Lake Chad Regional Economic Memorandum  |  Development for Peace Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram Remi Jedwab (George Washington University), Brian Blankespoor (World Bank), Takaaki Masaki (World Bank), and Carlos Rodríguez-Castelán (World Bank) 150 Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram 4.1 Introduction Violent conflicts present a formidable threat to 2019; Trebbi and Weese, 2019). But poor economies and regional economies. Throughout the world, border economic shocks also offer fertile grounds for conflicts regions in many countries are possibly impacted by the (e.g., Miguel et al., 2004; Bazzi and Blattman, 2014; cross-border economic effects of regional insurgencies in Burke et al., 2015).277 neighboring countries or national state failures, i.e. “bad neighbors”. This raises two questions. First, what is the While poverty traps and conflict traps can reinforce magnitude of the spill-over economic effects of foreign each other locally, spatio-dynamic spillovers can also conflict and what are the channels through which they be present  (Berman et al., 2017; König et al., 2017; operate? Second, what policies can governments adopt in Harari and Ferrara, 2018; Melnikov et al., 2020; Eberle the potentially exposed regions to mitigate such spill-over et al., 2020). Conflict in one location can beget conflict effects?276 in other locations, either via a direct expansion in space of conflict factors (e.g., armies) or because conflict in one In this paper, we adopt a difference-in-difference location increases poverty, and lowers the opportunity (DiD) framework leveraging the unexpected rise of cost of conflict labor, in other locations. Due to spill- the Boko Haram insurgency in Northeastern Nigeria overs that reinforce each other across locations, separately in 2009 to study its economic effects in neighboring identifying local and non-local effects is difficult areas in Cameroon, Chad and Niger that were not econometrically (Harari and Ferrara, 2018). directly targeted by Boko Haram activities. We find strong cross-border economic effects that are likely driven Studies on the impact of policies on conflict then focus by reduced trade activities, not the diffusion of conflict. on conflict prevention, management, resolution and/ Factors of local economic resilience to this foreign conflict or reconciliation in the conflict countries themselves shock then include trade diversification and political and (e.g., Nunn and Qian, 2014; König et al., 2017; Chiovelli  economic securitization. More generally, conflicts, if they et al., 2018; Sviatschi, 2018). Less is known about how have regional economic effects, may necessitate regional non-conflict countries can mitigate, in their border responses. regions, the local economic impact of foreign conflicts. The causal identification of the effects of conflict, To address some of these challenges, we exploit the and by extension of the mitigation effects of various exogenous rise of Boko Haram in Nigeria after 2009 locational factors, is complicated by a complex and estimate its local economic effects on neighboring endogenous relationship between conflict and socio- areas within Cameroon, Chad and Niger (CCN), so economic conditions  (Blattman and Miguel, 2010; outside Nigeria. Between 2009 and 2014, Boko Haram Djankov and Reynal-Querol, 2010). Conflicts impose became the world’s deadliest terrorist group ahead of ISIL, large economic tolls (e.g., International Monetary Fund, the Taliban and Al-Shabaab (Institute for Economics 276 Examples of regional insurgencies plausibly affecting other countries include the Cabo Delgado insurgency in Mozambique (Tanzania), the ISIL insurgency in Iraq and Syria (Turkey but also Jordan and Lebanon), the insurgency in the Maghreb (other countries in West Africa), the Taliban insurgency in Afghanistan (Iran, Pakistan, Tajikistan, Turkmenistan and Uzbekistan), etc. Examples of failed states with possible regional impacts include the Central Africa Republic, the Democratic Republic of the Congo, Somalia, South Sudan, Venezuela, Yemen, Zimbabwe, etc. 277 Other studies on the effects of (absolute or relative) poverty or income shocks on conflict or terrorism include, among many others, Krueger and Malečková (2003); Brückner and Ciccone (2010); Besley and Persson (2011); Ciccone (2011); Miguel and Satyanath (2011); Enders and Hoover (2012); Dube and Vargas (2013); Jia (2014); Couttenier and Soubeyran (2014, 2015); Berman and Couttenier (2015); Crost et al. (2016); Harari and Ferrara (2018); Berman et al. (2019);McGuirk and Nunn (2020); Eberle et al. (2020). Typically, poverty and negative income shocks are associated with individual incentives to engage in conflict as well as weakened state and counterinsurgency capacity. 