Connective financing: Chinese infrastructure projects and the diffusion of economic activity in developing countries

Bluhm, Richard et al. Article — Published Version Connective financing: Chinese infrastructure projects and the diffusion of economic activity in developing countries Journal of Urban Economics Provided in Cooperation with: Kiel Institute for the World Economy – Leibniz Center for Research on Global Economic Challenges Suggested Citation: Bluhm, Richard et al. (2025) : Connective financing: Chinese infrastructure projects and the diffusion of economic activity in developing countries, Journal of Urban Economics, ISSN 1095-9068, Elsevier, Amsterdam, Vol. 145, pp. 1-22, https://doi.org/10.1016/j.jue.2024.103730 This Version is available at: https://hdl.handle.net/10419/318203 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by/4.0/ Contents lists available at ScienceDirect Journal of Urban Economics journal homepage: www.elsevier.com/locate/jue Connective financing: Chinese infrastructure projects and the diffusion of economic activity in developing countries Richard Bluhma, Axel Dreherb,c,d,e,f,∗, Andreas Fuchsg,h, Bradley C. Parksi,j, Austin M. Strangek, Michael J. Tierneyl aInstitute of Economics and Law, University of Stuttgart, Germany bAlfred-Weber-Institute for Economics, Heidelberg University, Germany cCefES, Italy dKOF Swiss Economic Institute, Switzerland eCEPR, UK fCESifo, Germany gDepartment of Economics and Centre for Modern East Asian Studies, University of Göttingen, Germany hKiel Institute for the World Economy, Germany iAidData, Global Research Institute, William & Mary, United States of America jCenter for Global Development, United States of America kDepartment of Politics and Public Administration, The University of Hong Kong, Hong Kong, China lDepartment of Government, William & Mary, United States of America ARTICLE INFO JEL classification: F35 R11 R12 P33 O18 O19 Keywords: Development finance Transport costs Infrastructure Foreign aid Spatial concentration China ABSTRACT This paper studies the causal effect of transport infrastructure on the spatial distribution of economic activity within subnational regions across a large number of developing countries. To do so, we introduce a new global dataset of geolocated Chinese grantand loan-financed development projects from 2000 to 2014 and combine it with measures of spatial concentration based on remotely sensed data. We find that Chinesefinanced transportation projects decentralize economic activity within regions, as measured by a spatial Gini coefficient, by 2.2 percentage points. The treatment effects are particularly strong in regions that are less developed, more urbanized, and located closer to cities. 1. Introduction In 2009, the Export–Import Bank of China (China Eximbank) approved a loan to the Kenyan government to widen and improve the Nairobi–Thika Highway – a 50.4 km dual carriageway that extends from the center of Nairobi to the town of Thika. The project, locally known as the ‘‘Thika Super-Highway,’’ sought to reduce congestion and travel times between Nairobi and a set of satellite towns along a crit- ∗Corresponding author at: Alfred-Weber-Institute for Economics, Heidelberg University, Germany. E-mail addresses: [email protected] (R. Bluhm), [email protected] (A. Dreher), [email protected] (A. Fuchs), [email protected] (B.C. Parks), [email protected] (A.M. Strange), [email protected] (M.J. Tierney). ically important transportation corridor (African Development Fund, 2007). Upon completion in 2012, traffic flows increased by 45 percent, journey speeds rose from 8 km per hour to at least 45 km per hour in sections with the highest registered traffic, and average commuting times from Thika to Nairobi fell from 2–3 h to 30–45 min (KARA and CSUD,2012;African Development Bank,2014a,b,2016,2019). Economic activity spread out along the transport corridor and became substantially less concentrated in the core of Nairobi (see Fig. 1). The case of the Nairobi-Thika Highway fits within a broader pattern: https://doi.org/10.1016/j.jue.2024.103730 Received 20 February 2023; Received in revised form 31 October 2024 Journal of Urban Economics 145 (2025) 103730 Available online 9 December 2024 0094-1190/© 2024 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ). R. Bluhm et al. Starting with Baum-Snow (2007), a series of studies in a variety of countries show that major transport infrastructure investments can decentralize economic activity.1 The key contribution of this study is to examine whether and to what extent Chinese-financed infrastructure projects are decentralizing economic activity within developing countries. We conduct our analysis at the level of first-order administrative regions. First-order regions are one layer below the national level and correspond to provinces, states, oblasts, governorates, or emirates, depending on the administrative divisions in a given country.2Our analysis focuses specifically on the provision of transport infrastructure financing from Chinese state-owned entities, which has assumed a dominant role in the construction and rehabilitation of transportation infrastructure around the world during the 21st century. Most of our analysis centers on the concentration of economic activity within regions,3but we also present results relating to infrastructure financing and concentration across regions. Our study tests whether the results from the existing, usually country-specific, literature can be generalized across a large sample of developing countries that host projects supported by the world’s largest provider of infrastructure financing. Since 2000, China’s government has financed many of the largest transport infrastructure projects in the Global South. The shortand long-run consequences of China’s infrastructure financing activities – including those along the US$1 trillion Belt and Road Initiative (BRI) – are the subject of considerable debate in the media and within policy circles. A growing number of studies focus on the expected impact of the BRI in different regions (e.g., Perlez and Huang,2017; Bandiera and Tsiropoulos,2020;Bird et al.,2020;de Soyres et al., 2020;Lall and Lebrand,2020). Beijing’s critics claim that it finances poorly designed and hastily executed projects that provide few economic benefits, while Western donors and lenders have learned through decades of experience to design and implement infrastructure projects in more careful and sustainable ways. In response to mounting criticism that it finances politically motivated and economically unsustainable projects, the Chinese government has doubled down on its leadership role in the market for global infrastructure finance.4 Many developing countries have unmet infrastructure financing needs, and the leaders of these countries are quick to point out that China is willing and able to swiftly finance and build roads, bridges, railways, and ports at a time when Western donors and lenders are not (Swedlund,2017).5For example, during his tenure as the President of Senegal, Abdoulaye Wade admonished traditional donors and 1See, for example, Baum-Snow et al. (2017) and Banerjee et al. (2020) on China, Bayes (2007) on Bangladesh, Bird and Straub (2014) on Brazil, Donaldson (2018) on India, Henderson and Kuncoro (1996) on Indonesia, Garcia-Lopez et al. (2015) on Spain, Gibbons et al. (2019) on the United Kingdom, and Duranton and Turner (2012) on the United States. Redding and Turner (2015) as well as Baum-Snow and Turner (2017) provide surveys of this literature. 2In our sample, the average region’s size is 37,644 square kilometers, which roughly corresponds to the land area of South Carolina. It has 2.1 million inhabitants, which roughly corresponds to New Mexico. 3We use the terms spatial concentration and spatial centralization interchangeably as they both refer to changes in the distribution of people and output across space. 4At the 2017 Belt and Road Forum for International Cooperation, President Xi emphasized that ‘‘[i]nfrastructure connectivity is the foundation of development through cooperation. We should promote land, maritime, air and cyberspace connectivity, concentrate our efforts on key passageways, cities and projects and connect networks of highways, railways and sea ports [...]’’ (Xi, 2017). 5An important reason for these infrastructure financing gaps follows from the fact that ‘‘Western donors have by and large gotten out of hard infrastructure sectors [...] and [t]hey [instead] channel their assistance overwhelmingly to social sectors or to infrastructure sectors such as water supply and sanitation that have direct effects on household health’’ (Dollar,2008). creditors for their cumbersome bureaucratic procedures, noting that: ‘‘[w]ith direct aid, credit lines and reasonable contracts, China has helped African nations build infrastructure projects in record time. [...] I have found that a contract that would take five years to discuss, negotiate and sign with the World Bank takes three months when we have dealt with Chinese authorities’’ (Wade,2008). We introduce the first global dataset of geo-located Chinese government-financed projects that were undertaken in developing countries between 2000 and 2014.6The dataset includes 3485 projects in 6184 subnational locations across 138 countries during these 15 years. For our analysis, we focus on 269 Chinese government-financed transportation infrastructure projects undertaken in 1215 subnational locations across 86 countries. The lower bound for the total financial value of these projects is US$64 billion. We estimate the effects of these projects on the spatial concentration of economic activity – both within and across subnational jurisdictions – with satellite data on the geographical dispersion of nighttime light output (similar to Henderson et al.,2018). We identify the causal effect of Chinese government-financed transport projects on the spatial concentration of economic activity following the intuition of a generalized difference-in-differences design. Our instrumental variable relies on the availability of resources for the construction of such projects. This approach has the advantage that comparable data are available for a large number of countries, and plausibly exogenous instruments can be applied across these diverse empirical settings.7Specifically, we introduce an instrumental variable that uses an exogenous supply push variable interacted with a local exposure term: China’s domestic production of potential project inputs interacted with each recipient region’s probability of receiving projects. We use China’s annual production of aluminum, cement, glass, iron, steel, and timber to proxy its capacity to provide physical project inputs.8The intuition behind this approach is that the Chinese government has long considered these production materials as strategic commodities and, therefore, produced them in excess of domestic demand. This policy results in large surpluses, some of which China redirects to overseas infrastructure projects.9We therefore expect China to be more lenient towards countries that request financing for transport infrastructure projects in the years when such inputs are abundant and less lenient in the years when such inputs are scarce. We also expect subnational localities that frequently receive Chinese government-financed transport projects to be more heavily affected by year-to-year fluctuations in the supply of project inputs. We thus compare the effects of Chinese transport projects on spatial concentration induced by changes in China’s domestic production of potential project inputs across two groups: regions that are regular and irregular recipients of Chinese transport infrastructure financing. 6Though the BRI was not officially launched until late 2013, the Chinese government had already begun providing a significant number of large-scale financing for transport infrastructure in developing countries by the turn of the century. These pre-2014 projects share most of the characteristics of transport infrastructure projects that are now formally part of the BRI. 7The literature typically uses historical transport networks or other country-specific historical circumstances (such as minimum spanning trees connecting the largest cities). 8Exporting excess capacity in a variety of materials through infrastructure investments abroad is one of the secondary motives often ascribed to the Chinese government’s BRI initiative. For example, the Economist writes ‘‘Mr Xi [...] hopes to [...] export some of his country’s vast excess capacity in cement, steel and other metals’’ (see www.economist.com/the-economistexplains/2017/05/14/what-is-chinas-belt-and-road-initiative). Our approach extends the strategy proposed in Dreher et al. (2021a) which exclusively used the level of steel production. 