Policy Research Working Paper 11049 Forced Displacement, the Perpetuation of Autocratic Leadership, and Development in Origin Countries Nicolás Cabra-Ruiz Sandra V. Rozo María Micaela Sviatschi Development Economics Development Research Group January 2025 Policy Research Working Paper 11049 Abstract How does forced displacement shape development in origin forced displacement facilitates the perpetuation of auto- countries? This article examines the case of Venezuela, where cratic rule and hinders development: by weakening political nearly eight million people have been forcibly displaced. To opposition and enabling the growth of organized crime do this, it compares municipalities with varying shares of and illicit income sources. Using novel election data, the foreign-born populations before and after international oil article finds that areas affected by mass forced displacement price shocks accelerated forced displacement between 2014 experienced lower voter turnout and opposition support, and 2019. The findings show that municipalities with larger limiting political and social reforms. These areas also wit- foreign-born populations in 1990, which also exhibited nessed growth in organized crime and foreign non-state greater out-migration from Venezuela after 2014, experi- drug and human trafficking, which diminished incentives enced lower economic development and higher inequality. for economic change. The paper highlights a new mechanism through which This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at sandrarozo@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Forced Displacement, the Perpetuation of Autocratic Leadership, and Development in Origin Countries∗ as Cabra-Ruiz† Nicol´ Sandra V. Rozo‡ ıa Micaela Sviatschi§ Mar´ Keywords: Forced displacement, electoral outcomes, economic growth, Venezuela JEL Classification: O12, O15, O54 ∗ We are grateful to Roberto Mendoza and Gaspar Arias for excellent research assistant support and to Leah Boustan, Jos´ e Morales-Arilla, Salvador Traettino, Maria Esther Caballero, Vincenzo DiMaro, Daniel Pereira, Carolina Mej´ ın, Margaret Peters, Daniel Ortega, Juan Vargas, ıa, Jose Luis Espinoza, Pablo Querub´ Javier Romero, Carlos Rodr´ ıguez, Mark Thomas, and Dorothy Kronick for suggestions. We also thank participants at seminars at Oxford University, NOVA university, Bocconi University, the University of Turin, George Washington University, Villanova University, the University of Pittsburgh, John Hopkins University, the University of Chicago, EAFIT University, Universidad Carlos III de Madrid, Paris School of Economics, and World Bank seminars for their feedback. Rozo acknowledges support from the Research Support Budget of the World Bank. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank and its affiliated organizations or those of the Executive Directors of the World Bank or the governments they represent. † Pre-doctoral Scholar, Princeton University ‡ Development Research Group. World Bank, CEGA, EGAP, and IZA. Email: sandrarozo@worldbank.org § Princeton University and NBER. Email: msviatschi@princeton.edu “Exit has [been] shown to drive out voice.” Albert O. Hirschman, 1970 1 Introduction Over the past 50 years, the number of forced displacement crises has skyrocketed. By 2023, these involuntary relocations of people due to conflict, violence, persecution, disasters, and climate change had displaced more than 117 million people (UNHCR, 2024). This trend is likely to intensify, making the issue a pressing global development challenge. While recent research has studied how these flows affect destination countries (e.g., Dustmann et al., 2019; Steinmayr, 2021) and displaced individuals (e.g., Becker et al., 2020; Chiovelli et al., 2021), less is known about how they shape the development paths of origin countries. This knowledge gap is largely due to difficulties in obtaining reliable data from origin locations mired in intense conflict, economic instability, and social crises. Nonetheless, forced displacement can profoundly reshape the development trajectories of origin countries, especially those with poor institutions. In political terms, it can strengthen ruling parties if migrants would have supported opposition candidates; economically, it can thwart growth by depleting factors of production and reducing consumer demand. Forced displacement can also exacerbate income inequality. The poorest—who cannot migrate— are left behind, while the middle class departs and the richest—who benefit the most from autocratic regimes—remain. At the same time, potential positive externalities from migra- tion, such as knowledge diffusion, trade, and higher foreign direct investment (Javorcik et al. ezina 2018), may not occur in the context of weak institutions. Instead, 2011; Parsons and V´ negative externalities may emerge, as forced displacement could pave the way for more or- ganized crime and dependence on illicit economic resources (Kapur 2014). This happens when criminal networks coerce vulnerable forced migrants to participate in such activities, a scenario more attractive in origin countries with weak democracies or rule of law. 2 We investigate these hypotheses by assessing the role of forced displacement in shaping Venezuela’s development between 1992 and 2021. In response to economic and social turmoil avez and Nicol´ during the regimes of Hugo Ch´ as Maduro, at least 35 percent of Venezuela’s population—7.7 million individuals—has fled the country. Given the desperate social and economic conditions there and the size of the migration outflows, Venezuelans are recog- nized internationally as a forcibly displaced population, and are under the United Nations avez became president in 1998 and his govern- Refugee Agency protection. Although Ch´ ment introduced constitutional reforms that concentrated power in the executive branch and strongly weakened political accountability, forced displacement only surged after 2013, when his successor, Maduro, faced plummeting international oil prices in the world’s twelfth- largest oil-producing nation. This price decline, combined with detrimental economic policies that led to hyperinflation and international sanctions, deepened the crisis and contributed to the mass exodus. We employ a difference-in-difference methodology and exploit two sources of variation. First, annual variation is derived from the oil price shocks that occurred after 2013. Second, municipal variation is based on the share of foreigners living in each municipality in 1990, avez’s presidency. Foreigners in Venezuela in 1990 the last population census before Ch´ may have facilitated network support abroad and provided information on the migration process, potentially easing out-migration. In support of this hypothesis, we find that most foreigners living in Venezuela in 1990 were Colombians, and most Venezuelans who later migrated to Colombia originated from municipalities with a higher share of Colombians in 1990. Furthermore, Venezuela’s unique National Surveys on Living Conditions (ENCOVI in Spanish) that were collected between 2017 and 2021 show that most Venezuelan families had relatives living in Colombia.1 Indeed, using these surveys, we find that Venezuelan municipalities with a higher foreign share in 1990 had an increased likelihood of households reporting a relative in Colombia. 1 These data are available for 2014 to 2021, but the migration module is only available for 2017 to 2021. 3 Our empirical strategy compares economic outcomes in Venezuelan municipalities with vary- ing foreign settlement shares in 1990, before and after the shocks that induced mass forced displacement from 2013 to 2018. In particular, we exploited the fact that while the decline in oil prices and subsequent economic policies induced a national shock, Venezuelans from municipalities with more foreigners were more likely to be able to leave.2 Importantly, we show that municipalities with a large foreign share do not have different economic structures that could therefore be disproportionately affected by the oil shock. In addition, we show that national economic measures during the period of analysis did not differentially affect municipalities with a high foreign share. To explore these sources of variation, we use data on night-light density, household living conditions, elections, and criminal organizations for the period 1992–2021. First, we use night-light data as a proxy for economic growth and inequality.3 We measure inequality using a novel spatial measure that estimates the dispersion of night-light density at the parroquia (county) level within each municipality. We find that this measure positively correlates with traditional income-based inequality metrics, validated using census data from Venezuela and Colombia. Second, to examine effects on the perpetuation of political leadership, we use unique web-scraped data on the results from the last four presidential and six mayoral elections, the ENCOVI, a representative survey of Venezuelan migrants in Colombia, and data on the presence of several organized criminal groups from the Armed Conflict Location and Event Data (ACLED) and the Global Terrorism Database (GTD). Our first key result is that municipalities affected by forced displacement experienced sharp declines in development and higher inequality than other areas. Municipalities with a 1 percent higher share of foreign settlements in 1990 experienced nearly 12.6 percent lower 2 Additionally, examining previous trends, we can also test the identifying assumption—that is, munic- ipalities with a large foreign settlement share and those with a low share would have experienced similar economic development trends if migration had not increased after 2013. 3 We validate these measures for correlation with household-level income measures, using the ENCOVI. Moreover, we control for the presence of oil fields in our estimates to rule out concerns that luminosity driven by gas flares could affect our main results. 4 night-light density after 2013 than others.4 Inequality rose sharply in municipalities affected by forced displacement, a result validated by using per capita income data from the EN- COVI. Given that out-migration flows quintupled after 2013, our results suggest that in a municipality with a share of 10 percent of the total foreigners living in Venezuela in 1990, the estimated coefficients translate to a 20 percent decline in GDP. Regarding the timing of the effects, we observe that the drop in development was immediate, although it worsened after 2017 when the mass migration flows peaked. Although lower international oil prices weakened national economic conditions, the situation deteriorated in 2017 when monetary financing to cover the fiscal deficit led to hyperinflation. Using ENCOVI data, we find that particularly in the first years after the shock, the most productive and educated individuals left the country, taking their physical and human capital with them. This exodus contributed to the immediate decline in economic growth. Remittances did not offset these adverse effects, as they were received by a small share of households and primarily used for subsistence needs like food and housing rather than productive investments. As time passed and the crises intensified, we observed that migration is much more generalized and less driven by positively selected individuals.5 Furthermore, we find that individuals who stayed behind invested less in human capital. How does a political regime remain in power during such a profound crisis? Our second key finding is that forced displacement reinforced the medium-term continuation of autocratic leadership in two ways. First, we observe that forced migrants disproportionately backed opposition candidates.6 Thus, their displacement weakened the political opposition, further entrenching the incumbent and reducing momentum for social change. We also find that higher levels of forced displacement in affected municipalities caused a steep drop in electoral 4 To account for the possibility of gas flares affecting night-light measures, Table E.19 shows the estimation controlling by oil fields, using data from Sabbatino (2018). 5 While we can only validate this pattern using the ENCOVI survey, according to our conversations with governments in countries that received Venezuelan migrants, the same pattern is observed. 6 This is consistent with the evidence provided by Holland et al. (2024), in which 12.1 percent of all Venezuelan migrants in Colombia supported the left and only 0.1 percent supported Maduro’s government. 5 turnout and votes for opposition candidates after 2013. This mirrors the 2024 presidential election, where only 60,000 out of seven million Venezuelans abroad were allowed to vote.7 Second, we find that organized crime grew in Venezuelan municipalities affected by forced displacement, which could potentially benefit the autocratic regime, reducing the incentives for economic change. After 2013, these areas witnessed large gains in activities linked to crim- inal organizations that exploit migrants and are associated with the regime. These include drug-trafficking organizations from destination countries and local criminal organizations that specialize in human trafficking. They take advantage of irregular migration to recruit, enroll, and victimize migrants. Our quantitative findings align with abundant qualitative evidence from journalists and NGOs in the region, suggesting that forced displacement from Venezuela reduced the transactional costs of trafficking and allowed these groups to expand while benefiting the regime.8 These combined electoral and criminal effects of forced displacement exacerbate already negative impacts on development. Amid less pressure for reforms and more reliance on illicit activities, the incumbent government has less incentive to improve conditions for private and human capital investment. We leverage a mediation analysis to show that these two channels explain approximately one-third of the effect of forced displacement on development. Our findings are robust to several validity tests. Notably, municipalities with a higher share of foreigners in 1990 show similar trends in our main outcomes, even before the crisis began. Moreover, our core estimates remain consistent even after accounting for interactions be- tween linear time trends and a large set of observable baseline measures, providing evidence 7 Our findings hold when using opposition-reported 2024 election data and are consistent with results from the five Venezuelan municipal elections we analyzed. 8 Ample qualitative evidence highlights that the autocratic Venezuelan government maintains power by leveraging criminal rents, as seen in the Paraguan´ a Cartel and Cartel of the Sun cases (Insight Crime, 2022a,b). Politicians, security forces, and drug traffickers form symbiotic networks where officials shield traf- fickers from prosecution in exchange for financial support, political influence, and public services. Governors and mayors manipulate military and police appointments to ensure loyalty and facilitate drug trafficking. At the same time, traffickers finance campaigns, mobilize votes, and fill gaps left by a collapsing state. This interconnection of politics and crime bolsters both the regime’s power and organized crime. 6 that displacement effects are not confounded by unrelated dynamics specific to high-foreign share municipalities. Our results also remain robust when analyzing smaller geographic units, adjusting for spatial autocorrelation, approximating forced displacement using inverse distance to Colombia’s crossing points, and applying alternative difference-in-difference esti- mators based on recent methodological advances (Abadie, 2005; Sant’Anna and Zhao, 2020). Additionally, our findings are not influenced by contemporaneous government actions. Even when controlling for novel proxies of government intervention—such as data on expropria- tions of private firms, social program beneficiaries, and irregularities in electricity provision post-2019—our results hold. In fact, our results remain consistent when controlling for oil production, indicating that differential trends in oil-rich areas are not driving the effects. We further validate our findings through a strategy based on a measure of imputed outflows by year and municipality, calculated from the interaction of foreign settlement shares in 1990 and forced migration to Colombia. This measure closely tracks actual migration patterns, as shown by its strong correlation with the origins and arrival dates of Venezuelans interviewed in Colombia in 2018.9 Imputed outflows also positively correlate with the locations of rel- atives abroad, as seen in the ENCOVI data (2017–2021). Moreover, using road distances from municipalities to major border crossings to define affected areas yields similar results. One potential concern is the possible manipulation of electoral data. To address this, we conduct data checks for abnormal patterns and find no evidence of manipulation before 2018, in line with reports from the humanitarian and international organizations that validated those elections.10 However, we do find evidence of manipulation (or the effects of a political boycott by the opposition) in the 2018 presidential election. To validate our findings, we re-estimate the effects of forced displacement using opposition-collected data for the 2024 presidential election, which indicate that Maduro lost by a landslide. Additionally, we an- alyze mayoral elections across 335 municipalities, comprising five election years and 1,675 9 While the geographic scope of Colombian data limits econometric analysis, it still validates our measure. 10 Reports from the Carter Center (1998–2021) suggest no systematic manipulation before 2018. 