4.1 Introduction 151 Lake Chad Regional Economic Memorandum  |  Development for Peace and Peace, 2012–2020). Until 2014, Boko Haram trade. Indeed, the Boko Haram area historically served concentrated its terrorist activities in the Northeastern as a trade corridor between (relatively wealthier) areas of part of Nigeria close to the border with CCN but did Nigeria and (relatively poorer) Cameroon-Chad to the not enter these countries, mostly to avoid fighting at least East and Niger to the West.278 four government armies instead of one. As such, in CCN until 2014, the estimated effects must be due to the spill- We then use the same econometric framework to over effects described above. identify factors that can help mitigate the effects of foreign conflict shocks. We find stronger mitigation We use a simple DiD framework whereby we compare effects in those areas that were initially better connected to CCN areas “close to” and areas “farther away from” other markets either via trade networks or transportation the Boko Haram region in the years after 2009 versus infrastructure (thus benefiting from a more diversified set before 2009. We find a strong negative effect of Boko of potential trade partners), and more politically and/or Haram on regional economic activities (as proxied by economically “secured” by government consumption via changes in night light intensity)—particularly for areas defense-related facilities (e.g., military headquarters) or within 200 km from the Boko Haram region. More public employment (e.g., social services). precisely, we find an average decline of 10 percent for the post-2009 period, and a decline of about 20 percent for This paper makes four important contributions. First places closer to the shock (within 100 km). For all places of all, various studies show how economic shocks in some within 200 km, we find an overall effect of about 20 locations increase conflict there as well as in neighboring (50) percent by 2013 (2018), that is, 4 (9) years after the locations (Berman et al., 2017; König et al., 2017; Harari shock began. The effects appear to be driven by declines and Ferrara, 2018; Eberle et al., 2020; McGuirk and in per capita incomes rather than population outflows (or Nunn, 2020).279 Two channels explain spatial diffusion. refugees inflows since we control for it). We also show First of all, conflict factors can move (e.g., armies) or that the parallel trends assumption is verified in CCN. be moved (e.g., weapons) spatially. Secondly, due to Finally, we find no effect on local (i.e. non-Boko Haram) economic spillovers, poverty can increase in surrounding conflict in CCN. Therefore, the estimated spill-over locations, thus raising the likelihood of conflict there. effects are purely economic. We do not find any impact of Boko Haram activities in Nigeria on local (non-Boko Haram) conflict activities When studying the confidence intervals of the in CCN, and this despite significant income declines baseline effect, we find that the estimated effects in contexts where most individuals already lived close range from about -30 percent to -10 percent in 2013. to the subsistence level.280 The lack of conflict spillovers Thus, while most places within the 200 km region were does not appear to be due to the increase of government negatively affected, some were less affected than others, forces in the area. Our interpretation is that poverty which motivates us to analyze the heterogeneous effects disproportionately increased in trade-reliant urban of foreign conflict depending on initial (pre-2009) local locations. Even if the opportunity cost of conflict labor conditions. We find stronger effects for initially more decreased, other economic factors must have dominated developed locations, hence more urban locations, which the previous effect and prevented conflict. shows the potential importance of foreign conflict for 278 Likewise, we find weaker effects on “rural” outcomes. Our analysis shows no effects on measures of greenness (proxying for agricultural expansion) or land use. We, however, find effects on agricultural burning, which proxies for agricultural intensification in rural areas (Blankespoor et al., 2021), most likely as a result of reduced urban incomes. 279 There are related literatures on the determinants of the spatial diffusion of conflict (e.g., Bosker and de Ree, 2014; Novta, 2016) and economic shocks (e.g., Amarasinghe et al., 2020). These studies all highlight the role of ethnic networks. Such networks are particularly important for domestic and international trade in Africa (Fafchamps, 2003). 280 Boko Haram also had no incentive to enter CCN, at least until 2014. 