9Chinese infrastructure projects usually require construction inputs that are oversupplied in China, and Chinese state-owned banks usually obligate their borrowers to import these inputs on a preferential basis (Dreher et al.,2021b). Journal of Urban Economics 145 (2025) 103730 2 R. Bluhm et al. Fig. 1. Nairobi–Thika Highway, change in nighttime lights, 2008–2013. Notes: The figure illustrates the change in nighttime lights from 2008 to 2013 along the route of the Nairobi–Thika Highway in Kenya, which was constructed from January 2009 until October 2012. Major intersections and points of interest are highlighted along the highway. The change in nighttime lights is the difference between the F18 2013 image (in Digital Numbers from 0 to 63) and the F16 2008 image (in the same units). The differences have a range from −6 to 31 DN. The expansion of light around Nairobi is related to other infrastructure projects, many of which are also Chinese-financed but not highlighted here. Between 2008 and 2013, the geographical areas within a 4 km buffer of the highway experienced a 27 percent reduction in the spatial concentration of nighttime light intensity. Spatial concentration, as measured by the Gini coefficient introduced later in this paper, fell from 0.425 in 2008 to 0.31 in 2013 in the 4 km buffer. At the same time, land values doubled in Thika and rose even faster in areas closer to Nairobi (like Kasarani, where land values increased from US$46,000 per acre to US$500,000 per acre), farmgate prices for dairy products and horticulture rose due to increased access to markets, and trade and investment alongside the road corridor expanded (KARA and CSUD,2012;African Development Bank,2014b,2016,2019). Our results show that regions which are frequent recipients of projects receive larger amounts of Chinese government financing in years of overproduction than regions that infrequently receive Chinese government-financed transport projects. This difference presumably occurs because existing local capacity and relationships make it easier to implement additional projects. This estimate can be interpreted as a difference-in-differences estimate, similar to those reported in the ‘‘China shock’’ or aid and conflict literatures (e.g., Autor et al., 2013;Nunn and Qian,2014). We essentially compare the effects of Chinese transport projects induced by annual changes in the production of raw materials in subnational localities with a high probability of receiving such projects and subnational localities with a low probability of receiving such projects. We find that Chinese-financed connective infrastructure projects reduce spatial concentration within first-order regions and accelerate the diffusion of economic activity around cities (in line with Fig. 1 and Baum-Snow,2007). Specifically, we find that the Gini coefficient measuring the spatial concentration of economic activity is reduced by about 2.2 percentage points. Similar specifications for concentration between regions suggest effects of similar magnitude. However, these effects are less precisely estimated, suggesting that Chinese-financed infrastructure cannot be robustly linked to changes in the concentration of economic activity across regions. The absence of a ‘‘between effect’’ is in line with several recent studies that emphasize that transport infrastructure has heterogeneous and context-specific impacts on the distribution of economic activity across regions (Baum-Snow et al., 2020;He et al.,2020;Jedwab and Storeygard,2022;Fajgelbaum and Redding,2022). Our main results hold under various perturbations, such as the choice of control variables to strengthen identification or variations of the instrument. Consistent with predictions from land use theory, we find that transport projects shift activity from densely developed locations to less densely developed ones, that is, from the highest quintile of the light distribution to lower quintiles. We find no evidence that infrastructure projects increase the fraction of illuminated pixels in a recipient region or that a region’s light per capita increases. However, our results show a notable increase in overall light intensity, defined as the sum of light emissions within a region relative to its geographic area. This suggests that regions experience positive population growth in response to Chinese-financed infrastructure projects. The results also show that the impact of these projects on the concentration of activity within regions is heterogeneous. Chinese-financed transport infrastructure reduces concentration more strongly in regions with more urban areas, low travel time to cities, and higher road density. We take this as indirect evidence suggesting that our results are driven by a relocation of workers to the outskirts of cities rather than an increase in economic activity in peripheral cities of a region. We also provide evidence that these effects are largest in African countries and poorer regions Journal of Urban Economics 145 (2025) 103730 3 R. Bluhm et al. within developing countries, which tend to experience rapid population growth and have a high demand for infrastructure. The remainder of the paper proceeds as follows. Section 2briefly discusses what theory and the existing empirical literature suggest about the relationship between transport projects and the spatial concentration of economic activity within and across regions. Section 3introduces a subnationally georeferenced dataset of Chinese governmentfinanced projects around the world and discusses the remotely-sensed measure of spatial concentration.10 Section 4describes the empirical strategy. Section 5presents and discusses the results. Section 6 concludes. 2. Transport infrastructure and the concentration of economic activity Urban land use theory suggests that transport infrastructure should reduce spatial concentration within subnational regions if these jurisdictions primarily consist of urban areas and their surroundings. This is a key prediction of the canonical monocentric city model (Alonso et al.,1964;Mills,1967;Muth,1969), in which all workers commute to a single location in a central business district (CBD). In this model, agglomeration benefits and rents are highest in the city center but decline with distance from the CBD. Initially, many people choose to live near the center and pay higher rents to reduce their commuting times. Subsequent investments in transportation infrastructure increase transportation speed, reduce commuting costs, and increase the supply of readily accessible land, shifting this gradient outwards. Transportation infrastructure thus facilitates urban sprawl – the flow of people out of the city center – by turning a city’s agricultural surroundings into valuable locations to live in. The model also implies that people should spread out along newly created highways (Baum-Snow,2007), just as we document above for the case of Nairobi.11 Reality is not as stylized and most cities are not monocentric but rather characterized by a CBD that coexists with other employment subcenters. Polycentric cities arise when the location of employment and commercial activity are determined endogenously within the city (for early contributions, see Ogawa and Fujita,1980;Fujita and Ogawa, 1982;Henderson and Mitra,1996). In polycentric cities, agglomeration economies are smaller relative to the monocentric benchmark, but commuting costs – hence, dispersion forces – are also lower. New quantitative spatial models allow for complex patterns of agglomeration and dispersion forces, which can lead to various urban forms (see Redding and Rossi-Hansberg,2017, for a review). In these models, locations are not independent but connected via trade, commuting, and migration flows. Firms and residents make optimal choices based on these connections. The predictions of the frameworks depend on these spatial interactions and, hence, the data. Infrastructure improvements can reduce the concentration of activity by making peripheral areas more accessible, but they can also increase concentration by expanding the labor market for central firms. For example, early infrastructure improvements, such as the steam railway, permitted a specialization of city centers into workplaces and their surroundings into residences (Heblich et al.,2020). 10 We provide our code, most data, and a description of how to obtain the remaining data required to replicate all exhibits in our paper in Bluhm et al. (2024). 11 Firms have different incentives than residents since they face a more complex set of costs when leaving city centers. They trade agglomeration benefits against various costs (transportation costs of output, interaction costs, labor accessibility, etc.). This gives rise to a pattern where firms that benefit less from face-to-face interaction and knowledge spillovers, such as manufacturing firms, decentralize more than other firms (Rossi-Hansberg et al., 2009;Baum-Snow,2014). A large and growing body of empirical evidence supports the basic prediction of the monocentric city model that new or upgraded transportation infrastructure disperses economic activity away from urban agglomerations. Consider suburbanization in 20th-century America, where researchers have documented urban sprawl and strong population growth in cities with more developable surroundings (Burchfield et al.,2006;Saiz,2010). The construction of highways in the United States dramatically lowered commuting times and increased demand for suburban relative to urban residential space (Baum-Snow,2007). Congestion also plays an important role, particularly in contemporary settings. Allen and Arkolakis (2022) highlight that stronger congestion not only reduces the market access of central locations and shifts activity away from the core but that this effect increases with city size. There is also evidence for similar processes of diffusion around urban areas in developing countries (e.g., Bayes,2007;Zárate,2022). Baum-Snow et al. (2017) examine the effect of road and railway infrastructure on the spatial distribution of economic activity in China and find that ring road investments displaced 50 percent of industrial GDP from central cities to outlying areas. As Chinese-financed infrastructure projects in developing countries often represent a substantial proportion of local infrastructure investment in a given year, we similarly expect them to decentralize economic activity around urban areas. We also expect these projects to spur the formation of new sub-centers as businesses, workers, and other economic actors relocate into the periphery. Transport projects may also affect the concentration of economic activity across regions, a central concern of economic geography research. The classic core–periphery model stresses the role of increasing returns to scale when economic activity starts to concentrate in a particular region. When trade costs are high or prohibitive, firms are spread out evenly across regions to locate themselves close to consumer demand. When transport projects increase connectivity between leading and lagging regions, labor, and capital should move from the periphery to the better-connected core, creating a core–periphery split until there is almost complete specialization (Krugman,1991). However, some of these forces reverse at high levels of concentration. Puga (1999), for example, shows how a lack of migration with low trade costs implies that firms will again locate closer to final demand. This gives rise to a bell shape for regional inequalities in relation to trade costs. The advantage of being in a central location well-connected with other markets erodes when very low trade costs make it easy to reach the periphery. While the bell-shaped curve is a robust prediction, it requires additional heterogeneity in agricultural trade costs, urban congestion, or migration decisions that make the overall relationship difficult to identify. Empirical evidence on how transport costs shape regional concentration reflects this heterogeneity. Brülhart et al. (2020) show that the advantages of market potential are shrinking in the developed world but remain an important determinant of employment growth in developing countries. With respect to infrastructure investments, Bird and Straub (2014) find that investments in Brazil’s road network increased economic agglomeration in the already prosperous population centers of the South, while also facilitating economic agglomeration in less developed areas of the North. On balance, these investments reduced spatial inequality across the country’s municipalities.12 However, Faber (2014) provides evidence that China’s National Trunk Highway System – a major inter-regional transportation infrastructure project – reduced levels of economic activity in newly connected peripheral regions relative to non-connected peripheral regions. Given these mixed findings and the cross-national scope of our study, we do not have strong reasons 12 In a related study of Argentina’s steam railroad network and the agricultural sector, Fajgelbaum and Redding (2022) suggest that lower transport costs can enable economic actors located in remote, interior regions to participate in structural transformation. Journal of Urban Economics 145 (2025) 103730 4 R. Bluhm et al. to believe that Chinese-financed transportation projects will uniformly increase or decrease concentration between regions. Developing countries are an important and useful application of these theories. Most developing countries face major transportation infrastructure gaps in both urban and rural regions. Internal transport costs are four to five times higher within Ethiopia or Nigeria than within the United States (Atkin and Donaldson,2015). The total length of the road network per 1000 people is roughly 10 times lower for South Asia, East Asia and the Pacific, Sub-Saharan Africa, and the Middle East and North Africa than for North America (Andrés et al.,2014). Many developing countries have both rapidly expanding populations and underfunded, poorly designed transportation systems (Cervero, 2013). Major infrastructure financing gaps make it difficult for developing countries to overcome the spatial bottlenecks created by high levels of urban concentration and rural neglect. What is more, subnational regions within developing countries are often defined by dense central cities surrounded by underdeveloped hinterlands.13 Large cities in many African countries, for example, tend to be highly congested relative to overall levels of infrastructure, industry, and economic opportunity (Lall et al.,2017), while secondary cities tend to be isolated from world markets (Gollin et al.,2016). In these settings, high levels of urban congestion in developing economies are a latent force for spatial dispersion as new transportation options become available. 3. Data New geocoded dataset of Chinese government-financed projects The Chinese government considers the details of its overseas development program to be a ‘‘state secret’’ (Bräutigam,2009, p. 2). It does not publish a country-by-country breakdown of its aid expenditures or activities. Nor does it systematically publish project-level data on its less concessional and more commercially-oriented financial expenditures and activities in developing countries. To overcome this challenge, Dreher et al. (2021b) collaborated with AidData, a research lab at William & Mary, to build a global dataset of Chinese government-financed projects committed between 2000 and 2014. This project-level dataset uses a publicly documented method called Tracking Underreported Financial Flows (TUFF) to facilitate the collection of detailed and comprehensive financial, operational, and locational information about Chinese government-financed projects (Strange et al., 2017,2018). The TUFF method triangulates information from four types of open sources – English, Chinese, and local-language news reports; official statements from Chinese ministries, embassies, and economic and commercial counselor offices; the aid and debt information management systems of finance and planning ministries in counterpart countries; and case study and field research undertaken by scholars and non-governmental organizations (NGOs) – in order to minimize the impact of incomplete or inaccurate information.14 Economists, political scientists, and computational geographers have for the most part used these data to explain the nature, allocation, and effects of Chinese government-financed projects in Africa and in several country-specific studies outside of Africa (e.g., Hernandez,2017;Dreher et al.,2018; Isaksson and Kotsadam,2018a,b;Anaxagorou et al.,2020;Isaksson, 13 As Baum-Snow et al. (2017) point out, the urban distribution of economic activity in many developing countries today largely resembles that of early 20th century America, in which industry was initially overwhelmingly concentrated in urban centers. 