7 observations. The scale of these elections makes systematic manipulation unlikely. The estimates remain consistent and support the validity of our findings. This paper contributes to the literature on the relationship between migration and devel- opment. While the vast majority of research has focused on the effects of migration on destination economies (Card, 2001; Borjas, 2014; Abramitzky et al., 2014; Foged and Peri, 2016; Dustmann et al., 2017; Hanson et al., 2018), brain drain (Beine et al., 2008; Gibson and McKenzie, 2011; Docquier and Rapoport, 2012; Batista et al., 2012; Anelli et al., 2023), remittances (Amuedo-Dorantes and Pozo, 2006; Giuliano and Ruiz-Arranz, 2009; Portes, 2009; Ambler et al., 2015), cultural remittances (Melki et al., 2024), and workers’ outside options and bargaining power (Karadja and Prawitz, 2019), our study adds a new dimension by examining the impact of forced displacement on economic growth in origin countries, which are often governed by autocratic regimes. We offer empirical evidence for Hirschman’s hypothesis that in contexts with weak democracies, emigration may reduce pressure for re- forms (Hirschman 1970, 1978), in contrast to the well-documented link between emigration, higher democratization, and political change in well-functioning democracies (Spilimbergo 2009; Docquier et al. 2016; Karadja and Prawitz 2019; Grossmann et al. 2024).11 Further- more, we show that in weak democracies, the documented positive network effect of brain drain does not materialize (Gibson and McKenzie 2011; Docquier and Rapoport 2012). This may be because individuals are not inclined to invest in such places due to the high risks involved. In fact, “bad investors”—those who might benefit from a weak democracy and rule of law—are often the ones to move there. We also contribute to the literature by combining unique data sources to assess trends in a country experiencing mass forced migration and by exploring their impact on development outcomes. Such data are rarely available (Martinez 2022), as origin countries often undergo crises that make data collection extremely challenging. We integrate data from multiple 11 See Kapur (2014) for a detailed survey of the literature on international emigration and political outcomes in origin countries. 8 unique sources, including Venezuela’s annual ENCOVI (2014–2021). They provide insights into national poverty trends, sectoral recomposition, and public service provision despite being limited to the largest urban centers. Additionally, we use night-light density data at various scales—municipality, county, and grids of one square km—to approximate changes in economic growth and spatial inequality. We also web-scrape online electoral results from presidential and mayoral elections between 2004 and 2024, and combine them with pop- ulation censuses, web-scraped data on private-firm expropriations, international economic sanctions, and surveys of Venezuelan migrants in Colombia. This comprehensive approach enables us to assess the complexity of forced migration and its wide-ranging effects. alez Finally, we also contribute to recent work on the effects of authoritarian regimes (Gonz´ et al., 2023). However, rather than focus on the effects of the regimes themselves, we investi- gate the role of forced displacement flows and how they may reinforce the power of autocratic leaders in origin countries by reducing opposition and promoting reliance on illicit activities. 2 Local Context: Venezuela’s Unraveling Venezuela, historically reliant on oil, entered an economic recession in the late 1990s due to falling oil prices and a downturn in its non-oil sector, which fueled widespread discontent and led to the election of Hugo Ch´ ıguez 2014).12 Ch´ avez in 1998 (Hausmann and Rodr´ avez’s presidency brought sweeping changes, including constitutional reforms that concentrated power in the executive branch and weakened political accountability. His administration also expropriated private firms, spurring a sharp contraction of the private sector (Panel A, Figure A.2). Despite early political instability marked by a coup attempt and an oil strike, Venezuela eventually recovered and achieved sustained economic growth, thanks to rising avez’s tenure (Figure 1).13 These oil prices and expansionary government policies during Ch´ avez maintain widespread popularity until his death in 2013. factors helped Ch´ 12 The recession made already rampant inequality more salient and boosted the populist agenda. 13 For instance, government spending rose to 30 percent of GDP by 2006 (Weisbrot and Sandoval 2008). 9 Figure 1: Oil Prices and Production in Venezuela, 1992–2021 Notes: Oil price data come from the Global Commodity Prices dataset of the World Bank (2024). Annual oil production data come from Datosmacro (2024). 2.1 Venezuela’s Crisis, 2013 to Present avez—then terminally ill—appointed Nicol´ In 2013, Ch´ as Maduro as his successor. Following Maduro’s election, Venezuela suffered a sharp drop in international oil prices, triggering a severe external shock to the economy (Figure 1). In response, Maduro’s government sought financing to cover its deficit, leading to hyperinflation (Panel B, Figure A.2). Together, the oil price collapse, preexisting macroeconomic imbalances, and authoritarian policies induced a complete shutdown of international financing and the imposition of stringent sanctions (Panel C, Figure A.2). Consequently, the country plunged into crisis, with GDP sinking by over 70 percent between 2013 and 2018 (Morales-Arilla and Traettino 2023). ENCOVI data from 2014 to 2021 show a dramatic rise in poverty rates, with more than 94 percent of the population living below the poverty line and 74 percent below the extreme poverty line by 2021 (Figure A.3). The crisis extended beyond economic and political di- mensions to public services. As shown in Figure A.4, power and water supply interruptions 10 became widespread (Panels A and B). By 2021, at least 75 percent of the population had ceased to maintain a healthy diet and was subsisting on a limited variety of foods (Panel C). Health services were also decimated, with coverage dropping from 45.8 percent to just 3.4 percent between 2014 and 2021. By then, roughly 60 percent of those with chronic illnesses had little to no access to essential medications (Figure A.5). 2.2 Forced Displacement and Colombian Networks The events triggered by the 2013 crash in international oil prices have brought about unprece- dented forced displacement. According to the latest figures from the United Nations Refugee Agency (UNHCR), nearly eight million people have left Venezuela—more than 35 percent of the country’s population in 2013. Most Venezuelan migrants have settled in neighboring Colombia, with more than 2.8 million residing there by 2024 (Figure A.1). Figure 2 illustrates annual inflows of Venezuelan migrants to Colombia, which almost quintupled from 2013 to their peak in 2018. Data from the ENCOVI, which includes a migration module between 2017 and 2021, confirm that the relatives of Venezuelans living abroad are predominantly in Colombia (Table D.1). 11 Figure 2: Venezuelan Migration to Colombia, 2008–2021 Notes: Data on migration outflows between 2008 and 2021 come from the Colombian migration agency on Colombia, 2023). (Migraci´ 3 Data This section describes the data used in the study, broadly divided into three groups: remote- sensing data, electoral data, and observational data. Appendix B provides details on outcome construction and Appendix C presents descriptive statistics for the main variables employed in the analysis. Our analysis period is 1992–2021, in line with available remote-sensing data. 3.1 Municipal-Level Data To define which municipalities were mainly affected by out-migration, we use Venezuelan census data to quantify the presence of foreign residents in municipalities during 1990, the avez’s election. We construct the foreign settlement measure as the last census before Ch´ share of foreigners living in each municipality in 1990 as a percentage of the total number of avez’s foreigners in Venezuela that year. Since 1990 was the last population census before Ch´ 12 election, the location of foreigners at that time should be unaffected by the migration crisis.14 Figure 3 illustrates the geographic distribution of foreign settlement shares. Table D.1 presents results on the nationality of origin of foreigners living in Venezuela in 1990, as well as the host countries of relatives living abroad as reported by Venezuelan households interviewed in the ENCOVI between 2017 and 2021. The overwhelming majority of foreigners living in Venezuela in 1990 had been living in Colombia five years earlier. Moreover, these households reported that most of their relatives abroad were also living in Colombia. As such, our measure of foreign settlements aims to capture a push factor after 2013 that increased the likelihood of migration from municipalities with a higher share of foreigners by facilitating more network support abroad and providing more information on the migration process. In particular, we exploit the fact that most out-migration from Venezuela after 2013 came from municipalities with more foreigners in the 1990s. Section 4 validates this by examining if Venezuelans arriving in Colombia after 2013 were more likely to come from those areas. Tables D.2 and D.3 characterize foreigners living in Venezuela in 1990. In addition to pre- dominantly Colombian nationality, these individuals were mostly employed and had higher education levels and ages than Venezuelans (Table D.2). On average, they also had incomes three times greater than those of the Venezuelan population in 1990 (Table D.3).15 14 Although the census does not permit the direct identification of foreigners, it collects individual informa- tion on whether each person had been living outside of Venezuela in the last five years. We use this measure as a proxy for foreign nationality. 15 Since the main specification follows a difference-in-difference design, any imbalance in the outcome levels between high and low shares of foreign settlements in municipalities does not threaten the empirical strategy, because they are absorbed by the municipality fixed effect in the empirical specification. Nevertheless, we rule out the concern that these characteristics could trend differently by controlling for time trends in socioeconomic baseline characteristics. 13 Figure 3: Geographic Distribution of Foreign Settlements in 1990 Source: Data on administrative boundaries of Venezuela come from Instituto Geogr´ afico de Venezuela (2015), location data for migration posts are from Migraci´ on Colombia (2023), and data on the share of foreign residents are from IPUMS (2023). Legends express the number of foreign residents per municipality as a percentage of the total national foreign residents in Venezuela in 1990 multiplied by 100. 3.2 Time-Level Data We compile data on annual Venezuelan migration outflows to Colombia from 1992 to 2021, combining two data sources. For 1992–2002, we use information from the 1993 and 2005 Colombian population censuses, which record the year when each Venezuelan migrant arrived in Colombia. These data represent the number of Venezuelan nationals living in Colombia 14 each year, as reported by respondents in the retrospective censuses. From 2003 to 2021, our data are based on records from official Colombian migration checkpoints. 3.3 Remote-Sensing Data To create a proxy of Venezuela’s economic growth and inequality, we construct a municipality- year-level longitudinal dataset of satellite night-light density for 1992–2021. This information is processed by the National Oceanic and Atmospheric Administration using images collected by the U.S. Defense Meteorological Satellite Program Operational Linescan System (1992– 2013) and the Visible Infrared Imaging Radiometer Suite remote sensor (2012–2021). These sources are inconsistent due to differences in spatial and radiometric resolution, spectral re- sponses, the spread function of the sensors, local overpass time at night, radiance range, and on-board calibration (Li et al. 2017; Sahoo et al. 2020). Hence, we harmonize the night-light series for 1992–2021 following the process outlined by Li et al. (2020, 2017). Figure 4 shows the spatial distribution of night-light density in 2014 (right after the oil price decline) and in our last period of analysis (2021). It illustrates a stark reduction in night-light density over a remarkably short period of time.16 To measure spatial inequality, we construct a novel spatial Gini index that corresponds to the traditional Gini formula estimated for each municipality and year, using the night light of each parroquia (county) within each municipality as the unit of observation. Tables B.1 and B.2 show that our measure correlates with the traditional measure of income inequality constructed using population censuses for Venezuela (1990) and Colombia (1993 and 2005). We also use other remote-sensing variables as controls that are listed in Table C.1. These capture baseline economic characteristics and their linear trajectories. They include annual tree-cover loss from Global Forest Watch (2023) and data on urban coverage and areas 16 We confirm the validity of this measure as a proxy for economic growth in Table B.2 and test the sensitivity of our main results to controlling for oil production in order to account for concerns related to biases due to brightness from gas flares in the satellite data. 15 covered by water from MODIS Land Cover (2023) (see Appendix B for details). All variables are recorded in 2001, the first year in which they were available. 16 Figure 4: Geographic Distribution of Night-Light Density (a) 2014 (b) 2021 17 Source: Data on Venezuela’s administrative boundaries come from Instituto Geogr´ afico de Venezuela (2015), location data for migration posts are on Colombia (2023), and night-light density data come from Li et al. (2020). from Migraci´ 3.4 Electoral Data Presidential elections. We web-scrape information to construct longitudinal data at the municipal-election-year level for the four presidential elections from 2006 to 2018. The data come from the Consejo Nacional Electoral (2023) and the VE360 (2023) project. Specifically, we analyze three outcomes: turnout (total votes divided by the electoral census), incumbent avez or Maduro as a percentage of the electoral census), and opposition support (votes for Ch´ support (votes for other candidates as a percentage of the electoral census).17 Table B.3 describes the four presidential elections within the study period, the elected can- didate, and the type of election (ordinary or due to an extraordinary event). Appendix B provides detailed information on the outcome construction and data sources. Figures 5 and 6 present results for the geographic distribution of the outcomes in the elections before and after 2013, showing a sharp nationwide reduction in electoral turnout and opposition support. One relevant concern is the possibility of data manipulation. To address this, we search for abnormal data patterns consistent with manipulation, following the analysis first pro- posed by Klimek et al. (2012). The authors propose a test that consists of making two-way scatterplots of election results, illustrating the relationship between electoral turnout and avez support for the incumbent by municipality, which in our case means support for Ch´ and Maduro. Manipulation is evident graphically when there is a clear linear positive trend with concentration along the diagonal, indicating that for each vote added to the electoral turnout, support for the incumbent systematically increases by one vote as well. Elections with no manipulation typically show no obvious correlation between the variables. The results of this exercise, illustrated in Figure B.1, suggest there was no systematic ma- nipulation before 2018. In fact, the results are consistent with international electoral reports 17 We fix the electoral census at the year 2000 to keep the denominator constant in our estimates and facilitate interpretation of the coefficients. Nevertheless, we also verify that our main results are consistent when using the contemporaneous electoral census. 18 from the Carter Center for 1998–2021 (Carter Center, 2024). However, there seems to be ev- idence of potential electoral manipulation in 2018.18 Consequently, we re-estimate our spec- ifications for the 2024 presidential elections in order to validate our results. The opposition directly collected results of the 2024 elections and thus address concerns about manipulation by the incumbents.19 This exercise supports our main findings and we complement it with robustness checks using the results of municipal elections as outcomes. Mayoral elections. We web-scrape data for the 2004–2021 municipal elections, covering five election years in 335 municipalities and producing a total of 1,675 independent election observations. The original data come from the Consejo Nacional Electoral. We examine effects in the same two outcomes: electoral turnout and opposition support. To construct the share of opposition votes, we classify each candidate based on party affiliation and policy program. Appendix B provides details on each candidate’s classification process. 18 These results align with qualitative evidence that suggests the opposition boycotted those elections. 19 The data are publicly available at Resultados con VZLA (2024). 19 Figure 5: Geographic Distribution of Electoral Turnout (Presidential Elections) (a) 2006 (b) 2018 20 Source: Data on Venezuela’s administrative boundaries come from Instituto Geogr´ afico de Venezuela (2015), location data for migration posts are from Migraci´on Colombia (2023), and electoral data are from Consejo Nacional Electoral (2023) and VE360 (2023). Legends express the turnout percentage as a share of the electoral census in Venezuela’s presidential elections between 2006 and 2018. Figure 6: Geographic Distribution of Votes for the Opposition (Presidential Elections) (a) 2006–2013 (b) 2018 21 Source: Data on Venezuela’s administrative boundaries come from Instituto Geogr´ afico de Venezuela (2015), location data for migration posts are on Colombia (2023), and electoral data are web-scraped from Consejo Nacional Electoral (2023) and VE360 (2023). Legends express the from Migraci´ opposition percentage as a share of the electoral census in Venezuela’s presidential elections between 2006 and 2018. 3.5 Additional Survey and Observational Data We use a sample of approximately 3,000 Venezuelan households in Colombia, collected in 2018, to determine migrants’ municipality of origin in Venezuela before migration. This enables us to validate the measure of foreign settlements as a pull factor for Venezuelan migration to Colombia. Although the data are not representative of all Venezuelan migrants for use in the main analysis, they nonetheless are useful for verifying our hypothesis.20 We also use Venezuela’s unique ENCOVI (2014–2021). These annual cross-sectional surveys of independent households are drawn each year, representative at the national and state level, and available for the main urban centers. As previously noted, they include a migration module that characterizes the relatives of Venezuelans abroad between 2017 and 2021. We also construct an annual municipal panel spanning 1993–2024 that combines information on violent events and the presence of organized crime and non-state armed actors. These data come from the ACLED and the GTD. Last, we employ other data from multiple sources to examine the validity of the main results. These include municipal measures on the intensity of the 2019 energy blackouts, coverage of social welfare programs for 2016–2017, and measures of political repression under the avez regime in 2004.21 We also use annual municipal data on private firm expropriations Ch´ by the government and annual data on inflation and global sanctions for our whole period of analysis. All data sources are described in Appendix B. 20 anez et al. (2024), examined the effects of Colombia’s 2018 migrant regu- The survey, carried out by Ib´ larization program and was constructed to be representative of irregular migrants in Colombia at that time, whether or not they were eligible for the program. 