y <0 Differentiating (C.22) with respect to y , we obtain:   α 1 −1 ∂U (1 − Ln ) y α = −β n R − p∗ cf    α +1 f ∂y 1 1 xα + y α 1 1 1 x α y α −1 (1 − Ln )α + 2 (βn − 1) + βn R − p∗ f cf (1 + α) (C.28) α 1 1 A x +y α α Rearranging the terms in (C.28) we get: 1 1 1 ∂U y α −1 1 xα 1 + α (1 − Ln )α xα α = 1 1 (βn − 1) 1 1 + β n ( ) (R − p∗ f cf )( 1 1 − ) ∂y xα + y α α xα + y α α A xα + y α 1+α (C.29) From (C.29) we see that the effect of agricultural productivity growth on urbanization is negative in resource-poor countries where R−p∗ f cf < 0 and manufacturing productivity 1 xα α is relatively high ( 1 1 > 1+α ). In this case, the first and second terms in the square x α +y α brackets are negative and there is a shift in employment away from urban areas ( ∂U ∂y < 0). However, when manufacturing productivity is low, the first term is small and the second term is positive. In this case, the agricultural productivity shock spurs urbanization ( ∂U ∂y > 0). Differentiating (C.20) with respect to agricultural productivity y, we obtain: 1 ∂Lf 1 y α −1 1 (1 − Ln )α 1 1 = 2 (1 − βn )x α + βn R − p∗ f cf (αy − x ) α α (C.30) ∂y α 1 1 A x +y α α In resource-poor countries (i.e. R − p∗ f cf < 0) with sufficiently high manufacturing 1 1 productivity so that α y α − x α < 0, a productivity boost in agriculture shifts resources ∂Lf 1 1 into rural areas. In this case, ∂y > 0. Please note that whenever condition α y α − x α < 0 55 1 xα α is satisfied so is 1 1 > 1+α , which ensures that the productivity boost in agriculture x α +y α ∂U has an opposite effect on urbanization, i.e. ∂y < 0. However, when manufacturing 1 1 1 xα α productivity is low (i.e. α y α −x α > 0 and 1 1 < 1+α . the first terms in (C.29) x α +y α and (C.30) are small so the second terms dominate. In (C.29) the second term is positive ∂U implying a shift of labor into urban areas, i.e. ∂y > 0, while in (C.30) the second term is ∂Lf negative, implying a shift of labor away from agriculture, i.e. ∂y < 0. End of Proof: In sum, if the level of agricultural productivity is high enough, agricultural development leads to de-industrialization but enables urbanization and the shift from rural and urban tradables to urban non-tradables in “consumption cities”. C3. Industrial and/or Service Revolution and Production Cities We discuss how a manufacturing/FIRE revolution leads to production cities. Proposition 3 (Urbanization through industrialization and “production cities”) From (C.33), (C.34), (C.35), and (C.36) below, we have: ∂U ∂Ln ∂Lm ∂Lf ∗ > 0, ∗ > 0, ∗ > 0, <0 ∂pm Am ∂pm Am ∂pm Am ∂p∗ m Am so long as R − p∗ f cf < 0 and agricultural productivity is sufficiently high: 1 1 (α(p∗ ∗ m Am ) < (pf Af ) ). α α Proof: From (C.15) and the implicit function theorem and noticing that x = p∗ m Am : ∂Ln Fx =− (C.31) ∂x FLn The partial derivatives of F with respect to x is: 1 1 (1 − Ln )α x α −1 (1 − Ln )α x α −1 Fx = βn α+1 R− p∗ f cf = βn 1 1 R − p∗ f cf . (C.32) 1 1 x +y α α A x +yα α Using (C.16b), (C.31), and (C.32), we obtain the following result: 1 α x α −1 βn (1−Ln 1 1 ) R − p∗ f cf ∂Ln A x α +y α =− α−1 (C.33) ∂x 1 + βn α (1−LAn) R − p∗f cf The numerator in (C.33) is negative because R is low in resource-poor countries; as ∂Ln shown before, the denominator is positive. Thus, (C.33) is positive and ∂x > 0, implying that a positive productivity shock in manufacturing shifts