14 The method is organized in three stages: two stages of primary data collection (project identification and source triangulation) and a third stage to review and revise individual project records (quality assurance). The TUFF data collection and quality assurance procedures are described at length in Strange et al. (2017,2018). 2020;Martorano et al.,2020;Dreher et al.,2021b;Eichenauer et al., 2021;Horn et al.,2021;Gehring et al.,2022;Baehr et al.,2023). In this paper, we build on these project-level data to create a firstof-its-kind dataset of Chinese grantand loan-financed development project locations around the globe. In contrast to previous versions, our new data enable subnational analyses of Chinese-financed projects in five regions of the world (Africa, the Middle East, Asia and the Pacific, Latin America and the Caribbean, and Central and Eastern Europe) over 15 years (2000–2014). Our dataset takes all projects from Dreher et al. (2021b) that secured financial commitments from China and entered implementation or reached completion as a starting point.15 We then subject all of these projects to a double-blind geocoding process (Strandow et al.,2011), in which two trained coders independently employ a defined hierarchy of geographic terms and assign uniform latitude and longitude coordinates and standardized place names to each location where the project in question was active. Coders also specify a precision code for each location. Precision code 1 corresponds to an exact location; precision code 2 corresponds to locations within 25 kilometers of the exact project site; precision code 3 corresponds to a second-order region; and precision code 4 corresponds to a firstorder region.16 If the coordinates and precision codes do not match, a senior ‘‘arbitrator’’ identifies the source of the discrepancy and assigns a final set of geocodes for all sites. This double-blind coding process aims to minimize the risk of missed or incorrect locations.17 In total, the resulting dataset covers 3485 projects (worth at least US$273.6 billion in constant 2014 dollars) in 6184 discrete locations across 138 countries.18 To merge these geocoded project data with our outcome measures of spatial concentration within and across subnational regions, we aggregate all projects with precision codes 1–4 to first-order regions. Fig. 2shows the locations of projects that can be placed within firstorder regions over the 2000–2014 period. The resulting subsample includes 2140 Chinese government-financed projects at 4420 discrete locations (collectively worth US$201 billion) that were completed or being implemented in 883 first-order regions within 129 countries between 2000 and 2014.19 Our data can be disaggregated by financial flow class and sector. With respect to the former, we distinguish between Official Development Assistance (ODA) and other forms of concessional and non-concessional financing from Chinese government 15 We build on earlier georeferenced datasets that cover Africa, the Tropical Andes, and the Mekong Delta for fewer years only (BenYishay et al.,2016; Dreher et al.,2019a). Note that we exclude all suspended and canceled projects as well as projects that reached the (non-binding) pledge stage or (binding) official commitment stage but never reached implementation or completion during the period of study (2000–2014). 16 We exclude all projects from the regression analysis below that could not be geocoded with sufficient spatial precision to be included in the regionallevel data (e.g., country-wide projects, see https://docs.aiddata.org/ad4/files/ geocoding-methodology-updated-2017-06.pdf for details). 17 Note that the point-based method used to geocode these projects is not designed to measure the exact linear path of transportation infrastructure. This implies that one cannot ‘connect the dots’ and look for effects alongside the roads, railways, etc. However, it is useful for measuring the effects within treated subnational regions, as we do in the present paper. 18 For comparison, the Africa-specific data provided in Dreher et al. (2019a) include 1650 projects across 2969 locations in the 2000–2012 period. Note that, in contrast to our dataset, they also cover projects that have not (yet) reached implementation stage. 19 We only focus on low-income and middle-income countries. More precisely, we include countries that the World Bank does not classify as highincome countries in a given year (see https://datahelpdesk.worldbank.org/ knowledgebase/articles/906519-world-bank-country-and-lendinggroups, last accessed September 13, 2023). We also exclude small states with a population size below a threshold of 1,000,000 inhabitants. Table A-1 in the Online Appendix lists all countries included in the analysis. Journal of Urban Economics 145 (2025) 103730 5 R. Bluhm et al. institutions.20 For the purposes of the latter, we use the OECD’s threedigit sector classification scheme, which categorizes projects according to their primary objectives.21 269 projects were assigned to the transport and storage sector and implemented in 1215 locations. The combined value of these projects was at least US$64.1 billion (when counting those projects for which financial values are available) and amounts to an average of US$224.9 million per project.22 This is sizable in light of an average GDP of about US$4.4 billion per first-order region. The vast majority of these projects focused on building transportation infrastructure, such as roads, railways, bridges, seaports, and airports. With 651 project locations, long-distance roads are most frequent, followed by longdistance railways (245) and urban roads (123) (see Table A-2 in the Online Appendix).23 In our dataset, the average financial commitment from China for a long-distance road project was US$231.2 million compared to US$979.3 million for a long-distance railway project and US$148.3 million for an urban roads project. To better understand the nature of these projects, consider three projects from three different regions that are broadly indicative of the types of ‘‘treatments’’ that we analyze in our statistical analysis. First, in October 2014, China Eximbank provided a US$943.9 million loan to Montenegro’s Ministry of Finance to help finance the construction of a 169 km highway between Bar (the country’s main seaport in the south) and Boljare on the Montenegrin-Serbian border in the country’s north.24 The loan was worth approximately a quarter of the country’s GDP at the time that it was contracted. Upon completion, the highway was expected to reduce travel time between the capital of Podgorica and the northern city of Kolašin from 90 min to 30 min and facilitate economic development alongside the transport corridor. Second, in May 2013, China Eximbank issued a US$491.7 million loan to Djibouti’s Ministry of Finance for the construction of a 100 km segment of a 756 km railway that runs from Addis Ababa, the capital of Ethiopia, to the Doraleh seaport in Djibouti. At the time of its issuance, the loan represented approximately 40% of Djibouti’s GDP. Upon completion, the railway was expected to reduce travel time between Addis Ababa and Doraleh seaport from 7 days (on roads) to 10 h.25 Third, in 2009, China Eximbank provided a US$44.2 million loan to the Government of Tonga for phase 3 of the National Road Improvement Project. The loan was worth roughly 16% of GDP 20 More precisely, we code all Chinese government-financed projects as Official Development Assistance (‘‘ODA-like’’), Other Official Flows (‘‘OOFlike’’), or ‘‘Vague Official Finance.’’ Chinese ODA-like projects are financed by Chinese government institutions with development intent and a minimum level of concessionality (a 25 percent or higher grant element). Chinese OOF projects are financed by Chinese government institutions with commercial or representational intent and/or lack a grant element of 25 percent or more. Projects assigned to the Vague Official Finance category are Chinese governmentfinanced projects where there is insufficient information in the public domain about concessionality and/or intent to clearly determine whether the flows are more akin to ODA or OOF. Total Chinese Official Finance (OF) is, therefore, the sum of all projects coded as ODA-like, OOF-like, or Vague (Official Finance). For more detailed discussion of the distinction between these types of Chinese development finance, see Dreher et al. (2018). 21 There are 24 of these OECD sector codes (see www.oecd.org/ development/financing-sustainable-development/development-financestandards/purposecodessectorclassification.htm for details). 22 Many of these projects are implemented in multiple locations, for example, when they connect two cities with a railway line. The average amount committed per project location is US$49.8 million. 23 Transport projects are the ones we exploit for most of our analyses. We also use a larger sample of projects that supported economic infrastructure and services, which includes roads, railways, bridges, seaports, and airports but also power grids, power lines, cell phone towers, and fiber optic cable lines (514 projects at 1897 locations with a value of about US$165 billion). 24 See https://china.aiddata.org/projects/42330/ for details on the project. 25 See https://china.aiddata.org/projects/46183/ for details. at the time that it was contracted. The purpose of the project was to improve the 1 km Alipate Road on the island of Tongatapu, the 8 km road that runs from Kolonga to Talasiu on the island of Tongatapu, and the 4 km road that runs from Haveluliku to Lavengatonga on the island of Tongatapu.26 Fig. 2illustrates the global reach of China’s overseas development program in the 21st century. Consistent with earlier periods of Chinese aid giving (Dreher and Fuchs,2015), Chinese projects cover almost all developing countries (with countries recognizing the Chinese government in Taiwan as a notable exception).27 Chinese-financed development projects are densely concentrated in African and Asian countries. The figure also illustrates that many Chinese governmentfinanced projects are situated in coastal regions, including some of the highest-value transportation projects. Measuring concentration within and across subnational regions Reliably measuring local economic activity across the globe with official data is difficult. Few countries collect and report comprehensive data at the individual or plant/establishment level at regular intervals, and subnational GDP data are generally only available in highly developed countries. To circumvent this problem, we follow previous work that uses nighttime light intensity as a proxy for local economic activity (Henderson et al.,2012;Hodler and Raschky,2014;Michalopoulos and Papaioannou,2014). While nighttime lights were initially proposed as a measure of income for countries with weak statistical capacity, they were quickly adopted more broadly as a measure of subnational economic activity in developing countries. Subsequent studies have demonstrated that changes in light emissions correlate strongly with traditional welfare measures down to the village level (Weidmann and Schutte,2017;Bruederle and Hodler,2018). We follow Henderson et al. (2018), who use nighttime light intensity at the grid-cell level as a measure of aggregate economic activity – i.e., the product of population and light output per capita – and then calculate a spatial Gini coefficient based on the distribution of this proxy for total GDP. While we are primarily interested in whether and to what extent infrastructure investments relocate economic activity, we also investigate below whether such investments increase output per capita. We obtain data on nighttime light intensity from the Defense Meteorological Satellite Program’s (DMSP) Operation Line Scan satellites. The DMSP satellites circle the earth in sun-synchronous orbit and record evening lights between 8:30 and 9:30 pm on a 6-bit scale ranging from 0 to 63. The National Oceanic and Atmospheric Administration (NOAA) processes these data, creates annual composites of the daily images at a resolution of 30 arc seconds and makes them available to the general public. We use the so-called ‘‘stable lights’’ product, which filters out most background noise, forest fires, and stray lights. Even though there are well-known issues in these data with bottom and top coding (see Jean et al.,2016;Bluhm and Krause,2022), nighttime lights are measured in a consistent manner around the globe and avoid many of the measurement errors involved in more traditional survey data. We proceed in four steps to calculate our measure of spatial concentration. First, we divide the entire world into a grid of 6 arc minute cells (i.e., an area of about 9.3 km by 9.3 km at the equator) and align the grid with lights data.28 Second, we intersect this grid with the global first-order administrative boundaries, which creates regular cells in the 26 For details, see https://china.aiddata.org/projects/39199/. 27 For recent work that studies the allocation of China’s development finance across countries, see Dreher et al. (2022) or Hoeffler and Sterck (2022). 