152 4.1 Introduction Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram Indeed, the causal mechanism of economic shocks more urban locations. Among urban locations, the least leading to conflict is complex, and the type of economic developed locations were disproportionately impacted. shock and industries should mediate the effects on When studying which locations were more resilient conflict. Positive shocks to labor-intensive industries, economically to the foreign conflict shock, we find that such as agriculture, raise wages and reduce conflict (Dal more connected and more secure locations were better Bó and Dal Bó, 2011; Berman and Couttenier, 2015; able to “weather” some of the impact of the shock. These Harari and Ferrara, 2018).281 But positive shocks to results are, we believe, important for policy because it capital-intensive industries raise the likelihood of conflict identifies potential factors of resilience to foreign conflict because the capital intensive industry expands at the shocks. In contrast, other studies examine countries that expense of the labor intensive one, which lowers the are, or were, directly impacted by conflict (instead of cost of appropriation activities relative to the amount of indirectly via cross-border effects). In these countries, they appropriable resources. Commodity discoveries or price focus on policies aimed at conflict prevention, resolution increases then increase violence because there is more and/or management (de Ree and Nillesen, 2009; Berman to appropriate (Angrist and Kugler, 2008; Dal Bó and et al., 2011; Rohner et al., 2013a; Nunn and Qian, 2014; Dal Bó, 2011; Dube and Vargas, 2013).282 In our case, Crost et al., 2014; König et al., 2017; Chiovelli et al., the shock reduced the amount of appropriable resources 2018; Sviatschi, 2018; Hartman et al., 2018; Eberle et al., and disproportionately impacted sectors that were more 2020) or post-conflict reconciliation (Fearon et al., 2009, capital-intensive than agriculture. The trading sector is 2015; Blattman and Annan, 2015; Blattman et al., 2015). then particularly sensitive to the impact of conflict on The government interventions that our results highlight trade costs, and economic agents may internalize that differ from some of the policies that have been studied in (Martin et al., 2008b, 2012). In our context, urban the literature, partly because the affected border regions residents whose incomes decreased due to the shock do not directly suffer deaths and destruction.283 plausibly internalized that having community members engage in conflict would further reduce incomes. Finally, Thirdly, there is a large literature on the impact of the shock was for the main period of study (wrongly) seen conflict on local economic development  (e.g., Abadie as temporary, as it was believed that the Nigerian army, and Gardeazabal, 2003; Nunn and Wantchekon, 2011; helped by international allies, would eventually eradicate Besley and Reynal-Querol, 2014; Burger et al., 2015; Boko Haram. It may have prevented individuals from Brodeur, 2018; Melnikov et al., 2020).284 However, switching to more conflict-related activities. conflict often arises endogenously due to socio-economic conditions, making it difficult to measure truly causal Secondly, this paper sheds light on the heterogeneous, local economic effects. Our natural experiment has the not just average, effects of conflict on growth at the merit of being simple and allows us to estimate the effects subnational level. In particular, for a similar conflict of a non-local, more exogenous, conflict shock. However, “shock”, the local impact may differ depending on our shock is externally less valid than in some of the initial economic conditions. We find stronger effects for other studies since it measures a cross-border effect. Also, 281 As shown by McGuirk and Burke (2020), global food price shocks increase conflict in areas without crop agriculture where most workers are net consumers of food. In food-producing areas, higher food prices may simultaneously reduce conflict due to the higher incomes and increase conflict from workers whose real wages fall. 282 Related studies include Hodler (2006); Lei and Michaels (2014); Caselli et al. (2015); Berman et al. (2017); Chiovelli et al. (2018); Sviatschi (2018); Castillo et al. (2020); de la Sierra (2020); Adhvaryu et al. (2021). 283 The government interventions studied in the literature include, for example, diplomacy, different types and locations of military interventions, weapon embargoes, reforms to property rights, development programs, service provision, community engagement programs, and food aid in conflict-prone or conflict-ridden areas, and demining, development aid, employment programs, cash transfers, and therapy sessions in post-conflict areas. 284 Studies on the more individual-level effects of conflict include, for example, Bellows and Miguel (2009); Blattman and Annan (2010); Annan et al. (2011); Akresh et al. (2012); Bauer et al. (2016); Sviatschi (2018). 