resources into non-tradables. Differentiating (C.18) with respect to x, we obtain: 56 1 ∂Lm 1 x α −1 1 (1 − Ln )α 1 1 = 2 (1 − βn )y α + βn R − p∗ f cf (αx − y ) α α (C.34) ∂x α 1 1 A x +y α α The first term in the square brackets is positive. The second term is positive when the country is resource poor, i.e. R − p∗ f cf < 0, and agricultural productivity is high enough 1 1 so that α x α − y α < 0. The latter reflects the importance of the Green Revolution for industrial development. Industrialization in countries with low agricultural productivity is slower than in countries with higher agricultural productivity. Thus, the effect of a positive productivity shock in manufacturing is an expansion of employment in ∂Lm manufacturing and tradable services, i.e. ∂x > 0. This suggests that productivity growth in manufacturing and/or tradable services in resource poor countries fosters an expansion in the total employment of these sectors. Differentiating (C.22) with respect to x and using (C.34), gives us the following expression: ∂U 1 (1 − Ln )α = −β n x α −1 α+1 R − p∗ f cf ∂x 1 1 x +yα α 1 1 x α −1 1 (1 − Ln )α 1 1 + 2 (1 − βn )y + βn α R − p∗ f cf (αx − y ) α α (C.35) α 1 1 A x +yα α Both terms in (C.35) are positive when countries are resource-poor (R − p∗ f cf < 0) and agricultural productivity is high enough. In this case, productivity growth in manufacturing and/or tradable services fuels urbanization. Finally, from (C.20) we get: 1 ∂Lf 1 x α −1 1 (1 − Ln )α 1 1 = 2 (βn − 1) y + βn α R − p∗ f cf (y − αx ) α α (C.36) ∂x α 1 1 A x +y α α End of Proof: In (C.36), if agricultural productivity is sufficiently high, then both terms in the square brackets are negative. Thus, a productivity boom in manufacturing / tradable services leads to a shift of resources away from agriculture and into urban tradables. A shock that reduces the country’s relative level of manufacturing productivity should reduce manufacturing employment according to (C.34). For instance, it could 57 be that manufacturing productivity decreases (Am ) or that manufacturing productivity stays the same but other countries’ manufacturing productivity increases, thus lowering manufacturing prices (p∗ ∗ m ). In both cases, (x = pm Am ) would decrease. Of course, this applies to both manufacturing and FIRE. Various factors could account for a decrease in x. First, many countries, in particular in LAC, have adopted in the past ISI policies that artificially increased manufacturing productivity and employment at the expense of other sectors, and also raised the urbanization rate. When these policies were removed, productivity Am declined, but urbanization rates decreased little. Second, increased trade competition in the world, especially with the growth of China (e.g., in manufacturing) and India (e.g., in business services), reduced the world price levels of urban tradables. In countries where urban tradable productivity was initially unchanged, x likely decreased, resulting in the same effects as the removal of ISI policies. Third, the production functions of eq. (2) implicitly assume complementarities between technology and labor. However, new labor-saving technologies have appeared over time in urban tradable sectors, especially in more developed countries. While our model does not explicitly account for this mechanism, it could be interpreted in our model