28 Although the nominal resolution of the DMSP-OLS system is 30 arc seconds, geolocation errors and on-board processing of fine-resolution pixels lead to a true ground footprint of 5 km by 5 km (Elvidge et al.,2013). Taking about twice this resolution reduces the influence of this mechanical Journal of Urban Economics 145 (2025) 103730 6 R. Bluhm et al. Fig. 2. Locations of Chinese-financed projects, transport and non-transport, 2000–2014. Notes: The figure identifies all Chinese-financed transport (red) and non-transport projects (gray) which were committed and implemented in the period from 2000 to 2014. It shows a total of 2140 projects in 4420 discrete locations which have a precision accuracy of (at least) a first-order administrative division. 1345 projects have a precision accuracy less than the first-order region (not shown). Although there are ‘‘only’’ 269 transportation projects, 1211 of the 4420 locations shown in the figure have directly received (some part) of a larger transportation project. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) interior and ‘‘squiggly’’ cells along the regional borders.29 Third, for all regular and squiggly cells in this grid and all years in the nightlights data, we compute the sum of light (𝑠𝑙), the land area of each cell in km2 (𝑎𝑙), and the light intensity in the cell (𝑥𝑙=𝑠𝑙∕𝑎𝑙).30 We average the resulting light intensities whenever more than one satellite is available and turn off all pixels that do not fall on land before aggregating the lights to the grid level. Finally, we compute the Gini coefficient of light intensities (on a 0–1 scale) over all cells (including cells with zero light intensity) within an administrative region as Gini =∑𝑛 𝑙=1 𝑤𝑙∑𝑛 𝑚=1 𝑤𝑚|𝑥𝑙−𝑥𝑚| 2∑𝑛 𝑙=1 𝑤𝑙∑𝑛 𝑙=1 𝑤𝑙𝑥𝑙 ,(1) where 𝑤𝑙=𝑎𝑙 ∑𝑛 𝑙=1 𝑎𝑙 is an area-based weight and 𝑛is the total number of cells in a region. We also construct Gini coefficients for concentration between first-order regions. The formula remains the same, only that it is based on the average light intensity of a region (swapping 𝑥𝑙for 𝑥𝑙), and 𝑤𝑙is then defined as the land area of the entire region. Our spatial Gini coefficient can be interpreted as the average (weighted) difference between the light intensities of all possible pairs of cells within an administrative region. Geometrically, it is the area under the Lorenz curve plotting the cumulative distribution of weighted light intensities against the cumulative distribution of cell areas (in km2). Including cells with zero light intensity means the Lorenz curve will remain at zero before sloping up to one, but ensures that the Gini coefficient is a proper measure of economic concentration, which not only decreases when the distribution of light becomes more equal among already illuminated cells but also when new cells become illuminated. As can be seen from the long differences in the spatial Gini coefficient presented in a world map of first-order regions in Fig. 3, our dependent variable shows considerable variation over the time period under analysis, both within and across countries (2000–2013). spatial autocorrelation, reduces the influence of top coding and bottom coding, and limits the computational burden. The newer Visible Infrared Imaging Radiometer Suite (VIIRS) data, which have superior technical properties, do not span a significant portion of our sample. 29 We obtained the regional borders from the Database of Global Administrative Areas (GADM) vector dataset (version 2.8). We used the same data to geocode the Chinese-financed projects. 30 Dividing by the land area adjusts for the fact that 6 arc minute cells do not have a uniform area across the globe and may be covered by water. We calculate the land area of each cell using the Gridded Population of the World (v4) land/water raster. It is important to emphasize that the Gini coefficient captures the overall dispersion of economic activity, which is a product of the population distribution and the distribution of light per capita.31 Henderson et al. (2018) show that the cross-sectional variation in population density across administrative regions is substantially larger than the variation in income per capita. If this holds across time, then a significant proportion of observed changes in the within-region distribution of light intensities should be attributable to shifts in the population distribution rather than differences in per-capita income. This is precisely the type of variation we are interested in and expect to be affected by transport infrastructure investments. We prefer using nighttime lights over data for population density as our main outcome measure. Population data at comparable resolutions – such as the Global Human Settlement Layer, Gridded Population of the World, or Landscan – are based on rarely available censuses, which are then disaggregated in space and interpolated over time. They would not allow us to exploit annual variation in the commitment of transport projects and changes in economic activity, which are the basis of our identification strategy. Census data are also less frequently available in poorer developing countries that host many Chinese-financed development projects. 4. Empirical strategy We are interested in changes in the spatial concentration of economic activity caused by Chinese infrastructure investments. Denoting first-order administrative regions by 𝑗, countries by 𝑖,32 and years by 𝑡, our main equation relates our luminosity-based measure of spatial concentration, Gini𝑗 𝑖𝑡, to the total number of years in which transportation projects have been committed to a region up to 𝑡− 2, denoted 𝑁𝑗 𝑖,𝑡−2. We chose 𝑁𝑗 𝑖,𝑡−2 as the baseline treatment because transportation projects can vary widely in type and size, ranging from small bridges to 31 To see this, consider that 𝑥𝑖is defined as 𝑝𝑖 𝑎𝑖 ×𝑠𝑖 𝑝𝑖 , where 𝑝𝑖 𝑎𝑖 is population density and 𝑠𝑖 𝑝𝑖 is light per capita in each cell. 32 Going below the first-order level would change many sample characteristics. We would lose a significant share of projects that have only been accurately coded to first-order regions. The exposure variable we introduce below would also be based on substantially fewer projects per region. What is more, in countries with a smaller land mass, second-order regions often correspond to administrative boundaries of cities, which would effectively exclude their surroundings from a within-unit analysis. Journal of Urban Economics 145 (2025) 103730 7 R. Bluhm et al. Fig. 3. Long differences in spatial concentration, within first-order regions, 2000–2013. Notes: The figure illustrates the cross-regional and temporal variation in spatial concentration. It shows long differences in the Gini coefficient for spatial concentration within first-order regions, that is, a region’s value in 2013 minus the value in 2000. Only countries not classified as high-income countries by the World Bank are shown. Missing values occur when there are too few lit cells to compute the Gini coefficient in the initial or final period. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) (segments of) extensive multiregional highways.33 To assess the impact of any infrastructure project or bundle of projects in a region in a given year, this variable aggregates over all years in which at least one project was committed (see the discussion of the differenced version below). In robustness checks, we use the number of project locations or financial values (where available). By aggregating over all years, our specification assumes that the level of concentration depends on the entire history of projects. Of course, the effect of projects could fade over time so that projects committed in the distant past would no longer affect concentration. Since our sample only covers the period since 2000, and Chinese development finance was comparatively low before that time, we focus on the medium-term effects of new projects rather than those committed in the distant past. We lag this variable by two years to account for the difference between the commitment date and the expected completion date of a project.34 We begin with a flexible specification that allows the effects of Chinese-financed projects to be arbitrarily correlated with regionspecific fixed effects and region-specific time trends: Gini𝑗 𝑖𝑡 =𝛽 𝑁𝑗 𝑖,𝑡−2 +𝜇𝑗 𝑖+𝜃𝑗 𝑖×𝑡+𝜆𝑖𝑡 +𝜖𝑗 𝑖𝑡,(2) where 𝜇𝑗 𝑖are region-fixed effects, 𝜃𝑗 𝑖×𝑡are region-specific linear time trends, and 𝜆𝑖𝑡 are country-year fixed effects that absorb a variety of potential shocks to all regions of a country in a particular year. A more tractable and intuitive version of this model can be estimated in first differences: 𝛥Gini𝑗 𝑖𝑡 =𝛽 𝛥𝑁𝑗 𝑖,𝑡−2 +𝜃𝑗 𝑖+𝜏𝑖𝑡 +𝛥𝜖𝑗 𝑖𝑡,(3) where 𝜏𝑖𝑡 =𝜆𝑖𝑡 −𝜆𝑖,𝑡−1 represents a new set of country-year fixed effects obtained via differencing. 𝜃𝑗 𝑖now captures region-specific trends in 33 Another technical reason is that projects are often co-located, e.g., different sections of a highway, but may have distinct project IDs in our data because they are financed through different financial tranches. This does not necessarily capture the intensive margin of infrastructure investments but reflects the definitions adopted during the geocoding process. 34 Start and end dates are available for a subset of the projects in our dataset. Across these approximately 1100 Chinese government projects, the average time from start to completion is about 2.1 years. Historical data on Chinese development projects also reveal a median of two years between project start and completion (Dreher et al.,2021b, based on data from Bartke,1989). The two-year lag we use for the main analysis thus allows projects to register effects directly after (expected) completion. As we show below, the effects of Chinese transport projects materialize quickly. When longer lags are used, estimates become smaller and less precisely estimated. levels. In the differenced form, it becomes 𝜃𝑗 𝑖×𝑡−𝜃𝑗 𝑖× (𝑡− 1), which represents the change in the region-fixed effect over time. The specification in differences shows that 𝛽estimates the persistent effect of a new transport project (or a bundle of transport projects committed in the same year) on spatial concentration two years later.35 The model nests less flexible approaches with a strict parallel-trends assumption since all 𝜃𝑗 𝑖could be zero. Our preferred measure of transportation infrastructure, 𝛥𝑁𝑗 𝑖,𝑡−2, is thus a binary variable indicating that at least one new project is committed to a particular region in a year. Since the size of the projects is not homogeneous across locations, the effects of projects on spatial concentration might differ along the intensive margin. Unfortunately, we lack information on the financial values for more than a third of these projects (see Dreher et al.,2021b), which is why we prefer the binary indicator and present additional results using (the log of) aggregate dollar values for comparison. Moreover, we define our dependent variable based on commitment years rather than actual disbursement dates as comprehensive data on disbursements are not available and virtually impossible to estimate through open-source data collection. With first differences, a two-year lag, and a lack of nighttime lights data after 2013 (from the DMSP-OLS system), our sample effectively covers the period from 2002 until 2013.36 We allow a wide range of dependency structures to occur in 𝛥𝜖𝑗 𝑖𝑡. Transportation infrastructure projects often connect more than one administrative region. Clustering standard errors on the country level permits arbitrary spatial and temporal correlation among all regions within a country. To account for connections across countries, we also report Conley errors with a spatial cutoff of 500 km and a heteroskedasticityand autocorrelation-consistent (HAC) structure with a lag cutoff of 1000 years in the time-series dimension.37 While our 35 The event-study equivalent would be a permanent change of size 𝛽 starting in 𝑡= 2for a project committed in 𝑡= 0. 36 The original DMSP-OLS data are only available through 2013. Nighttime observations by NPP-VIIRS have replaced this system, but they record light intensities with different sensors and on a different scale. Though researchers have aimed to harmonize the two series, we only lose one year in our analysis and therefore prefer to use data from a single, consistent source. 37 Using a long lag cutoff in the HAC part of the errors implies that the weight of a time-series shock is almost constant, which is equivalent to clustering on regions in the time-series dimension. We also used higher spatial cutoffs to estimate Conley errors but found no substantive changes beyond 500 km (and a very small subset of standard errors could not be computed beyond 500 km). Alternatively, we clustered at the level of regions or at the level of regions and years. None of this affects our qualitative results. Journal of Urban Economics 145 (2025) 103730 8 R. Bluhm et al. Table 4 Identification: Altering the instrument, within first-order regions, 2002–2013. Overprod. All inputs Both factors Leave-one-out shares Cold War shares Africa US steel placebo Det. steel (1) (2) (3) (4) (5) (6) (7) Panel (a) 2SLS estimates Projects (𝛥𝑁𝑗 𝑖,𝑡−2)−0.0319 −0.0170 −0.0133 −0.0243 −0.0563 0.0952 −0.0262 (0.0118)*** (0.0109) (0.0067)** (0.0099)** (0.0288)* (0.1188) (0.0111)** [0.0103]*** [0.0081]** [0.0066]** [0.0077]*** [0.0508] [0.1112] [0.0089]*** Panel (b) First-stage estimates Instrument 1 0.3361 – 0.4819 0.4881 0.0778 −0.0647 0.3495 (0.0597)*** (0.0853)*** (0.0860)*** (0.0350)** (0.0740) (0.0642)*** [0.0612]*** [0.0772]*** [0.0785]*** [0.0301]*** [0.0737] [0.0631]*** Instrument 2 0.1748 (0.1026)* [0.0778]** Additions/Modifications Modified shock All six instruments Modified shock Modified shares Modified shares Placebo instrument Modified shock J-test (𝑝-value) – 0.34 0.09 – – – – First stage F-stat 31.73 36.63 17.38 32.20 4.94 0.76 29.63 Observations 27,162 27,162 27,162 27,162 8401 27,162 27,162 Regions 2406 2406 2406 2406 729 2406 2406 Countries 122 122 122 122 48 122 122 Notes: The table reports regressions results within first-order administrative regions. Panel a shows two-stage least-squares fixed-effects regressions where the dependent variable is the first difference of the Gini coefficient of light intensity within first-order administrative regions. Panel b shows the corresponding first-stage regressions where the dependent variable is a binary indicator for new project commitments (𝛥𝑁𝑗 𝑖,𝑡−2) in a region. ‘Overprod.’ implies that the factor inputs were residualized by running a regression of each input on the log of Chinese GDP at constant local currency units before the first factor was extracted. ‘Det. steel’ uses the linearly detrended log of Chinese steel production as the time-series shock. ‘All inputs’ uses interactions with detrended Chinese aluminum, cement, glass, iron, steel, and timber production time series as separate instruments. ‘Both factors’ uses the second principal factor of the detrended inputs as a second instrument. ‘Leave-one-out shares’ subtracts the value of 𝛥𝑁𝑖,𝑡−2 from the cross-sectional average in each period. The probability is now time-varying, so we control for it in both stages of the regression. ‘Cold War shares Africa’ uses the probability of receiving a project with data from the Cold War period from Bartke (1989) geocoded for Africa by Dreher et al. (2021a). ‘US placebo’ uses a US steel production index from FRED (IPN3311A2BS) as part of a placebo instrument. All specifications include region-fixed effects and country-year fixed effects. Standard errors clustered at the country level are reported in parentheses. Conley errors with a spatial cutoff of 500 km and a time-series HAC with a lag cutoff of 1000 years are reported in brackets. *** p<0.01, ** p<0.05, * p<0.1. 2000–2014 period in which a Chinese non-transport project has been committed (𝐹𝑡−3 × 𝑀𝑗 𝑖). We find that our results for transport projects hardly change once we control for non-transport projects. This implies that our results do not reflect the effect of all projects but those of transport projects specifically.56 The final column 5 of Table 3includes the three time-varying shocks and the indicator for non-transport projects in concert. Again, our results remain unchanged. The coefficient of transport projects remains significant and hardly varies in magnitude. While this does not rule out that other shocks that correlate with our instrument and spatial concentration bias our coefficient, given that the arguably most obvious sources of bias hardly change our results, we expect such bias to be negligible. The next set of tests focus on the instrument itself. Christian and Barrett (2024) suggest tests to probe the validity of the assumptions underlying our instrumental variable approach. In addition to visual inspection of trends in Fig. 4and Figure A-1, we conduct a randomization inference test where we reassign the transport project indicator and instrumental variable to different countries and years in the sample. As can be seen from the results of 999 Monte Carlo simulations shown in Figure A-2, the resulting coefficient estimates center around zero. According to an exact Fisher test, the coefficient from our main estimate (introduced below and indicated by the vertical dashed line) significantly differs from the randomized coefficients (𝑝-value =0.019). The same holds when we break the timing structure required for identification less radically and instead randomize (i) the entire time series between regions, (ii) years within regions, and (iii) regions within years (also shown in Figure A-2). Thus, omitted variables are unlikely to correlate with our key variables in a way that spuriously brings about our main result. 56 For brevity, we only report the first-stage results for transport projects but note that the first-stage F-statistic remains relatively high (for an equation with two endogenous variables), the estimated coefficient on our primary instrumental variable hardly changes, while the instrument for non-transport projects only predicts those types of projects. Table 4presents tests of robustness to modifications in the timeseries (‘shock’) and cross-sectional (‘share’) components of our instrument. Since our identification strategy leverages exogenous shocks, we first perturb the time series. Column 1 modifies our detrending method and residualizes each input series via a regression on the log of GDP at constant local currency units. Intuitively, this results in a measure of input growth that is faster or slower than GDP growth, as opposed to deviating from a linear trend. Column 2 (‘All inputs’) uses detrended Chinese aluminum, cement, glass, iron, steel, and timber production rather than their first principal component, each interacted with the same share component as before, as separate instruments.57 This modification delivers estimates that are one standard error smaller but, more importantly, does not reject a J-test that tests whether these instruments identify the same LATEs. In the same spirit, column 3 includes the second factor we derived from our factor analysis of China’s input materials (interacted with the original share) as part of a second instrument. The first-stage result suggests some information is contained in the second component, but the strength of the relationship drops and the confidence intervals of the 2SLS estimate overlap with that of our baseline. In columns 4 and 5, we test the robustness of alternative measures of the share component of our instrument – a region’s probability of receiving a project. While we follow the approach of Nunn and Qian (2014), researchers typically use shares calculated before the shock in the broader shift-share literature. A concern might be that computing the probability of receiving a development project using all available years might bias our estimates (even though the fixed effects control for the endogenous probability). We approach this concern in two ways. First, in column 4, we re-estimate our baseline model but exclude contemporaneous projects when we calculate probabilities, excluding 𝛥𝑁𝑗 𝑖,𝑡−2 from the cross-sectional average, and denote this the ‘leaveone-out’ instrument.58 Second, column 5 replaces the probability of 57 We do not report the six first-stage coefficients in the table to reduce clutter. 58 While this removes the contemporaneous correlation between the instrument and project commitments, the probability is now time-varying and is Journal of Urban Economics 145 (2025) 103730 15 R. Bluhm et al. Table 5 Timing of effects, within first-order regions, 2004–2013. Lag structure for 𝛥𝑁𝑖,𝑡−𝑠and 𝐹𝑡−𝑠−1 𝑠= 0𝑠= 1𝑠= 2𝑠= 3𝑠= 4 (1) (2) (3) (4) (5) Panel (a) 2SLS estimates, changes in 𝑠for project and IV Projects (𝛥𝑁𝑖,𝑡−𝑠)−0.0223 −0.0188 −0.0181 −0.0045 0.0026 (0.0187) (0.0128) (0.0114) (0.0067) (0.0079) [0.0167] [0.0110]* [0.0081]** [0.0073] [0.0088] Panel (b) 2SLS estimates, only changes in s for IV Projects (𝛥𝑁𝑖,𝑡−2)−0.0563 −0.0195 −0.0181 −0.0043 0.0026 (0.0501) (0.0131) (0.0114) (0.0062) (0.0082) [0.0568] [0.0113]* [0.0081]** [0.0069] [0.0089] First-stage F-stat panel a 6.08 6.74 16.56 20.43 25.79 First-stage F-stat panel b 2.40 12.55 16.56 15.86 11.81 Observations 22,445 22,445 22,445 22,445 22,445 Regions 2389 2389 2389 2389 2389 Countries 121 121 121 121 121 Notes: The table reports 2SLS regressions where the dependent variable is the first difference of the Gini coefficient of light intensity within first-order administrative regions. Both panels show two-stage least-squares fixed-effects regressions where the dependent variable is the Gini coefficient of light intensities within first-order administrative regions. Panel a fixes the lag structure for the first difference of projects and the instrument but varies years of commitment as indicated in the column header. For example, column 1 estimates the effects of projects committed in 𝑡, instrumented with the interaction including inputs produced in 𝑡− 1. Panel b fixes the timing of projects but varies the lag between projects and input production. For example, column 1 reports the effect of projects committed in 𝑡− 2, instrumented with the interaction including inputs produced in 𝑡− 2. All specifications include region-fixed effects and country-year fixed effects. Standard errors clustered at the country level are reported in parentheses. Conley errors with a spatial cutoff of 500 km and a time-series HAC with a lag cutoff of 1000 years are reported in brackets. *** p<0.01, ** p<0.05, * p<0.1. receiving a project with data from the Cold War period. We take these data from Dreher et al. (2021a), who provide geocoded information on 688 Chinese project locations completed in 47 African countries over the 1956–1987 period. Column 4 of Table 4shows that our results remain similar when we instrument projects with the ‘leave-one-out’ instrument. The results are still broadly in line with our main findings when we use Cold War projects in a subsample of African countries as the ‘share’ part of our instrument (in column 5), but the first stage is weaker. This is not surprising as pre-sample probabilities have a less direct relationship with the number of projects in any given year, making them a less potent predictor. Column 6 reports the results of a placebo regression, using (detrended) US steel production rather than Chinese production values. This presents a falsification test as US steel production should be unrelated to Chinese-financed infrastructure projects. To facilitate comparison, we show the analogous regression using Chinese steel production as part of our instrument in column 7.59 As expected, the first stage of the placebo regression collapses and is extremely weak, with a Kleibergen–Paap F-statistic of just 0.76, while the 2SLS estimate is positive and imprecisely estimated. In short, US steel production does not help to predict the commitment of Chinese-financed transport projects. On the contrary, as can be seen from column 7, first-stage and second-stage estimates are similar to our baseline results when we use China’s steel production as shift component of our instrument. Taken together, these estimates suggest that our results do not hinge on the specific choice of how we define Chinese production shocks and the probability to receive Chinese projects and that spurious trends do not drive the time-series component. Table 5investigates the timing of the diffusion effects of Chinese transport projects in more detail. Recall that we have determined the two-year lag duration for our analysis based on Dreher et al. (2021b), who provide start and end dates for 300 projects.60 We continue to thus no longer absorbed by the region-fixed effects. Instead, we control for it in both stages of the regression. We have also calculated the probability based on time periods before the respective commitment and using projects committed to neighboring regions rather than a region itself. In both cases, first-stage F-statistics are low. 59 Note that none of these results control for the other ‘‘China shocks’’ from the previous table, but the results hardly change if we include them. focus on first differences but vary the timing by which we allow projects to affect spatial concentration as well as the timing with which we assume construction materials to affect projects. Panel a fixes the oneyear lag between the instrument and Chinese transport projects but shifts both forward and backward in time. Column 1 starts with the effects of projects in the year of commitment (𝑠= 0) instrumented by the first lag of project inputs. Column 5 ends with the effect of projects committed four years earlier (𝑠= 4), instrumented by the fifth lag of project inputs. To allow for a meaningful analysis of the effect of the lag structure, we hold the sample constant across all regressions, resulting in fewer observations compared to the baseline regression. Column 3 reports results analogous to our baseline specification. The results are somewhat weaker compared to those reported above but remain qualitatively similar. Estimates become smaller for longer lags (see columns 4 and 5) and are imprecisely estimated, even though the first-stage relationship between inputs and commitments is strong for deeper lags. Column 2 shows the effects on concentration one year after commitment. The coefficients are similar to our baseline regression in magnitude but are estimated less precisely, and the first-stage relationship becomes weak. When we turn to contemporary effects in the year of commitment in column 1, the first stage turns out even weaker and the estimates are no longer significant at conventional levels.61 Taken together, these results support our choice of a two-year lag and show the robustness to using a one-year lag, in line with the notion that much of the impact occurs early. Panel b of Table 5examines the lag between our instrument and project commitments. We focus on transport projects committed in 𝑡− 2, but vary the lag of the instrument between one and four years (corresponding to 𝑠= 0to 𝑠= 4). We do not claim that variations 60 More precisely, the average observed project duration in this subsample is 664 days. Dreher et al. (2021b) point out that this lag choice aligns with the prevailing belief held by development practitioners and government officials in host countries that Chinese development projects are executed swiftly. 