21 These data come from Hsieh et al. (2011). 22 4 Empirical Design 4.1 Using Past Settlements of Foreigners as a Migrant Pull Factor Migrants are often said to “vote with their feet,” meaning they leave struggling areas in search of more prosperous and stable regions. These behaviors are even more pronounced in contexts of forced displacement, where migrants must leave due to sudden, intense political and economic crises. Consequently, we cannot simply compare municipalities with higher versus lower outflows of forcibly displaced populations, as this comparison would be biased and likely suggest large negative impacts of forced displacement on economic activity. Moreover, such comparisons are typically restricted by data limitations, especially in origin locations that are undergoing social, political, and economic upheaval. We address these issues by exploiting quasi-exogenous variation from changes in outcomes observed by municipalities with varying levels of foreign settlements in 1990, before and after the onset of the international oil price and migration crises in 2013. Specifically, we estimate the following equation: ymt = γm + αt + β I(t>=2013) × Foreigners Sharem1990 + η (z × αt ) + εmt , (1) ′ z ⊂Xm where m stands for the municipality and t denotes the year. ymt represents the main outcomes of analysis, including economic growth, spatial inequality, and electoral outcomes. I (t >= 2013) represents an indicator variable equal to one after 2013, and Foreigners Sharem1990 is the share of foreign settlements in each municipality in 1990, constructed as the ratio of foreigners in the municipality over total foreigners in the country, multiplied by 100 to ′ facilitate interpretation. Additionally, Xm is a vector of baseline control variables (before the beginning of the crises in 2013). These pre-shock municipal characteristics are interacted with year fixed effects to flexibly control for differential municipal trends. The variables in this 23 vector include baseline night-light density measured in 1992; urban coverage, water bodies, and tree-cover loss observed in 2001; and proxies for political repression collected in 2004.22 γm and αt are municipality and year fixed effects, respectively. Standard errors are clustered at the municipality level to account for potential serial correlation within municipalities. As such, β , our coefficient of interest, measures the change in outcomes when the share of foreign settlements increases by one percent, before and after the onset of the international oil price and migration crises in 2013. To quantify the effects, we exploit Colombian data on the number of Venezuelan migrants per year. In particular, we estimate the same specification in equation (1) but replace the year dummy with a continuous measure of imputed outflows: ymt = γm + αt + β Imputed Outflowsmt + η (z × αt ) + εmt , (2) ′ z ⊂Xm where Imputed Outflowsmt = Outflowst × Foreigners Sharem1990 . (3) Imputed Outflows is our quasi-experimental variation on forced displacement outflows, con- structed as the interaction of total annual outflows of forcibly displaced migrants from Venezuela to Colombia and the share of foreigners living in each municipality in Venezuela in 1990. Foreigners Share is multiplied by 100 to facilitate interpretation. The variable is scaled by the total population of 1990 to approximate the share of individuals leaving each municipality as a percentage of its total population in 1990.23 Our coefficient of interest, β , measures the change in outcomes when the share of imputed outflows increases by one 22 As described in Appendix B, these measures correspond to the municipal count of the individuals who opposed the Ch´ avez regime in an open letter and subsequently received fewer economic opportunities through lower employment and access to the social welfare system. The data come from Hsieh et al. (2011). 23 Even if migrants do not plan to stay in Colombia, they are likely to leave Venezuela through Colombia as the other routes are more difficult (i.e., they need to cross the Amazon Rainforest). 24 percent of the municipal population of 1990. Our estimates are valid as long as there are no time-varying covariates which are not con- trolled for that correlate with the share of foreign settlements and might affect our outcomes of interest. We test the validity of this assumption by verifying that municipalities with a high and low share of foreign settlements in 1990 showed similar time trends in the outcomes of interest before the onset of the crises in 2013. We show that this is the case graphically in Figures 7 and 8.24 In addition, for equations (1) and (2) to identify the effect of forced displacement on the outcomes of interest, the 2013 shock must have significantly increased the number of outflows in municipalities with a more extensive share of foreign settlement. We start by examining this issue in the next section. 4.2 Validity of Imputed Outflows In this subsection, we show that our measure of Imputed Outflows (equation (3)) correctly approximates the variation in actual forced displacement outflows from Venezuela. First, we use data on Venezuelan migrants in Colombia to show that our measure correlates both positively and statistically significantly with the origin municipalities and arrival dates of Venezuelans interviewed in Colombia in 2018 (Panel A of Table D.4).25 We then use household data from Venezuela in which individuals report if they had a member who mi- grated abroad and in which year. Table D.4 shows that our measure of imputed outflows also correlates with the number of households reporting a relative living abroad in the ENCOVI data between 2017 and 2021.26 24 We estimate equation (1) replacing I (t > 2013) with year dummies (the omitted category is 2012). 25 While the Colombian dataset on Venezuelan migrants enables us to validate our measure, it does not allow us to conduct a rigorous econometric analysis due to its limited geographic scope (in particular, these data do not cover individuals who enter Colombia from illegal crossings). 26 As with the other surveys, the ENCOVI allows validation of the measure of imputed outflows but has too few municipalities to be useful for a more rigorous econometric analysis. 25 5 Forced Displacement and Development Table 1 depicts the results of estimating equations (1) and (2) using three outcomes: night- light density (column (1)), the logarithm of night-light density (column (2)), and the spatial Gini measure based on night-light (column 3). Panel A illustrates the estimates of equation (1), and Panels B and C illustrate the estimates of equation (2) with and without con- trols. Consistently, all the estimates show that municipalities with a higher share of foreign settlements saw sharp reductions in night-light density after 2013. Table 1: Forced Displacement and Development Night Light Log(Night Light) Spatial Gini Panel A: Diff-in-diff estimates including controls (1) (2) (3) I(Year>=2013) × Foreigners Share -0.356*** -0.126*** 0.012* (0.120) (0.038) (0.006) Panel B: Imputed outflows, including baseline controls × time trends Imputed Outflows -0.036*** -0.016*** 0.001* (0.009) (0.003) (0.001) Panel C: Imputed outflows, excluding controls Imputed Outflows -0.037*** -0.015*** 0.001* (0.009) (0.004) (0.001) Additional controls for all panels Observations 10,020 9,974 10,020 Dependent Mean 1992 3.77 0.034 0.27 Municipality Fixed Effects ✓ ✓ ✓ Year Fixed Effects ✓ ✓ ✓ Notes: The table illustrates the estimated coefficients of equation (1). Imputed Outflows is defined in equation (3) as the product of the share of foreign settlement in each municipality in 1990 and annual outflows of Venezuelans to Colombia. It is rescaled by the total municipal population of 1990. Controls in the baseline are interacted with time trends and include urban coverage, water bodies, and tree-cover loss for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Standard errors clustered by municipality are in parentheses. ***p<0.01, **p <0.05, *p<0.1. 26 Our preferred results are those in column (2) and Panel A. They suggest that municipalities with a 1 percent higher share of foreign settlements in 1990 experienced a 12.6 percent average reduction in night-light density after 2013, relative to the prior period. These estimates align with those in Panels B and C, which suggest that when imputed outflows increase by 1 percent of the total municipal population, night-light density decreases by approximately 1.6 percent. These are substantial effects. For example, if we consider a municipality with a 10 percent foreign settlement share (which translates into about a 5 percent migration flow of Venezuelans to Colombia; see Table E.22), given that imputed outflows rose by 500 percent between 2013 and 2018, the total reduction in night-light density caused by migration outflows between 2013 and 2018 is 80 percent (=500 × 0.016 × 10). This decrease is equivalent to a contraction of 22 percent (=0.8 × 0.28) in GDP (Henderson et al. 2012). Additionally, municipalities with a 1 percent higher share of foreign settlements in 1990 experienced a 0.012 increment in spatial inequality after 2013, relative to the prior period. These are large effects considering the baseline spatial inequality in 1992 was 0.27. In terms of timing, Figure 7 shows that before 2013, there were no differences in economic development across municipalities with different shares of foreign settlement. In line with the idea that out-migration mostly occurred in municipalities with a large foreign network, we observe an immediate decline in economic conditions after 2013. Moreover, the effects are accentuated after 2017, when out-migration peaked. Our core results remain consistent across a number of robustness tests, including estimating the effects for smaller units, even grids of one square km (Table E.1); correcting the standard errors for spatial auto-correlation by assuming correlations at different distances, including 300, 150, and 50 km (Tables E.3, E.4, and E.6); approximating the municipal variation of imputed outflows with the linear and road inverse distance of each municipality to the main entry points in Colombia (Tables E.8 and E.10); using alternative difference-in-difference estimators, in line with the latest methods literature (Table E.18); and employing only 27 the Colombian foreign shares in Venezuela in 1990 to approximate the variation in foreign settlement share and imputed outflows (Table E.25). Moreover, to validate our main results, we also estimate the effects of mass out-migration using annual ENCOVI data to calculate income per capita and inequality by state and year between 2017 and 2021.27 The results point to a reduction in real total and per capita income at the household level (Table E.12), supporting our main findings. However, the coefficients are smaller given that real income is extremely low for the period for which microdata exist (2017–2021). In fact, income nationwide declined dramatically and hyperinflation caused a complete generalized loss of purchasing power in this period. As noted earlier, by 2021, most individuals (94 percent) had an income below the poverty line (Figure A.3). Finally, we examine if the effects emerge from differential trends in municipalities with higher or lower oil production. The results show that our main estimates do not change when we control for the location of oil fields interacted with time trends (Table E.19).28 27 We cannot use data beginning in 2014 due to the low number of cities covered for those initial years. 28 We use the Global Oil and Gas Features Database, which includes information on the development of global oil and gas infrastructure, from Sabbatino (2018). The data standardize and integrate information on disparate oil and gas infrastructure from over 380 sources worldwide, encompassing more than 4.8 million features for 2018. This control also addresses concerns about the effect of gas flaring on night-light density. 28 Figure 7: Forced Displacement and Development, Event Study (a) Night-Light Density (b) Spatial Inequality Notes: The figure illustrates an event study of the change in night-light density and spatial Gini between 1992 and 2021 for municipalities with varying levels of foreign settlement shares in 1990. Foreign Shares is constructed for each of the 335 municipalities using data from the 1990 population census; it corresponds to the ratio of foreigners living in each municipality to the total foreigners living in Venezuela in 1990. All estimates include fixed effects by year and municipality. Bars illustrate 95 percent confidence intervals, and 2012 is the omitted (reference) year. Standard errors are clustered at the municipality level. 29 5.1 Who is Leaving the Country? The most intuitive explanation of the observed effects of forced displacement on development is that the most educated and productive individuals migrated, which might have greatly af- fected aggregate productivity and long-term institutional development. For instance, highly educated professionals are often better-positioned to challenge the status quo and are typi- cally key capital holders, investors, and primary drivers of innovation and productivity. The ENCOVI enables us to profile the relatives of households that emigrated each year by age, gender, and education (Table 2). In line with the above hypothesis, we find that these migrants were disproportionately adults of working age (15–50 years old), predominantly male, and initially included individuals with higher education levels than those who stayed. However, over time, less-educated individuals also left Venezuela, primarily due to the na- tional economic policies that followed the 2013 oil shock that led to a aggregate deterioration of the economy. This ultimately led to a reduction of 35 percent of Venezuelan population compared to the population observed before the election of Hugo Chavez. As described in Section 2.1, although the Venezuelan migration crisis did not originate principally from a conflict or a climate disaster, it is still considered a forced displacement crisis considering the desperate conditions of Venezuelans, including the inaccessibility of food and health services and the size of the migration flows. 30 Table 2: Sociodemographic Characteristics of Migrants (%) Variables 2017 2018 2019 2021 Age 0-15 6.55 5.98 6.33 5.68 15-29 53.45 53.06 49.11 48.96 30-49 32.00 31.60 39.64 40.32 50+ 4.91 4.16 4.70 4.95 Missing 3.09 5.20 0.22 0.09 Sex Female 49.82 46.16 45.03 42.34 Male 49.64 53.84 54.97 57.66 Missing 0.55 0.00 0.00 0.00 Educational Attainment No Education 0.42 0.61 1.01 1.33 Primary 5.83 6.89 8.85 11.03 Secondary 35.21 45.18 53.00 51.06 Graduate and Post 53.75 43.34 37.15 36.58 Missing 4.79 3.98 0.00 0.00 Country of Migration Colombia 35.82 38.10 41.75 44.55 United States 10.55 4.94 2.48 2.30 Ecuador 6.18 7.15 8.74 9.50 Peru 7.09 18.47 22.17 21.44 Panama 8.00 3.77 1.35 0.77 Chile 9.45 8.45 9.94 9.10 Spain 3.82 1.95 2.88 3.20 Mexico 1.45 1.82 0.44 0.50 Argentina 5.09 3.64 3.02 1.89 Brazil 0.00 0.00 4.33 3.87 Other country 11.64 9.23 2.91 2.88 Missing 0.91 2.47 0.00 0.00 Notes: The panels depicting age and sex represent the proportions of migrants categorized by age groups and gender, respectively. Panel C considers only migrants over the age of 18, as the majority of this demographic has completed secondary education, and Panel D provides a tabulation of migrants according to their country of migration. Source: ENCOVI. Using ENCOVI data for 2017-2021, we also looked at the characteristics of stayers. We find a sharp decline in the share of Venezuelans with higher education (Table A.1) and a major shift from professional and technical occupations to elementary occupations. Additionally, 31 nearly every economic sector contracted between 2017 and 2021 except for agriculture and other services, indicating a shift from formal and secondary sectors to informal and primary ones (Table 3). Table 4 confirms this, depicting a steep decline in salaried work offset by a rise in self-employment. Table 3: Characterizing the Local Labor Force Remaining in Venezuela, 2017–2021 (%) Variables 2017 2018 2019 2021 Type of Occupation Managers 2.37 2.00 1.52 1.59 Professionals and technicians 27.06 26.62 21.52 18.65 Machine operators and related trades workers 13.54 12.36 9.04 8.70 Service and sales workers 25.54 26.14 24.05 21.71 Elementary occupations 25.68 27.49 41.36 46.90 Armed forces occupations 1.98 2.15 2.51 2.45 Missing 3.83 3.24 0.00 0.01 Economic Sector Agriculture, forestry and fishing 6.68 8.09 7.86 12.04 Mining 1.41 1.76 0.71 0.87 Manufacturing 4.79 3.71 2.73 1.88 Electricity, gas and water 1.26 1.29 1.46 1.51 Construction 5.81 5.34 4.64 4.81 Wholesale and retail 22.33 23.92 17.43 19.53 Transportation and communications 11.56 10.03 14.03 11.59 Financial, real estate, and scientific services 6.76 4.07 5.73 5.34 Public administration, education and health 36.77 27.06 19.98 17.20 Other services (*) - 12.65 25.39 25.20 Missing 2.62 2.10 0.04 0.02 Notes: Panel B presents a tabulation of the employed population aged 15–65, categorized by the main occupation classification according to a modified version of the A10 aggregation in the International Stan- dard Industrial Classification (ISIC). The A10 aggregation represents an advanced taxonomic classification, consisting of amalgamations of divisions from the Fourth Revision of the International Standard Industrial Classification of All Economic Activities (ISIC Rev. 4). The category “Other services (*)” includes repairs, cleaning, hairdressing, funeral, and domestic services. Additionally, in 2017, the questionnaire introduced the category “other services” within the sectors of public administration, education, and health. Similarly, Panel A shows the classification of main occupations based on aggregations of the major group classifications from the International Standard Classification of Occupations (ISCO-08). We have aggregated the categories of professionals, technicians, associate professionals, and clerical support workers into “professionals and tech- nicians.” Additionally, craft and related trades workers, plant and machine operators, and assemblers are combined into “machine operators and related trades workers”; and skilled agricultural, forestry, and fishery workers are consolidated with elementary occupations into the single category “elementary occupations.” Source: ENCOVI. 