4.1 Introduction 153 Lake Chad Regional Economic Memorandum  |  Development for Peace in CCN, our shock did not directly lead to deaths and (e.g. Glaeser, 2014; Gollin et al., 2016; Haslop et al., destruction (unlike most conflicts).285 2021a), demographic growth (e.g. Jedwab et al., 2017b; Jedwab and Vollrath, 2019), or climate shocks (Barrios Next, one of the mechanisms through which conflict et al., 2006; Henderson et al., 2017; Kocornik-Mina et affects economic development is by reducing trade and al., 2020; Haslop et al., 2021b). To our knowledge, the increasing economic uncertainty. While the linkages literature on conflict (or terrorism) and city growth is between conflict and trade have already been studied, more limited (e.g. Glaeser and Shapiro, 2002; Voigtländer empirical evidence is mostly cross-national (Blomberg and Voth, 2012; Dincecco and Onorato, 2016).288 and Hess, 2006; Martin et al., 2008b,a, 2012; Glick and Taylor, 2010; Qureshi, 2013; Rohner et al., 2013b; Seitz Lastly, our analysis has limitations. The three et al., 2015; De Sousa et al., 2018).286 Our focus instead countries of study are among the poorest in the world.289 lies in understanding how, within a country, spatial Understanding the economic effects of foreign conflict proximity to a foreign conflict affects local economic in such contexts is particularly important. However, development, most likely via trade disruptions as in data infrastructure and finances to collect and produce Chiovelli et al. (2018) who study the effects of post- data can be challenging;290 no consistent panel data on conflict demining on market access and local economic within-country variation in trade and migration flows, development. Other within-country studies such as production, wages, consumer prices and amenities are Berman and Couttenier (2015), Berman et al. (2017) available. or McGuirk and Burke (2020) then exploit trade shocks (from international commodity or food prices) to study The rest of this paper is structured as follows. Section the causal effects of income on conflict rather than the 4.2 provides information on the context while Section effects of conflict on income via trade or the role of trade 4.3 and Section 4.4 describe the data and the empirical diversification in mitigating the local economic impact of strategy, respectively. Sections 4.5 and 4.6 discuss the foreign conflict.287 estimated average and heterogeneous effects of the Boko Haram shock as well as its effects on local (non-Boko Fourthly, this work contributes to a body of literature Haram) conflict. Section 4.7 concludes. on the drivers of city growth in poor countries. Other studies on Africa have focused on the impact of transportation investments (e.g. Storeygard, 2016; Jedwab and Moradi, 2016; Jedwab et al., 2017a; Jedwab and Storeygard, 2020), trade more generally 285 Other studies of the effects of Boko Haram are mostly qualitative. Exceptions include Adelaja and George (2019); Bertoni et al. (2019). However, they study its effects in Nigeria, which complicates causal identification. 286 Martin et al. (2008b,a); Seitz et al. (2015) study the role of trade in conflict. Other studies examine the reverse relationship. Fenske and Kala (2017) study the relationship between historical African conflict and the slave trade. Emran et al. (2019) examine the long-lasting effects from temporary trade restrictions on the spatial distribution of employment and resource allocation exploiting the disrupted change in routes to the international market for two neighboring landlocked countries as the result of the the civil war in Côte d’Ivoire. 287 Berman and Couttenier (2015) find strong effects of negative income shocks (from lower international demand for a location’s crops) on conflict. They find a weaker effect for more remote locations. Because remote locations are more disconnected from international markets, their shock is smaller, hence their effect is smaller. Our analysis differs because we study how trade diversification is a factor of resilience for a given (conflict-driven) economic shock. 288 There is, however, a literature on the economic impact of refugees on cities (e.g., Lewis and Peri, 2015; Alix-Garcia et al., 2018; Fallah et al., 2019; Rozo and Sviatschi, 2021). In our analysis, the number of refugees received by each location is a control, not the main outcome of study. Indeed, we aim to capture the economic spill-over effects of conflict via trade disruptions mostly, instead of the direct reallocation of populations from Nigeria to CCN. 289 According to the World Economic Outlook database of the International Monetary Fund, Niger, Chad and Cameroon are in 2021 the 8th, 10th and 38th poorest countries in the world, respectively. 290 For household survey data collection, Kilic et al. (2017) find significantly higher survey implementation cost per household in Africa compared to other regions of the world. 