61 We have also tested whether future projects predict past concentration (not reported). Significant estimates of future projects on past concentration would substantially weaken the credibility of our estimation strategy. As expected, the ‘‘effect’’ of projects one or two years in the future on today’s concentration is estimated very imprecisely, with first-stage F-statistics below one. Journal of Urban Economics 145 (2025) 103730 16 R. Bluhm et al. Table 6 Light intensity and quintile shares, within first-order regions, 2002–2013. Moments of spatial concentration Light Light Extensive Quintile shares density per capita margin 0%–20% 20%–40% 40%–60% 60%–80% 80%–100% (1) (2) (3) (4) (5) (6) (7) (8) Panel (a) 2SLS estimates Projects (𝛥𝑁𝑗 𝑖,𝑡−2) 0.1462 −0.0005 0.0122 0.0028 0.0032 0.0102 0.0191 −0.0353 (0.0476)*** (0.0028) (0.0173) (0.0015)* (0.0027) (0.0040)** (0.0075)** (0.0114)*** [0.0516]*** [0.0028] [0.0104] [0.0019] [0.0024] [0.0043]** [0.0078]** [0.0116]*** Panel (b) First-stage estimates IV (𝐹𝑡−3 × 𝑁𝑗 𝑖) 0.4505 0.4498 0.4505 0.4394 0.4394 0.4394 0.4394 0.4394 (0.0732)*** (0.0732)*** (0.0732)*** (0.0749)*** (0.0749)*** (0.0749)*** (0.0749)*** (0.0749)*** [0.0688]*** [0.0688]*** [0.0688]*** [0.0716]*** [0.0716]*** [0.0716]*** [0.0716]*** [0.0716]*** First-stage F-stat 37.86 37.72 37.86 34.44 34.44 34.44 34.44 34.44 Observations 28,037 28,025 28,037 26,877 26,877 26,877 26,877 26,877 Regions 2440 2439 2440 2379 2379 2379 2379 2379 Countries 122 122 122 122 122 122 122 122 Notes: The table reports 2SLS regressions where the dependent variable is the first difference of the Gini coefficient of light intensity within first-order administrative regions. Panel a shows two-stage least-squares fixed-effects regressions where the dependent variable is indicated in the column header. Panel b shows the corresponding first-stage regressions where the dependent variable is a binary indicator for new project commitments (𝛥𝑁𝑗 𝑖,𝑡−2) in a region. We use the inverse hyperbolic sine transformation in columns 1 and 2, which is defined as 𝑖ℎ𝑠(𝑧) =𝑙 𝑜𝑔(𝑧+√𝑧2+ 1), for light density and light per capita to include regions with zero light and retain an interpretation similar to logs. The extensive margin in column 3 is defined as the (untransformed) fraction of pixels with a non-zero light density. All specifications include region-fixed effects and country-year fixed effects. Standard errors clustered at the country level are reported in parentheses. Conley errors with a spatial cutoff of 500 km and a time-series HAC with a lag cutoff of 1000 years are reported in brackets. *** p<0.01, ** p<0.05, * p<0.1. of our instrumental variable are excludable for different years but consider this as a predictive exercise to find the lag structure for which the relationship is strongest. For example, projects committed in 2010 are likely to be driven by input materials not only in 2009, but also in other adjacent years. When we vary the lag between projects and input materials, the first-stage relationship is strongest for our preferred one-year lag between inputs and commitments and declines for all other timings. The first-stage relationship between inputs and projects completely breaks down when we predict projects with inputs produced one year in the future (column 1). Overall, these results support our choice of a one-year lag between inputs and commitments. Extensions Our finding that Chinese-financed transport projects reduce the concentration of economic activity within subnational regions raises the question of where exactly this diffusion takes place. The monocentric city model implies that we should observe a shift in activity from cities to their immediate periphery (and, hence, expect some heterogeneity with respect to the level of urbanization). The model has little to say about whether this should also increase overall activity in a region or whether this kind of development occurs by leap-frogging into undeveloped areas or integrating less densely developed areas. Other frameworks offer some guidance. Heblich et al. (2020) study how transportation technology affects the specialization of locations within a city into a workplace and residence in a large range of quantitative urban models. A key finding is that improvements in transport technology lead to faster growth of suburbs relative to the central city by developing open fields and building out preexisting villages. Core–periphery models, in turn, offer predictions on overall activity as regions become better connected. For example, Faber (2014) shows that the effect of infrastructure improvements is negative for peripheral regions but heterogeneous in the level of pre-existing trade integration and relative market size between the core and periphery. We take these results as motivation to study the effects of Chinese infrastructure investment for different moments of the distribution of nighttime lights, different regions of the world, and different levels of urbanization and trade integration. Table 6examines different moments of the distribution. Columns 1 and 2 show a strong effect of transport projects on overall economic activity but not on our proxy for per capita incomes. A new transport project increases the average light density in a region by about 15 percent (column 1), which is both economically and statistically significant. When we instead focus on light per capita (column 2), we cannot reject the null hypothesis that Chinese transport projects have no effect on changes in light per capita and estimate a coefficient close to zero. This aligns with the results of Dreher et al. (2022), who also report a null effect of Chinese development finance in general on lights per capita in a global sample. The insignificant coefficient for the world sample stands in contrast to results for the African continent, where previous work finds positive effects of aid on development (Dreher et al.,2021a,2022). Since the per capita data use interpolated population data in the denominator, we cannot rule out that the imprecise estimate occurs due to added noise in the dependent variable. Column 3 uses the fraction of illuminated pixels as the dependent variable. The insignificant result suggests that economic activity does not seem to expand primarily into previously undeveloped areas. To summarize, columns 1 to 3 show that Chinese transport projects increase economic activity in the receiving region. This may represent an increase in population rather than welfare and appears to be primarily occurring in already somewhat developed areas. The remaining columns of Table 6focus on relative changes in economic activity across quintiles of the light distribution. This allows us to directly trace which type of changes reduce the Gini coefficient. The pattern is consistent with predictions from urban land use theory. We find that Chinese transport projects significantly reduce nighttime lights in the highest quintile, while they raise the share of activity taking place in the lower quintiles (though estimated imprecisely for the second quintile and with borderline significance for the first). It thus appears that Chinese transport projects gradually redistribute activity from the most densely developed parts of regions, that is, the city centers, to less densely developed places. The magnitudes of the estimated coefficients suggest that this process benefits the higher quintiles more relative to the least developed parts of a region. Table 7presents a more direct approach to measuring from where to where the relocation of activity takes place. We report a series of regressions that split the sample along the median of several variables typically linked with rapid urban growth. The results provide further support for the conjecture that these effects occur around cities. We find a sizable decentralization of activity in regions with belowmedian travel time to cities, high urbanization rates, high road density, and above-median proximity to the coast. The estimated effects are substantial in these sub-samples. By contrast, they are imprecisely estimated and typically of the opposite sign or lower in magnitude in Journal of Urban Economics 145 (2025) 103730 17 R. Bluhm et al. Table 7 Sample splits, within first-order regions, 2002–2013. Splitting at the median of … Travel time Urbanization Road Distance Light to cities rate density to coast per capita (1) (2) (3) (4) (5) Panel (a) Below median, 2SLS estimates Projects (𝛥𝑁𝑗 𝑖,𝑡−2)−0.0306 0.0044 −0.0101 −0.0207 −0.0276 (0.0147)** (0.0087) (0.0099) (0.0154) (0.0089)*** [0.0104]*** [0.0103] [0.0107] [0.0090]** [0.0056]*** Panel (b) Above median, 2SLS estimates Projects (𝛥𝑁𝑗 𝑖,𝑡−2) 0.0003 −0.0222 −0.0262 −0.0232 −0.0288 (0.0105) (0.0180) (0.0121)** (0.0096)** (0.0287) [0.0097] [0.0129]* [0.0112]** [0.0098]** [0.0312] First-stage F-stat a 11.74 20.15 22.60 23.23 21.01 First-stage F-stat b 39.68 12.11 16.90 16.33 13.76 Observations (a) 13,016 13,642 13,527 13,509 13,142 Observations (b) 13,875 13,254 13,451 13,545 13,672 Notes: The table reports 2SLS regressions where the dependent variable is the first difference of the Gini coefficient of light intensity within first-order administrative regions. Panel a shows two-stage least squares fixed effects regressions for first-order regions with below median values of the variable indicated in the column header. Panel b shows two-stage least squares fixed effects regressions for first-order regions with above median values of the variable indicated in the column header. ‘Travel time to cities’ is measured as the travel time to the nearest city of 50,000 or more people in the year 2000 (Nelson,2008). The ‘urbanization rate’ is measured as the fraction of land in the region which is defined as an urban cluster or urban center in 2000 by the Global Human Settlement Layer (Pesaresi et al.,2019). ‘Road density’ is measured as the total road length over the area of the region where road length is derived from the gROADS data set (CIESIN and ITOS,2013). ‘Distance to coast’ is the average ‘‘as-the-crow-flies’’ distance to the nearest coastline (from Natural Earth). ‘Light per capita’ is the sum of light in a region divided by its population in 2000 (from the Global Human Settlement Layer). All specifications include region-fixed effects and country-year fixed effects. Standard errors clustered at the country level are reported in parentheses. Conley errors with a spatial cutoff of 500 km and a time-series HAC with a lag cutoff of 1000 years are reported in brackets. *** p<0.01, ** p<0.05, * p<0.1. the other sub-samples (the exception being below-median proximity to the coast, with similar estimated magnitudes and standard errors in both samples). Last but not least, we also find that the effect seems to be driven by relatively poor regions as measured by below-median light per capita. This is not surprising, given that some of the poorest regions have some of the highest population growth rates and are home to many of the fastest-growing cities. The evidence presented here aligns with the literature focusing on individual countries or regions discussed above. For example, in their study of the expansion of China’s highway system, Baum-Snow et al. (2017) find that reductions in spatial concentration were larger within coastal and richer central regions. Similarly, although they do not focus on decentralization within regions per se, studies focusing on the spatial impact of the Belt and Road Initiative (BRI) typically estimate that coastal regions, border crossings, and urban hubs will benefit more (Lall and Lebrand,2020).62 Next, we investigate major world regions separately. China’s global infrastructure footprint is uneven and most of its transportation projects are located in Africa and Asia (recall Fig. 2). Urban population growth is rapid and infrastructure constraints are most severe in these regions. As can be seen from the sub-samples in columns 1 to 3 of Table 8, our main findings are driven by regions in African countries, where the effect is larger than our baseline estimates. The coefficient on Chinese transport projects is insignificant or substantially smaller for Asia and the Americas, although the first stage remains about equally powerful in all three regions. This is not surprising given that Africa lags behind the other two world regions in terms of infrastructure development. It is also the region where urban primacy is most pronounced and where deficiencies in urban infrastructure have been linked to slower economic growth at the national level (CastellsQuintana,2017). Chinese-financed projects in Africa, therefore, appear to mitigate congestion, which, eventually, could enable cities to reap the benefits of agglomeration economies. Column 4 restricts the sample to countries classified by the World Bank as low-income economies in 2000. It highlights that the diffusion 62 This literature suggests that BRI projects will lead to an increasing specialization among regions and hence more concentration of economic activity in regions with better access to world markets but does not consider the distribution of activity within regions. effects of Chinese transport projects also occur in the poorest countries of the world. Finally, in column 5, we restrict our analysis to only those subnational regions that have received at least one transport project from China over the entire sample period. This addresses one last identification challenge that would arise if regions that received any development-related project from China experience different nonlinear trends than those which did not. Our results become substantially stronger. Finally, Table A-6 in the Online Appendix explores the issue of colocation with other types of projects. Our results remain similar when we control for the presence of World Bank projects in the transport sector (or in any sector) and for Chinese-financed projects in other sectors (also recall that our results are robust to including any non-transport project from China, as shown in column 4 of Table 3). 