32 Table 4: Characteristics of Economically Active Population Remaining in Venezuela (%) Variables 2014 2015 2016 2017 2018 2019 2021 Type of Employment Employee (public sector) 34.29 27.44 27.16 31.87 31.40 24.60 20.49 Employee (private company) 25.16 32.44 26.77 22.11 22.53 22.10 20.11 Employer 3.12 3.93 3.03 2.87 1.85 2.82 2.85 Self-employed worker 28.92 29.53 37.58 36.21 39.27 44.95 51.37 Member of cooperatives 1.12 1.27 0.94 1.74 0.43 0.73 0.59 Paid / unpaid family 1.77 2.35 1.71 1.65 1.42 2.93 2.93 Domestic service 0.97 1.97 0.78 0.84 1.03 1.87 1.65 Missing 4.65 1.07 2.03 2.71 2.07 0.00 0.00 Notes: The table presents a detailed tabulation of the employed population aged 15–65, categorized by employment type. Source: ENCOVI. 5.1.1 Remittances A key question concerns the remittances sent by Venezuelans abroad to their families at home (Amuedo-Dorantes and Pozo 2006; Giuliano and Ruiz-Arranz 2009; Portes 2009; Ambler et al. 2015). In Table 5, we find that remittances have increased in municipalities affected by mass migration. While remittances may help partially offset the economic loss of the most productive members of the labor force, broader effects depend on how recipients use these funds. Using data from the 2019 and 2021 ENCOVI, we analyze how households used remittances. These are the only two nationally representative surveys that categorize such spending. As illustrated in Figure E.1, the vast majority of households used remittances for subsis- tence needs such as food and housing rather than for investment or business activities. Consequently, these remittances were less likely to stimulate long-term economic growth. 33 Table 5: Forced Displacement and Household Remittances Probability to Receive Remittances Panel A: Remittances (1) Imputed Outflows 0.00069*** (0.00017) Observations (State and Year) 120,014 Notes: Imputed Outflows is defined as explained in equation (2). Controls in baseline interacted with time trends include urban coverage, water bodies, and tree-cover loss for year 2001; night-light density for 1992; social welfare program (Carnet de la Patria ) participants 2016–2017; number of parroquias (counties) with rationed energy at the municipality level (April 2019) and blackout intensity (March 2019); and number of enterprises acquired by the Venezuelan state. Controls from the ENCOVI include gender of household head and members, age and marital status, and number of household members. To calculate the probability of receiving remittances we create a dummy variable equal to 1 if the household has received remittances between 2017-2021. ***p<0.01, **p <0.05, *p<0.1. 5.1.2 Forced Displacement and Human Capital Investments As forced displacement intensifies and economic opportunities become scarce, investments in human capital at origin locations might be disproportionately discouraged. This could be due to changes in the returns to education locally, to the substitution of subsistence activities for education, or to the desire to invest only in education abroad. We explore this possibility by using data from the ENCOVI and assessing the role of imputed outflows on the average years of education for individuals older than 18 years, and school attendance for children and adolescents.29 The results in Table 6 support the idea that areas with higher imputed outflows also experienced lower education investments and school attendance. Moreover, in Table 7 we document that these effects are concentrated on children ages 12-14 years old who might contribute economically to household’s income or even join illicit crime groups as documented by Sviatschi (2022). Moreover, these results are not driven by selective out- migration of the individuals who are the most educated out of Venezuela as the results are only observed for individuals on schooling age during the mass forced migration shock. In fact, we do not observe the same effects of forced migration on school attendance for those individuals younger than 6 or older than 17 years of age in the ENCOVI sample in 2013 29 We cannot estimate equation (1) because the ENCOVI data are only available after 2014. 34 (Table E.24). Table 6: Forced Displacement and Reduced Human Capital Investments Years of Education School Attendance Panel A: Education Outcomes (1) (2) Imputed Outflows -0.07435*** -0.00120*** (0.00722) (0.00025) Observations (State and Year) 73,997 46,813 Dependent Mean 2014 7.93 0.65 Notes: Imputed Outflows is defined as explained in equation (2). Controls in baseline interacted with time trends include urban coverage, water bodies, and tree-cover loss for year 2001; night-light density for 1992; social welfare program (Carnet de la Patria ) participants 2016–2017; number of parroquias (counties) with rationed energy at the municipality level (April 2019) and blackout intensity (March 2019); and number of enterprises acquired by the Venezuelan state. Controls from the ENCOVI include gender of household head and members, age and marital status, and number of household members. For column (1), the estimates include only individuals over 23 years old, and for column (2), they only include individuals aged 6–17 in 2013. Bootstrap standard errors are in parentheses. ***p<0.01, **p <0.05, *p<0.1. Table 7: Forced Displacement and School Attendance by Schooling Age 6-14 6-11 12-14 6-17 Panel A: School Attendance by Schooling Age (1) (2) (3) (4) Imputed outflows -0.00078** 0.00055 -0.00344*** -0.00072** (0.00035) (0.00044) (0.00055) (0.00030) Observations (State and Year) 23,118 15,715 7,403 30,058 Notes: Imputed Outflows is defined as explained in equation (2). Controls in baseline interacted with time trends include urban coverage, water bodies, and tree-cover loss for year 2001; night-light density for 1992; social welfare program (Carnet de la Patria ) participants 2016–2017; number of parroquias (counties) with rationed energy at the municipality level (April 2019) and blackout intensity (March 2019); and number of enterprises acquired by the Venezuelan state. Controls from the ENCOVI include gender of household head and members, age and marital status, and number of household members. For column (1), the estimates included only individuals between 6 and 14 years old in 2013, column (2) only included individuals aged 6-11 in 2013, column (3) only included individuals aged 12-14 in 2013, and column (4) only included individuals aged 6-17 in 2013. Bootstrap standard errors are in parentheses. ***p<0.01, **p <0.05, *p<0.1. 5.2 Ruling Out the Influence of Government Actions on Foreigner- Dense Municipalities One relevant concern is that our results might be affected by governmental actions that could disproportionately target municipalities with more or less foreign settlement shares in 1990. 35 For example, the government might have used social programs to gain more political support or disproportionately alienated the private sector in municipalities with more foreigners in 1990. Three salient economic policies and events took place during our analysis period: the Venezuelan social welfare program (centralized through an ID called Carnet de la Patria ), mass energy blackouts, and expropriations from the private sector. We collect approximate municipal temporal variation measures for each of these from multiple sources and control for their variation in Tables E.15 and E.16.30 Our main results remain unchanged, suggesting that these policies and events did not alter the principal effects of out-migration. 6 Forced Displacement and Autocratic Leadership In this section, we study the conditions that help autocratic leaders endure despite economic and social crises. We show that in the medium term, forced displacement aids the perpetua- tion of such rulers by weakening political opposition and abetting the expansion of organized crime and illicit economic activities. These two effects further dampen pressure for economic and social change. 6.1 Weakening the Political Opposition Forced migrants are likely to oppose governments they perceive as responsible for the crises they are fleeing. However, their departure may deplete opposition support and diminish popular will to fight for change. We explore this idea by examining the effects of forced displacement on electoral turnout and support for the political opposition, using them as outcomes in the specifications in equations (1) and (2). Presidential turnout is defined as the total votes in each election divided by the electoral turnout of 2000 (the earliest available before the 2013 oil price shock). We keep the electoral turnout fixed as it may be affected by migration flows; hence, we only measure 30 See Appendix B for a description of all data sources. 36 changes in the numerator. Opposition support is defined as the ratio of opposition votes avez or Maduro) to the electoral turnout of 2000.31 (votes for candidates besides Ch´ Table 8: Forced Displacement and Presidential Electoral Outcomes Turnout Opposition Panel A: Diff-in-diff estimates including controls (1) (2) I(Year>=2013)× Foreigners Share -4.471*** -2.624*** (1.416) (0.705) Panel B: Imputed outflows, including baseline controls × time trends Imputed Outflows -0.663*** -0.358*** (0.158) (0.073) Panel C: Imputed outflows, excluding controls Imputed Outflows -0.649*** -0.351*** (0.158) (0.076) Additional controls for all panels Observations 1,324 1,324 Dependent mean 2006 31.4 10.2 All controls ✓ ✓ Municipality FE ✓ ✓ Year FE ✓ ✓ Notes: The table illustrates the estimated coefficients of equation (1). Turnout in column (1) is defined as the total votes for each presidential election held in Venezuela between 2006 and 2018, divided by the electoral census of 2000. In column (2), opposition is the total votes of unofficial parties divided by the electoral census of 2000. Imputed Outflows is defined in equation (3) as the product of shares of foreign settlement in each municipality in 1990 and annual outflows of Venezuelans to Colombia. Controls in baseline are interacted with time trends and include urban coverage, water bodies, and tree-cover loss for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador, and standard errors clustered by municipality are in parentheses. ***p<0.01, **p <0.05, *p<0.1. Table 8 presents results. The findings in Panel A suggest that municipalities with a 1 percent higher share of foreign settlements in 1990 experienced an average drop of 4.5 percentage 31 2000 is the first year available to conduct this exercise. Forced displacement was negligible at this time as oil prices in Venezuela were surging. No estimates change if we use the contemporaneous electoral census. However, in our view, leaving the denominator fixed in 2000 makes interpretation easier and provides a lower bound of the estimates. 37 points in electoral turnout and 2.62 percentage points in opposition support after 2013, relative to the previous time frame. Similarly, the results from Panels B and C suggest that when imputed migration outflows rose by 1 percent of the population in 1990, presidential turnout and opposition support fell by about 0.66 and 0.34 percentage points, respectively. Due to concerns about manipulation in the 2018 elections, we also estimate the effects of mass forced migration using the last presidential elections of 2024, for which we have results for over 83.5 percent of the country. The opposition parties collected and published these results, which should attenuate these concerns.32 The results don’t change when we add data from these elections to our main estimates (Table E.13). We also find consistent effects for mayoral elections (see Table E.14). Although the co- efficients indicate smaller effects, they are negative and statistically significant. This is a remarkable consistency test since these estimates evaluate forced displacement outflows over five years of elections in 335 municipalities, totaling 1,675 independent municipal elections. Overall, our results consistently suggest that mass forced migration lessened both opposition to the incumbent regime and political turnout, both of which are key factors for social change. 32 The data come from Resultados con VZLA (2024). 38 Figure 8: Forced Displacement and Presidential Electoral Outcomes, Event Study (a) Electoral Turnout (b) Opposition Support Notes: The figure illustrates an event study of the change in turnout and opposition support in the presi- dential elections (measured in votes for candidates other than Maduro or Ch´ avez over the electoral census of 2000) between 2006 and 2021 for municipalities with varying levels of foreign settlement shares in 1990. Foreign Shares is constructed for each of the 335 municipalities using data from the 1990 Venezuelan pop- ulation census, and it corresponds to the ratio of foreigners living in each municipality to the total number of foreigners living in Venezuela in 1990. All estimates include fixed effects by year and municipality. Bars illustrate 95 percent confidence intervals, and 2012 is the omitted (reference) year. Standard errors are clustered at the municipality level. 39 6.2 The Expansion of Organized Crime and Illicit Economic Ac- tivities Migration typically advances connections between individuals in origin countries, those who leave, and those in destination countries. Previous work has shown that migration networks can promote knowledge diffusion, trade, and foreign direct investment globally (Javorcik ezina 2018). However, these positive effects may be absent in origin et al. 2011; Parsons and V´ locations with weak institutions and rule of law. Instead, forced displacement may expedite connections for organized criminal networks at both origin and destination locations (Kapur 2014). This occurs, for example, when criminals exploit and recruit migrants by capitalizing on their vulnerability or relying on them to help fund, inform, and expand illegal activities. In fact, abundant qualitative evidence suggests that human trafficking is a significant crim- inal enterprise in these contexts, with forced migrants—especially women and children— becoming victims (Insight Crime, 2023d, 2021b,a). For instance, one article suggests that “migrants using irregular border crossings also represent a source of income for armed groups and predatory criminals, who extort the migrants as they cross their territory and, in some cases, also rob or kidnap them or force them into roles such as human couriers for drug trafficking. Human traffickers have also capitalized on the crisis, tailoring their recruitment of Venezuelan women and girls by offering false offers of jobs, scholarships, or even religious charity as bait for what ends up being coerced sexual exploitation.” (Insight Crime 2023c). Journalistic reports also report that forced displacement outflows go hand in hand with the expansion of drug-trafficking groups from Colombia to Venezuela. Oftentimes, this expansion comes from employing forced migrants in the lower ranks of these organizations or from charging them extortion fees en route (Insight Crime, 2023a,b, 2019). For example, one report states that “(...) Venezuelans have to choose between hunger or joining the ranks of organized crime groups, which is helping them strengthen and reorganize their criminal structures while also facilitating the spread of their illegal activities into Venezuela.” (Insight 40 Crime, 2018). We examine these hypotheses using longitudinal municipal-level data on violent and crime events by actor type, sourced from the GTD (1992–2021) and the ACLED (2018–2024). Specifically, we examine the effect of mass forced migration outflows on the growth of Colom- bian and Venezuelan organized criminal groups and non-state armed actors. Table 9 presents the results.33 The findings in Panel A largely support these hypotheses. They imply that mass forced migration is associated with an increased number of events by the largest non- state armed actors from Colombia, including the Revolutionary Armed Forces of Colombia (FARC) and the National Liberation Army (ELN), both of which are largely involved in drug trafficking. The findings also indicate an increase in violent events linked to Venezue- lan organized criminal groups, including gangs such as El Tren de Aragua, which is known for its involvement in human trafficking.34 Table 9: Forced Displacement and the Expansion of Organized Crime FARC ELN Organized Crime Total Events Associated with (1) (2) (3) Imputed Outflows 0.002** 0.008** 0.052** (0.001) (0.004) (0.022) Observations 11,022 11,022 11,022 Dependent mean baseline 0.009 0.003 2.10 Exclude outlier municipality (Libertador) ✓ ✓ ✓ Municipality FE ✓ ✓ ✓ Year FE ✓ ✓ ✓ Notes: Column (1) represents the total number of conflict events involving the Revolutionary Armed Forces of Colombia (FARC) in Venezuela from 1992 to 2024. Column (2) shows the total conflict events associated with the Colombian National Liberation Army (ELN) during the same period. Column (3) captures the total conflict events carried out by non-terrorist but criminal armed groups—such as gangs, colectivos, sindicatos, and drug-trafficking cartels—between 2018 and 2024. Standard errors clustered by municipality are in parentheses. ***p<0.01, **p <0.05, *p<0.1. The validity of the estimated effects are further illustrated in Figure 9, where all data are 33 We only present estimates using imputed outflows as the source of quasi-exogenous variation, as detailed actor-level disaggregation in the data is only available from 2018 onward to enable this analysis. 34 We also examine the effects on irregular armed groups—such as small street gangs and criminal organi- zations not related to drugs, human smuggling, or the regime—and find no effects, demonstrating that the results are not due to an increase in overall crime. 41 aggregated to have a longer time series.35 The figure shows a spike in the likelihood of an event being associated with non-state actors in municipalities with higher concentrations of foreign settlements after 2017, coinciding with the peak of Venezuelan forced displacement to Colombia. Although the first waves of migrants after 2013 were positively selected, over time hyperinflation and the collapse of the social safety net prompted more vulnerable people to leave. Thus, the resulting mass displacement created conditions that allowed these groups to extend their influence in areas with higher concentrations of foreign settlements, especially after 2017, when migrants were highly vulnerable and less educated. In fact, there is ample qualitative evidence that highlights how the rise of the Venezuelan gang Tren de Aragua coincided with the peak of the Venezuelan economic crisis in 2018, which pushed millions of vulnerable Venezuelans to migrate. Figure 9: Indicator Variable for Non-State Actors Notes: The figure illustrates an event study of the change in the probability that an event associated with non-state actors occurred between 1993 and 2024 with varying levels of foreign settlement shares in 1990. Foreign Shares is constructed for each of the 335 municipalities using data from the 1990 Venezuelan population census, and it corresponds to the ratio of foreigners living in each municipality to the total number of foreigners living in Venezuela in 1990. All estimates include fixed effects by year and municipality. Bars illustrate 95 percent confidence intervals, and 2012 is the omitted (reference) year. Standard errors are clustered at the municipality level. 