154 4.1 Introduction Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram 4.2 Background: Studying Boko Haram as a Foreign Conflict In Nigeria. A decade-long insurgency posed by Boko Map 4.1:  Boko Haram Area and the Three Countries of Study Haram in Northeastern Nigeria (2009–present) is a case in point where its devastating economic and humanitarian impact has spilled over to its neighboring countries of Cameroon, Chad and Niger, killing tens of thousands of people and displacing 2.6 million globally (Tayimlong, 2020). According to the Global Terrorism Index of the Institute for Economics and Peace (2012– 2020), Boko Haram became during our main period of study—2009–2013—the world’s deadliest terrorist group, ahead of ISIL, the Taliban and Al-Shabaab. It is still the second deadliest terrorist group as of 2020. The group was founded in 2002. Boko Haram’s radicalization dates back to 2009 when state security forces killed 800 of its members, including its founder M. Yusuf (Kimenyi et al., 2014). At its peak (2015), the group seized a large swath of territories in Northeastern Nigeria, including major cities. 15 million people have been severally affected by the insurgency and the Notes: This figure shows the main Boko Haram area (defined as the area corresponding to the states of Adamawa, Borno and Yobe that is between the Komadugu Yobe river in the Yobe and the Benue river in Adamawa, and where most Boko Haram conflict events counterinsurgency efforts (Vanda Felbab-Brown, 2018). in 2009–2013 were located). It also shows the three countries of study (Cameroon, Chad and Niger) as well as the area within these three countries that is within 200 km from the Boko Haram has continued to engage in killing and Boko Haram area. abducting civilians, forcibly marrying off women and girls to its fighters, and conducting terrorist attacks Adamawa, most conflict events occurred north of the against government property, markets, refugee camps, Benue river. In this figure and in the rest of the paper, and mosques (Omenma et al., 2020). Anecdotal evidence we thus define the core Boko Haram area as the area of abounds to suggest that regional trade has been severely Borno, Yobe and Adamawa that is between the Yobe river disrupted by the insurgency, which has resulted in in the North (in Yobe) and the Benue river in the South repeated temporary border and road closures hampering (Adamawa). the mobility of people, goods and services in the whole Lake Chad region (World Food Program, 2016; Opoku Seen from space, the rise of Boko Haram after 2009 et al., 2017; Foyou et al., 2018; OECD/SWAC, 2020). is strongly associated with a rapid (relative) decline in the level of economic activity in Northeastern As seen in Map 4.1, most of the attacks between Nigeria, as measured based on changes in night- 2009 and 2013—our main period of study—were time light intensity (NTL). As explained in the next geographically concentrated in a few states in the section, this data comes from the U.S. Air Force Defense Northeastern corner of Nigeria, essentially Borno Meteorological Satellite Program (OLS-DMSP, 1992– (60 percent of all Boko Haram conflict events), but 2013). Using data for 7,761 0.1×0.1 degree grid cells also Yobe and Adamawa. However, within Yobe, most (≈ 11x11km at the equator) in Nigeria for the years conflict events took place south of the Yobe river. Within 2000–2013 (N = 108,654), and relying on a simple 4.2 Background: Studying Boko Haram as a Foreign Conflict 155 Lake Chad Regional Economic Memorandum  |  Development for Peace panel difference-in-difference (panel-DiD) framework to Exogeneity. Within-Nigeria effects are not necessarily account for cell and year effects, we find that the level causal given that the rise of Boko Haram might have not of NTL decreased by 6 percent on average between been independent of local socio-economic conditions. 2000–2008 (pre) and 2009–2013 (post) (not shown).291 That said, the timing of the insurgency—2009—could be By 2013, the correlation was -7.5 percent (Figure A4.1 pointed as exogenous. Boko Haram was founded in 2002 shows the coefficient of the Boko Haram area dummy and existed more or less peacefully as a sect for seven years in each year with 2000 being the omitted year). If we (Cook, 2011). When in 2009 the government started restrict the panel-DiD to 3,717 cells that were lit at any investigating Boko Haram’s activities and members were point between 2000 and 2013, we get -8.5 percent and arrested, deadly clashes took place and the insurrection -10 percent economic decline respectively (not shown).292 broke out. For many observers, it was surprising that We focus on the period 2000–2013 because Boko Haram the Nigerian government waited so long before cracking had not yet entered Cameroon, Chad and Niger. In down on the movement. For others, it was surprising that addition, night lights data from DMSP is only available the government finally decided to act in 2009. Thus, the until 2013. However, Li et al. (2020) combine night light government’s investigation could have started anytime data from two satellite series—OLS-DMSP (1992–2013) prior to, or after, 