6. Conclusion This article examined whether and to what extent transport infrastructure projects decentralize economic activity in recipient regions across the Global South. We overcome the challenge of missing geolocalized data on comparable infrastructure projects across countries and how to estimate their causal effects by focusing on infrastructure projects financed by the Chinese government – a single but massive source of infrastructure financing across the developing world. While many scholars and policymakers are skeptical about the quality and effects of Chinese development projects, China’s commitment to financing overseas infrastructure was unambiguous during the first two decades of the 21st century. Transport projects, such as roads, highways, railways, harbors, and airports, are at the heart of this approach, and Chinese state-owned entities have financed hundreds of them in developing countries since 2000 (Dreher et al.,2022). One of our key contributions is to provide a new geocoded dataset of China’s growing development footprint around the world, much of which comes in the form of large-scale infrastructure investments but extends across multiple sectors. While our data cover the period from 2000 to 2014 and thus mostly precede the BRI, the projects we focus on share many characteristics with infrastructure built during the first decade of the BRI. Using these data, we test whether infrastructure projects influence the spatial concentration of economic activity Journal of Urban Economics 145 (2025) 103730 18 R. Bluhm et al. Table 8 Regional variation, within first-order regions, 2000–2013. Regional subsets and related sample perturbations Africa Asia Americas Low income  𝑁all >0 (1) (2) (3) (4) (5) Panel (a) 2SLS estimates Projects (𝛥𝑁𝑗 𝑖,𝑡−2)−0.0252 −0.0134 −0.0084 −0.0204 −0.0301 (0.0076)*** (0.0244) (0.0079) (0.0117)* (0.0073)*** [0.0100]** [0.0112] [0.0019]*** [0.0058]*** [0.0100]*** Panel (b) First-stage estimates IV (𝐹𝑡−3 × 𝑁𝑗 𝑖) 0.4413 0.4171 0.7981 0.4327 0.4348 (0.1123)*** (0.1020)*** (0.2285)*** (0.0897)*** (0.0753)*** [0.0997]*** [0.1096]*** [0.3355]** [0.0826]*** [0.0744]*** First-stage F-Stat 15.45 16.72 12.20 23.28 33.31 Observations 8401 9191 4954 11,357 8639 Regions 729 791 430 982 735 Countries 48 34 22 60 92 Notes: The table reports 2SLS regressions. Panel a shows two-stage least squares fixed effects regressions where the dependent variable is the first difference of the Gini coefficient of light intensity within first-order administrative regions. Panel b shows least squares fixed effects regressions where the dependent variable is a binary indicator for new project commitments (𝛥𝑁𝑗 𝑖,𝑡−2) in a region. Columns 1 to 3 report regional subsets as indicated in the column header. Column 4 uses only countries classified as low-income economies by the World Bank in 2000. Column 5 uses only regions that have received any transport or non-transport project from China over the entire period. All specifications include region-fixed effects and country-year fixed effects. Standard errors clustered at the country level are reported in parentheses. Conley errors with a spatial cutoff of 500 km and a time-series HAC with a lag cutoff of 1000 years are reported in brackets. *** p<0.01, ** p<0.05, * p<0.1. within and between recipient regions. Our identification strategy relies on commodity inputs produced in China that affect the availability of projects over time in tandem with a variable that measures the likelihood that countries receive a smaller or larger share of China’s projects. Our results show that Chinese government-financed transportation projects reduce the concentration of economic activity within regions in developing countries. While we find similar effect sizes for concentration between regions, these effects are estimated less precisely. Our within-region results imply that the Gini coefficient measuring the spatial concentration of economic activity is reduced by 2.2 percentage points within first-order regions. These results are robust in a large number of different specifications, to the choice of control variables and variations of the instrumental variable. The effect increases for completed projects, holds for projects financing economic infrastructure more broadly, and is largest in poor regions and African countries, which most need infrastructure financing. In line with urban land use theory, we find that our results are driven by changes in economic activity in and around urban areas. In financing major transport projects, the Chinese government appears to be helping cities and regions in developing countries transform from dense, crowded, and unproductive places into productive hubs. While these results are encouraging, they do not imply that Chinese government-financed transport infrastructure projects only have positive effects. There is growing evidence that Chinese development projects also produce negative externalities. For example, in related work, we have shown that China’s ‘‘aid on demand’’ approach is vulnerable to domestic political capture wherein incumbent government leaders steer Chinese development projects towards their home regions, often at the expense of poorer regions with greater material need (Dreher et al.,2019a). There are many other concerns about the consequences of China’s development finance, ranging from their impact on the environment and debt sustainability (Horn et al.,2021; Baehr et al.,2023). In short, Chinese-financed transportation projects may help deal with congestion in developing countries, but our study should not be read as a comprehensive assessment of their costs and benefits. There is considerable scope for future research in this area. CRediT authorship contribution statement Richard Bluhm: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Axel Dreher: Writing – review & editing, Writing – original draft, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Andreas Fuchs: Writing – review & editing, Writing – original draft, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Bradley C. Parks: Writing – review & editing, Writing – original draft, Methodology, Investigation, Funding acquisition, Data curation, Conceptualization. Austin M. Strange: Writing – review & editing, Writing – original draft, Methodology, Investigation, Funding acquisition, Data curation, Conceptualization. Michael J. Tierney: Project administration, Funding acquisition, Conceptualization. Acknowledgments We owe a debt of gratitude to the large team of research assistants – including Melanie Aguilar-Rojas, Bilal Asad, Zach Baxter, Rachel Benavides, Ellie Bentley, Liliana Besosa, Allison Bowers, Ariel Cadby-Spicer, Emma Cahoon, Bree Cattelino, Alex Chadwick, Ava Chafin, Tina Chang, Yuning Chen, Mengfan Cheng, Tiffanie Choi, Miranda Clarke, Kate Connors, McKay Corbett, Graeme Cranston-Cuebas, Catherine Crowley, Alex DeGala, Hannah Dempsey, Rohan Desai, Justin DeShazor, Joseph Dobbels, Isabel Docampo, Weiwei Du, Ashton Ebert, Caleb Ebert, Aili Espigh, Claire Etheridge, Jordan Fox, Robert Francis, Ze Fu, Melanie Gilbert, Sara Gomez, Liz Hall, Thompson Hangen, Lauren Harrison, Michael Hathaway, Collin Henson, Jasmine Herndon, Elizabeth Herrity, Keith Holleran, Weijue Huang, Daniel Hughes, Torey Jackson, Jiaorui Jiang, Qi Jiang, Emmaleah Jones, Amar Kakirde, Rachel Kellogg, Connor Kennedy, Ciera Killen, Ian Kirkwood, Warren Kirkwood, Emily Koerner, Dylan Kolhoff, Lidia Kovacevic, Mirian Kreykes, Isabella Kron, Karthik Kumarappan, Daniel Lantz, Caroline Lebegue, Jade Li, Xiao Liu, Steven Livingston, Yaseen Lofti, Adriane Lopez, Flynn Madden, Dominick Margiotta, Sarah Martin, Emily McLenigan, Marie Mullins, Will Nelson, Qiuyan Ni, Jack Nicol, Alexandra Pancake, Carol Peng, Grace Perkins, Sophia Perrotti, Victor Polanco, Laura Preszler, Emily Qiu, Kamran Rahman, Sarah Reso, David Rice, Sara Rock, Ann Rogers, Elizabeth Saccoccia, Natalie Santos, Dominic Sanzotta, Faith Savaiano, Dominic Scerbo, Rebecca Schectman, Leigh Seitz, Ryan Septon, Lu Sevier, Kaitlan Shaub, Andrea Soleta, Lauren Su, Joanna Tan, Emily Tanner, Nate Tanner, Brittany Tennant, Rebecca Thorpe, Austin Trotta, Anna Umstead, Jessica Usjanauskas, Julia Varoutsos, Emily Walker, Yale Waller, Katherine Walsh, Xinyi Wang, Matt Westover, Tom Westover, James Willard, (Jiacheng) Jason Xi, Hanyang Xu, Darice Xue, Erya Yang, Antonio Tianze Ye, Jack Zhang, Yue Zhang, Echo Zhong, Joana Zhu, and Junrong Zhu—who helped to assemble and geo-locate Journal of Urban Economics 145 (2025) 103730 19 R. Bluhm et al. the dataset of Chinese development projects used in this study. We also thank Seth Goodman and Miranda Lv for their spatial data quality assurance and integration assistance, as well as Gerda Asmus for sharing code for aggregating the project-level data. This study was made possible with generous financial support from John D. and Catherine T. MacArthur Foundation, United States, Humanity United, United States, the William and Flora Hewlett Foundation, United States, the Academic Research Fund of Singapore’s Ministry of Education, the United Nations University World Institute for Development Economics Research (UNU-WIDER), Finland, the German Research Foundation (DFG projects DR 640/5-1&3 and FU 997/1-1&3), and William & Mary, United States. We also acknowledge that this study was indirectly made possible through a cooperative agreement (AID-OAA-A-12-00096) between USAID’s Global Development Lab and AidData at William & Mary under the Higher Education Solutions Network (HESN) Program, as it supported the creation of a spatial data repository and extraction tool which we used to execute our data analysis. The views expressed here do not necessarily reflect the views of any of our funders. Last but not least, this paper received many helpful comments at research seminars, workshops, and conferences. We thank Bruce Bueno de Mesquita, Vivien Foster, Erik Haustein, Robert Inklaar, Mathilde Lebrand, Ferdinand Rauch, seminar audiences at New York University (New York City, USA, September 2018), the Graduate Institute of International and Development Studies (Geneva, Switzerland, April 2019), the University of Milano-Bicocca (Milan, Italy, May 2019), the University of Hohenheim (Stuttgart, Germany, July 2019), the European Stability Mechanism (Luxembourg, November 2019), the University College Dublin (Dublin, Ireland, November 2019), the University of Groningen (Groningen, Netherlands, January 2020), and the University of British Columbia (Vancouver, Canada, February 2020); as well as conference participants of the HSU-IfW Workshop in Development and Environmental Economics (Hamburg, Germany, November 2018), the TEDE Workshop ‘‘Topics in Development and Environmental Research’’ at the University of Birmingham (Birmingham, UK, February 2019), the Annual Meeting of the European Public Choice Society at the Hebrew University (Jerusalem, Israel, April 2019), the GeoData in Economics Workshop at the University of Hamburg (Hamburg, Germany, May 2019), the European Meeting of the Urban Economics Association at the Vrije Universiteit (Amsterdam, Netherlands, June 2019), the Annual Conference of the Verein für Socialpolitik Research Committee Development Economics at the DIW (Berlin, Germany, June 2019), the Annual Bank Conference on Development Economics at the World Bank (Washington DC, USA, June 2019), the China Economics Summer Institute at Peking University (Beijing, China, August 2019), the Annual Economic Research Southern Africa Workshop on ‘‘Structural Constraints on the Economy, Growth and Political Economy’’ at the University of the Witwatersrand (Johannesburg, South Africa, September 2019), the Biennial Conference of the Economic Society of South Africa (Johannesburg, South Africa, September 2019), and the Annual Meeting of the International Political Economy Society at the University of California San Diego (San Diego, USA, November 2019) for comments on earlier versions of this paper. Appendix A. Supplementary data Supplementary material related to this article can be found online at https://doi.org/10.1016/j.jue.2024.103730. References Adão, R., Kolesár, M., Morales, E., 2019. Shift-share designs: Theory and inference. Q. J. Econ. 134 (4), 1949–2010. African Development Bank, 2014a. Kenya: Nairobi-Thika Highway Improvement Project. African Development Bank, Abidjan, Côte d’Ivoire, Available at https://www.afdb.org/fileadmin/uploads/afdb/Documents/Project-andOperations/Presidential_awards_2014_-_Kenya_-_Nairobi_-_Thika_Highway_ Improvement_Project.pdf. (Accessed 3 April 2020). African Development Bank, 2014b. Study on Quality of Bank Financed Road Projects. African Development Bank, Abidjan, Côte d’Ivoire, Available at https://www.afdb.org/fileadmin/uploads/afdb/Documents/Events/ATFforum/ Study_on_the_quality_of_Bank_Financed_Road_Projects_-_AfDB.pdf. (Accessed 3 April 2020). African Development Bank, 2016. PCR Evaluation Note for Public Sector Operations: Nairobi-Thika Highway Improvement Project. African Development Bank, Abidjan, Côte d’Ivoire, Available at https://evrd.afdb.org/documents/docs/EN_PN10706.pdf. (Accessed 3 April 2020). African Development Bank, 2019. Nairobi-Thika Highway Improvement Project: Project Completion Report. African Development Bank, Abidjan, Côte d’Ivoire, Available at https://www.afdb.org/sites/default/files/documents/projects-andoperations/kenya_-_thika_highway_improvement_project.pdf. (Accessed 3 April 2020). African Development Fund, 2007. Appraisal Report: Nairobi–Thika Highway Improvement Project. African Development Fund, Tunis, Tunisia, Available at https: //www.afdb.org/fileadmin/uploads/afdb/Documents/Project-and-Operations/ Kenya_-_Nairobi-Thika_Highway_Improvement_Project_-_Appraisal_Report.PDF. (Accessed April 2020). Allen, T., Arkolakis, C., 2022. The welfare effects of transportation infrastructure improvements. Rev. Econ. Stud. 89 (6), 2911–2957. Alonso, W., et al., 1964. Location and Land Use: Toward a General Theory of Land Rent. Harvard University Press, Cambridge, MA. Anaxagorou, C., Efthyvoulou, G., Sarantides, V., 2020. Electoral motives and the subnational allocation of foreign aid in Sub-Saharan Africa. Eur. Econ. Rev. 127, 103430. Andrés, L., Biller, D., Dappe, M.H., 2014. Infrastructure Gap in South Asia: Infrastructure Needs, Prioritization, and Financing. The World Bank, Washington, DC. Atkin, D., Donaldson, D., 2015. Who’s Getting Globalized? The Size and Implications of Intra-national Trade Costs. NBER Working Paper 21439, National Bureau of Economic Research, Cambridge, MA. Autor, D.H., Dorn, D., Hanson, G.H., 2013. The China syndrome: Local labor market effects of import competition in the United States. Amer. Econ. Rev. 103 (6), 2121–2168. Autor, D.H., Dorn, D., Hanson, G.H., 2016. The China shock: Learning from labor-market adjustment to large changes in trade. Annu. Rev. Econ. 8, 205–240. Baehr, C., BenYishay, A., Parks, B., 2023. Highway to the forest? Land governance and the siting and environmental impacts of Chinese government-funded road building in Cambodia. J. Environ. Econ. Manag. 