35 Before 2018, the data excluded information on gangs to allow for their separation in analysis. 42 6.3 Mediating Effects of these Channels on Development In this subsection we examine the roles of weak political opposition and strong organized criminal groups to explain the negative effects of forced migration on development outcomes. To do this, we apply the methodology introduced by Acharya et al. (2016), which estimates the average controlled direct effect (ACDE) of a treatment. The ACDE is the effect of forced displacement on development after partialing out the effect of forced displacement on political opposition and organized crime. This exercise is estimated by partialing out the effect of political opposition and organized crime and then estimating the ACDE by regressing the de-mediated night-light density on imputed outflows. This is done through a two-stage model as follows: Ymt = δ0 + δ1 Imputed Outflowsmt + δ2 Oppositionmt + δ3 Organized Crimemt + η (z × αt ) + ϵmt , ′ z ⊂Xm (4) ˆmt = γ0 + γ1 Imputed Outflowsmt + Y η (z × αt ) + υmt , (5) ′ z ⊂Xm which follows the same notation as the one used in our main analysis. In the first stage, Ymt denotes night-light density, Oppositionmt denotes opposition support (estimated as the share of votes for the opposition), and Organized Crimemt is an indicator variable for the presence ˆmt is the de-mediated night-light density of organized criminal groups. In the second stage, Y estimated as Y ˆ ˆmt = [Ymt − (δ ˆ 2 Oppositionmt + δ3 Organized Crimemt )], and υmt is the error term estimated through bootstrapping. Figure 10 presents the results of this exercise, which suggest that the standard estimated effect of imputed outflows on night-light density falls from –0.037 to –0.022 when the variation explained by the decline in political opposition and increase in organized crime is accounted 43 for.36 These two channels therefore explain approximately one-third of the effect of forced displacement on development. Figure 10: Mediation Analysis: The Role of Political Opposition and Organized Crime Notes: This figure presents coefficients for the two mediators, political opposition and organized crime, represented by squares. The triangle indicates the direct effect of imputed migration outflows on night-light density, excluding controls. Standard errors are clustered at the municipality level. 7 Discussion This paper examines the impact of forced displacement on the development of origin coun- tries, a topic historically constrained by limited quality data because these nations are often in crisis. To overcome these challenges in the case of Venezuela, we leverage satellite data, web-scrape online data, employ observational data, and use unique individual surveys. To identify causal changes resulting from mass forced out-migration, we compare municipalities with a higher share of foreign settlements before and after the onset of the 2013 oil price 36 Table E.23 presents the point estimates. 44 shock. Our findings show that these areas had a disproportionate number of Colombian for- eigners who provided crucial information and network support abroad, enabling Venezuelans to migrate more easily to Colombia, where the majority of those migrants now live. Our first key finding is that municipalities with a higher proportion of foreign settlements, which we also document as having higher outflows of forced displacement, experienced sig- nificant reductions in economic growth and increased inequality after 2013, relative to other areas. We find that forced migrants tend to be positively selected and disproportionately represent the productive labor force. Moreover, we document that Venezuela experienced a dramatic shift from formal to informal and illicit activities. How does a regime stay in power amid such appalling economic conditions? We document two main ways in which forced migration perpetuates autocratic leadership. First, we find that forced migrants disproportionately backed the opposition and thus their departure re- duced voter turnout and diluted opposition strength. Second, forced displacement created fertile ground for growth in organized crime and illicit income sources, particularly for those involved in drug and human trafficking. Both of these dynamics reinforced the political status quo and deflated pressure for socioeconomic reform in the medium term. Our results underscore the significant impact of forced displacement on the countries where these flows originate and on their political leadership. Our findings suggest that political leaders and ruling parties in these contexts may actually benefit from these migration flows and thus have little incentive to restrict or prevent them. However, since forced displacement depletes factors of production, it comes at a grave cost: steep declines in economic growth. Historically, many governments have promoted the emigration of high-profile individuals and larger groups as a way to relieve political tensions. Examples include Japan, which dispro- portionately encouraged emigration from a handful of southwest regions in the nineteenth century (Endoh 2010), the Soviet Union’s exile of Aleksandr Solzhenitsyn, China’s exile of Chen Guangcheng, and Cuba’s and Zimbabwe’s strategy of “venting disgruntled groups” 45 through emigration (Kapur 2014). Another example is Russia, where 1.1 million individuals left in the 1990s and another 1.25 million left in the 2000s, reflecting the alienation of profes- sionals and entrepreneurs. Yet, the “Kremlin couldn’t care less if the most talented, the most active Russians are emigrating, because their exodus lifts the social and political tension in the country and weakens the opposition” (Loiko, 2011). In the words of Hirschman (1970), “exit has [been] shown to drive out voice.” Our study documents the price of this strategy: overall economic growth in these countries is likely to be notably lower. As this paper reports, forced displacement flows have the power to transform cultural values, economic systems, and political landscapes in both origin and destination countries. As more time passes and additional data become available, future research can explore the medium- to long-term impacts of these flows and offer valuable insights into their broader consequences for the development paths of affected countries. References Abadie, A. (2005). Semiparametric Difference-in-Differences Estimators. The Review of Economic Studies, 72(1):1–19. Abramitzky, R., Boustan, L. P., and Eriksson, K. (2014). A nation of immigrants: Assimila- tion and economic outcomes in the age of mass migration. Journal of Political Economy, 122(3):467–506. Acharya, A., Blackwell, M., and Sen, M. (2016). Explaining causal findings without bias: Detecting and assessing direct effects. American Political Science Review, 110(3):512–529. Ambler, K., Aycinena, D., and Yang, D. (2015). Channeling remittances to education: A field experiment among migrants from El Salvador. American Economic Journal: Applied Economics, 7(2):207–232. 46 Amuedo-Dorantes, C. and Pozo, S. (2006). Migration, remittances, and male and female employment patterns. American Economic Review, 96(2):222–226. Anelli, M., Basso, G., Ippedico, G., and Peri, G. (2023). Emigration and entrepreneurial drain. American Economic Journal: Applied Economics, 15(2):218–252. Batista, C., Lacuesta, A., and Vicente, P. C. (2012). Testing the ‘brain gain’ hypothesis: Micro evidence from Cape Verde. Journal of Development Economics, 97(1):32–45. ander, N., and Zhuravskaya, E. (2020). Forced Becker, S. O., Grosfeld, I., Grosjean, P., Voigtl¨ migration and human capital: Evidence from post-WWII population transfers. American Economic Review, 110(5):1430–1463. Beine, M., Docquier, F., and Rapoport, H. (2008). Brain drain and human capital formation in developing countries: Winners and losers. The Economic Journal, 118(528):631–652. Borjas, G. J. (2014). Immigration economics. Harvard University Press. Card, D. (2001). Immigrant inflows, native outflows, and the local labor market impacts of higher immigration. Journal of Labor Economics, 19(1):22–64. Carter Center (2024). Venezuela: Monitoring elections. https://www.cartercenter.org/ countries/venezuela.html]. Chiovelli, G., Michalopoulos, S., Papaioannou, E., and Sequeira, S. (2021). Forced displace- ment and human capital: Evidence from separated siblings. National Bureau of Economic Research, (w29589). Colombian Spatial Data Infrastructure (2023). Colombia shapefiles. Consejo Nacional Electoral (2023). Resultados electorales 1998-2018. on de petr´ Datosmacro (2024). Venezuela - Producci´ oleo. 47 Docquier, F., Lodigiani, E., Rapoport, H., and Schiff, M. (2016). Emigration and democracy. Journal of Development Economics, 120:209–223. Docquier, F. and Rapoport, H. (2012). Globalization, brain drain, and development. Journal of Economic Literature, 50(3):681–730. onberg, U., and Stuhler, J. (2017). Labor supply shocks, native wages, and Dustmann, C., Sch¨ the adjustment of local employment. The Quarterly Journal of Economics, 132(1):435– 483. Dustmann, C., Vasiljeva, K., and Piil Damm, A. (2019). Refugee migration and electoral outcomes. The Review of Economic Studies, 86(5):2035–2091. Endoh, T. (2010). Exporting Japan: Politics of Emigration to Latin America. University of Illinois Press. Foged, M. and Peri, G. (2016). Immigrants’ effect on native workers: New analysis on longitudinal data. American Economic Journal: Applied Economics, 8(2):1–34. Friedl, M. A., Sulla-Menashe, D., Tan, B., Schneider, A., Ramankutty, N., Sibley, A., and Huang, X. (2010). Modis collection 5 global land cover: Algorithm refinements and char- acterization of new datasets. Remote Sensing of Environment, 114(1):168–182. Gibson, J. and McKenzie, D. (2011). Eight questions about brain drain. Journal of Economic Perspectives, 25(3):107–128. Gibson, J., Olivia, S., Boe-Gibson, G., and Li, C. (2021). Which night lights data should we use in economics, and where? Journal of Development Economics, 149:102602. Giuliano, P. and Ruiz-Arranz, M. (2009). Remittances, financial development, and growth. Journal of Development Economics, 90(1):144–152. Global Forest Watch (2023). Annual tree cover loss 2001-2021. Global Sanctions Database (2023). The global sanctions data base (gsdb) 2023. 48 alez, F., Mart´ Gonz´ noz, P., and Prem, M. (2023). Higher education and mor- ınez, L. R., Mu˜ tality: Legacies of an authoritarian college contraction. Journal of the European Economic Association, page jvad066. ˇ and Roesel, F. (2024). Forced Migration, Staying Minorities, Grossmann, J., Jurajda, S., and New Societies: Evidence from Postwar Czechoslovakia. American Journal of Political Science, 68(2):751–766. Hanson, G. H., Kerr, W. R., and Turner, S. (2018). High-skilled migration to the United States and its economic consequences. University of Chicago Press. ıguez, F. (2014). Why did Venezuelan growth collapse? Venezuela Hausmann, R. and Rodr´ before Chavez: Anatomy of an economic collapse, pages 30–56. Henderson, J. V., Storeygard, A., and Weil, D. N. (2012). Measuring economic growth from outer space. American Economic Review, 102(2):994–1028. Hirschman, A. O. (1970). Exit, voice, and loyalty: Responses to decline in firms, organiza- tions, and states. Harvard University Press. Hirschman, A. O. (1978). Exit, voice, and the state. World Politics, 31(1):90–107. Holland, A., Peters, M. E., and Zhou, Y.-Y. (2024). Left out: How political ideology affects support for migrants in Colombia. The Journal of Politics, 86(4):1291–1303. Hsieh, C.-T., Miguel, E., Ortega, D., and Rodriguez, F. (2011). The price of political opposition: evidence from Venezuela’s Maisanta. American Economic Journal: Applied Economics, 3(2):196–214. anez, A. M., Moya, A., Ortega, M. A., Rozo, S. V., and Urbina, M. J. (2024). Life out Ib´ of the shadows: The impacts of regularization program on the lives of forced migrants. Journal of European Economic Association, Forthcoming. 49 Insight Crime (2018). Colombia’s ELN, Ex-FARC Mafia Recruiting Hungry Venezuela Mi- grants. Insight Crime. Insight Crime (2019). Colombia’s Armed Groups Prey on Venezuelan Migrants. Insight Crime. Insight Crime (2021a). Human Trafficking Accompanies Illegal Mining in Venezuela’s Orinoco. Insight Crime. Insight Crime (2021b). Venezuela’s Other Plight: Sex Trafficking in Trinidad and Tobago. Insight Crime. a Cartel: Drug Trafficking and Political Power in Insight Crime (2022a). The Paraguan´ Venezuela. Insight Crime. Insight Crime (2022b). Venezuela’s cocaine revolution. Insight Crime. Insight Crime (2023a). Dutch Caribbean Remains a High-Risk Route for Venezuelan Mi- grants. Insight Crime. Insight Crime (2023b). Venezuelan migrants remain easy prey for organized crime. Insight Crime. Insight Crime (2023c). Venezuelan security policy: The criminal exploitation of the migrant crisis. Insight Crime. Insight Crime (2023d). Venezuela’s corrupt officials key to human trafficking networks. Insight Crime. afico de Venezuela (2015). Venezuelan roads shapefiles. Instituto Geogr´ International Monetary Fund (2024). World Economic Outlook 2024. on y vivienda 1990 - IPUMS subset. IPUMS (2023). XII censo general de poblaci´ 50 ¨ Javorcik, B. S., Ozden, ¸ ., Spatareanu, M., and Neagu, C. (2011). Migrant networks and C foreign direct investment. Journal of Development Economics, 94(2):231–241. Kapur, D. (2014). Political effects of international migration. Annual Review of Political Science, 17(1):479–502. Karadja, M. and Prawitz, E. (2019). Exit, voice, and political change: Evidence from Swedish mass migration to the United States. Journal of Political Economy, 127(4):1864–1925. Klimek, P., Yegorov, Y., Hanel, R., and Thurner, S. (2012). Statistical detection of systematic election irregularities. Proceedings of the National Academy of Sciences, 109(41):16469–16473. Li, X., Li, D., Xu, H., and Wu, C. (2017). Intercalibration between DMSP/OLS and VIIRS night-time light images to evaluate city light dynamics of Syria’s major human settlement during Syrian Civil War. International Journal of Remote Sensing, 38(21):5934–5951. Li, X., Zhou, Y., Zhao, M., and Zhao, X. (2020). A harmonized global nighttime light dataset 1992–2018. Scientific Data, 7(1):168. Loiko, S. (2011). Russians are leaving the country in droves. Los Angeles Times. Martinez, L. R. (2022). How much should we trust the dictator’s gdp growth estimates? Journal of Political Economy, 130(10):2731–2769. Melki, M., Rapoport, H., Spolaore, E., and Wacziarg, R. (2024). Cultural remittances and modern fertility. Technical report, National Bureau of Economic Research. on Colombia (2023). Migration flow records 2012-2022. Migraci´ MODIS Land Cover (2023). Modis land cover type yearly global 500m. Morales-Arilla, J. (2021). Autocrats in crisis mode: Strategic favoritism during eco- nomic shocks. Technical report, Mimeo. https://scholar. harvard. edu/josemorales- arilla/publications/% E2 . . . . 51 Morales-Arilla, J. and Traettino, S. (2023). Vulnerability, fear of discrimination and clien- telism: Organizer identity and the backfire of solidarity. Working Paper. National Survey on Living Conditions (2021). Encovi 2014 - 2021. OCHA (2023). Venezuelan shapefiles at administrative levels. ezina, P.-L. (2018). Migrant networks and trade: The vietnamese boat Parsons, C. and V´ people as a natural experiment. The Economic Journal, 128(612):F210–F234. Portes, L. S. V. (2009). Remittances, poverty and inequality. Journal of Economic Devel- opment, 34(1):127. on interagencial para refugiados y migrantes. R4V (2024). La plataforma de coordinaci´ Resultados con VZLA (2024). Resultados Elecciones Presidenciales - Venezuela 2024. Rios-Avila, F., Sant’Anna, P. H. C., and Naqvi, A. (2021). DRDID: Doubly Robust Difference-in-Differences Estimators for Stata. https://www.stata.com/symposiums/ economics21/slides/Econ21_Rios-Avila.pdf. SSC Symposium Presentation. Sabbatino, M. (2018). Global oil gas features database. Sahoo, S., Gupta, P. K., and Srivastav, S. K. (2020). Inter-calibration of DMSP-OLS and SNPP-VIIRS-DNB annual nighttime light composites using machine learning. GIScience Remote Sensing, 57(8):1144–1165. Sant’Anna, P. H. and Zhao, J. (2020). Doubly robust difference-in-differences estimators. Journal of Econometrics, 219(1):101–122. Spilimbergo, A. (2009). Democracy and foreign education. American Economic Review, 99(1):528–543. Steinmayr, A. (2021). Contact versus exposure: Refugee presence and voting for the far right. Review of Economics and Statistics, 103(2):310–327. 52 Sviatschi, M. M. (2022). Making a narco: Childhood exposure to illegal labor markets and criminal life paths. Econometrica, 90(4):1835–1878. UNHCR (2024). Global trends report, 2024. UNHCR. Van Kerm, P. (2020). Sgini: Stata module to compute generalized gini and concentration coefficients, gini correlations and fractional ranks. VE360 (2023). Elecciones 2000 - 2015. Technical report, Venezuela 360. Vendata (2024). Base de datos: Empresas en Propiedad del Estado Venezolano. avez Weisbrot, M. and Sandoval, L. (2008). Update: the Venezuelan economy in the Ch´ years. Center for Economic and Policy Research, 2. World Bank (2024). World Bank Commodity Price Data. Zhao, M., Zhou, Y., Li, X., Zhou, C., Cheng, W., Li, M., and Huang, K. (2019). Building a series of consistent night-time light data (1992–2018) in Southeast Asia by integrating DMSP-OLS and NPP-VIIRS. IEEE Transactions on Geoscience and Remote Sensing, 58(3):1843–1856. 53 Part Appendix Table of Contents A Appendix A: Characterizing the Venezuelan Crisis 56 A.1 Characterizing the Migrants Leaving Venezuela . . . . . . . . . . . . . . . 62 B Appendix B: Data Sources and Variable Construction 63 B.1 Satellite Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 B.2 Electoral Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 B.3 Manipulation Test for the Presidential Elections . . . . . . . . . . . . . . 68 B.4 Other municipal data sources . . . . . . . . . . . . . . . . . . . . . . . . . 70 C Appendix C: Descriptive Statistics, Variables Employed in the Main Analysis 71 D Appendix D: Characterizing Foreign Settlements of 1990 72 E Appendix E: Robustness Tests 74 E.1 Lower grid units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 E.2 Conley Standard Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 E.3 Alternative Municipal Variation: Inverse Network and Linear Distance . . 