via a lower x, with again the same consequences. In the end, regardless of the “origin” of the reduction in x, production cities see their sectoral composition change as employment in urban tradables declines. If we assume that urban residents do not migrate to rural areas, for example because skills acquired in the urban sectors have no value in the agriculture sector or because agricultural productivity is high, a negative shock to manufacturing will not shift resources from urban to rural areas (as in Proposition 3). Instead, it will shift resources from urban tradables to non-tradables, resulting in the transformation of a production city into a consumption city. Thus, we formulate Proposition 4. D Data Creation: Aggregate Data Sample. We focus on 116 countries that were still “developing” countries (i.e., had not reached high income status) in 1960. We obtain data every 5 years between 1960 and 2010. The full sample thus consists of 116 countries times 13 years = 1,508 observations. GDP Share of Manufacturing and Services. When available we obtain the GDP share of manufacturing and the GDP share of manufacturing and services from the Beta version 58 of the World Development Indicators (WDI) database of the World Bank (2021).53 More recent versions of the WDI do not report these GDP shares for earlier decades, only the older versions of the WDI do. The Beta version has the merit of showing all available yearly estimates simultaneously for all versions of the WDI. For each country-year, we then take the mean of the available estimates. To maximize the number of available estimates for the years 1960, 1965, ..., 2015, 2020, and in order to minimize fluctuations due to year-specific measurement issues, we rely on five-year moving averages. After doing so, of the 1,508 observations in our data, for 189, 159 and 195 observations we still do not have an available estimate for the GDP share of MFG, services (SERV), and MFG+SERV, respectively. For the 2010-2020 period, we complete the data using estimates from Central Intelligence Agency (2021) and reports from international organizations or governmental agencies. Even after doing this, for 188, 153 and 189 observations we still do not have an available estimate, respectively. For the long-difference regressions, we need data c. 1960-1970. For these years, we use United Nations (1960-1980). However, for 87, 75, and 88 observations, we do not have an available estimate for the GDP share of MFG, SERV, and MFG+SERV, respectively. The System of National Accounts (SNA) - Analysis of Main Aggregates (AMA) database of United Nations (2020c) reports the GDP share of aggregated sectors, including MFG and services, for all countries from 1970 to 2020. When needed, we use this database to complete the missing country-years of our main data set, after verifying that the newly added estimates are consistent with the estimates that we already had for other years.54 GDP Share of FIRE.The National Accounts Official Country Data database of United Nations (2020a) reports when available the GDP share of various sectors – using both the ISIC Revision 3 and Revision 4 – from the 1960s to date. The data are patchy, however, and we employ them only to obtain the GDP share of FIRE c. 2020 (observations from 2015-2020). 