The World Bank Economic Review, 39(1), 2025, 61–84 https://doi.org10.1093/wber/lhae018 Article Free Trade and Subnational Development: Economic Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 Activity and Human Welfare John Cruzatti C. ABSTRACT This paper delves into the subnational relationship between free trade agreements (FTAs) and human develop- ment worldwide. Utilizing a difference-in-differences and an event-study approach with high-spatial-resolution land-cover data and a comprehensive time series of national-level FTA indicators for 207 countries, the study quantifies the effects of FTAs on subnational development. The findings indicate a small negative impact of FTAs on the Human Development Index but a notable positive impact on economic activity, with urbanized regions benefiting the most. Unequal and more vulnerable regions grapple with declining human development indica- tors. The depth of FTAs does not sway these outcomes. These patterns raise questions about the inclusivity and equitable distribution of the benefits of trade liberalization. While prior literature has examined the national implications of FTAs, this paper provides insight into the subnational repercussions of FTAs and emphasizes the role of inequality in hindering holistic developmental benefits from FTAs. JEL classification: F13, F63, O15, R1 Keywords: FTAs, human development, economic activity, inequality 1. Introduction In an era of unprecedented global interconnectivity, the role of free trade agreements (FTAs) in shaping international trade dynamics takes on heightened significance. The prevailing academic and economic theories propose that FTAs serve as catalysts for economic growth, promising mutual benefits for par- ticipating nations (Amiti and Konings 2007).1 Yet this optimistic narrative is increasingly countered by concerns over worsening inequalities and a growing public disquiet toward these agreements (Astorga John Cruzatti C. is an assistant professor at the International Institute of Social Studies (ISS), Erasmus University Rotterdam, the Netherlands; and an associated researcher at Facultad de Ciencias Sociales y Humanísticas, Escuela Superior Politécnica del Litoral, Km 30.5 Vía Perimetral, Guayaquil, 09-01-5863 Ecuador; email: cruzatticonstantine@iss.nl. I thank the editor and two anonymous referees for their constructive comments. I also thank Axel Dreher, Andreas Fuchs, Matthias Rieger, Maria José Mendoza, Lennart Kaplan, Rajesh Ramachandran, Valentin Lang, Angelika Büdjan, Paul Schaudt, Sarah Langlotz, Leonardo Baccini, Thierry Madies, Neil Foster-McGregor, Johannes Matzat, Leon Barth, 2018 BBQ participants, 2018 IPES participants, 2018 Alliance Paris participants, IPWSD 2019 participants, GIC 2019 participants, PEARL 2019 participants, SCG 2019 participants, and LACEA-LAMES 2019 participants for all the valuable comments. I thank Roland Hodler and Paul Raschky for generously sharing their birth region data for this project. A supplementary online appendix is available with this article at The World Bank Economic Review website. 1 Classical trade theory posits that even when trade liberalization produces losers, compensatory measures can redress such losses, driving the system toward Pareto optimality (Hicks 1939; Kaldor 1939). C The Author(s) 2024. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs licence (https://creativecommons.org/licenses/by- nc- nd/4.0/), which permits non-commercial reproduction and distribution of the work, in any medium, provided the original work is not altered or transformed in any way, and that the work is properly cited. For commercial re-use, please contact journals.permissions@oup.com 62 Cruzatti C. 2010; Furceri et al. 2018).2 Such complexities give rise to a compelling paradox; do FTAs function as genuine enablers of human development, or do they inadvertently exacerbate global disparities?3 This research examines the following question: What are the impacts of free trade agreements on human development, especially at the subnational level? Amid the competing narratives of economic growth and overall welfare, this paper aims to illustrate the multifaceted repercussions of FTAs. Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 To explore this question, the study employs a difference-in-differences and an event-study methodology, combining high-resolution land-cover information with a time series of national-level FTA indicators across 207 countries. This approach circumvents limitations of previous research by comparing FTA impacts on subnational regions characterized by differing and naturally determined land exploitability, and using fixed effects and time-variant local controls. This methodology alleviates the usual concerns of endogeneity, thereby enriching the discourse on the effects of national-level trade policies. The meta-analysis by Stevens et al. (2015) shows that less than five percent of the literature shows a negative effect of FTAs on national trade volumes. Inherently, however, studies at the national level over- look the nuances of subnational transmission mechanisms and localized disparities, necessitating granular, local-level data for a real exploration of these dimensions. While existing studies have touched upon sub- national impacts, these works are often constrained by their focus on specific, non-generalizable settings (Pierce and Schott 2020; Dix-Carneiro and Kovak 2017). Local analyses are not always straightforward given the uneven measurement quality across countries, yet compelling studies have become more at- tainable with the increasing accessibility to remote-sensing data. Perhaps the most relevant example of the compelling use of these data, and closely associated with this study, is the work by Henderson et al. (2018). Their work argues that economic development (proxied by night-light satellite imagery) derives from the interplay of determinants such as trade intensity, geographical traits (e.g., distance to partner, altitude, temperature, relative distance to the coast), and a path-determined human capital that divides the globe between early- and late-developed countries. This paper uncovers the link between FTAs and development. While night-lights serve as a tangible, immediate indicator of economic activity and growth, particularly in agriculturally rich and urbanized re- gions, the Human Development Index (HDI) offers a more comprehensive view, encapsulating the broader socioeconomic, education-, and health-related advancements or detriments in the region. The integration of these metrics is crucial to discerning whether the economic growth spurred by FTAs is commensu- rate with advancements in the broader spectrums of human life, and to understanding the distributional aspects of the benefits derived from FTAs. Raw trends indicate a significant boost in economic development, particularly in urban and agricul- tural regions, as evidenced by increased night-light emissions. However, such benefits do not translate into proportional advancements in human development indices, raising questions about the equitable distribution of FTA-related gains. Subsequent analyses reveal an intricate relationship between economic growth and human development. FTAs exhibit a robust positive impact on economic indicators such as night-light emissions, yet their effect on broader human development remains negligible. Further explo- ration uncovers how regional factors—particularly, inequality levels—mediate and moderate the impact of FTAs on human development. While less industrialized and more unequal regions experience lower 2 Critics argue that FTAs protect specific commercial interests and exacerbate inequality both within and across countries, thereby diminishing their developmental benefits (Diwan and Rodrik 1991; Caliendo et al. 2015; Adão et al. 2022). 3 Public demonstrations against FTAs are widespread and span both developed and developing nations, attesting to ex- isting anti-FTA sentiment (Kriesi et al. 2012; Flesher Fominaya 2014). The widespread mobilizations in Europe against the Transatlantic Trade and Investment Partnership (TTIP) in 2015, and the street demonstrations that have taken place since 2004 in South Korea and China against a trilateral FTA with Japan, are just some examples of an existing anti- FTA attitude in the developed world (Teney, Lacewell, and De Wilde 2014). Mexico and Colombia have often also been explored as case studies for the negative consequences of FTAs (Otero 2011; Salamanca, Gomez, and Landínez 2009). The World Bank Economic Review 63 returns in terms of human development improvement, the economic growth spurred by FTAs remains largely consistent across regions with differing sectors and inequality levels. This research enriches the existing body of literature on FTAs and development by incorporating a global perspective with a focus on subnational dynamics. Unlike previous studies confined to specific national contexts, this paper utilizes global, comparable subnational data to uncover granular, local-level Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 insights. Moreover, the study introduces a critical dimension by examining the role of inequality as a pivotal factor in translating economic prosperity into genuine human welfare.4 A set of robustness tests further validate the study’s findings, encompassing varying sample structures, and geographic, political, and demographic controls. From a policy standpoint, the results underscore the imperative for a holistic approach to trade agreements—one that considers economic growth and broader dimensions of human development. My study suggests that equitable distribution of the benefits of eco- nomic activity should be a paramount concern in crafting trade policies that genuinely enhance human development. However, the study acknowledges potential limitations, including the inherent complexities and interactions within FTAs, as well as possible data constraints.5 The paper is structured as follows: the next section details the empirical strategy, econometric specifi- cation, and data sources. Subsequently, the results’ section presents key and complementary analyses. The last section, summarizes the findings and outlines implications for both research and policy. 