80 E.4 ENCOVI Estimates, 2014–2021 . . . . . . . . . . . . . . . . . . . . . . . . 87 E.5 Electoral Outcomes Including 2024 Presidential Election . . . . . . . . . . 88 E.6 Robustness to the Inclusion of Additional Control Variables . . . . . . . . 90 E.7 Impacts on Crime Events share per 100,000 inhabitants . . . . . . . . . . 92 54 E.8 The Role of Remittances . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 E.9 Alternative DiD Estimators . . . . . . . . . . . . . . . . . . . . . . . . . . 94 E.10 Controlling by Oil Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 E.11 Electoral Outcomes Using Electoral Rolls . . . . . . . . . . . . . . . . . . 97 E.12 Elasticity between Foreign Share and PEP Share . . . . . . . . . . . . . . 98 E.13 Mediation Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 E.14 Placebo Test: School Attendance on No Schooling Age . . . . . . . . . . . 99 E.15 Colombian foreigners share . . . . . . . . . . . . . . . . . . . . . . . . . . 99 55 A Appendix A: Characterizing the Venezuelan Crisis Figure A.1: Main Host Countries of Venezuelan Migrants in Latin America Notes: Data on Venezuelan migrant outflows to Latin American countries until May 2024 by host country come from R4V (2024) (downloaded on June 10, 2024). 56 Figure A.2: Expropriations, Inflation Rate, and International Sanctions (a) Number of expropriations (b) Inflation rate (c) Sanctions Notes: Records on the number of expropriations to private firms came from Vendata (2024). Inflation rate was estimated as the annual change in the average consumer price taken from International Monetary Fund (2024). Data on the number of international sanctions imposed on Venezuela were obtained from the Global Sanctions Database (2023). 57 Figure A.3: Poverty Rates, 2014–2021 Notes: The figure depicts the proportion of individuals whose per capita household monetary income fell below the official moderate and extreme poverty thresholds over the specified years. The poverty and extreme poverty lines were calculated by the World Bank. Source: Authors’ estimates using data from ENCOVI. 58 Figure A.4: Public Service Provision and Accessibility, 2014–2021 (a) Frequency of Water Supply (b) Frequency of Electricity Interruptions (c) Food Security Notes: The top panel shows the proportion of households according to the frequency of water supply from the pipeline over the last three months. Similarly, the second panel illustrates the frequency of electric power interruptions during the same period. Finally, the third panel presents the proportion of households in which at least one adult has had to stop a healthy diet, consume a limited variety of food, consume less food, or experience hunger due to a lack of money or resources. Source: ENCOVI. 59 Figure A.5: Health and Medicine Access (a) Proportion of Individuals with Health Insurance (b) Proportion of Individuals with a Chronic Disease Notes: The top panel displays the proportion of the population affiliated with public or private health insurance. Additionally, the bottom panel depicts the proportion of individuals with chronic diseases who have limited or no access to medication. We examine the following chronic diseases: hypertension, rheumatoid arthritis, hypothyroidism, diabetes, epilepsy, and cardiovascular conditions. Source: ENCOVI. 60 Table A.1: Sociodemographic Characteristics of the Local Population (%) Variables 2014 2015 2016 2017 2018 2019 2021 Age 0-15 26.69 28.75 29.59 29.51 29.50 29.01 28.35 15-29 24.42 25.96 25.90 26.29 26.02 21.58 21.90 30-49 27.94 27.23 25.00 24.63 24.67 26.70 26.81 50+ 20.95 18.06 19.51 19.57 19.81 22.72 22.95 Sex Female 51.25 50.26 49.90 49.90 49.76 51.44 50.84 Male 48.75 49.74 50.10 50.10 50.24 48.56 49.16 Educational attainment No Education 4.62 3.33 2.14 2.90 3.03 0.32 0.22 Primary 18.60 18.09 13.65 17.68 17.88 24.70 24.79 Secondary 39.47 38.50 38.33 41.34 43.50 43.57 44.78 Graduate and Post 35.49 38.74 35.36 36.43 33.96 27.25 25.37 Missing 1.82 1.35 10.53 1.65 1.63 4.17 4.85 Employment status Unemployed 4.19 4.40 4.51 5.48 5.18 2.34 1.66 Employed 54.21 53.64 56.05 55.08 53.92 58.26 52.15 Inactive 41.60 41.96 39.44 39.44 40.90 39.40 46.19 Mission or Social Program No 86.33 71.91 70.86 56.45 37.83 - - Yes 7.59 25.39 27.93 42.91 60.17 - - Missing 6.08 2.71 1.22 0.63 2.00 - - Household with emigrated members No - - - 93.20 87.57 81.31 88.44 Yes - - - 6.80 12.43 18.69 11.56 Notes: The first two panels depict the proportion of individuals categorized by age groups and gender, re- spectively. Additionally, the panel on educational attainment considers only adults over the age of 18, as the majority of these have completed secondary education. In the employment status panel, the labor force is defined as individuals aged between 15 and 65 years. Unemployment is identified as the proportion of indi- viduals who are not currently employed and are not actively seeking employment. Conversely, employment refers to the proportion of individuals who are presently employed. Moreover, inactivity describes those who are neither employed nor actively seeking employment. The panel on mission or social program indicates the proportion of households with at least one member affiliated with a mission or social program, and the last panel displays the proportion of households with at least one member who has emigrated within the past five years. Source: ENCOVI. 61 A.1 Characterizing the Migrants Leaving Venezuela Table A.2: Occupational Categories of Migrants in 2019 (%) Before leaving \After leaving Managers Professionals Machine Service Elementary Armed No technicians operators sales workers occupations forces Occupation Managers 33.33 14.29 4.76 19.05 23.81 0.00 4.76 Professionals and technicians 0.40 51.50 4.39 20.16 18.56 0.80 4.19 Machine operators 0.54 3.76 63.98 15.05 14.52 0.54 1.61 Service and sales workers 0.52 3.37 1.55 79.02 11.40 0.52 3.63 Elementary occupations 0.00 2.82 4.58 8.92 77.82 2.11 3.76 Armed forces occupation 1.96 1.96 9.80 15.69 58.82 7.84 3.92 No occupation 0.00 7.43 2.83 27.79 33.63 1.59 26.73 Notes: The table presents a percentage-based comparative analysis of the types of occupations held by individuals before and after migration. The classification of occupations is derived from the major groups in the International Standard Classification of Occupations (ISCO-08). We have aggregated the categories of professionals, technicians, associate professionals, and clerical support workers under “professionals and technicians.” Similarly, craft and related trades workers, plant and machine operators, and assemblers are combined into “machine operators and related trades workers.” Finally, skilled agricultural, forestry, and fishery workers are grouped with elementary occupations under the category “elementary occupations.” Source: ENCOVI. Table A.3: Occupational Categories of Migrants in 2021 (%) Before leaving \After leaving Managers Professionals Machine Service Elementary Armed No technicians operators sales workers occupations forces Occupation Managers 58.62 20.69 6.90 6.90 0.00 0.00 6.90 Professionals and technicians 0.29 57.76 3.74 18.10 12.93 1.44 5.75 Machine operators 0.00 2.46 60.66 7.38 25.41 0.82 3.28 Service and sales workers 0.29 6.92 2.59 71.76 10.95 1.15 6.34 Elementary occupations 0.00 2.31 4.94 11.57 75.62 1.70 3.86 Armed forces occupation 0.00 3.64 7.27 20.00 50.91 14.55 3.64 No occupation 0.00 7.45 3.17 35.57 25.33 0.56 27.93 Notes: The table presents a percentage-based comparative analysis of the types of occupations held by individuals before and after migration. The classification of occupations is derived from the major groups in the International Standard Classification of Occupations (ISCO-08). We have aggregated the categories of professionals, technicians, associate professionals, and clerical support workers under “professionals and technicians.” Similarly, craft and related trades workers, plant and machine operators, and assemblers are combined into “machine operators and related trades workers.” Finally, skilled agricultural, forestry, and fishery workers are grouped with elementary occupations under the category “elementary occupations.” Source: ENCOVI. 62 B Appendix B: Data Sources and Variable Construc- tion B.1 Satellite Data B.1.1 Night-Light Density Data on night-light luminosity comes from two sources: the Defense Meteorological Satel- lite Program Operational Linescan System (DMSP-OLS) and the Visible Infrared Imaging Radiometer Suite (VIIRS). The former spans data from 1992 to 2013, and the latter, from 2012 to the present. Despite the availability of data, both series are inconsistent due to differences in spatial and radiometric resolution, spectral responses, the spread function of the sensors, local overpass time at night, radiance range, and on-board calibration (Li et al., 2017; Sahoo et al., 2020). Further, the DMSP-OLS sensor measures night-light density an- nually in digital numbers (DN), while the VIIRS sensor measures monthly in radiance (units of nanoWatts/cm2 /sr) (Gibson et al., 2021). Due to the night-light problem outlined above, we use the harmonized night-light series (1992–2021) elaborated by Li et al. (2020). The authors follow four steps to intercalibrate DMSP-OLS and VIIRS sensors: DMSP-OLS calibration, the annual composition of VIIRS, and VIIRS conversion like DMSP-OLS. In the first step, the authors calibrated the stable DMSP-OLS night lights from 1992 to 2013. This calibration process aimed to ensure con- sistency and accuracy in the DMSP-OLS data. In the second step, the authors addressed the noise in the VIIRS data caused by factors such as clouds, auroras, and temporary lights like fires and boats. They applied noise-removal techniques to improve the quality of the VIIRS data. In the third step, the authors converted the higher resolution of VIIRS (15 arc-seconds) to match the resolution of DMSP-OLS (30 arc-seconds) using the Kernel den- sity approach, which is similar to the method described in Li et al. (2017). This conversion 63 ensured consistency between the two datasets. Finally, to convert the processed data into digital numbers (DN), the authors employed a sigmoid function proposed by Zhao et al. (2019). As a result, consistent and calibrated night-light data are accessible. Night-light density at various administrative levels for Venezuela. Spatial coordinates for the second administrative levels (municipios ) and third administrative levels (parroquias ) were obtained from OCHA (2023). The global rasters of harmonized night-light series, developed by Li et al. (2020), were clipped to the boundaries of Venezuela’s administrative levels. This ensured that we analyzed only the relevant region of interest. Subsequently, we computed night-light density by calculating the simple mean of night-light luminosity across all pixels within each administrative level. This approach enabled the derivation of a consistent and calibrated measure of night-light luminosity for 1992 to 2021. The processing of night-light data at the grid cell level in Venezuela involved two distinct steps. First, we created grid cells covering the entire territory of Venezuela. Then, we calcu- lated the night-light luminosity for each grid cell. In the initial step, the spatial coordinates of Venezuela’s third administrative levels (parroquias) were transformed from their original WGS84 degree measurements into metric coordinates using the UTM zone 18N projection. This conversion facilitated the subsequent segmentation of each parroquia’s polygon into 4 km square grids, which was accomplished using the Geopandas library in Python. As a result, a total of 75,605 grids encompass the entirety of Venezuela’s territory. In the second step, we clipped the annual global raster dataset of harmonized night lights to match the boundaries of Venezuela’s territory. Finally, we calculated the night-light density for each grid cell by computing the simple mean of the night-light density across all pixels contained within the respective grid cell. This process provided a comprehensive assessment of the night-light intensity within each grid cell. Night-light density for various administrative levels for Colombia. The second administrative level in Colombia is known as municipio while its territorial subdivision is called vereda. We 64 obtained the spatial information at municipio level, in the form of shapefile, from OCHA (2023). For the vereda level, spatial data come from two sources, OCHA (2023) and Colom- bian Spatial Data Infrastructure (2023). Similarly to the process described above, we com- puted the night-light density by calculating the simple mean of the night-light density across all pixels within each municipio and vereda, respectively. 65 B.1.2 Spatial Inequality Inequality is approximated at the municipal and annual levels by calculating a Gini index for each municipality and year using the NLD at the parroquia level as a unit of observation. Particularly, we use the S-Gini user-written Stata command STATA that computes classic relative (scale-invariant) Gini indices of inequality by default but can be requested to produce absolute (translation-invariant) indices or aggregate welfare S-Gini indices (Van Kerm, 2020). Table B.1 shows that the spatial measure of inequality is correlated with the traditional measure of income inequality constructed using the 1990 Venezuelan population census. Moreover, Table B.2 shows that our spatial measure of inequality also correlates with tradi- tional measures of municipal income inequality constructed using the Venezuelan population censuses of 1993 and 2005. Table B.1: Correlation between Gini and NLD Inequality Measures (1) (2) (3) Correlation Pre-Ch´avez Post-Ch´avez All years Income gini using 1990 census data 0.721*** 0.722*** 0.722*** (0.124) (0.056) (0.051) Observations 2,010 8,040 10,050 Notes: Income Gini is only available for 1990 and is constructed using Venezuela’s census data from IPUMS (2023) 66 Table B.2: Correlation between Gini and NLD Inequality Measures NLD Gini Correlation (1) Income Gini (sum all years) 1.486*** (0.158) Income Gini (first year 1993) 0.442** (0.216) Income Gini (last year 2005) 2.781*** (0.226) Notes: Income Gini at the municipal level is only available in Colombia for 1993 and 2005 when population censuses were collected. B.1.3 Type of Land Cover Areas of urban land and water bodies come from MODIS Land Cover (2023). MODIS Land Cover Type data cover a longer period, from 2001 to 2020, and have a coarser resolution of 500 meters (Friedl et al., 2010). Additionally, this dataset incorporates several supervised classification methodologies, including the International Geosphere-Biosphere Programme (IGBP), University of Maryland (UMD) Leaf Area Index (LAI), BIOME-Biogeochemical Cycles (BGC), and Plant Functional Types (PFT). For the analysis, we computed the area of land type (hectare units) at the second administrative level (municipio ) from 2001 to 2020. We do this using the Google Earth Engine code editor. B.1.4 Deforestation We downloaded an Excel dataset of annual tree-cover loss for the period 2001–2021 from Global Forest Watch (2023). This dataset provides information on annual tree-cover loss measured in hectares at both the national and the second administrative levels (municipios). Finally, we merged this data with the municipality’s shapefile using the first and second administrative-level names and fuzzy matching methods. 67 B.2 Electoral Data B.2.1 Presidential elections For electoral outcomes, we use data from all Venezuelan presidential elections between 1998 and 2018 from the Consejo Nacional Electoral (2023) and the VE360 (2023) project (see Table B.3). Specifically, we web-scraped data on the electoral census, total votes, and votes for ruling and opposition parties at the municipal level. Data for the period 2000–2013 are from the VE360 project whereas data for the 1998. The current web portal of Consejo Nacional Electoral (2023) does not include data at municipal level for this year, but by using WayBackMachine, we were able to recover the records except for the electoral census. Data for the 2018 elections are from the CNE. Table B.3: Presidential Elections in Venezuela 1998–2018 Year Elected president Type of election 2006 avez Frias Hugo Ch´ Ordinary 2012 avez Frias Hugo Ch´ Ordinary 2013 as Maduro Moros Nicol´ avez’s death Ch´ 2018 as Maduro Moros Nicol´ Ordinary B.2.2 Mayoral elections We web-scraped data from the Venezuelan National Electoral Council (CNE) website, collect- ing information on total votes and the electoral register at both the candidate and political- party levels for mayoral elections between 2006 and 2021. Afterward, we cleaned and clas- sified the data, categorizing each political party based on whether it supported or opposed avez and Nicol´ the governments of Hugo Ch´ as Maduro. B.3 Manipulation Test for the Presidential Elections 68 Figure B.1: Manipulation test 69 Notes: This test follows the methodology outlined by Klimek et al. (2012). B.4 Other municipal data sources Inflation. We estimated inflation rate as the annual change in the average consumer price taken from International Monetary Fund (2024). It corresponds to annual variation for the whole period of analysis. Sanctions. We obtained data on the number of international sanctions imposed on Venezuela from the Global Sanctions Database (2023). They correspond to annual variation in the number of sanctions for the whole period of analysis. Expropriations. Records on the number of expropriations from private firms come from Vendata (2024). We have complete records on the location of each firm that we impute to each municipality and the year. The comprises our whole period of analysis. Energy blackout, 2019. Data on blackouts correspond to the number of parroquias that reported an energy blackout within each municipality in the large blackout that occurred in March/April 2019. We have discrete variation in the number of parroquias and in the average duration of the blackout. The data come from Morales-Arilla (2021). Social welfare program coverage, 2016/2017. Data on the coverage by number of individ- uals of the social protection programs grouped under an ID called Carnet de la Patria by municipality come from Morales-Arilla and Traettino (2023). This ID centralizes all the government welfare programs. It is available only for one time period. Repression, 2004. Records of repression come from Hsieh et al. (2011). They correspond to avez regime with the names of several millions of voters a list published in 2004 by the Ch´ who attempted to remove him from office. The authors show that these individuals saw a 5 percent drop in earnings and a 1.3 percentage point drop in employment rates after the voter list was released. We have data on the number of individuals listed in each municipality for 2004. 