122, 102898. Bandiera, L., Tsiropoulos, V., 2020. A framework to assess debt sustainability under the Belt and Road Initiative. J. Dev. Econ. 146, 102495. Banerjee, A., Duflo, E., Qian, N., 2020. On the road: Access to transportation infrastructure and economic growth in China. J. Dev. Econ. 145, 102442. Bartke, W., 1989. The Economic Aid of the PR China to Developing and Socialist Countries, second ed. K. G. Saur, Munich, Germany. Baum-Snow, N., 2007. Did highways cause suburbanization? Q. J. Econ. 122 (2), 775–805. Baum-Snow, N., 2014. Urban Transport Expansions, Employment Decentralization, and the Spatial Scope of Agglomeration Economies. Brown University, Providence, Unpublished Manuscript. Baum-Snow, N., Brandt, L., Henderson, J.V., Turner, M., Zhang, Q., 2017. Roads, railroads, and decentralization of Chinese cities. Rev. Econ. Stat. 99 (3), 435–448. Baum-Snow, N., Henderson, J.V., Turner, M.A., Zhang, Q., Brandt, L., 2020. Does investment in national highways help or hurt hinterland city growth? J. Urban Econ. 115, 103–124. Baum-Snow, N., Turner, M.A., 2017. Transport infrastructure and the decentralization of cities in the People’s Republic of China. Asian Dev. Rev. 34 (2), 25–50. Baum-Snow, N., et al., 2007. Suburbanization and transportation in the monocentric model. J. Urban Econ. 62 (3), 405–423. Bayes, A., 2007. Impact Assessment of Jamuna Multipurpose Bridge Project (JMBP) on Poverty. Japan Bank for International Cooperation, Dhaka, Bangladesh. BenYishay, A., Parks, B., Runfola, D., Trichler, R., 2016. Forest Cover Impacts of Chinese Development Projects in Ecologically Sensitive Areas. AidData Working Paper 32, AidData at William & Mary, Williamsburg, VA. Berman, N., Couttenier, M., 2015. External shocks, internal shots: The geography of civil conflicts. Rev. Econ. Stat. 97 (4), 758–776. Bird, J., Lebrand, M., Venables, A.J., 2020. The Belt and Road Initiative: Reshaping economic geography in Central Asia? J. Dev. Econ. 144, 102441. Bird, J., Straub, S., 2014. The Brasilia Experiment: Road Access and the Spatial Pattern of Long-Term Local Development in Brazil. World Bank Policy Research Working Paper 6964, The World Bank, Washington, DC. Bluhm, R., Dreher, A., Fuchs, A., Parks, B.C., Strange, A.M., Tierney, M.J., 2024. Replication Data for: Connective Financing. Chinese Infrastructure Projects and the Diffusion of Economic Activity in Developing Countries. Mendeley Data, V1, http://dx.doi.org/10.17632/235v7ksk8y.1. Bluhm, R., Krause, M., 2022. Top lights: Bright cities and their contribution to economic development. J. Dev. Econ. 157, 102880. Borusyak, K., Hull, P., Jaravel, X., 2022. Quasi-experimental shift-share research designs. Rev. Econ. Stud. 89 (1), 181–213. Journal of Urban Economics 145 (2025) 103730 20 R. Bluhm et al. Bräutigam, D., 2009. The Dragon’s Gift: The Real Story of China in Africa. Oxford University Press, Oxford, UK. Brazys, S., Elkink, J.A., Kelly, G., 2017. Bad neighbors? How co-located Chinese and World Bank development projects impact local corruption in Tanzania. Rev. Int. Organ. 12 (2), 227–253. Brazys, S., Vadlamannati, K.C., 2021. Aid curse with Chinese characteristics? Chinese development flows and economic reforms. Public Choice 188, 407–430. Bruederle, A., Hodler, R., 2018. Nighttime lights as a proxy for human development at the local level. PloS One 13 (9), e0202231. Brülhart, M., Desmet, K., Klinke, G.-P., 2020. The shrinking advantage of market potential. J. Dev. Econ. 147, 102529. Burchfield, M., Overman, H.G., Puga, D., Turner, M.A., 2006. Causes of sprawl: A portrait from space. Q. J. Econ. 121 (2), 587–633. Castells-Quintana, D., 2017. Malthus living in a slum: Urban concentration, infrastructure and economic growth. J. Urban Econ. 98, 158–173. Cervero, R., 2013. Linking urban transport and land use in developing countries. J. Transp. Land Use 6 (1), 7–24. Che, Y., He, X., Zhang, Y., 2021. Natural resource exports and African countries’ voting behaviour in the United Nations: Evidence from the economic rise of China. Can. J. Econ./Rev. Can. Écon. 54 (2), 712–759. Christian, P., Barrett, C.B., 2024. Spurious regressions and panel IV estimation: Revisiting the causes of conflict. Econom. J. 134 (659), 1069–1099. CIESIN and ITOS, 2013. Global Roads Open Access Data Set, Version 1 (gROADSv1). NASA Socioeconomic Data and Applications Center (SEDAC). de Chaisemartin, C., d’Haultfoeuille, X., 2020. Two-way fixed effects estimators with heterogeneous treatment effects. Amer. Econ. Rev. 110 (9), 2964–2996. de Soyres, F., Mulabdic, A., Ruta, M., 2020. Common transport infrastructure: A quantitative model and estimates from the Belt and Road Initiative. J. Dev. Econ. 143, 102415. Dollar, D., 2008. Supply Meets Demand: Chinese Infrastructure Financing in Africa. World Bank Blog, 10 July 2008. Accessed at https://blogs.worldbank.org/ eastasiapacific/supply-meets-demand-chinese-infrastructure-financing-in-africa. Donaldson, D., 2018. Railroads of the Raj: Estimating the impact of transportation infrastructure. Am. Econ. Rev. 108 (4–5), 899–934. Dreher, A., Fuchs, A., 2015. Rogue aid? An empirical analysis of China’s aid allocation. Can. J. Econ. 48 (3), 988–1023. Dreher, A., Fuchs, A., Hodler, R., Parks, B.C., Raschky, P.A., Tierney, M.J., 2019a. African leaders and the geography of China’s foreign assistance. J. Dev. Econ. 140, 44–71. Dreher, A., Fuchs, A., Hodler, R., Parks, B.C., Raschky, P.A., Tierney, M.J., 2021a. Is favoritism a threat to Chinese aid effectiveness? A subnational analysis of Chinese development projects. World Dev. 139, 105291. Dreher, A., Fuchs, A., Langlotz, S., 2019b. The effects of foreign aid on refugee flows. Eur. Econ. Rev. 112, 127–147. Dreher, A., Fuchs, A., Parks, B.C., Strange, A., Tierney, M.J., 2018. Apples and dragon fruits: The determinants of aid and other forms of state financing from China to Africa. Int. Stud. Q. 62, 182–194. Dreher, A., Fuchs, A., Parks, B., Strange, A., Tierney, M., 2021b. Aid, China, and growth: Evidence from a new global development finance dataset. Am. Econ. J.: Econ. Policy 13 (2), 135–174. Dreher, A., Fuchs, A., Parks, B., Strange, A., Tierney, M.J., 2022. Banking on Beijing: The Aims and Impacts of China’s Overseas Development Program. Cambridge University Press, Cambridge, MA. Dreher, A., Langlotz, S., 2020. Aid and growth: New evidence using an excludable instrument. Can. J. Econ. 53 (3), 1162–1198. Duranton, G., Turner, M.A., 2012. Urban growth and transportation. Rev. Econ. Stud. 79 (4), 1407–1440. Eichenauer, V.Z., Fuchs, A., Brückner, L., 2021. The effects of trade, aid and investment on China’s image in Latin America. J. Comp. Econ. 49 (2), 483–498. Elvidge, C.D., Baugh, K.E., Zhizhin, M., Hsu, F.-C., 2013. Why VIIRS data are superior to DMSP for mapping nighttime lights. Proc. Asia-Pac. Adv. Netw. 35, 62–69. Faber, B., 2014. Trade integration, market size, and industrialization: Evidence from China’s National Trunk Highway System. Rev. Econ. Stud. 81 (3), 1046–1070. Fajgelbaum, P., Redding, S.J., 2022. Trade, structural transformation, and development: Evidence from Argentina 1869–1914. J. Polit. Econ. 130 (5), 1249–1318. Fujita, M., Ogawa, H., 1982. Multiple equilibria and structural transition of non-monocentric urban configurations. Reg. Sci. Urban Econ. 12 (2), 161–196. Garcia-Lopez, M.-A., Holl, A., Viladecans-Marsal, E., 2015. Suburbanization and highways in Spain when the Romans and the Bourbons still shape its cities. J. Urban Econ. 85, 52–67. Gehring, K., Kaplan, L., Wong, M.H., 2022. China and the World Bank: How contrasting development approaches affect the stability of African states. J. Dev. Econ. 158, 102902. Gibbons, S., Lyytikäinen, T., Overman, H.G., Sanchis-Guarner, R., 2019. New road infrastructure: The effects on firms. J. Urban Econ. 110, 35–50. Goldsmith-Pinkham, P., Sorkin, I., Swift, H., 2020. Bartik instruments: What, when, why, and how. Am. Econ. Rev. 110 (8), 2586–2624. Gollin, D., Jedwab, R., Vollrath, D., 2016. Urbanization with and without industrialization. J. Econ. Growth 21 (1), 35–70. Guillon, M., Mathonnat, J., 2020. What can we learn on Chinese aid allocation motivations from available data? A sectorial analysis of Chinese aid to African countries. China Econ. Rev. 60, 101265. He, G., Xie, Y., Zhang, B., 2020. Expressways, GDP, and the environment: The case of China. J. Dev. Econ. 145, 102485. Heblich, S., Redding, S.J., Sturm, D.M., 2020. The making of the modern Metropolis: Evidence from London. Q. J. Econ. 135 (4), 2059–2133. Henderson, J., Kuncoro, A., 1996. Industrial centralization in Indonesia. World Bank Econ. Rev. 10 (3), 513–540. Henderson, V., Mitra, A., 1996. The new urban landscape: Developers and edge cities. Reg. Sci. Urban Econ. 26 (6), 613–643. Henderson, J., Squires, T., Storeygard, A., Weil, D., 2018. The global distribution of economic activity: Nature, history, and the role of trade. Q. J. Econ. 133, 357–406. Henderson, J., Storeygard, A., Weil, D., 2012. Measuring economic growth from outer space. Am. Econ. Rev. 102, 994–1028. Hernandez, D., 2017. Are ‘‘new’’ donors challenging World Bank conditionality?. World Dev. 96, 529–549. Hodler, R., Raschky, P.A., 2014. Regional favoritism. Q. J. Econ. 129 (2), 995–1033. Hoeffler, A., Sterck, O., 2022. Is Chinese aid different? World Dev. 156, 105908. Horn, S., Reinhart, C.M., Trebesch, C., 2021. China’s overseas lending. J. Int. Econ. 133, 103539. Humphrey, C., Michaelowa, K., 2019. China in Africa: Competition for traditional development finance institutions? World Dev. 120 (C), 15–28. Iacoella, F., Martorano, B., Metzger, L., Sanfilippo, M., 2021. Chinese official finance and political participation in Africa. Eur. Econ. Rev. 136, 103741. Isaksson, A.-S., 2020. Chinese aid and local ethnic identification. Int. Organ. 74 (4), 833–852. Isaksson, A.-S., Kotsadam, A., 2018a. Chinese aid and local corruption. J. Public Econ. 159, 146–159. Isaksson, A.-S., Kotsadam, A., 2018b. Racing to the bottom? Chinese development projects and trade union involvement in Africa. World Dev. 106, 284–298. Jean, N., Burke, M., Xie, M., Davis, M., Lobell, D., Ermon, S., 2016. Combining satellite imagery and machine learning to predict poverty. Science 353, 790–794. Jedwab, R., Storeygard, A., 2022. The average and heterogeneous effects of transportation investments: Evidence from Sub-Saharan Africa 1960–2010. J. Eur. Econom. Assoc. 20 (1), 1–38. KARA and CSUD, 2012. The Social/Community Component of the Analysis of the Thika Highway Improvement Project. Kenya Alliance of Resident Associations (KARA) and the Center for Sustainable Urban Development (CSUD), Accessed at http://csud.ei.columbia.edu/files/2012/11/KARA-report_FINAL.pdf. Krugman, P., 1991. Increasing returns and economic geography. J. Political Econ. 99 (3), 483–499. Lall, S.V., Henderson, J.V., Venables, A.J., 2017. Africa’s Cities: Opening Doors to the World. World Bank, Washington, DC. Lall, S.V., Lebrand, M., 2020. Who wins, who loses? Understanding the spatially differentiated effects of the Belt and Road Initiative. J. Dev. Econ. 146, 102496. Lang, V., 2021. The economics of the democratic deficit: The effect of IMF programs on inequality. Rev. Int. Organ. 16 (3), 599–623. Lessmann, C., Seidel, A., 2017. Regional inequality, convergence, and its determinants – A view from outer space. Eur. Econ. Rev. 92, 110–132. Lujala, P., Ketil Rod, J., Thieme, N., 2007. Fighting over oil: Introducing a new dataset. Confl. Manag. Peace Sci. 24 (3), 239–256. Marchesi, S., Masi, T., Paul, S., 2024. Project aid and firm performance. Econom. Dev. Cult. Chang. URLs: Marchesi: https://doi.org/10.1086/730829, Wellner: https: //doi.org/10.1086/729539, forthcoming. Martorano, B., Metzger, L., Sanfilippo, M., 2020. Chinese development assistance and household welfare in Sub-Saharan Africa. World Dev. 129, 104909. Michalopoulos, S., Papaioannou, E., 2014. National institutions and subnational development in Africa. Q. J. Econ. 129 (1), 151–213. Mills, E.S., 1967. An aggregative model of resource allocation in a metropolitan area. Am. Econ. Rev. 57 (2), 197–210. Muth, R.F., 1969. Cities and Housing: The Spatial Pattern of Urban Residential Land Use. Graduate School of Business, University of Chicago, Chicago, IL. Nelson, A., 2008. Estimated Travel Time to the Nearest City of 50,000 or More People in Year 2000. European Commission, Joint Research Centre (JRC). Nunn, N., Qian, N., 2014. U.S. food aid and civil conflict. Am. Econ. Rev. 104 (6), 1630–1666. Ogawa, H., Fujita, M., 1980. Equilibrium land use patterns in a nonmonocentric city. J. Reg. Sci. 20 (4), 455–475. Olea, J.L.M., Pflueger, C., 2013. A robust test for weak instruments. J. Bus. Econom. Statist. 31 (3), 358–369. Perlez, J., Huang, Y., 2017. Behind China’s $1 trillion plan to shake up the economic order. N. Y. Times. Pesaresi, M., Florczyk, A., Schiavina, M., Melchiorri, M., Maffenini, L., 2019. GHS Settlement Grid, Updated and Refined REGIO Model 2014 in Application to GHSBUILT R2018A and GHS-POP R2019A, Multitemporal (1975–1990-2000–2015) R2019A. European Commission, Joint Research Centre (JRC). Ping, S.-N., Wang, Y.-T., Chang, W.-Y., 2022. The effects of China’s development projects on political accountability. Br. J. Political Sci. 52 (1), 65–84. Journal of Urban Economics 145 (2025) 103730 21 R. Bluhm et al. Puga, D., 1999. The rise and fall of regional inequalities. Eur. Econ. Rev. 43 (2), 303–334. Redding, S.J., Rossi-Hansberg, E., 2017. Quantitative spatial economics. Annu. Rev. Econ. 9 (1), 21–58. Redding, S.J., Turner, M.A., 2015. Transportation costs and the spatial organization of economic activity. In: Handbook of Regional and Urban Economics, vol. 5, Elsevier, pp. 1339–1398. Rossi-Hansberg, E., Sarte, P.-D., Owens III, R., 2009. Firm fragmentation and urban patterns. Internat. Econom. Rev. 50 (1), 143–186. Saiz, A., 2010. The geographic determinants of housing supply. Q. J. Econ. 125 (3), 1253–1296. State Council, 2011. White Paper on China’s Foreign Aid. Xinhua/Information Office of the State Council, People’s Republic of China, Beijing, China. Strandow, D., Findley, M., Nielson, D., Powell, J., 2011. The UCDP-AidData Codebook on Geo-Referencing Foreign Aid. Version 1.1. Uppsala Conflict Data Program, Uppsala University, Uppsala, Sweden. Strange, A.M., Dreher, A., Fuchs, A., Parks, B., Tierney, M.J., 2018. Tracking underreported financial flows: China’s development finance and the aid–conflict nexus revisited. J. Confl. Resolut. 61 (5), 935–963. Strange, A.M., Ghose, S., Russel, B., Cheng, M., Parks, B., 2017. AidData’s Methodology for Tracking Underreported Financial Flows. Version 1.3. AidData at William & Mary, Williamsburg, VA. Swedlund, H.J., 2017. The Development Dance: How Donors and Recipients Negotiate the Delivery of Foreign Aid. Cornell University Press, Ithaca, NY. Wade, A., 2008. Time for the West to Practise What It Preaches. Financial Times, (January 23). Weidmann, N.B., Schutte, S., 2017. Using night lights for the prediction of local wealth. J. Peace Res. 54 (2), 125–140. Wellner, L., Dreher, A., Fuchs, A., Parks, B.C., Strange, A., 2024. Can aid buy foreign public support? Evidence from Chinese development finance. Econom. Dev. Cult. Chang. forthcoming. Xi, J., 2017. Work together to build the silk road economic belt and the 21st century maritime silk road. A keynote speech at the opening ceremony of the belt and road forum (BRF) for international cooperation in Beijing, China. Zárate, R.D., 2022. Spatial Misallocation, Informality, and Transit Improvements: Evidence from Mexico City. Policy Research Working Paper Series 9990, The World Bank, Washington, DC. Zeitz, A., 2021. Emulation or differentiation? China’s development finance and traditional donor aid in developing countries. Rev. Int. Organ. 16, 265–292. Journal of Urban Economics 145 (2025) 103730 22