2. Empirical Strategy and Data This study employs a difference-in-differences (DID) and an event-study methodology to examine the effect of free trade agreements on subnational development. The empirical strategies follow the main rationale of a standard DID that exploits both cross-sectional and time variations.6 Specifically, I utilize local, naturally determined, land-cover characteristics to differentiate between treated and control groups (cross-sectional variation) and use the FTA status of each country to demarcate pre- and post-treatment periods (time variation). The land-cover data are sourced from a high-resolution (30 m) global data set by ESA (2017), which delineates the predominant type of land—e.g., cropland, urban, or bare land— covering subnational regions from 1992 to 2015. Trade agreement data, which detail the FTA status of up to 207 countries from 1990–2015, are obtained from Dür, Baccini, and Elsig (2014). The identifying assumption rests on the parallel trends assumption; conditional on relevant controls, the evolution of development outcomes in subnational areas with and without naturally determined ex- ploitable land is presumed to be the same in the absence of FTAs. Given the model and control variables employed, this strategy might render the effect of FTAs on local development as conditionally exogenous.7 The baseline sample comprises countries participating in FTAs over the past three decades. Within these countries, I create a subnational grid with divisions measuring 111 km by 111 km to analyze the local developmental impact of FTAs. The study includes 749 FTAs and covers 19,033 unique grid cells, representing 2,078 provinces/states across 207 countries during the 1990–2015 time frame. 4 The literature connecting trade policy and development is vast. At the subnational level particularly, some of these works have already focused on development variables like inequality or poverty (Topalova 2010; McCaig 2011; Kovak 2013). Similarly, others have explored such relationships with a concrete interest in labor, education, and health out- comes (Edmonds and Pavcnik 2005; Edmonds, Topalova, and Pavcnik 2009; Edmonds, Pavcnik, and Topalova 2010; Bombardini and Li 2020). As for a joint discussion on economic growth, inequality, and human welfare, one can refer to, for instance, the work of Artuc, Porto, and Rijkers (2019) or Cingano (2014). 5 Particularly those related to causal inference. 6 While similar to a standard DID, my DID and (in general) staggered event studies take advantage of multiple activation periods specific to each unit of analysis, thereby enriching the temporal dimension of the treatment effect. 7 The high spatial resolution of the data set allows for enhanced statistical robustness and the incorporation of finer- grained fixed effects, thereby mitigating potential omitted variable bias. 64 Cruzatti C. The main empirical specification is formulated as Developmenti,t = β1 Post j,t × Treati + β2 Zi,t −1 + β3 η j,t + β4 γi + i,t . (1) In this equation, Developmenti, t represents the average development level in subnational region i at time t. The binary variable Postj, t indicates whether the time period is pre- or post-FTA activation in country j, Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 while Treati identifies regions with exploitable land. The variable Zi, t − 1 includes various regional polit- ical and economic controls, detailed subsequently. The fixed effects ηj, t and γ i capture country-year and regional variations, respectively. To account for potential spatial and temporal autocorrelation in the error terms, standard errors i, t are clustered at the regional and country-year levels. The interaction term Postj, t × Treati aims to isolate the effect of FTAs on subnational development, purged of other time-variant or region-specific factors. In the specified model, both Postj, t and Treati are not separately included as they are inherently absorbed by the fixed effects ηj, t and γ i , respectively. The variable Developmenti, t is assessed in several ways. First, I evaluate the effect of FTAs on the sub- national Human Development Index (HDI) (Kummu, Taka, and Guillaume 2018). The HDI portrays the degree of overall accomplishment in fundamental development dimensions considered by the United Na- tions: health, education, and income (UNDP 2017). These dimensions are measured by jointly assessing the life expectancy at birth, the years of schooling, and the gross national income per capita of the regions under study. The index ranges from 0 to 1, with 1 representing the highest level of development and 0 sig- nifying the lowest.8 The subnational HDI includes global data between 1990 and 2015, with a (roughly) 10 by 10 km spatial resolution.9 Second, this study also looks into proxies of each development dimension within the HDI—given data of the actual variables used for the construction of HDI are not available at the spatial and time dimensions used throughout this study—namely, night-lights (economic dimension) in the results’ section, and infant mortality (health dimension) and mothers’ education years (education dimension) in supplementary online appendix S2. Nighttime light emissions (night-lights) are among the most standardized proxies for economic activity. Apart from their panel and global nature, which adds to comparability, they reduce the recurrent measurement error in local data production expected in de- veloping regions of the world.10 The night-light data come from satellite imagery generated by the Earth Observation Group, part of the National Oceanic and Atmospheric Administration of the United States (NOAA 2015). The dataset covers the whole world from 1992-2013 period and has a spatial resolution of 1 by 1 km. Similarly, infant mortality and mothers’ education years data are plausible proxies of life expectancy and years of schooling (Yang et al. 2010; Kong et al. 2015). Moreover, such variables have comparable geo-referenced data around the world. They come from the work produced by the USAID on their Demographic and Health Surveys (DHS) (ICF 2017). This DHS data compile precise geo-referenced information—slightly displaced to protect the anonymity of the observed units (see ICF 2013 for more 8 For more detail on the construction of the HDI, see the technical notes of UNDP (2016). 9 For more technical detail on the subnational HDI, see Kummu, Taka, and Guillaume (2018). For a visual representation, see fig. 1. 10 The works by Sutton and Costanza (2002) and Sutton, Elvidge, and Tilottama (2007) canonized the use of remote sensing data by using night-light emissions to proxy levels of economic activity at the local level. Henderson, Storeygard, and Weil (2012) and Jean et al. (2016) furthered the use of remote sensing data in development studies by proposing the prediction of rates of growth and poverty via the use of geographically detailed data, e.g., altitude, temperature, geo- location. Upon these works, the literature has expanded with new ways of assessing satellite-imagery quality (Chen and Nordhaus 2011; Chen 2016; Mellander et al. 2015) and associating it with development. Night-lights have now not only been shown to be correlated to economic activity but also to figures of wealth, health, and education (Noor et al. 2008; Weidmann and Schutte 2017; Bruederle and Hodler 2018). For a visual representation, see fig. 2. The World Bank Economic Review 65 Figure 1. Gridded Human Development Index over Time. Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 Source: Author’s analysis. Note: The dimensions of grids are roughly 111 km by 111 km. The grids are clipped to land. Red tones mark the least developed regions, blue tones the most developed ones. details)—on the education and health variables mentioned for almost 60 countries between 1981 and 2015.11 The data on FTAs come from the work of Dür, Baccini, and Elsig (2014), who construct country-level indicators for the depth or conditions added to 1,002 FTAs since 1948.12 The depth value ranges between zero and seven, and it is an additive indicator of the provisions that a particular FTA includes—see table 11 The DHS data are not as comprehensive as the sample used for HDI, and night-lights—which englobe almost the entire world—and therefore, my analysis on them is somewhat limited. Moreover, the data set is an unbalanced panel as surveys take place (mostly) every five years. The results on these indicators then should be taken with a grain of salt. 12 I consider “accessions” as separate FTAs while they add a new country to the deal. 66 Cruzatti C. Figure 2. Gridded Night-Lights over Time. Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 Source: Author’s analysis. Note: The dimensions of grids are roughly 111 km by 111 km. The grids are clipped to land. Darker tones denote less night-lights, brighter tones more. S3.1 for more detail.13 Thus, for Postj, t , if the average FTAs’ depth includes substantial provisions on—at least—tariffs and quotas for the exchange of goods (i.e., FTA depth ≥ 1), then for all years since t, Postj, t will be set to 1, and 0 otherwise. Land exploitability data stem from the remote-sensing repository of the European Space Agency (ESA) and the Climate Change Institute (ESA 2017). The data classify the predominant land cover of subnational regions worldwide from 1992 to 2015 using the Land Cover Classification System (LCCS) designed by the 13 As an initial illustration, while FTAs with a depth ≥ 1 refer to FTAs with almost no barriers on tariffs and quotas for the exchange of goods, the FTAs with a depth ≥ 2 describe agreements that, on top of eliminating almost all barriers on tariffs and quotas, include the elimination of most constraints on the exchange of services. The same logic follows for deeper FTAs. For a visual representation, see fig. 3. The World Bank Economic Review 67 Figure 3. Free Trade Agreements Depth over Time. Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 Source: Author’s analysis. Note: National variation of FTAs’ depth. Darker tones denote deeper FTAs signed on average, brighter tones denote shallower FTAs. Food and Agriculture Organisation (FAO).14 The spatial resolution of the data is mostly of 30 meters.15 I define exploitable regions (Treati = 1) as those areas that predominantly have cropland, urban land, other forms of natural vegetation, or consolidated bare land. Conversely, I define as non-exploitable regions 14 The LCCS is thoroughly detailed in table S3.2 of the supplementary online appendix. 15 To define the predominant category of LC in region i, I use the mode. For instance, as detailed in table S3.2 and visually operationalized in fig. 4, if the mode LC value of region i is 190 (urban), I define urban land as the predominant LC of region i. Similarly, if region i’s mode LC value is between 20 and 30 (more than 50 percent cropland), I define agricultural land as the predominant LC in the region. 68 Cruzatti C. Figure 4. Subnational Land-Cover Categories in the World. Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 Source: Author’s analysis. Note: The figure shows the world’s subnational land divided into grids, and into four cover types: non-exploitable land, agricultural, urban, and other. The dimensions of grids are roughly 111 km by 111 km. Figure 5. Subnational Land-Cover Categories of Northern Utah. Source: Author’s analysis. Note: The figure shows the northeast of the state of Utah, USA. The blue-shaded areas correspond to non-exploitable lands, and the gray to exploitable ones. (Treati = 0) the ones with mostly unconsolidated bare land, water, or permanent snow and ice covering them.16 Figure 5 shows an example of how I operationalize the land classification. In this figure, around 30 percent of the subnational regions in northeast Utah are marked as “non-exploitable.” These regions are mainly covered by water, frozen, or bare unconsolidated land, which makes it very difficult or impossible for productive activities or factories to operate there.17 The regions that are “exploitable” in northeast Utah have mostly shrubland and trees, which provide more opportunities for new ventures to emerge and 16 For a visual overview of the variation of the main variables exploited (HDI, night-lights, FTAs, land cover) see figs 1 to 4. For more detail about the dummies used throughout, see tables S3.1 and S3.3 of supplementary online appendix S3. 17 I acknowledge that fishing industries can settle in areas with a lot of water, so I run robustness tests in table 4 where I code regions with mostly water bodies as “exploitable.” The results are qualitatively the same. The World Bank Economic Review 69 grow. I argue that the non-exploitable regions are a plausible and relevant control group for the local effects of FTAs, as the activities that trade liberalization would encourage will likely concentrate in areas with favorable geographic characteristics—as suggested by Henderson et al. (2018)—given their relatively more advantageous conditions to produce, e.g., farmable soil, better access, better climate, and greater social visibility. These conditions are unlikely to vary over time, thus the non-exploitable regions would Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 remain so even in the presence of wide economic shocks like FTAs. Notwithstanding, as I will show later, I also create subcategories of exploitable regions based on the type of land they predominantly have. This differentiation stems from the anticipation that, given their degree of exploitability, areas might be increasingly susceptible to trade agreements—for instance, I anticipate urban regions to be the most susceptible. The vector of covariates Zi, t − 1 mainly includes a control for linear trends at the individual level (i.e., regional level), and three political economy and geographic controls: temperature, leaders’ birthplaces, and World Bank aid. Being under a particular FTA is arguably correlated with factors that affect development differently in regions with high or low exploitability. For instance, the degree of exploitability could correlate with geographic time-varying patterns such as changes in the local mean temperature. This would imply that any development differences between exploitable and non-exploitable regions could result from year-to-year changes in the quality of the land rather than from the inherent, time-invariant natural endowments of exploitable land that a region could have. Indeed, temperature has been shown as an overall good proxy of productivity (Pregitzer and King 2005; Nishar 2017). To account for this time- changing quality of the land, I use the data on local temperatures from the PRIO-GRID vector grid as the mean degrees Celsius within region i in any given year t (Tollefsen, Strand, and Buhaug 2012). I also use two political-economy controls, namely the birth region information of leaders of the executive (Hodler and Raschky 2014) and aid disbursements by the World Bank (AidData 2017).18 The birth region variable arguably captures the role of political favoritism in a region’s development. Hodler and Raschky (2014) show that leaders seem to favor their birth regions as suggested by higher night-light emissions. Following Hodler and Raschky’s rationale, I expect significant results on development a year after the leader took office. I thus construct a dummy indicating whether the leader of a country j is in office by year t − 1 and was born in region i. Similarly, Dreher et al. (2019) show that one of the channels of favoritism is aid. Moreover, Cruzatti C., Dreher, and Matzat (2023) show a relevant impact of aid on health indicators, a key dimension of development. In this study I use the geo-referenced data constructed by AidData (2017) on World Bank aid from 1995–2015 and calculate the region’s yearly mean World Bank aid disbursements in constant 2014 USD. I only use projects that have coordinates with exact location information, with a maximum of 25 km noise, or refer to the country’s second-order administrative division—depending on the country, this would be referring to province or state.19 The inclusion of country-year (ηj, t ) and regional (γ i ) fixed effects (FE) as controls serves the purpose of addressing latent factors inherent to national and regional dynamics that influence subnational develop- ment. An illustrative instance lies in the context of free trade agreements, where initiating such agreements typically hinges upon a nation’s pre-existing economic and political strengths or weaknesses.20 This ren- ders FTAs intrinsically tied to a subnational region’s development, thus challenging their exogeneity. Traits like the economic or the political, however, while varying over time are not so transitional as to change re- peatedly within a year. Herein, the country-year FE plausibly captures nation-year-specific characteristics that could impact the development of all subnational regions of a particular country in any given year—an exemplar among these attributes being the very introduction of an FTA. Yet the realm of unobservable influences extends beyond national boundaries, with structural determinants of development residing at 18 Hodler and Raschky directly provided the data on leaders’ birth regions. 19 Tables S3.4 and S3.5 in the supplementary online appendix show the study variables’ sources, definitions, and descriptive statistics. 20 Even supranational characteristics—such as belonging to blocs like the European Union. 70 Cruzatti C. Figure 6. Trends of Human and Economic Development. Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 Source: Author’s analysis. Note: The figures show the subnational trends for the Human Development Index (HDI) and night-lights 10 years before and 10 after the introduction of free trade agreements. Figures (a) and (b) show two categories: exploitable land (in black), and non-exploitable land (in gray). the subnational level presenting particular concern. In the context of national-level models of trade, for instance, a cornerstone principle stipulates that international exchange is directly proportional to size and inversely proportional to distance—an axiom widely known as the “gravity model” (see Rauch 2016). These models argue that the geographic distance between regions and their sizes are relevant explanatory variables of trade. However, note that the distance between regions and their sizes vary across regions but not over time. Consequently, the inclusion of regional fixed effects within equation (1) effectively encompasses these time-invariant subnational unobservables, alongside others.21 3. Results The plots presented in fig. 6 display the subnational, raw trends of HDI and night-lights. These visual- izations shed light on the evolution of both human (HDI) and economic development (night-lights) in the periods leading up to and following the activation of FTAs. Graphs (a) and (b) also compare the trends for exploitable and non-exploitable regions. Within these representations, it is evident that the critical parallel-trends assumption foundational to a DID analysis remains largely consistent; particu- larly, the trends between regions dominated by non-exploitable land and those characterized by primarily exploitable land maintain their parallelism from t − 10 until t = 0 (Christian and Barrett 2017; Roth et al. 2022). The graphs show that exploitable regions tend to have higher levels of human and economic development than non-exploitable regions—even though this pattern is subtler for human development in more distant pre-periods. These graphs, however, only display raw trends and do not control for any relevant covariates as in equation (1). 