53 For manufacturing, we use the series “Manufacturing, value added (% of GDP)”. For services, we use as our baseline the series “Services, etc., value added (% of GDP)”. When estimates of the service share are not available, we rely on another WDI series: “Services, value added (% of GDP)”. 54 We do not use SNA-AMA as our baseline database. Indeed, when comparing WDI + the yearbooks and SNA-AMA, it appears that many SNA-AMA estimates were extrapolated. 59 Figure D.1: Map of Production, Neutral, and Consumption Cities, Asia, c. 2000 This figure shows production (Prod., blue), neutral (grey) and consumption cities (Cons., yellow-red). Figure D.2: Map of Production, Neutral, and Consumption Cities, Africa, c. 2000 This figure shows production (Prod., blue), neutral (grey) and consumption cities (Cons., yellow-red). 60 Figure D.3: Map of Production, Neutral, and Consumption Cities, Europe, c. 2000 This figure shows production (Prod., blue), neutral (grey) and consumption cities (Cons., yellow-red). Figure D.4: Map of Production, Neutral, and Consumption Cities, North America, c. 2000 This figure shows production (Prod., blue), neutral (grey) and consumption cities (Cons., yellow-red). 61 Figure D.5: Map of Production, Neutral, and Consumption Cities, South America, c. 2000 This figure shows production (Prod., blue), neutral (grey) and consumption cities (Cons., yellow-red). Figure D.6: GDP Share of MFGSERV vs. GDP Share of MFG+FIRE, c. 2020 (N = 78) (a) Share of MFGSERV vs. MFG + FIRE (b) Share of Services vs. FIRE Notes: The left panel shows that countries with a high GDP share of MFG+FIRE today are countries with a high GDP share of MFGSERV today, thus validating this proxy. Indeed, the right panel shows that countries with a high GDP share of FIRE today are countries with a high GDP share of services today. Figure D.7: World Map of Production Cities in Manufacturing or FIRE, World, c. 2000 62 Notes: This figure shows the location of production cities based on manufacturing only (MFG, cities in various shades of purple), neutral production cities (grey), and production cities based on FIRE only (FIRE, cities in various shades of green). 63 Table D.1: Employment Share of Urban Tradables by City Size, Cross-Section, c. 2000 Dep. Var. = MFGFIREa,00 Coef. SE Coef. SE Capital Citya (CAP) -5.46** (2.44) URBc 0.03 (0.05) Pop. Size CATa = 2 -0.66 (1.42) URBc * Pop. Size CATa = 2 0.07** (0.03) Pop. Size CATa = 3 1.03 (2.35) URBc * Pop. Size CATa = 3 0.08* (0.04) Pop. Size CATa = 4 3.69 (3.99) URBc * Pop. Size CATa = 4 0.04 (0.06) Pop. Size CATa = 5 5.79 (4.41) URBc * Pop. Size CATa = 5 0.01 (0.07) Pop. Size CATa = 6 13.09*** (4.10) URBc * Pop. Size CATa = 6 -0.06 (0.06) Pop. Size CATa = 7 8.59* (4.50) URBc * Pop. Size CATa = 7 0.03 (0.07) Pop. Size CATa = 8 11.70* (5.98) URBc * Pop. Size CATa = 8 0.05 (0.10) Pop. Size CATa = 9 21.76*** (5.35) URBc * Pop. Size CATa = 9 -0.09 (0.08) Constant 16.24*** (4.06) Notes: Obs. = 6,865 urban agglomerations. R2 = 0.20. The dependent variable is the employment share of MFG+FIRE in each urban agglomeration a belonging to country c circa 2000. The other variables are also defined in 2000. Robust SE clustered at the country level in parentheses. * p<0.10, ** p<0.05, *** p<0.01. Table D.2: Additional Correlations for the Long-Difference Regressions Dependent Variable: Urbanization Rate (%) in 2020 (1) (2) (3) NRXGDP (%) (Mean 1960-2020) 1.02*** 0.74*** 