2009. Likewise, such investigation and SNPP-VIIRS (2012–2018)—to generate global could have been successful without resulting in an DMSP NTL time-series data for the whole period 1992– insurrection, or the insurrection might have been swiftly 2018 (DMSP is used as the baseline until 2013). As seen contained instead of dragging on for years.294 Finally, in Figure A4.2, Boko Haram areas have experienced “control” locations outside the Boko Haram area were an even bigger relative decline in night light intensity also affected by Nigeria losing control of almost one fifth between 2014 and 2018. Note that we use the same model of its territory. as just described (N = 7,761 cells) but for the full period 2000–2018. While there are still apparent comparability Focusing on Cameroon, Chad, and Niger (henceforth issues between DMSP and VIIRS, the figure suggests that “CCN”). To bypass these identification issues as well night light intensity might have decreased by as much as as focus on the spill-over effects of foreign conflict, we 60 percent by 2018.293 restrict our analysis to grid cells in CCN. Indeed, it was not until 2014 that Boko Haram expanded its terrorist As expected, the negative correlation between the activities outside the territory of Nigeria and into the Boko Haram area dummy and economic development territory of CCN (Figure 4.1 shows the trends in the decreased in 2015 and 2016 w  hen a coalition of West number of conflict events by country for the period African forces managed to regain part of the territory that 2009–2018). Indeed, Boko Haram did not want to have Boko Haram had captured. However, attacks by Boko to face four government armies. It is only when Boko Haram have since escalated and Boko Haram remains in Haram had no choice that it did, in particular after the control of large swaths of Northeastern Nigeria. Nigeria government dramatically intensified its military campaign against Boko Haram, forcing the movement to 291 The dependent variable is the log of mean light intensity (sum of lights divided by area + 1) in cell c in year t. We include cell c fixed effects, year t fixed effects, and interact the Boko Haram area dummy c (equal to one if the cell is within the Boko Haram area or if its centroid is within 10 km from the area’s border) with a post-2009 (incl.) dummy t. The coefficient of interest is the coefficient of the interacted dummy. To account for spatial autocorrelation, standard errors are clustered at the Local Government Area (LGA; N = 721). With 7,761 cells, there are 11 cells per LGA. 292 Also excluding 89 cells with top-coded pixels (whose maximum value is 63), we get -8.5 percent and -10.5 percent, respectively. 293 The harmonized night light data 1992–2018 is comparable by design, but harmonizing two disparate night light data sets from two different satellite series does rely on model estimates subject to error/assumptions. 294 Using the full sample and the same panel as before and interacting the Boko Haram dummy with a dummy for each year 2001–2013, we find that the negative effect of Boko Haram appears after 2009 (see Figure A4.1). Interestingly, we observe a slight positive effect in 2009, most likely due to the increased military presence in the area. 156 4.2 Background: Studying Boko Haram as a Foreign Conflict Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram move some of its activities to neighboring countries. The suggests that trade volumes severely diminished as borders population of the broader Lake Chad region has since were intermittently closed and major trade routes became been subject to an increasing number of attacks by Boko less accessible or even inaccessible (UNHCR and World Haram, which is now linked to al-Qaeda in the Islamic Bank, 2016; World Food Program, 2016) as well as local Maghreb as well as the Islamic State (Enobi and Johnson- markets (Blankespoor, 2021). Rokosu, 2016; Daouda, 2020). Figure 4.1: Number of Boko Haram Events, 2009–2018 700– 600– 500– 400– 300– 200– 100– 0– 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 Period of analysis After period of analysis J Cameroon J Chad J Niger J Nigeria Notes: This figure shows for Nigeria and each of the countries of study the number of Boko Haram conflict events in each year. As can be seen, the Boko Haram conflict was restricted to Nigeria until 2013 (incl.). The Boko Haram insurrection represented a major economic shock at the “doorstep” of the affected regions in CCN. While the Boko Haram area of Nigeria was about twice poorer (based on mean night light intensity) than the rest of Nigeria in 2008, it was on average almost 10 percent wealthier than the whole sample of CCN (ibid.). In the region—defined as the Boko Haram area plus CCN’s areas within 200 km from the Boko Haram area (see Map 4.1)—the Boko Haram area contributed more than 50 percent of the total sum of night lights in 2008. The economic shock caused by