3.1. DID Table 1 shows the results for the impact of FTAs on local development, namely on the Human Devel- opment Index and night-lights. Columns 1 to 6 report the estimates for equation (1) and include fixed effects, linear trends, and other controls progressively for each development variable. In columns 1 to 3, I report the results for HDI, while in columns 4 to 6, I show the results for night-lights. The preferred 21 Culture, religion, or ethnicity are some of the other plausibly time-invariant subnational characteristics controlled for with the regional fixed effects. The World Bank Economic Review 71 Table 1. FTAs: Average Effects on Exploitable Regions (1) (2) (3) (4) (5) (6) HDI HDI HDI Night- Night- Night- lights lights lights Treati × Postjt −0.016 −0.070∗ 0.723∗∗∗ 0.712∗∗∗ 0.656∗∗∗ Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 0.003 (0.033) (0.032) (0.039) (0.130) (0.126) (0.143) Observations 449,786 449,786 238,453 389,968 389,968 209,911 Adjusted R-squared 0.997 0.997 0.998 0.965 0.965 0.970 Region FE Yes Yes Yes Yes Yes Yes Country-year FE Yes Yes Yes Yes Yes Yes Linear trends No Yes Yes No Yes Yes Other controls No No Yes No No Yes Mean of dependent variable 70.57 70.57 72.45 1.63 1.63 1.62 Countries 190 190 175 200 200 176 Regions 18,375 18,375 16,719 18,046 18,046 16,392 Source: Author’s analysis. Note: All HDI values are scaled (HDI×100). Standard errors are clustered at the country-year and regional levels and are shown in parentheses. Significance levels: ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Figure 7. Free Trade Agreements’ Effects on Development: Country Level. Source: Author’s analysis. Note: The figures on the Human Development Index are scaled (HDI×100). Night-lights are logged. Infant mortality is computed as the number of children that died before 12 months old per every 1,000 live births. The gray lines represent the upper and lower bounds of the 95 percent confidence interval. specifications correspond to columns 3 and 6, which include the full set of controls.22 The columns re- port a consistent pattern for HDI and night-lights. Even when accounting for time-variant national and subnational characteristics, the effects of FTAs on HDI are negative yet minor, whereas such effects on 22 The discrepancy between the number of observations in the specifications with and without all control variables mostly stems from the missing values of the temperature variable. 72 Cruzatti C. night-lights are positive and considerable—statistically and in size.23 Columns 1 to 3 show a mostly neg- ative yet always close to zero effect for HDI—0.07 percentage points at 10 percent conventional level, being the most significant in size—while the effect shown in the most stringent specification of night-light emissions (column 6) reports, relative to its standard deviation, a 1 percent level statistically significant 15.5 percent increase.24 Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 The modest negative effect of FTAs on human development might be explained by an inequality nar- rative.25 Resources could be concentrated in specific sectors or spaces within countries. As argued by Cingano (2014) and Artuc, Porto, and Rijkers (2019), while globalization has brought about clear eco- nomic progress in sectors such as trade and technology, especially for more urbanized regions, it has neglected other less developed sectors such as the ones present in rural regions, i.e., agriculture.26 Sim- ilarly, Balland et al. (2020), Auer (2015), Hidalgo (2015), Van den Berg (2012), and Hausmann and Hidalgo (2010) show how free trade can be particularly beneficial for high-skilled regions that specialize in the provision of services and predominantly operate in highly developed cities. Even if FTAs indeed bring more overall economic activity to a region, such benefits might be concentrated in a sectoral or geographic minority, hindering the potential positive spillovers of such economic development elsewhere. Thus, the existing levels of inequality within the regions may very well play a role in the negligible (or even negative) impact of FTAs on human development.27 In table 2, I test the inequality mechanism. Columns 1 and 2 examine the direct effects of FTAs on local inequality, specifically in relation to night-light inequality measures. I construct such an inequality indi- cator by following Elvidge et al. (2012).28 As highlighted by Salvati et al. (2017), however, the inequality indicator can be less accurate in areas with low population density. In response, column 2 introduces a heterogeneity test where I differentiate between regions with larger and smaller populations. As seen, the 23 Note that I use the figures of night-lights at the level. To account for the skewness of the night-light distribution— i.e., most grids having night-light values close to zero—I also compute the most stringent regression of column 6 with the inverse hyperbolic sine function of night-lights as a dependent variable. The result, while smoother (12.7 percent increase), is qualitatively comparable and is available upon request. 24 For context, the average inter-annual change of HDI at the country level between 1990 and 2015 was 0.0051 UNDP (2017). The effect would represent approximately 13.7 percent of the historical yearly change average. 25 Another straightforward explanation could stem from the fact that the HDI is a composite of three distinct dimensions of development (health, education, income), and the effects of FTAs on such dimensions might be heterogeneous, as one can already see in the post-FTA national trends depicted in fig. 7. Thus, even if the effects on local economic activity seem to be consistently positive, the negative or null effects on the dimensions of health and education might be dominating the economic ones, explaining why the overall effects on the composite HDI are slightly negative. I explore this potential mechanism in supplementary online appendix S2 using DHS data. These data come from mostly developing countries and are not a panel, which makes them structurally different from the main sample used so far, so the results should be taken with a grain of salt. Results, while showing no statistically significant impact of FTAs on the measures of health and education overall, do portray how more unequal regions are the ones solely experiencing negative effects on their health dimension. This then, might very well explain the conflicting HDI and night-light results explored in table 1. 26 Similarly, Otero (2011), and Wise (2009, 2014) argue how the signing of an FTA can heavily influence the agricultural industry of developing and developed countries. Both Wise and Otero show how NAFTA negatively affected the wheat and grain production in Mexico to the benefit of their counterparts in the United States. 27 There is extensive literature on inequality and its relationship to development. Perhaps one of the most recent relevant works is the one by Aiyar and Cummins (2021) who link the lingering effects of income/economic activity distribution with indicators of human development like health, and show how parents can correct an otherwise negative health trend in their children via parental investments. These parental investments can only be afforded by a minority with the economic means to do it. Moreover, recognizing the myriad of connections made in the literature, the UN has even developed an HDI measure that accounts for income/consumption inequality: IHDI. The measure adjusts each dimension (health, education, income) for the level of existing inequality. For more detail, see UNDP (2016). 28 The inequality measure uses the Lorenz curve principle to plot the cumulative distribution of night-lights against the cumulative distribution of population density. See more details in supplementary online appendix S1. The World Bank Economic Review 73 Table 2. Mechanism: Inequality Local effect Country split Local split (1) (2) (3) (4) (5) (6) Inequality: Inequality: HDI Night- HDI Night- Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 Full sample Population split lights lights Treati × Postjt 0.703 1.937 – – – – (1.205) (1.748) Treati × Postjt × BigRegionit – −2.253 – – – – (1.646) Treati × Postjt – – −0.141∗∗∗ 0.838∗∗∗ 0.055 0.690∗∗∗ (0.048) (0.229) (0.052) (0.167) Treati × Postjt × UnequalCountryjt − 1 – – −0.071 0.036 – – (0.084) (0.054) Treati × Postjt × UnequalRegionit − 1 – – – – −0.242∗∗∗ −0.095 (0.076) (0.106) Observations 197,855 197,855 125,669 108,349 202,880 194,729 Adjusted R-squared 0.693 0.693 0.997 0.967 0.997 0.970 Mean of dependent variable 64.54 64.54 75.46 1.58 71.64 1.64 Countries 173 173 130 123 173 172 Regions 16,111 16,111 14,705 14,116 16,179 15993 Source: Author’s analysis. Note: All HDI values are scaled (HDI×100). All columns include linear trends, country-year and regional fixed effects, and all controls. Standard errors are clustered at the country-year and regional levels and are shown in parentheses. Significance levels: ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. impact of FTAs on local inequality is both statistically and substantively negligible. Subsequently, columns 3 to 6 explore the influence of FTAs across varying dimensions of inequality.29 Namely, columns 3 to 6 assess the role of existing national (columns 3 and 4) and local inequality (columns 5 and 6) in elucidating the impacts of FTAs.30 The findings are consistent with the prevailing literature and (this paper’s) empirical expectations. While FTAs exhibit little to no influence on local inequality, they present minor negative effects on HDI and pronounced positive effects on night-lights when impacts are observed. Yet nuances in the inequality analysis deserve mention.31 Specifically, columns 3 and 4 suggest that national inequality does not capture the varied impacts of FTAs on HDI and night-light emissions, as evidenced by the lack of statistical differences between the terms Treati × Postjt and Treati × Postjt × UnequalCountryjt − 1 . Meanwhile, columns 5 and 6 offer a deeper dive. In regions marked by significant inequality, symbolized by Treati × Postjt × UnequalRegionit − 1 there is a discernible drop in HDI. Yet when examining night-light emissions, the disparity between more (Treati × Postjt ) and less egalitarian regions is not statistically significant. Overall, table 2 highlights that regional inequalities play a significant role in shaping the effects of FTAs on economic and human development. While FTAs consistently boost economic development, the pre-existing inequalities cause these benefits to be concentrated among a select few. As a result, despite general economic growth, the broader influence of FTAs on human development remains limited. 