0.74*** [0.239] [0.236] [0.275] MFGSERV (%) (2020) 1.09*** [0.195] DEINDU (%) (1980-2020) -0.04 -0.70 -0.88 [0.343] [0.480] [0.628] MFG (%) (2020) 1.57*** 1.06*** [0.279] [0.395] SERV (%) (2020) 0.41* [0.244] FIRE (%) (2020) 0.69** [0.327] SERV (non-FIRE) (%) (2020) 0.57* [0.303] Beta Coef. MFGSERV 0.57 Beta Coef. MFG 0.62 0.22 Beta Coef. SERV 0.16 Beta Coef. FIRE 0.14 Beta Coef. SERV (non-FIRE) 0.16 Controls Y Y Y Notes: Obs. = 115. We control for initial conditions c. 1960 – i.e., the urban share and the value of the variables c. 1960 (except for FIRE and SERV (non-FIRE) since there is no existing data for the FIRE sector in the 1960s – and add the controls for area, population, small islands, and urban definitions. Robust SE in parentheses. 64 Table D.3: Timing of the Correlations btw Urbanization & the Measures, 10-Year Panel Dependent Variable: Urbanization Rate (%) in Year t Correlations with: Baseline 1 Lead l Lag 2 Lags 2 Lags NRXGDP (%) t 0.05 [0.062] NRXGDP (%) t-10 0.19** 0.18*** 0.12* 0.03 [0.089] [0.065] [0.065] [0.080] NRXGDP (%) t-20 0.23*** 0.23*** [0.072] [0.068] NRXGDP (%) t-30 0.14*** [0.046] MFGSERV (%) t+10 0.15 [0.096] MFGSERV (%) t 0.43*** 0.26*** 0.33** 0.37** 0.37** [0.141] [0.090] [0.142] [0.159] [0.168] MFGSERV (%) t-10 0.26** 0.16** 0.17** [0.103] [0.077] [0.078] MFGSERV (%) t-20 0.31** 0.32** [0.127] [0.125] DEINDU (%) 1980-t+10 0.27 [0.238] DEINDU (%) 1980-t 0.42 0.15 0.37 0.11 0.10 [0.266] [0.152] [0.252] [0.271] [0.272] DEINDU (%) 1980-t-10 0.01 -0.09 -0.08 [0.295] [0.170] [0.162] DEINDU (%) 1980-t-20 0.35 0.35 [0.443] [0.440] FMXGDP (%) t-10 0.05 [0.090] FMXGDP (%) t-20 0.23*** [0.069] FMXGDP (%) t-30 0.14*** [0.047] AGXGDP (%) t-10 -0.04 [0.160] AGXGDP (%) t-20 0.23 [0.195] AGXGDP (%) t-30 0.15 [0.161] Sum for NXGDP 0.35*** 0.40*** [0.12] [0.15] Sum for MFGSERV 0.60*** 0.85*** 0.85*** [0.20] [0.29] [0.29] Sum for DEINDU 0.37 0.37 0.37 [0.32] [0.46] [0.46] Sum for FMXGDP 0.42*** [0.15] Sum for AGXGDP 0.33 [0.34] Cntry FE, Yr FE, Ctrls; Obs Y; 694 Y; 578 Y; 578 Y; 462 Y; 462 Notes: Robust SEs are clustered at the country level. * p<0.10, ** p<0.05, *** p<0.01. 65 Table D.4: Resources, Industrialization & Sectoral Employment / Informality, Panel Dep.Var. = (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Empl. Sh. of ... (t) MFG FIRE SUM NTR NTR2 NTR3 GOVT GOVT2 WAGE SELF MFGSERVt 0.34** 0.06 0.40 -0.29 -0.54** -0.51*** -0.10 -0.15** 0.14 -0.27 [0.154] [0.111] [0.243] [0.239] [0.233] [0.099] [0.083] [0.063] [0.337] [0.271] NRXGDPt−10 0.30* -0.04 0.26 0.21 0.49*** 0.38*** 0.18** 0.04 -0.31 0.30 [0.166] [0.071] [0.217] [0.153] [0.152] [0.104] [0.076] [0.167] [0.257] [0.228] DEINDU1980−t -0.42 -0.03 -0.45 -0.23 -0.47*** -0.12 0.10 -0.17 0.30 -0.51*** [0.368] [0.204] [0.563] [0.161] [0.153] [0.191] [0.126] [0.127] [0.199] [0.17] NRXGDP -0.04 -0.10 -0.14 0.51 1.03*** 0.90*** 0.28* 0.19 -0.45 0.57 - MFGSERV [0.19] [0.13] [0.29] [0.35] [0.30] [0.15] [0.14] [0.18] [0.49] [0.42] DEINDU -0.76*** -0.10 -0.85** 0.06 0.07 0.40* 0.20 -0.03 0.16 -0.24 - MFGSERV [0.27] [0.14] [0.41] [0.20] [0.15] [0.21] [0.15] [0.13] [0.28] [0.23] Observations 124 124 124 124 124 124 120 93 99 99 Country, Year FE Y Y Y Y Y Y Y Y Y Y Controls Y Y Y Y Y Y Y Y Y Y Notes: This table shows the correlation between the employment share of each sector/type of employment in urban areas in t and measures of natural resource exports (t-10), industrialization/FIRE-ization (t), and deindustrialization (1980-t). Robust SE clustered at the country level. * p<0.10, ** p<0.05, *** p<0.01. 