70 C Appendix C: Descriptive Statistics, Variables Em- ployed in the Main Analysis Table C.1: Descriptive Statistics, Main Analysis Variable Mean SD Min Max Obs Years Outcomes Night light density mean 5.19 6.77 0.00 45.13 10050 1992-2021 Log light mean 0.74 1.72 -9.92 3.81 10004 1992-2021 Gini measure from parroquia level 1992-2021 0.21 0.18 0.00 0.86 10050 1992-2021 Turnout % 67.76 14.12 13.87 87.89 1662 2000, 2006, 2012, 2013, 2018 Turnout % calculated by electoral census 2000 39.92 26.21 0.00 591.04 1990 1998, 2000, 2006, 2012, 2013, 2018 Electoral census 2000 92585 175341 848 1700000 9960 1992-2021 Presidential total votes 34824 77726 0 1100000 1996 1998, 2000, 2006, 2012, 2013, 2018 Presidential electoral census 58256 127412 0 1700000 1663 2000, 2006, 2012, 2013, 2018 % Oficialism share divided by total votes by year 58.40 12.62 13.01 97.51 1995 1998, 2000, 2006, 2012, 2013, 2018 Presidential votes to chavism 19109 40423 0 584221 1996 1998, 2000, 2006, 2012, 2013, 2018 Opposition votes as % of total votes 38.97 12.05 2.33 83.32 1995 1998, 2000, 2006, 2012, 2013, 2018 Opposition total votes 14879 37204 0 549722 1996 1998, 2000, 2006, 2012, 2013, 2018 Identification variables Number of foreign residents in Venezuela (1990) 474 1196 1 10465 10050 1992-2021 Foreign residents in Venezuela 1990/ total foreign residents 0.00 0.01 0.00 0.19 10050 1992-2021 Inflows of venezuelan citizens 225830 317756 947 1400000 10050 1992-2021 Total migrants outflows divided by municipality pop. 1990 10.25 19.12 0.00 178.60 10050 1992-2021 Municipality population in 1990 53997 120325 7614 1800000 10050 1992-2021 Migrants Imputed Outflows 0.01 0.07 0.00 3.78 10050 1992-2021 Interaction of high migration dummy and foreign settlements 0.00 0.00 0.00 0.19 10050 1992-2021 Control and Robustness Variables % Venezuelan inflows by year / total inflows of vn migrants 3.33 4.69 0.01 20.07 10050 1992-2021 Rate of venezuelan inflows per 1,000 citizens 7.72 10.69 0.04 45.59 10050 1992-2021 Interaction of inflows share and inverse net distance 0.00 5.75 -24.01 65.33 10050 1992-2021 Interaction of inflows 1k rate and inverse net distance 0.00 13.16 -54.54 148.40 10050 1992-2021 Linear distance mean in km 539.36 210.92 276.36 1115.38 10050 1992-2021 Interaction of high migration dummy and inverse network dist 0.00 0.32 -1.20 3.25 10050 1992-2021 Network distance mean in km 1277.38 421.11 832.39 2477.89 10050 1992-2021 Sumatory of weights k x linear distance mk 532.05 296.67 118.68 1266.93 10050 1992-2021 Sumatory of weights x network distance 830.00 603.74 172.60 2477.89 10050 1992-2021 1/ Sumatory of weights k x linear distance mk 0.00 0.00 0.00 0.01 10050 1992-2021 1/ Sumatory of weights k x NETWORK distance mk 0.00 0.00 0.00 0.01 10050 1992-2021 Standarized: 1/ Sumatory of weights k x linear distance mk 0.00 1.00 -1.05 3.08 10050 1992-2021 Standarized: 1/ Sumatory of weights k x NETWORK distance mk 0.00 1.00 -1.20 3.25 10050 1992-2021 Lost forest hectares baseline (2001) 373.31 770.74 0.00 5569.50 10050 1992-2021 Urban and Built-up Lands (square km) into Municipio 15.96 27.09 0.00 259.79 10050 1992-2021 Water Bodies Area (square km) into Municipio 26.83 116.96 0.00 1324.15 10050 1992-2021 Light mean (Baseline 1992) 59.46 116.54 0.00 1228.80 10050 1992-2021 Note: This table presents summary statistics of most raw variables used in the analysis. The information comes from diverse sources. See Appendix A for more details. 71 D Appendix D: Characterizing Foreign Settlements of 1990 Table D.1: Foreign Settlements in 1990 vs Migrants’ Hosting Countries in 2017–2021 Country % of total foreigners % of total migrants (Census 1990) by host country (2017-2021) Colombia 59.02% 40.87% Peru 2.48% 19.70% Chile 1.50% 9.21% Ecuador 2.38% 8.40% United States 0.85% 3.35% Brazil 0.32% 3.19% Spain 6.90% 2.89% Argentina 0.64% 2.82% Panama 0.11% 1.98% Mexico 0.25% 0.70% Other 2.78% 4.34% Unknown 1.02% 3% No information in both surveys 22% - Notes: Authors’ estimates using data from IPUMS (2023) and National Survey on Living Conditions (2021). Table D.2: Characterizing Foreigners in Venezuela in 1990 Mean Mean Difference Test Foreigners Nationals (p-value) Age 36.88 23.68 0.000 Years of schooling 5.91 4.65 0.000 Earned income ($BOL) $ 678,196.00 $ 223,045.00 0.000 72 Table D.3: Characterizing Foreigners in Venezuela in 1990 (cont’d) Percent of foreigners Percent nationals Sex Male 53.26% 51.78% Female 46.74% 48.22% Employment status Employed 55.67% 27.48% Unemployed 3.25% 4.18% Inactive 35.23% 35.38% Missing 5.85% 32.96% Literacy Literate 87.14% 71.28% Illiterate 10.76% 18.13% Missing 2.10% 10.59% Table D.4: Correlation Between Imputed Outflows and Colombian Survey of Venezuelans Panel A: Correlation by sources at municipality level Imputed Outflows Migration records from PEP 0.329*** Panel B: Correlation by sources at state level Migration records from ENCOVI 2017 and 2021 0.250*** Migration records from ENCOVI (bianually) 0.205*** Migration records from ENCOVI 2017, 2019 and 2021 0.137*** Notes: Migration records from PEP are defined as the number of Venezuelan migrants in the Colombian Survey of Venezuelans registry per year and municipality since 1992. The Colombian survey of Venezuelans was conducted in 2018 and is representative of the migrants living in Colombia that year. The survey was collected to evaluate the effects of a large regularization program enacted in Colombia in 2018. The data come from Ib´ anez et al. (2024). Migration records from ENCOVI are defined as the number of individuals reported as migrants (living abroad) by their relatives in the ENCOVI surveys (2017–2021). The ’2017 and 2021’ label indicates that outflows by state and year were reconstructed using ENCOVI 2021 for 2017–2021 and ENCOVI 2017 for 2012–2016. The ’Biannual’ label refers to the use of biannual ENCOVI data. The ’2017, 2019, and 2021’ label means ENCOVI 2017 was used for 2012–2016, ENCOVI 2019 for 2017–2019, and ENCOVI 2021 for 2020–2021. 73 E Appendix E: Robustness Tests E.1 Lower grid units Table E.1: Effects of Outmigration on Economic Growth and Inequality Measures 1-Square-Km-Grid Units Night Light Log(Night Light) Spatial Gini Panel A: Difference-in-difference estimates including all controls (1) (2) (3) I(Year>=2013)× Foreign Settlements -0.423*** -0.538*** 0.024*** (0.088) (0.091) (0.008) Panel B: Including baseline controls × time trends Imputed outflows -0.034*** -0.036*** 0.003*** (0.006) (0.004) (0.001) Observations 2,265,990 2,214,162 2,214,162 Dependent mean 1992 3.84 0.04 0.27 All controls ✓ ✓ ✓ Exclude outlier municipality (Libertador) ✓ ✓ ✓ Municipality FE ✓ ✓ ✓ Year FE ✓ ✓ ✓ Note: The table illustrates the estimated coefficients of equation (1). Imputed Outflows is defined in equation (3) as the product of municipal foreign settlement shares in 1990 and annual outflows of Venezuelans to Colombia. Controls in the baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1.. 74 E.2 Conley Standard Errors E.2.1 Cutoff 250 km Table E.2: Effects of Migration on Development and Inequality Measures (Conley Standard Errors) Night Light Log(Night Light) Spatial Gini Panel A: Diff-in-diff estimates including controls (1) (2) (3) I(Year>=2013) × Foreign Share -0.356*** -0.126*** 0.012*** (0.068) (0.025) (0.002) Panel B: Imputed outflows, including baseline controls × time trends Imputed Outflows -0.036*** -0.016*** 0.001*** (0.013) (0.003) (0.000) Panel C: Imputed outflows, excluding controls Imputed Outflows -0.037*** -0.015*** 0.001*** (0.013) (0.004) (0.000) Observations 10,020 9,974 10,020 Dependent Mean 1992 3.77 0.034 0.27 Conley Standard Errors (cutoff 250km) ✓ ✓ ✓ Municipality Fixed Effects ✓ ✓ ✓ Year Fixed Effects ✓ ✓ ✓ Note: The table illustrates the estimated coefficients of equation (1). Imputed Outflows is defined in equation (3) as the product of municipal foreign settlement shares in 1990 and annual outflows of Venezuelans to Colombia. Controls in the baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 75 Table E.3: Effects of Migration on Electoral Outcomes (Conley Standard Errors) Turnout Opposition Panel A: Difference-in-difference estimates including all controls (1) (2) I(Year>=2013)× Foreign Share -4.471* -2.624* (2.704) (1.517) Panel B: Including baseline controls × time trends Imputed Outflows -0.663*** -0.358*** (0.117) (0.049) Panel C: Imputed outflows, excluding controls Imputed Outflows -0.649*** -0.351*** (0.129) (0.058) Observations 1,324 1,324 Dependent mean 2006 31.4 10.2 All controls ✓ ✓ Conley Standard Errors (cutoff 250km) ✓ ✓ Municipality FE ✓ ✓ Year FE ✓ ✓ Note: The table illustrates the estimated coefficients of equation (1). Turnout in column (1) is defined as the total votes in each Venezuelan presidential election between 2006 and 2018 divided by the electoral census of 2000. (2) Opposition is the total votes of non-ruling parties divided by the electoral census of 2000. Imputed Outflows is defined in equation (3) as the product of municipal foreign settlement shares in 1990 and annual outflows of Venezuelans to Colombia. Controls in the baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 76 E.2.2 Cutoff 150 km Table E.4: Effects of Migration on Development and Inequality Measures (Conley Standard Errors) Night Light Log(Night Light) Spatial Gini Panel A: Diff-in-diff estimates including controls (1) (2) (3) I(Year>=2013) × Foreign Share -0.356*** -0.126*** 0.012*** (0.074) (0.024) (0.003) Panel B: Imputed outflows, including baseline controls × time trends Imputed Outflows -0.036*** -0.016*** 0.001*** (0.011) (0.003) (0.000) Panel C: Imputed outflows, excluding controls Imputed Outflows -0.037*** -0.015*** 0.001*** (0.010) (0.004) (0.000) Additional controls for all panels Observations 10,020 9,974 10,020 Dependent Mean 1992 3.77 0.034 0.27 Conley Standard Errors (cutoff 150km) ✓ ✓ ✓ Municipality Fixed Effects ✓ ✓ ✓ Year Fixed Effects ✓ ✓ ✓ Note: The table illustrates the estimated coefficients of equation (1). Imputed Outflows is defined in equation (3) as the product of municipal foreign settlement shares in 1990 and annual outflows of Venezuelans to Colombia. Controls in the baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 77 Table E.5: Effects of Migration on Electoral Outcomes (Conley Standard Errors) Turnout Opposition Panel A: Difference-in-difference estimates including controls (1) (2) I(Year>=2013)× Foreign Share -4.471* -2.624* (2.674) (1.450) Panel B: Including baseline controls × time trends Imputed Outflows -0.663*** -0.358*** (0.144) (0.054) Panel C: Imputed outflows, excluding controls Imputed Outflows -0.649*** -0.351*** (0.143) (0.059) Observations 1,324 1,324 Dependent mean 2006 31.4 10.2 All controls ✓ ✓ Conley Standard Errors (cutoff 150km) ✓ ✓ Municipality FE ✓ ✓ Year FE ✓ ✓ Note: The table illustrates the estimated coefficients of equation (1). Turnout in column (1) is defined as the total votes in each Venezuelan presidential election between 2006 and 2018 divided by the electoral census of 2000. (2) Opposition is the total votes of non-ruling parties divided by the electoral census of 2000. Imputed Outflows is defined in equation (3) as the product of municipal foreign settlement shares in 1990 and annual outflows of Venezuelans to Colombia. Controls in the baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 78 E.2.3 Cutoff 50 km Table E.6: Effects of Migration on Development and Inequality Measures (Conley Standard Errors) Night Light Log(Night Light) Spatial Gini Panel A: Diff-in-diff estimates including controls (1) (2) (3) I(Year>=2013) × Foreign Share -0.356*** -0.126*** 0.012*** (0.064) (0.022) (0.003) Panel B: Imputed outflows, including baseline controls × time trends Imputed Outflows -0.036*** -0.016*** 0.001*** (0.007) (0.002) (0.000) Panel C: Imputed outflows, excluding controls Imputed Outflows -0.037*** -0.015*** 0.001*** (0.007) (0.003) (0.000) Observations 10,020 9,974 10,020 Dependent Mean 1992 3.77 0.034 0.27 Conley Standard Errors (cutoff 50km) ✓ ✓ ✓ Municipality Fixed Effects ✓ ✓ ✓ Year Fixed Effects ✓ ✓ ✓ Note: The table illustrates the estimated coefficients of equation (1). Imputed Outflows is defined in equation (3) as the product of municipal foreign settlement shares in 1990 and annual outflows of Venezuelans to Colombia. Controls in the baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 79 Table E.7: Effects of Migration on Electoral Outcomes (Conley Standard Errors) Turnout Opposition Panel A: Difference-in-difference estimates including all controls (1) (2) I(Year>=2013)× Foreign Share -4.471** -2.624** (2.093) (1.171) Panel B: Including baseline controls × time trends Imputed Outflows -0.663*** -0.358*** (0.161) (0.070) Panel C: Imputed outflows, excluding controls Imputed Outflows -0.649*** -0.351*** (0.156) (0.072) Observations 1,324 1,324 Dependent mean 2006 31.4 10.2 All controls ✓ ✓ Conley Standard Errors (cutoff 50km) ✓ ✓ Municipality FE ✓ ✓ Year FE ✓ ✓ Note: The table illustrates the estimated coefficients of equation (1). Turnout in column (1) is defined as the total votes in each Venezuelan presidential election between 2006 and 2018 divided by the electoral census of 2000. (2) Opposition is the total votes of non-ruling parties divided by the electoral census of 2000. Imputed Outflows is defined in equation (3) as the product of municipal foreign settlement shares in 1990 and annual outflows of Venezuelans to Colombia. Controls in the baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. E.3 Alternative Municipal Variation: Inverse Network and Linear Distance As an additional exercise, we leverage the fact that migrants often choose destinations with lower migration costs. Consequently, we concentrate on municipalities located closer to road and major river networks, as these areas typically witness higher migration rates. We then explore the relationship between this geographical proximity and national migration 80 from Venezuela to Colombia. Since national migration patterns do not perfectly mirror municipal migration trends, this creates an exogenous factor that we can employ as a shift- share instrument. In particular, we estimate the following specification: ym t = γm + αt + βOutf lowst × InvN etDistm + η ( z × α t ) + εm t (6) ′ z ⊂Xm Where InvN etDistm is defined by the following equation: 1 InvN etDistm = 7 (7) i=1 ωi ∗ netdistmi In Equation 6, ym t is the economic growth, inequality or electoral outcome of interest for municipality m in year t, Outf lowst is the outflow of Venezuelan migrants to Colombia for year t, InvN etDistm is the inverse network distance for each municipality m defined by ′ Equation 7, Xm is a control-variables vector at baseline and γm , αt are municipality and time fixed effects, respectively. Pre-shock municipal characteristics are interacted with year fixed effects to flexibly control for differential municipal trends. Robust standard errors (εmt ) are clustered at the municipality level to account for potential serial correlation within municipalities. In Equation 7, ωi is the weight of Venezuelan migrants’ entry point i, calculated with the share of total entries between 2012 and 2022. netdistmi is the network distance from centroid afico de of municipality m to entry point i using the road lines collected by Instituto Geogr´ Venezuela (2015). This index is standardized by subtracting the mean and dividing it by its standard deviation. To calculate the network distance netdistmi , we use an Origin Destination (OD) cost-matrix analysis. This method finds and measures the least costly paths along a network from mul- tiple origins to multiple destinations. We use the Venezuelan road and major rivers network 81 afico de Venezuela (2015) to connect and calculate the shapefile created by Instituto Geogr´ minimum distance in kilometers from all centroids of municipalities to each of the seven entry points. Additionally, we calculate the inverse distance using the linear distance from each municipality’s centroid to the nearest entry point. Some municipalities do not have a full connected network road from their centroids to each entry point. Thus, we assign the maximum network distance with a penalty of 20 percent of network distance mean for disconnected municipalities. This exercise basically amounts to replacing our municipal measure of foreign settlements with the inverse distance of each municipality to the main crossing points in Colombia. 