29 I also refine the assessment of the effects of FTAs on local inequality in table S1.1. All results are qualitatively comparable. 30 As seen in figure S.2, there is suggestive evidence of high heterogeneity of inequality levels around the world. In columns 3 and 4, I divide nations between those below and above a time-variant world median of inequality using the national income Gini index. In columns 5 and 6, I divide regions between those below and above a time-variant local inequality using the night-lights Gini index explained in supplementary online appendix S1. 31 The total number of observations varies across the explored specifications due to data availability on inequality. Tests constrained to the smallest set of non-missing data across all specifications confirm that results are qualitatively consis- tent. These findings are available upon request. 74 Cruzatti C. Figure 8. Stages of Development by 1950. Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 Source: Author’s analysis. Note: The overall distinction evaluates three dimensions: education, Gross Domestic Product per capita (GDPpc), and urbanization. Late-developed countries do not pass any of the cutoffs proposed by Henderson et al. (2018) for each dimension by 1950. Early-developed countries surpass at least one cutoff of any of the three indicators in the same period. Table 3 examines the effects of FTAs on both human and economic development for different groups. In columns 1 and 2, I look into the impact heterogeneity across different economic sectors as they break down the exploitable regions into agricultural, urban, and other.32 Columns 1 and 2, as expected by the exploitability assumption explained in the empirical strategy and data section, portray how the positive and larger effects concentrate in the more exploitable regions, i.e., in the agricultural and urban areas. The estimates also reveal that urban-associated productive regions perform better than other exploitable regions. These results align with the above-mentioned regional studies showing structural development differences in favor of urbanized areas. Indeed, skill-intensive regions like the urbanized benefit the most from trade liberalization processes. Columns 3 and 4 differentiate between FTAs signed with Global South partners (PostSouthi, t ) and those linked with Global North counterparts (PostNorthi, t ). This categorization aligns with the early- and late-developer framework by Henderson et al. (2018), depicted in fig. 8. Several scholars, such as Diwan and Rodrik (1991), Marchetti and Mavroidis (2011), and Sell (2011), suggest that trade agree- ments frequently favor developed nations over their developing counterparts. This perspective is evident in the coefficients presented in column 3 for PostSouthi, t and PostNorthi, t . Notably, column 1 reveals that while FTAs with northern countries slightly depress HDI, southern FTAs do not. Further insights are drawn from columns 5 and 6. Here, the analysis pivots from trade partners to the country of primary study. Relying on the most recent (t − 1) national GDP figures, I label countries as either big (PostBigGDPi, t ) if their GDP figures are above the global median, or small (PostSmallGDPi, t ) if they are under. Cumula- tively, the evidence from columns 3 to 6 strengthens the claim that FTAs seem to primarily favor global 32 The sectoral differentiation is based on the prevailing land type within the analyzed subnational area: agricultural, i.e., cropland as the predominant land cover, urban, i.e., urban land as the predominant land cover, and other, i.e., natural vegetation or consolidated bare land as the predominant land covers. Table 3. Heterogeneity: Sectors, Partners, and Depth Sectors Partner type FTAs’ depth (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) HDI Night- HDI Night- HDI Night- HDI Night- HDI Night- lights lights lights lights lights TreatAgrici × Postjt 0.061 1.450∗∗∗ – – – – – – – – (0.065) (0.352) TreatUrbani × Postjt −0.514 3.254 – – – – – – – – The World Bank Economic Review (0.340) (2.163) TreatOtheri × Postjt −0.106∗∗∗ 0.414∗∗∗ – – – – – – – – (0.040) (0.107) Treati × PostSouthjt – – 0.060 0.251 – – – – – – (0.041) (0.163) Treati × PostNorthjt – – −0.122∗∗ 0.491∗∗ – – – – – – (0.056) (0.198) Treati × Post × SmallGDPjt – – – – −0.345∗ 0.582∗∗∗ – – – – (0.179) (0.185) Treati × Post × BigGDPjt – – – – 0.302 0.074 – – – – (0.192) (0.116) Treati × PostClassicFTAjt – – – – – – −0.062 0.428∗ – – (0.070) (0.230) Treati × PostDeepFTAjt – – – – – – −0.009 0.285 – – (0.078) (0.224) Treati × NumClassicFTAjt – – – – – – – – −0.001 −0.00005 (0.001) (0.0003) Treati × NumDeepFTAjt – – – – – – – – 0.002 0.004 (0.003) (0.012) Observations 238,453 209,911 238,453 209,911 238,453 209,911 238,453 209,911 101,359 91,379 Adjusted R-squared 0.998 0.970 0.998 0.970 0.998 0.970 0.998 0.970 0.997 0.973 Mean of dependent variable 72.45 1.62 72.45 1.62 72.45 1.62 72.45 1.62 73.46 2.04 Countries 175 176 175 176 175 176 175 176 161 161 Regions 16,719 16,392 16,719 16,392 16,719 16,392 16,719 16,392 15,841 15,035 Diff (p-val) 0.018 0.001 0.039 0.466 0.079 0.063 0.708 0.737 0.387 0.744 Source: Author’s analysis. Note: All Human Development Index values are scaled (HDI×100). All columns include linear trends, country-year and regional fixed effects, and all controls. Standard errors are clustered at the country-year and regional levels and are shown in parentheses. “Diff” tests if coefficients are equal. Significance levels: ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. 75 Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 76 Cruzatti C. Figure 9. Free Trade Agreements’ Depth Evolution. Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 Source: Author’s analysis. Note: The graph shows the average depth of free trade agreements (FTAs) between 1990 and 2015. The definition of FTAs’ depth corresponds to the work of Dür et al. (2014). north/bigger countries, often at the expense of global south/smaller nations. Yet this disparity may not solely be attributed to the development characteristics of countries. Major players, such as the United States and the European Union, have a track record of protecting critical sectors in their trade agreements.33 One could posit that these developed nations have an upper hand when it comes to negotiations, sculpting regulations, and ensuring adherence. For instance, criti- cisms abound over how patent and IPP regulations in FTAs have made essentials, such as medicines, pro- hibitively expensive in the Global South.34 Thus, while FTAs’ depth should not, in principle, discriminate between partner types, the benefits accrued by developed countries at the cost of those in the developing world could indeed be instrumentalized via deeper FTA provisions (Rodrik 2018; Sakyi et al. 2017).35 Dür, Baccini, and Elsig (2014) argue that over the past three decades, free trade agreements have evolved into intricate constructs, incorporating a rising number of provisions that go beyond the traditional tar- iff and quota stipulations of earlier FTAs, as evident in fig. 9. Some of these supplementary provisions contain critical criteria related to services exchange and investments, while others center on establishing shared regulation and law enforcement in sensitive domains such as competition rules, product standards, and intellectual property rights. According to Limão (2016), by 2011, a substantial 76 percent of exist- ing preferential trade agreements were subject to at least one aspect of investment standardization, 61 percent included intellectual property rights protection, and 46 percent mandated adherence to environ- mental regulations. Therefore, as Rodrik (2018) argues, a plausible explanation for the poor impact of FTAs on human development might lie in the depth of the FTA a country enters into, rather than in the specific country or country partners’ characteristics. The effectiveness of an FTA could be determined by the extent of conditionality or comprehensiveness outlined in such agreements. 33 A notable example is agriculture (Wise 2009, 2014; Otero 2011; Grochowska and Ambroziak 2018; Grennes 2018; Kareem et al. 2018). 34 This perspective finds resonance with the core-periphery literature that underscores the unequal global distribution of benefits and detriments from globalization (Hirst 1997; Wallerstein 1976, 2005). 