66 Table D.5: Resources, Industrialization & Employment, Cross-Section, I2D2 Database Dep.Var. = (1) (2) (3) (4) (5) (6) (7) (8) Empl. Sh. (c. 2005): MFG FIRE MFGFIRE NTRI NTRI2 WAGE SELF UNPAID Panel A: Vars. in 2000 MFGSERV2000 0.12*** -0.04 0.08* 0.09 0.14 0.44** -0.48*** -0.15 [0.035] [0.039] [0.043] [0.127] [0.144] [0.180] [0.136] [0.118] NRXGDP1960−2000 -0.06 0.03 -0.03 0.18 0.03 0.41 -0.30 -0.15 [0.056] [0.045] [0.076] [0.127] [0.209] [0.294] [0.221] [0.121] DEINDU1980−2000 -0.26 0.14 -0.12 0.12 1.42** -0.77 1.08* 0.54 [0.167] [0.105] [0.204] [0.364] [0.597] [0.715] [0.557] [0.326] NRXGDP -0.18*** 0.07 -0.11 0.08 -0.11 -0.03 0.18 -0.00 - MFGSERV [0.04] [0.05] [0.07] [0.17] [0.20] [0.27] [0.22] [0.15] DEINDU -0.38** 0.18 -0.20 0.03 1.28** -1.21 1.56*** 0.69* - MFGSERV [0.16] [0.12] [0.21] [0.40] [0.60] [0.74] [0.57] [0.36] Panel B: Vars. in 2010 MFGSERV2010 0.10 -0.03 0.07 0.09 0.26 0.69*** -0.62*** -0.23 [0.061] [0.043] [0.061] [0.145] [0.199] [0.217] [0.157] [0.142] NRXGDP1960−2010 -0.13** 0.04 -0.10 0.14 0.16 0.42* -0.28 -0.13 [0.065] [0.045] [0.070] [0.143] [0.223] [0.244] [0.195] [0.134] DEINDU1980−2010 -0.09 0.14 0.05 0.10 1.11** -0.02 0.47 0.23 [0.134] [0.085] [0.159] [0.309] [0.503] [0.652] [0.564] [0.242] NRXGDP -0.23*** 0.07 -0.16*** 0.05 -0.09 -0.27 0.34* 0.09 - MFGSERV [0.05] [0.07] [0.06] [0.20] [0.20] [0.22] [0.18] [0.18] DEINDU -0.19 0.17 -0.02 0.02 0.86 -0.71 1.09* 0.45 - MFGSERV [0.15] [0.10] [0.17] [0.36] [0.55] [0.69] [0.57] [0.30] Observations 93 90 90 94 91 94 93 93 Controls Y Y Y Y Y Y Y Y Notes: This table shows the correlation between the employment share of each sector/type of employment in urban areas c. 2005 and measures of natural resource exports, industrialization/FIRE-ization, and deindustrialization, defined with respect to 2000 or 2010. Robust SE. * p<0.10, ** p<0.05, *** p<0.01. 67 Table D.6: Resources, Industrialization & Growth of FUAs, 1975-2015, Cross-Section Dependent Variable Log FUA Pop. 2015 - Log FUA Pop. 1975 Capital + Largest City: Top 1 Top 2 Top 5 Top 0 Top 1 Top 2 Top 5 Top 0 (1) (2) (3) (4) (5) (6) (7) (8) MFGSERV*TOP -0.02 -0.01 0.00 -0.01 -0.02 -0.01 0.00 -0.01 [0.025] [0.024] [0.025] [0.025] [0.026] [0.025] [0.026] [0.026] NRXGDP*TOP -0.01 -0.01 -0.01 -0.00 [0.029] [0.023] [0.018] [0.039] DEINDU*TOP 0.04 0.03 0.01 0.05 0.04 0.03 0.01 0.05 [0.047] [0.046] [0.044] [0.051] [0.047] [0.046] [0.044] [0.051] FMXGDP*TOP -0.01 -0.01 -0.01 -0.00 [0.031] [0.024] [0.019] [0.042] AGXGDP*TOP -0.06 -0.04 -0.05 -0.05 [0.057] [0.052] [0.046] [0.061] TOP*(NRXGDP - MFGSERV) 0.00 0.00 -0.01 0.00 [0.05] [0.04] [0.04] [0.06] TOP*(DEINDU - MFGSERV) 0.06 0.04 0.01 0.06 0.06 0.04 0.01 0.06 [0.07] [0.06] [0.06] [0.07] [0.07] [0.07] [0.06] [0.07] TOP*(FMXGDP - MFGSERV) 0.01 0.00 -0.01 0.01 [0.05] [0.04] [0.04] [0.06] TOP*(AGXGDP - MFGSERV) -0.04 -0.03 -0.05 -0.04 [0.05] [0.05] [0.04] [0.05] Observations 7,422 7,422 7,422 7,422 7,422 7,422 7,422 7,422 Notes: Robust SE in parentheses. * p<0.10, ** p<0.05, *** p<0.01. Table D.7: Resources, Industrialization & Growth of FUAs, 1975-2015, Panel Dependent Variable: Log FUA Pop. in Year t Extra Lags Included 1 Extra Lag 2 Extra Lags Capital + Largest City: Top 1 Top 2 Top 5 Top 0 Top 1 Top 2 Top 