82 E.3.1 Inverse Network Distance Table E.8: Effects of Migration on Development and Inequality Measures Night Light Log(Night Light) Panel A: Difference-in-difference estimates including all controls (1) (2) I(Year>=2013)× Network distance -0.359*** -0.185*** (0.106) (0.029) Panel B: Including baseline controls × time trends Outflows share × Network distance -0.032*** -0.016*** (0.008) (0.002) Panel C: Using inflows 1k rate, including all controls Outflows 1k rate × Network distance -0.014*** -0.007*** (0.004) (0.001) Observations 10,020 9,974 Dependent Mean 1992 3.77 0.034 Exclude Libertador municipality ✓ ✓ Municipality Fixed Effects ✓ ✓ Year Fixed Effects ✓ ✓ Note: The table illustrates the estimated coefficients of equation (6). Network distance is defined in equation (7). Controls in the baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 83 Table E.9: Effects of Migration on Electoral Outcomes Turnout Opposition Panel A: Difference-in-difference estimates including all controls (1) (2) I(Year>=2013)× Network distance -3.407*** -1.351*** (0.998) (0.406) Panel B: Including baseline controls × time trends Outflows share × Network distance -0.376*** -0.161*** (0.113) (0.048) Panel C: Using inflows 1k rate, including all controls Outflows 1k rate × Network distance -0.167*** -0.071*** (0.050) (0.021) Observations 1,324 1,324 Dependent mean 2006 31.4 10.2 Exclude Libertador municipality ✓ ✓ Municipality FE ✓ ✓ Year FE ✓ ✓ Note: The table illustrates the estimated coefficients of equation (6). Network distance is defined in equation (7). Controls in the baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 84 E.3.2 Inverse Linear Distance Table E.10: Effects of Migration on Development and Inequality Measures (using Inverse Linear Distance) Night Light Log(Night Light) Panel A: Difference-in-difference estimates including all controls (1) (2) I(Year>=2013)× Linear Distance -0.397*** -0.158*** (0.109) (0.027) Panel B: Including baseline controls × time trends Outflows share × Linear distance -0.034*** -0.014*** (0.008) (0.002) Panel C: Using outflows 1k rate, including all controls Inflows 1k rate × Linear distance -0.015*** -0.006*** (0.004) (0.001) Observations 10,020 9,974 Dependent Mean 1992 3.77 0.034 Exclude Libertador municipality ✓ ✓ Municipality Fixed Effects ✓ ✓ Year Fixed Effects ✓ ✓ Note: The table illustrates the estimated coefficients of equation 6 but using the inverse linear distance.Linear distance is analogue to the standardized measure of network inverse distance used in Equation 7 but using the linear distance from the municipality’s centroid to each entry point. Controls in the baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 85 Table E.11: Effects of Migration on Electoral Outcomes Turnout Opposition Panel A: Difference-in-difference estimates including all controls (1) (2) I(Year>=2013)× Linear Distance -3.871*** -1.582*** (1.062) (0.422) Panel B: Including baseline controls × time trends Outflows share × linear distance -0.427*** -0.187*** (0.120) (0.049) Panel C: Using outflows 1k rate, including all controls Outflows 1k rate × linear distance -0.189*** -0.083*** (0.053) (0.022) Observations 1,324 1,324 Dependent mean 2006 31.4 10.2 Exclude Libertador municipality ✓ ✓ Municipality FE ✓ ✓ Year FE ✓ ✓ Note: The table illustrates the estimated coefficients of equation 6 but using the inverse linear distance. Linear distance is analogue to the standardized measure of network inverse distance used in Equation 7 but using the linear distance from the municipality’s centroid to each entry point. Controls in the baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night- light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 86 E.4 ENCOVI Estimates, 2014–2021 Table E.12: Impacts of Mass Out-migration in State-Level Income and Inequality Real total Income (IHS) Real total Income per capita (IHS) Panel A: No controls (1) (2) Imputed Outflows -0.00131*** -0.00128*** (0.00015) (0.00014) Panel B: Including baseline controls Imputed Outflows -0.00155*** -0.00151*** (0.00014) (0.00013) Observations (State and Year) 139,328 139,328 Dependent Mean 2014 915.80 0.34 State Fixed Effects ✓ ✓ Year Fixed Effects ✓ ✓ Notes: The term Imputed Outflows is defined as explained in Equation (2). Controls in baseline interacted with time trends include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; Carnet de la Patria holders (2016–2017); number of parroquias with rationed energy at munici- pality level (April 2019) and intensity of blackout (March 2019); and number of enterprises acquired by the Venezuelan state. Controls from ENCOVI: head of household gender, education level, marital status, and number of household members. Total and per capita income coefficients were estimated at the state level. Bootstrap standard errors in parentheses. ***p<0.01, **p <0.05, *p<0.1. 87 E.5 Electoral Outcomes Including 2024 Presidential Election Table E.13: Effects of Mass Out-migration in Presidential Electoral Outcomes Turnout Opposition Panel A: Difference-in-difference estimates including all controls (1) (2) I(Year>=2013)× Foreign Share -4.845** -3.623*** (1.913) (1.370) Panel B: Including baseline controls × time trends Imputed Outflows -0.612*** -0.330*** (0.140) (0.060) Panel C: Imputed outflows, excluding controls Imputed Outflows -0.576*** -0.309*** (0.123) (0.052) Observations 1,648 1,648 Dependent mean 2006 31.4 10.2 All controls ✓ ✓ Municipality FE ✓ ✓ Year FE ✓ ✓ Notes: The table illustrates the estimated coefficients of equation (1). Turnout in column (1) is defined as the total votes for each Venezuelan presidential election between 2006 and 2018 divided by the electoral census of 2000. (2) Opposition is the total votes of non-ruling parties divided by the electoral census of 2000. Imputed Outflows is defined in equation (3) as the product of municipal foreign settlement shares in 1990 and annual outflows of Venezuelans to Colombia. Controls in baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 88 Table E.14: Forced Displacement Weakens Political Opposition, Mayoral Elections Turnout Opposition Panel A: Diff-in-diff estimates including controls (1) (2) I(Year>=2013)× Foreigners Share -2.198*** -1.137* (0.731) (0.654 ) Observations 1,649 1,652 Dependent mean 2004 32.5 14.59 All controls ✓ ✓ Exclude outlier municipality (Libertador) ✓ ✓ Municipality FE ✓ ✓ Year FE ✓ ✓ Notes: Turnout in column (1) is defined as the total votes for each mayoral election divided by the electoral census of 2000, (2) Opposition is the total votes of non-ruling parties divided by the electoral census of 2000. Imputed Outflows is defined in equation (3) as the product of municipal foreign settlement shares in 1990 and annual outflows of Venezuelans to Colombia. Controls in baseline are interacted with time trends. They include: urban coverage, water bodies, and forest- loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 89 E.6 Robustness to the Inclusion of Additional Control Variables E.6.1 Carnet de la Patria, Energy Blackouts, and Expropriation records Table E.15: Development Outcomes: Controlling for Governmental Actions Night Light Log(Night Light) Spatial Gini Imputed Outflows (1) (2) (3) Imputed Outflows -0.043*** -0.018*** 0.001* (0.012) (0.003) (0.0006) Observations 10,020 9,974 10,020 Dependent mean 5.19 0.74 0.27 All controls ✓ ✓ ✓ Exclude outlier municipality (Libertador) ✓ ✓ ✓ Municipality FE ✓ ✓ ✓ Year FE ✓ ✓ ✓ Note: NLD in column (1) is defined as Night-Light Density, Log NLD in column (2) is the Log Night-Light Density, Gini in column (3) is the level of night-light concentration using Gini index. The term Imputed Outflows is defined as explained in equation 2. Controls in baseline interacted with time trends: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; repression records for 2004; Carnet de la Patria holders (2016–2017); number of parroquias with rationed energy at the municipality level (April 2019) and intensity of blackout (March 2019); number of enterprises acquired by the Venezuelan state. Clustered standard errors by municipality in parentheses. ***p<0.01, **p <0.05, *p<0.1. 90 Table E.16: Electoral Outcomes: Controlling for Governmental Actions Turnout Opposition Imputed Outflows (1) (2) Imputed Outflows -0.699*** -0.371*** (0.181) (0.082) Observations 1,324 1,324 Dependent mean 2006 31.4 10.2 All controls ✓ ✓ Exclude outlier municipality (Libertador) ✓ ✓ Municipality FE ✓ ✓ Year FE ✓ ✓ Note: Turnout in column (1) is defined as the total votes for each presidential election held in Venezuela between 2006 and 2018 divided by the electoral census of 2000; (2) Opposition is the total votes of non- ruling parties divided by the electoral census of 2000. Imputed Outflows is defined as explained in equation 2. Controls in baseline interacted with time trends: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; repression records for 2004; Carnet de la Patria holders (2016–2017); number of parroquias with rationed energy at the municipality level (April 2019) and intensity of blackout (March 2019); number of enterprises acquired by the Venezuelan state. Clustered standard errors by municipality are in parentheses. ***p<0.01, **p <0.05, *p<0.1. 91 E.7 Impacts on Crime Events share per 100,000 inhabitants Table E.17: Effects of Mass Out-migration on Crime FARC ELN Organized Crime Panel A: Total events share per 100,000 inhabitants (1) (2) (3) Imputed outflows 0.005* 0.022* 0.155** (0.003) (0.012) (0.060) Protest Regime Sindicato Panel B: Total events share per 100,000 inhabitants (1) (2) (3) Imputed outflows -0.017 0.109** -0.003* (0.045) (0.0506) (0.001) Observations 11,022 11,022 11,022 Dependent mean baseline 0.036 0.007 3.68 All controls ✓ ✓ ✓ Exclude outlier municipality (Libertador) ✓ ✓ ✓ Municipality FE ✓ ✓ ✓ Year FE ✓ ✓ ✓ Notes: In Panel A, column (1) represents the total number of conflict events involving the Revolutionary Armed Forces of Colombia (FARC) in Venezuela from 1992 to 2024, normalized by the Venezuelan munici- pality population in 1990 and expressed per 100,000 inhabitants. Column (2) reports the total conflict events associated with the Colombian National Liberation Army (ELN) during the same period, also normalized by the 1990 population and presented per 100,000 inhabitants. Column (3) captures the total conflict events carried out by non-terrorist but criminal armed groups—such as gangs, colectivos, sindicatos, and drug- trafficking cartels—between 2018 and 2024, similarly normalized by the 1990 population and expressed per 100,000 inhabitants. In Panel B, column (1) reports the total number of both peaceful and violent protests from 2018 to 2024, adjusted by the 1990 population of Venezuelan municipalities and presented per 100,000 inhabitants. Column (2) records the total number of events involving actors linked to the Venezuelan po- litical regime between 1992 and 2024, also normalized by the 1990 population and expressed per 100,000 inhabitants. Finally, column (3) reflects the total number of events involving irregular armed groups engaged in illegal mining from 2018 to 2024, adjusted by the 1990 population and presented per 100,000 inhabitants. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 92 E.8 The Role of Remittances Figure E.1: Household Utilization of Remittances, 2019 and 2021 Notes: The table describes the categories of expenditures of remittances by households inter- viewed in the ENCOVI of 2019 and 2021. The categories are non-exclusionary; households could mention one or more categories on which they spent the majority of the remittances they received from relatives abroad. 93 E.9 Alternative DiD Estimators Table E.18: Alternative DiD estimators - Development Outcomes Night Light Log(Night Light) Spatial Gini Panel A: Doubly Robust Improved Estimator (1) (2) (3) I(Year>=2013)× I(Imputed Outflows>= P80 ) –1.264*** -0.178*** 0.028*** (0.17) (0.05) (0.008) Panel B: Abadie (2005) IPW Estimator I(Year>=2013)× I(Imputed Outflows>= P80 ) -2.852*** -.569*** 0.028** (0.59) (0.099) (0.012) Panel C: Standardized IPW estimator I(Year>=2013)× I(Imputed Outflows>= P80 ) -1.577*** -0.341*** 0.028** (0.466) (0.076) (0.012) Observations 10,020 9,974 10,020 Dependent Mean 1992 3.77 0.034 0.27 Notes: Inverse Probability Weighted (IPW) estimators from Abadie (2005) and Sant’Anna and Zhao (2020). Doubly Robust Improved Estimator from Sant’Anna and Zhao (2020). We use STATA command drdid from Rios-Avila et al. (2021) for estimations. ***p<0.01, **p <0.05, *p<0.1. E.10 Controlling by Oil Fields We use the Global Oil Gas Features Database pertaining to the development of the Global Oil and Gas Infrastructure (GOGI) geodatabase from Sabbatino (2018). This dataset stan- dardizes and integrates disparate oil and gas infrastructure data from over 380 sources worldwide, encompassing more than 4.8 million features for 2018. The project employed both manual searches by experts and machine learning algorithms to gather global data on oil and gas infrastructure, including production, transportation, and storage. The result is a comprehensive geodatabase that provides users with spatially explicit data. The GOGI database offers a unified platform to assess and visualize global oil and gas infrastructure, 94 addressing key uncertainties and information gaps across nations that produce and consume hydrocarbons. There are 194 countries in the GOGI geodatabase. The geodatabase contains physical in- formation such as Fields, Mines, Platforms, Wells, Underground Storage, Pipelines, Ports, Railways, Basins, LNG facilities, Power Plants, Processing Plants, Refineries, Stations, and Storage Units. Table E.19: Effects of Forced Displacement on Development Outcomes Including controls for oil field locations Night Light Log(Night Light) Spatial Gini Panel A: Diff-in-diff estimates including controls (1) (2) (3) I(Year>=2013) × Foreign Share -0.234** -0.127*** 0.011* (0.100) (0.038) (0.006) Panel B: Imputed outflows, including baseline controls × time trends Imputed Outflows -0.025*** -0.017*** 0.001 (0.008) (0.003) (0.001) Additional controls for all panels Observations 10,020 9,974 10,020 Dependent Mean 1992 3.77 0.034 0.27 Municipality Fixed Effects ✓ ✓ ✓ Year Fixed Effects ✓ ✓ ✓ Notes: The table illustrates the estimated coefficients of equation (1). Imputed Outflows is defined in equation (3) as the product of municipal foreign settlement shares in 1990 and annual outflows of Venezuelans to Colombia. Controls in baseline interacted with time trends: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; political repression records for 2004; Carnet de la Patria holders (2016–2017); number of parroquias with rationed energy at the municipality level (April 2019) and intensity of blackout (March 2019); number of enterprises acquired by the Venezuelan state; and oil field area as percentage of total in 2018. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 95 Table E.20: Effects of Mass Out-migration on Presidential Electoral Outcomes Turnout Opposition Panel A: Diff-in-diff estimates including controls (1) (2) I(Year>=2013)× Foreign Share -4.445*** -2.621*** (1.419) (0.702) Panel B: Imputed outflows, including baseline controls × time trends Imputed Outflows -0.705*** -0.369*** (0.162) (0.072) Additional controls for all panels Observations 1,324 1,324 Dependent mean 2006 31.4 10.2 All controls ✓ ✓ Municipality FE ✓ ✓ Year FE ✓ ✓ Notes: The table illustrates the estimated coefficients of equation (1). Turnout in column (1) is defined as the total votes for each presidential election held in Venezuela between 2006 and 2018 divided by the electoral census of 2000. (2) Opposition is the total votes of non-ruling parties divided by the electoral census of 2000. Imputed Outflows is defined in equation (3) as the product of municipal foreign settlement shares in 1990 and annual outflows of Venezuelans to Colombia. Controls in baseline interacted with time trends: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; political repression records for 2004; Carnet de la Patria holders (2016–2017); number of parroquias with rationed energy at the municipality level (April 2019) and intensity of blackout (March 2019); number of enterprises acquired by the Venezuelan state; and oil field area as percentage of total in 2018. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 96 E.11 Electoral Outcomes Using Electoral Rolls Table E.21: Effects of Mass Out-migration on Presidential Electoral Outcomes Turnout Opposition Panel A: Diff-in-diff estimates including controls (1) (2) I(Year>=2013)× Foreign Share -3.613*** -4.389*** (0.787) (0.867) Panel B: Imputed outflows, including baseline controls × time trends Imputed Outflows -0.706*** -0.591*** (0.075) (0.069) Panel C: Imputed outflows, excluding controls Imputed Outflows -0.703*** -0.588*** (0.081) (0.076) Additional controls for all panels Observations 1,323 1,323 Dependent mean 2006 31.4 10.2 All controls ✓ ✓ Municipality FE ✓ ✓ Year FE ✓ ✓ Notes: The table illustrates the estimated coefficients of equation (1). Turnout in column (1) is defined as the total votes for each presidential election held in Venezuela between 2006 and 2018 divided by the electoral census. (2) Opposition is the total votes of non-ruling parties divided by the electoral census. Imputed Outflows is defined in equation (3) as the product of municipal foreign settlement shares in 1990 and annual outflows of Venezuelans to Colombia. Controls in baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. 97 E.12 Elasticity between Foreign Share and PEP Share Table E.22: Correlation Between Foreigners’ Share and Colombian Survey of Venezuelans Panel A: Correlation PEP and foreign residents share Log Migration records from PEP Log Foreign Residents Share 0.445*** Notes: The Colombian survey of Venezuelans was conducted in 2018 and is representative of the migrants living in Colombia that year. The survey was collected to evaluate the effects of a large regularization program enacted in Colombia in 2018. The data come from Ib´anez et al. (2024). E.13 Mediation Analysis Table E.23: Development and Forced Displacement - Mediation Analysis Night Light Log(Night Light) Spatial Gini Panel A: Coefficient Equation 1, excluding controls (1) (2) (3) Imputed outflows -0.037*** -0.015*** 0.001* (0.009) (0.004) (0.001) Panel B: Coefficient without the variation explained by crime and opposition Imputed outflows -0.022*** -0.011*** 0.001* (0.007) (0.002) (0.000) Observations Panel A 10,020 9,974 10,020 Observations Panel B 1,984 1,976 1,984 Dependent mean 3.77 0.034 0.27 Notes: The table illustrates the estimated coefficients of equation (1). Imputed Outflows is defined in equation (3) as the product of municipal foreign settlement shares in 1990 and annual outflows of Venezuelans to Colombia. Controls in the baseline are interacted with time trends. They include: urban coverage, water bodies, and forest-loss area for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Clustered standard errors by municipality are presented in parentheses. ***p<0.01, **p <0.05, *p<0.1. 98 E.14 Placebo Test: School Attendance on No Schooling Age Table E.24: School Attendance on No Schooling Age Age < 6 or > 17 Panel A: School Attendance (1) Imputed Outflows -0.00030 (0.00019) Observations (State and Year) 38,238 Dependent Mean 2014 7.93 Notes: Imputed Outflows is defined as explained in equation (2). For column (1), the estimates include only individuals in non-schooling age (less than 6 and more than 17 years old in 2013). Bootstrap standard errors are in parentheses. ***p<0.01, **p <0.05, *p<0.1. E.15 Colombian foreigners share Table E.25: Forced Displacement and Development Night Light Log(Night Light) COL - Night Light COL - Log(Night Light) Panel A: Diff-in-diff estimates including controls (1) (2) (3) (4) I(Year>=2013) × Foreigners Share -0.356*** -0.126*** -0.297*** -0.111*** (0.120) (0.038) (0.082) (0.038) Panel B: Imputed outflows, including baseline controls × time trends Imputed Outflows -0.036*** -0.016*** -0.019*** -0.010*** (0.009) (0.003) (0.005) (0.002) Panel C: Imputed outflows, excluding controls Imputed Outflows -0.037*** -0.015*** -0.019*** -0.008*** (0.009) (0.004) (0.005) (0.002) Additional controls for all panels Observations 10,020 9,974 10,020 9,974 Dependent Mean 1992 3.77 0.034 3.77 0.034 Municipality Fixed Effects ✓ ✓ ✓ ✓ Year Fixed Effects ✓ ✓ ✓ ✓ Notes: The table illustrates the estimated coefficients of equation (1). In column (1) and (2) Imputed Outflows is defined in equation (3) as the product of the share of foreign settlement in each municipality in 1990 and annual outflows of Venezuelans to Colombia. It is rescaled by the total municipal population of 1990. In columns (3) and (4) the share of the foreign settlement in each municipality is replaced by the share of Colombian citizens living in Venezuela before 1990. Controls in the baseline interacted with time trends, including urban coverage, water bodies, and tree-cover loss for 2001; night-light density for 1992; and political repression records for 2004. All estimates exclude the outlier municipality Libertador. Standard errors clustered by municipality are in parentheses. ***p<0.01, **p <0.05, *p<0.1. 99 100