35 See, for example, statements by World Health Organization Director-General Dr. Margaret Chan criticizing the Trans- Pacific Partnership (Germanos 2015). The World Bank Economic Review 77 In columns 7 and 8, I separate FTAs that include classic FTA provisions such as tariffs and quotas (PostClassicFTA) from those that involve conditions beyond such aspects (PostDeepFTA). The results do not show a differential impact for the types of FTAs, with no statistical effect on HDI and a slightly significant (at the 10 percent conventional level) positive effect on night-lights. In columns 9 and 10, I run an extension test that uses the total number of each type of FTA signed—instead of dummies. Results Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 reveal that the number of FTAs—of either type—does not explain the variation in my sample, as the coefficients of interest are both statistically insignificant. Overall, the results in columns 7 to 10 portray the lack of empirical support for the hypothesis that FTAs’ increasing depth/conditionality plays a critical role in the FTAs’ impacts on human and economic development at the subnational level. 3.2. Robustness This work addresses endogeneity in various forms, most notably through the natural selection of treated (exploitable regions) and control units (non-exploitable regions) as previously discussed. If the demar- cation between treated and control groups were artificially constructed rather than natural, we would expect to observe systematic man-made alterations in the regions’ primary land cover over time. Yet data suggest otherwise. Out of 19,033 regions in the sample, only 4.10 percent (780 regions) transitioned from one land category to another at any point during the study period. Even more telling, 99.23 percent (774 out of 780) of these regions reverted to their most frequently observed land category shortly thereafter. Based on these facts, the likelihood of artificial, non-random determination of the treated and control groups seems minuscule. I nevertheless conduct tests that allow for alternative categorization of the treated and control units. These tests are detailed in panel A of table 4. Initially, to address concerns that regions undergoing land changes might obscure certain localized, time-specific phenomena, I have omitted regions from the sam- ple that witnessed any predominant land alterations during the studied duration. Further, the regression sample is narrowed to only encapsulate countries with at least one non-exploitable region. This facilitates direct comparisons with exploitable regions within the same nation. Similarly, countries where the dif- ferentiation between exploitable and non-exploitable regions is ambiguous are omitted from the sample. This ambiguity is particularly evident in nations that conduct their oil exploitation activities in desert regions, such as those located in the Arabian desert.36 The outcomes from the three modifications reveal no significant deviations from the principal findings presented in table 1. Moreover, it is pertinent to recognize the potential for fishing industries to gravitate toward water-rich areas. As such, in row 4 of panel A, regions predominantly characterized by water bodies are designated as “exploitable.” An additional concern might arise from potential treatment contamination on the “non- treated” regions. At the macroscale, regions in non-FTA nations that share borders with FTA countries might be indirectly impacted by the trade spillovers emanating from their FTA neighbors, as postulated by Khan (2020). Extending this logic, one could hypothesize that similar spillover effects manifest at the micro level, where non-exploitable regions adjacent to exploitable ones are indirectly influenced. Such spillovers could result in the treatment contamination of the control group.37 However, as demonstrated in rows 4 and 5, when potential contamination is accounted for—either by excluding non-exploitable regions neighboring exploitable ones, or by excluding non-exploitable regions with populations comparable to exploitable regions—the primary results remain consistent. Additional tests were conducted to explore alternative configurations of the primary model (1). These variations involved either the introduction of new control variables or modifications to the primary FTA dummy variable and are systematically detailed in panel B of table 4. To begin with, a population variable 36 The excluded countries encompass Saudi Arabia, Iraq, Jordan, Qatar, the United Arab Emirates (UAE), Oman, and Yemen. 37 More than likely this scenario would lead to an underestimation of the estimator, considering that trade shocks (from FTAs) would impact both exploitable and non-exploitable regions in a similar manner. 78 Cruzatti C. Table 4. Robustness: Other (1) (2) HDI Night-lights Panel A − 0.089∗∗ 0.627∗∗∗ Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 No changing regions (0.040) (0.138) Non-exploitable countries − 0.070∗ 0.656∗∗∗ (0.039) (0.144) Non-desert-exploitable − 0.070∗ 0.656∗∗∗ (0.039) (0.143) Water regions − 0.079∗∗ 0.656∗∗∗ (0.039) (0.143) Neighbors − 0.152∗∗∗ 0.700∗∗∗ (0.042) (0.139) Similar populations − 0.038 0.237∗∗∗ (0.030) (0.084) Panel B Population − 0.071∗ 0.651∗∗∗ (0.040) (0.143) 55 km − 0.002 0.662∗∗∗ (0.025) (0.138) Generalized DID − 0.006 − 0.074 (0.018) (0.086) FTA depth 0.021∗ 0.155∗∗∗ (0.012) (0.059) Number FTAs − 0.001 − 0.001 (0.001) (0.001) Source:Author’s analysis. Note: All Human Development Index values are scaled (HDI×100). All columns include linear trends, country-year and regional fixed effects, and all controls. Standard errors are clustered at the country- year and regional levels and are shown in parentheses. Significance levels: ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. was incorporated into the specification.38 Given that short-term migration fluctuations within a region can impact average access to goods and services, this inclusion aimed to account for welfare changes, especially in densely populated developing regions. However, the core findings remained unaltered by this addition. Given the subnational units’ preferred grid size (111 by 111 km) could potentially obscure even more localized FTA effects, I performed additional tests on more granular grids (55 by 55 km) obtaining quali- tatively comparable results. Subsequently, I employed a “generalized” difference-in-differences approach. Unlike the primary method, where the Postj, t dummy remains active post its first FTA activation, this approach allows the dummy to adjust annually. When compared to the outcomes in table 1, it seems apparent that FTAs exert their impact in the years following their introduction and not in the same year of activation. The concluding tests in panel B, which used both the mean depth and the annual count of FTAs as the core variables, go mostly in line with the main findings of table 3. 3.3. Event Study Finally, I take note of the recent advances in the two-way fixed effects DID literature compellingly sum- marized in the work of Roth et al. (2022) and put to the test the, arguably, most salient event-study model 38 These population data are sourced from the History Database of the Global Environment (HYDE), as documented by Goldewijk et al. (2011), with a spatial resolution of approximately 10 km2 . The World Bank Economic Review 79 of such works (i.e., Callaway and Sant’Anna 2021). Roth et al. (2022)’s work visits the contributions of, among others, Callaway and Sant’Anna (2021), Gardner (2021), Sun and Abraham (2021) on several imputation methods to build an estimator that is robust to the heterogeneity of the effect among groups or over time. Their methods vary, yet most coincide in staging the computations on the non-treated or not-yet treated first (to compute a propensity score or a set of group and period fixed effects) and later Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 imputing them into the computations using the treated groups. In this way, one can relax the parallel trends assumption, account for treatment weights, and reduce the overall possibility of bias. These esti- mators can account for heterogeneity and potentially arbitrary effect dynamics by relaxing the perhaps more comprehensive yet less realistic assumption of a simple linear model. The results obtained in the event study mostly align with the DID model in the previous section and can be seen in detail in fig. 10. Graph (a), (b), and (c) show the effect of FTAs over HDI, night-lights, and local inequality, respectively. The graphs show the magnitude of the effects from 10 years before the FTAs were introduced (t = −10) until 10 years after (t = 10). As seen, as one should expect from placebo periods (Goldsmith-Pinkham, Sorkin, and Swift 2020; Borusyak, Hull, and Jaravel 2021; Borusyak, Jaravel, and Spiess 2021), in the pre-period (t − 10 to t) there are no real effects as all estimations are around 0. These findings strongly suggest the absence of developmental pre-trends that might predict the occurrence of FTAs in subsequent years, or in other words, none of the development indicators significantly correlate with future FTAs. In the post-period, on the other hand, there are important trends to note. The effects on HDI become progressively negative, reaching a peak decrease of 3 percentage points, 10 years post-introduction. Mean- while, the impacts on both night-lights and local inequality shift to the positive side. This corresponds to a maximum surge in economic activity of 0.7 night-light units (roughly 15.6 percent, relative to the control group standard deviation) and a peak increase of 4 percentage points in the local Gini coefficient. These findings align with the results from the DID analysis but add a time perspective. Specifically, while FTAs markedly enhance economic activity, they exert detrimental effects on