5 Top 0 (1) (2) (3) (4) (5) (6) (7) (8) Sum of Lags for MFGSERV 0.01 0.00 0.00 0.00 0.02** 0.01* 0.01* 0.01 [0.01] [0.01] [0.01] [0.01] [0.01] [0.01] [0.01] [0.01] Sum of Lags for NRXGDP 0.01 0.01 0.01 0.01 0.02 0.02 0.02 0.03 [0.02] [0.01] [0.01] [0.02] [0.02] [0.02] [0.02] [0.02] Sum of Lags for DEINDU -0.02 -0.01 -0.01 -0.02 -0.01 -0.01 -0.00 -0.02 [0.02] [0.02] [0.02] [0.01] [0.01] [0.01] [0.01] [0.01] Observations 29,681 29,681 29,681 29,681 22,259 22,259 22,259 22,259 City FE Y Y Y Y Y Y Y Y Country-Year FE Y Y Y Y Y Y Y Y Notes: Robust SE clust. at the city level in parentheses. * p<0.10, ** p<0.05, *** p<0.01. 68 Table D.8: Country-Level Production City-Ness (PCC): Average and Gini Index c. 2000 Panel A: Average Production City-Ness for Each Country Rank Country PCC Rank Country PCC Rank Country PCC 1 Romania 24.6 26 South Africa 1.4 51 Peru -4.7 2 Lesotho 20.3 27 Togo 1.0 52 Ecuador -5.0 3 Slovenia 18.6 28 Jamaica 0.7 53 Myanmar -5.3 4 Mauritius 14.7 29 Papua New Guinea 0.2 54 Haiti -5.6 5 Ireland 12.4 30 Morocco 0.2 55 Ethiopia -5.6 6 Belarus 12.0 31 Thailand -0.1 56 Egypt, Arab Rep. -5.7 7 Honduras 10.8 32 India -0.6 57 Malawi -6.2 8 China 8.8 33 Philippines -0.9 58 Venezuela, RB -6.4 9 Malaysia 8.2 34 Chile -1.0 59 Rwanda -7.5 10 Costa Rica 8.2 35 Dominican Rep. -1.2 60 Zambia -7.7 11 Israel 6.7 36 Benin -1.4 61 Cameroon -8.3 12 Poland 6.2 37 Italy -1.4 62 Colombia -9.7 13 France 6.0 38 United States -1.9 63 Cambodia -10.2 14 Portugal 5.9 39 Guinea -2.0 64 Sudan -10.3 15 Spain 5.5 40 Armenia -2.3 65 Uganda -10.3 16 Guatemala 4.5 41 Botswana -2.5 66 Indonesia -10.5 17 El Salvador 4.3 42 Paraguay -3.1 67 Mozambique -10.8 18 Lao PDR 3.8 43 Brazil -3.1 68 Mali -11.0 19 Fiji 3.6 44 Bolivia -3.2 69 Liberia -11.3 20 Nepal 2.9 45 Argentina -3.3 70 Senegal -12.1 21 Turkey 2.7 46 Panama -3.6 71 Sierra Leone -14.5 22 Vietnam 2.7 47 Iran, Islamic Rep. -3.6 72 Tanzania -14.8 23 Canada 2.3 48 Ghana -3.8 73 Nigeria -15.0 24 Mexico 2.2 49 Jordan -3.8 74 Iraq -17.2 25 Nicaragua 1.9 50 Kyrgyz Republic -4.2 Panel B: Gini Index of Production City-Ness for Each Country Rank Country Gini Rank Country Gini Rank Country Gini 1 Nigeria 0.30 26 Honduras 0.12 51 France 0.06 2 Colombia 0.25 27 Italy 0.12 52 Canada 0.06 3 Indonesia 0.20 28 Mozambique 0.11 53 Dominican Rep. 0.06 4 India 0.20 29 Nicaragua 0.11 54 Armenia 0.06 5 Egypt, Arab Rep. 0.18 30 Peru 0.11 55 Poland 0.06 6 Iran, Islamic Rep. 0.17 31 Tanzania 0.11 56 Belarus 0.06 7 Guatemala 0.16 32 Myanmar 0.11 57 Nepal 0.05 8 Venezuela, RB 0.15 33 Sierra Leone 0.10 58 Senegal 0.05 9 Sudan 0.15 34 Panama 0.10 59 Portugal 0.04 10 Turkey 0.15 35 Philippines 0.10 60 Ghana 0.04 11 Brazil 0.14 36 Vietnam 0.10 61 Haiti 0.04 12 Ethiopia 0.14 37 Ecuador 0.10 62 Uganda 0.03 13 Mexico 0.14 38 United States 0.09 63 Mali 0.03 14 Malaysia 0.13 39 Iraq 0.09 64 Israel 0.03 15 Kyrgyz Republic 0.13 40 Thailand 0.09 65 Ireland 0.03 16 Papua New Guinea 0.13 41 Jamaica 0.09 66 Togo 0.03 17 Morocco 0.13 42 Chile 0.09 67 Guinea 0.02 18 Lao PDR 0.13 43 Bolivia 0.08 68 Slovenia 0.02 19 Spain 0.12 44 Benin 0.08 69 Rwanda 0.02 20 China 0.12 45 South Africa 0.07 70 Costa Rica 0.01 21 Botswana 0.12 46 El Salvador 0.07 71 Liberia 0.01 22 Argentina 0.12 47 Romania 0.07 72 Lesotho 0.00 23 Cameroon 0.12 48 Cambodia 0.07 73 Mauritius 0.00 24 Jordan 0.12 49 Paraguay 0.07 74 Fiji 0.00 25 Zambia 0.12 50 Malawi 0.07