broader metrics of human develop- ment. The reasons can be traced back to the influence of FTAs on local inequality patterns, as alluded to in earlier sections. This disparity becomes increasingly pronounced post-FTA introduction. Consequently, while FTAs boost economic productivity, the advantages that could potentially benefit other facets of hu- man development are concentrated among a select minority, leaving the vast majority, broadly speaking, in a less favorable position following the implementation of FTAs. 4. Conclusions The role of free trade agreements in global economics remains a topic of active discussion across developed and developing regions, given the ongoing debate regarding their tangible benefits. The evolution of trade agreements, a cornerstone of economic globalization, can be traced back to pivotal junctures such as the GATT meeting in 1947. Since 1990, there has been a notable rise in both the quantity and complexity of FTAs. While a significant majority of economists advocate for free trade over protectionism, the influence and implications of these trade agreements extend beyond their sheer prevalence. Indeed, their power lies not only in their ubiquity but also in their pivotal role in shaping the regulations governing global trade. In this research endeavor, I delve into the impacts of FTAs on human and economic development at the subnational scale, attempting to reconcile their effects with the diverse perspectives they elicit. In my empirical strategy, I harnessed global high-resolution land-cover data, highlighting the primary, naturally defined land types across subnational territories. This was contrasted with a national-level FTA dummy, encompassing up to 207 countries from 1992 to 2015. Employing a DID and an event-study design, I facilitated an interaction between the natural land data set and the FTA indicators, aiming to capitalize on subnational temporal variations. The premise behind this approach rests on the assumption that the diverging trends between naturally determined exploitable and non-exploitable regions can only 80 Cruzatti C. Figure 10. The Impact of Free Trade Agreements over Time. Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 Source: Author’s analysis. Note: The figures show the average effect of free trade agreements on the Human Development Index (HDI), night-lights, and local inequality in each period. The estimation follows the work of Callaway and Sant’Anna (2021). HDI and inequality are scaled (×100). The shadows represent 95 percent confidence intervals. The World Bank Economic Review 81 be explained due to the introduction of FTAs. In essence, this approach aimed to pinpoint the effects of FTAs at a subnational dimension. The main findings indicate that the local impact of FTAs on human development is small yet negative, while their influence on economic development is statistically significant, positive, and sizeable. A central theme emerging from this analysis relates to inequality within the examined sectors and regions. The data Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 suggest that economic gains are fairly universal, with more pronounced benefits observed in developed sectors such as urbanized areas. Conversely, human development appears to face challenges, especially within less developed sectors and regions with stark inequalities. Dissecting these effects further, distin- guishing between northern/bigger and southern/smaller nations reveals a discernible negative bearing on human development for the latter group. In terms of the depth of FTAs, the research finds no difference between the impacts of deeper agreements and their shallower counterparts. As per further roads of improvement, one would need to find ways to overcome potential limitations of the empirical strategy. Note that the effect of free trade agreements could be correlated with other local relevant economic and political shocks. For instance, a local government with greater ambition may be more inclined to enhance economic conditions and foster free trade initiatives. Similarly, uncovering natural resources could potentially catalyze increased trade activity within specific regions. While the model and robustness address some of these issues, there is still room for improvement. This research illuminates the multifaceted implications of FTAs across various developmental indices. Drawing from a vast data set, the study offers a more expansive and generalizable analytical framework compared to earlier works. This study contributes to our understanding of the subnational effects of FTAs on developmental metrics, granting insights into the nuanced local variations and underlying mechanisms shaping the FTA landscape. For policymakers, this research holds deep significance. It not only elucidates the sectors, regions, partners, and depths that should be at the forefront of FTA negotiations, but also underscores the pivotal dimensions that ought to be prioritized in crafting these agreements. Ultimately, recognizing the contexts influenced by inequality is instrumental in ensuring that economic gains from FTAs translate meaningfully into advancements in human development. Data availability statement The data underlying this article are available in the article and in its online supplementary material. References Adão, R., P. Carrillo, A. Costinot, D. Donaldson, and D. Pomeranz 2022, “Imports, Exports, and Earnings Inequality: Measures of Exposure and Estimates of Incidence.” Quarterly Journal of Economics 137 (3): 1553–1614. AidData. 2017. World Bank Geocoded Research Release Level1 v1.4.2 Geocoded Dataset. Williamsburg, VA and Washington, DC: AidData. Aiyar, A., and J. R. Cummins 2021, “An Age Profile Perspective on Two Puzzles in Global Child Health: The Indian Enigma & Economic Growth.” Journal of Development Economics 148: 102569. Amiti, M., and J. Konings 2007, “Trade Liberalization, Intermediate Inputs, and Productivity: Evidence from Indone- sia.” American Economic Review 97(5): 1611–38. Artuc, E., G. Porto, and B. Rijkers 2019, “Trading Off the Income Gains and the Inequality Costs of Trade Policy.” Journal of International Economics 120: 1–45. Astorga, P. 2010, “A Century of Economic Growth in Latin America.” Journal of Development Economics 92(2): 232–43. Auer, R. A. 2015, “Human Capital and the Dynamic Effects of Trade.” Journal of Development Economics 117: 107–18. Balland, P.-A., C. Jara-Figueroa, S. G. Petralia, M. P. A. Steijn, D. L. Rigby, and C. A. Hidalgo 2020, “Complex Economic Activities Concentrate in Large Cities.” Nature Human Behaviour 4(3): 248–54. 82 Cruzatti C. Bombardini, M., and B. Li 2020. “Trade, Pollution and Mortality in China.” Journal of International Economics 125: 103321. Borusyak, K., P. Hull, and X. Jaravel 2021. “Quasi-Experimental Shift-Share Research Designs.” Review of Economic Studies 89 (1): 181-213. Borusyak, K., X. Jaravel, and J. Spiess 2024. “Revisiting Event Study Designs: Robust and Efficient Estimation.” The Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 Review of Economic Studies rdae007. Bruederle, A., and R. Hodler 2018. “Nighttime Lights as a Proxy for Human Development at the Local Level.” PLoS One 13(9): e0202231. Caliendo, L., R. Feenstra, J. Romalis, and A. Taylor 2015. “Tariff Reductions, Entry, and Welfare: Theory and Evidence for the Last Two Decades.” NBER Working Paper 21768, National Bureau of Economic Research. Callaway, B., and P. H. C. Sant’Anna 2021. “Difference-in-Differences with Multiple Time Periods.” Journal of Econo- metrics 225(2): 200–30. Chen, X. 2016. “Addressing Measurement Error Bias in GDP with Nighttime Lights and an Application to Infant Mortality with Chinese County Data.” Sociological Methodology 46(1): 319–44. Chen, X., and W. D. Nordhaus 2011. “Using Luminosity Data as a Proxy for Economic Statistics.” Proceedings of the National Academy of Sciences 108(21): 8589–94. Christian, P. J., and C. B. Barrett 2017. “Revisiting the Effect of Food Aid on Conflict: A Methodological Caution.” Policy Research Working Paper Series 8171, The World Bank. Cingano, F. 2014. “Trends in Income Inequality and its Impact on Economic Growth.” Working paper, OECD. Cruzatti, C. J., A. Dreher, and J. Matzat 2023. “Chinese Aid and Health at the Country and Local Level.” World Development 167: 106214. Diwan, I., and D. Rodrik 1991. “Patents, Appropriate Technology, and North-South Trade.” Journal of International Economics 30(1-2): 27–47. Dix-Carneiro, R., and B. K. Kovak 2017. “Trade Liberalization and Regional Dynamics.” American Economic Review 107(10): 2908–46. Dreher, A., A. Fuchs, R. Hodler, B. C. Parks, P. A. Raschky, and M. J. Tierney 2019. “African Leaders and the Geog- raphy of China’s Foreign Assistance.” Journal of Development Economics 140: 44–71. Dür, A., L. Baccini, and M. Elsig 2014. “The Design of International Trade Agreements: Introducing a New Dataset.” Review of International Organizations 9(3): 353–75. Edmonds, E. V., and N. Pavcnik 2005. “The Effect of Trade Liberalization on Child Labor.” Journal of International Economics 65(2): 401–19. Edmonds, E. V., N. Pavcnik, and P. Topalova 2010. “Trade Adjustment and Human Capital Investments: Evidence from Indian Tariff Reform.” American Economic Journal: Applied Economics 2(4): 42–75. ———. 2009. “Child Labor and Schooling in a Globalizing World: Some Evidence from Urban India.” Journal of the European Economic Association 7(2-3): 498–507. Elvidge, C. D., K. E. Baugh, S. J. Anderson, P. C. Sutton, and T. Ghosh 2012. “The Lumen Gini Coefficient: A Satellite Imagery Derived Human Development Index.” Social Geography Discussions 8(1): 27–59. ESA. 2017. “Land Cover CCI Product User Guide Version 2.0.7.” Tech. Rep.Available at https://maps.elie.ucl.ac.be/ CCI/viewer/download/ESACCI- LC- Ph2- PUGv2_2.0.pdf. Flesher Fominaya, C. 2014. Social Movements and Globalization: How Protests, Occupations and Uprisings Are Changing the World. London, United Kingdom. Palgrave Macmillan. Furceri, D., S. Hannan, J. Ostry, and A. Rose 2018. “Macroeconomic Consequences of Tariffs.” NBER Working Paper 25402, National Bureau of Economic Research. Gardner, J. 2022. “Two-Stage Differences in Differences.” arXiv preprint. arXiv:2207.05943. Germanos, A. 2015. “WHO Head Joins Chorus Denouncing TPP’s Blow to Public Health, Boon to Big Pharma.” https://www.commondreams.org/news/2015/11/12/who- head- joins- chorus- denouncing- tpps- blow- public- health- boon- big- pharma. Accessed May 2, 2018. Portland, US. Goldewijk, K. K., A. Beusen, G. V. Drecht, and M. D. Vos 2011. “The HYDE 3.1 Spatially Explicit Database of Human-Induced Global Land-Use Change over the Past 12,000 Years.” Global Ecology and Biogeography 20(1): 73–86. Goldsmith-Pinkham, P., I. Sorkin, and H. Swift 2020. “Bartik Instruments: What, When, Why, and How.” American Economic Review 110(8): 2586–624. The World Bank Economic Review 83 Grennes, T. 2018. “An Economic Analysis of the Jones Act (05/02/2017)” MERCATUS RESEARCH. Available at ht tps://ssrn.com/abstract=3211671 or http://dx.doi.org/10.2139/ssrn.3211671. Grochowska, R., and A. Ambroziak 2018. “Selected Aspects of Protectionism of EU Member States in the EU Agri- ´ Food Market.” Problems of World Agriculture / Problemy Rolnictwa Swiatow ego 18(33): 82–94. Hausmann, R., and C. Hidalgo“Country Diversification, Product Ubiquity, and Economic Divergence.” 2010. Work- Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 ing Paper Series RWP10-045, John F. Kennedy School of Government, Harvard University.. Henderson, J. V., T. Squires, A. Storeygard, and D. Weil 2018. “The Global Distribution of Economic Activity: Nature, History, and the Role of Trade.” Quarterly Journal of Economics 133(1): 357–406. Henderson, J. V., A. Storeygard, and D. N. Weil 2012. “Measuring Economic Growth from Outer Space.” American Economic Review 102(2): 994–1028. R Hicks, J. 1939. “The Foundations of Welfare Economics.” Economic Journal 49(196): 696. Hidalgo, C. A. 2015. Why Information Grows: The Evolution of Order, from Atoms to Economies(1 ed.). New York: Basic Books. Hirst, P. 1997. “The Global Economy—Myths and Realities.” International Affairs 73(3): 409–25. Hodler, R., and P. A. Raschky 2014. “Regional Favoritism.” Quarterly Journal of Economics 129(2): 995–1033. ICF. 2004-2017. “Demographic and Health Surveys (Various) [Datasets].” Funded by USAID. Rockville, Maryland: ICF [Distributor]. ———. 2013. Incorporating Geographic Information into Demographic and Health Surveys: A Field Guide to GPS Data Collection. Calverton, MD: ICF International. Jean, N., M. Burke, M. Xie, W. M. Davis, D. B. Lobell, and S. Ermon 2016. “Combining Satellite Imagery and Machine Learning to Predict Poverty.” Science 353(6301): 790–4. Kaldor, N. 1939. “Welfare Propositions of Economics and Interpersonal Comparisons of Utility.” Economic Journal 49(195): 549. Kareem, F. O., I. Martínez-Zarzoso, and B. Brümmer 2018, “Protecting Health or Protecting Imports? Evidence from EU Non-Tariff Measures.” International Review of Economics & Finance 53: 185–202. Khan, N. S. 2020. “Spillover Effects of Trade Shocks in the Central and Eastern European and Baltic Countries.” Journal of Economic Integration 35(1): 39–68. Kong, F., Z. Chen, S. Xue, X. Wang, and J. Liu 2015. “Mother’s but Not Father’s Education Predicts General Fluid Intelligence in Emerging Adulthood: Behavioral and Neuroanatomical Evidence.” Human Brain Mapping 36 (11). Kovak, B. K. 2013. “Regional Effects of Trade Reform: What Is the Correct Measure of Liberalization?” American Economic Review 103(5): 1960–76. Kriesi, H., E. Grande, M. Dolezal, M. Helbling, D. Höglinger, S. Hutter, and B. Wüest 2012. Political Conflict in Western Europe. Cambridge: Cambridge University Press. Kummu, M., M. Taka, and J. H. A. Guillaume 2018. “Gridded Global Datasets for Gross Domestic Product and Human Development Index over 1990–2015.” Scientific Data 5(1). Limão, N. 2016. “Preferential Trade Agreements.” NBER Working Paper 22138, National Bureau of Economic Re- search. Marchetti, J. A., and P. C. Mavroidis 2011. “The Genesis of the GATS (General Agreement on Trade in Services).” European Journal of International Law 22(3): 689–721. McCaig, B. 2011. “Exporting Out of Poverty: Provincial Poverty in Vietnam and US Market Access.” Journal of International Economics 85(1): 102–13. Mellander, C., J. Lobo, K. Stolarick, and Z. Matheson 2015. “Night-Time Light Data: A Good Proxy Measure for Economic Activity?” PLoS ONE 10(10): e0139779. Navaei, A. Health Effects of Trade Liberalization: Evidence from US Counties 2021. European Journal of Multidisci- plinary Sciences 6(1): 120–143. Nishar, A. 2017. “Temperature Effects on Biomass and Regeneration of Vegetation in a Geothermal Area.” Frontiers in Plant Science 8. NOAA. 2015. “Earth Observation Group - Defense Meteorological Satellite Progam, Boulder — ngdc.noaa.gov.” ht tps://ngdc.noaa.gov/eog/dmsp/downloadV4composites.html. Accessed 2017-04-19. Noor, A. M., V. A. Alegana, P. W. Gething, A. J. Tatem, and R. W. Snow 2008. “Using Remotely Sensed Night-Time Light as a Proxy for Poverty in Africa.” Population Health Metrics 6(1). 84 Cruzatti C. Otero, G. 2011. “Neoliberal Globalization, NAFTA, and Migration: Mexico’s Loss of Food and Labor Sovereignty.” Journal of Poverty 15(4): 384–402. Pierce, J. R., and P. K. Schott 2020. “Trade Liberalization and Mortality: Evidence from US Counties.” American Economic Review: Insights 2(1): 47–64. Pregitzer, K., and J. King 2005. “Effects of Soil Temperature on Nutrient Uptake.” In Nutrient Acquisition by Plants: Downloaded from https://academic.oup.com/wber/article/39/1/61/7667624 by WORLDBANK THIRDPARTY user on 05 February 2025 An Ecological Perspective, 277–310. Berlin, Heidelberg: Springer. Rauch, F. 2016. “The Geometry of the Distance Coefficient in Gravity Equations in International Trade.” Review of International Economics 24(5): 1167–77. Rodrik, D. 2018. “What Do Trade Agreements Really Do?” Journal of Economic Perspectives 32(2): 73–90. Roth, J., P. H. C. Sant’Anna, A. Bilinski, and J. Poe 2023. “What’s Trending in Difference-in-Differences? A Synthesis of the Recent Econometrics Literature.” 235 (2): 2218–2244 Sakyi, D., J. Villaverde, A. Maza, and I. Bonuedi 2017. “The Effects of Trade and Trade Facilitation on Economic Growth in Africa.” African Development Review 29(2): 350–61. Salamanca, L. J. G., F. B. Gomez, and I. C. Landínez 2009. “Impact of the US-Colombia FTA on the Small Farm Economy in Colombia.” Research report, Bogotá: Oxfam International. Salvati, L., A. Guandalini, M. Carlucci, and F. M. Chelli 2017. “An Empirical Assessment of Human Development through Remote Sensing: Evidences from Italy.” Ecological Indicators 78: 167–72. Sell, S. K. 2011. “TRIPS Was Never Enough: Vertical Forum Shifting, FTAS, ACTA, and TTP.” Journal of Intellectual Property Law 18(2): 447–78. Stevens, C., M. Irfan, I. Massa, and J. Kennan 2015. “The Impact of Free Trade Agreements between Developed and Developing Countries on Economic Development in Developing Countries.” Rapid Evidence Assessment, Overseas Development Institute, London. Sun, L., and S. Abraham 2021. “Estimating Dynamic Treatment Effects in Event Studies with Heterogeneous Treat- ment Effects.” Journal of Econometrics 225(2): 175–99. Sutton, P., C. Elvidge, and G. Tilottama 2007. “Estimation of Gross Domestic Product at Sub-national Scales Using Nighttime Satellite Imagery.” International Journal of Ecological Economics & Statistics 8(S07).5–21. Sutton, P. C., and R. Costanza 2002. “Global Estimates of Market and Non-market Values Derived from Nighttime Satellite Imagery, Land Cover, and Ecosystem Service Valuation.” Ecological Economics 41(3): 509–27. Teney, C., O. P. Lacewell, and P. De Wilde 2014. “Winners and Losers of Globalization in Europe: Attitudes and Ideologies.” European Political Science Review 6(4): 575–95. Tollefsen, A. F., H. Strand, and H. Buhaug 2012. “PRIO-GRID: A Unified Spatial Data Structure.”Journal of Peace Research 49 (2): 363–374. Topalova, P. 2010. “Factor Immobility and Regional Impacts of Trade Liberalization: Evidence on Poverty from India.” American Economic Journal: Applied Economics 2(4): 1–41. UNDP (United Nations Development Programme). 2016. “Human Development for Everyone”. Human Development Report 2016: New York. ———. 2017. “Human Development Reports Database.” United Nations Development Programme. United Nations, Geneva. Van den, BergH,. 2012. “Complexity and the Culture of Economics: A Sociological and Inter-disciplinary Analysis.” Journal of Philosophical Economics 5(2): 38–63. Wallerstein, I. 1976. “A World-System Perspective on the Social Sciences.” British Journal of Sociology 27(3): 343–52. ———. 2005. “After Developmentalism and Globalization, What?” Social Forces 83(3): 1263–78. Weidmann, N. B., and S. Schutte 2017. “Using Night Light Emissions for the Prediction of Local Wealth.” Journal of Peace Research 54(2): 125–40. Wise, T. A. 2009. “Agricultural Dumping under NAFTA: Estimating the Costs of U.S. Agricultural Policies to Mexican Producers.” GDAE working papers, Boston: GDAE, Tufts University. ———. 2014. “How Beer Explains 20 Years of NAFTA’s Devastating Effects on Mexico.” http://www.globalpost.c om/dispatches/globalpost- blogs/global- pulse/nafta- 20- years- mexican- beer. Accessed April 20, 2014. World Bank. 2020. “World Development Indicators.” Accessed February 15, 2020. https://datacatalog.worldbank.or g/dataset/world- development- indicators. Yang, S., Y. Khang, S. Harper, G. D. Smith, D. Leon, and J. Lynch 2010. “Understanding the Rapid Increase in Life Expectancy in South Korea.” American Journal of Public Health 100(5): 896–903.