the insurrection was then amplified by the fact that the Boko Haram area offered a major trade corridor between the other three countries. The state capital of Maiduguri is the principal trade hub in Northeastern Nigeria and also between Niger and Cameroon-Chad. Anecdotal evidence 4.2 Background: Studying Boko Haram as a Foreign Conflict 157 Lake Chad Regional Economic Memorandum  |  Development for Peace 4.3 Sample and Main Data for Cameroon, Chad and Niger We focus on estimating the spill-over effects of the period 1992–2013. Each 30-arcsecond pixel (≈ 1x1km) Boko Haram-driven economic shock on the Boko in each satellite-year contains a digital number (DN), an Haram area’s neighboring areas in CCN. Our full integer between 0 and 63, inclusive, that represents an sample consists of 25,491 0.1*0.1 degree grid cells in average of lights in all nights after sunlight, moonlight, CCN for the period 2000–2013 (N = 356,874). Our aurorae, forest fires, and clouds have been removed baseline analysis relies on a subsample of cells that were lit algorithmically, leaving mostly human settlements. This (NTL>0) at any point between 2000–2013, which yields data is typically subject to the issue of top-coding. In a sample of 1,546 cells and a total of 21,644 observations our case, however, this is not an issue. In fact, among (1,546 cells x 14 years). the 1,546 cells of our main analysis, only 11 have some top-coding. Indeed, the three countries of interest are Conflict Data and Boko Haram (BH) Area. We define among the poorest countries in the world. Among these the (core) BH area as the area of Borno, Yobe and Adamawa 11 cells, the mean share of top-coded pixels is then only that is between the Yobe river in the North (in Yobe) and 0.05.296 Finally, to study long-term effects we rely on the Benue river in the South (Adamawa) (see Map 4.1). the harmonized NTL data (1992–2018) from Li et al. For each CCN cell, we obtain their centroid’s Euclidean (2020).297 distance to the BH area. Our main conflict data is from the Armed Conflict Location & Event Data Project Rural Outcomes. NTL may not perform well in (ACLED) (Raleigh et al., 2010). We will also use data capturing economic activities in rural areas which remain from the Uppsala Conflict Database (UCD) (University, largely dark at night. We thus turn to other measures 2020) and the Social Conflict Analysis Database (SCAD) proxying for agricultural economic development in (CCAPS, 2020). rural areas. The first indicator of such activities is the Normalized Difference Vegetation Index (NDVI)—or Nighttime Lights (NTL). Since there is no reliable data Greenness Index—from NASA (2020b) and we calculate that measures income or economic activities at a fine its monthly mean at the grid level from 2001 to 2018. spatial level, we rely on satellite data on light emitted Higher values indicate denser vegetation. From European into space at night.295 Satellites from the U.S. Air Force Space Agency (2017, 2019), we then obtain the share of Defense Meteorological Satellite Program (DMSP) land that can be classified as “cropland”, “mosaic”, “other” have been recording data on lights at night using their or “urban” (available in 2000–2018).298 Operational Linescan System (OLS) sensor since the mid-1960s, with a global digital archive beginning in A common agricultural practice in the region is the 1992. Since two satellites are recording in most years, 30 burning of fields  (Kull and Laris, 2009; Nwaga et al., satellite-years worth of data are available for the 22-year 2010). Thick layers of biomass burning aerosols, generated 295 Henderson et al. (2011) and Bruederle and Hodler (2018) demonstrate the utility of it as a local measure of GDP and human development, respectively. See Michalopoulos and Papaioannou (2013, 2014) for studies on Africa. 296 We could have used instead the radiance calibrated data from NOAA 2015 which has the advantage of not being top coded. However, this data stops in 2011 whereas DMSP-OLS stops in 2013 and we need to study 2009–2013. 297 Li et al. (2020) combine night light data from OLS-DMSP (1992–2013) and SNPP-VIIRS (2012–2018). The nighttime lights from the SNPP satellite, carrying VIIRS, series brings unprecedented information compared to the previous OLS series, including improvements such as spatial resolution (15 arc seconds or 500m) and measurement (14 bit quantization) with a wider dynamic range and lower detection limits (Elvidge et al., 2017). 298 Cropland corresponds to rain-fed, irrigated or post-flooding. Mosaic corresponds to mosaic cropland (>50 percent) or natural vegetation (tree, shrub, herbaceous cover) (<50 percent). Other corresponds to all remaining land cover. 158 4.3 Sample and Main Data for Cameroon, Chad and Niger Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram mainly by agricultural burning during the dry season, can be detected across the Sahel region of Africa (Johnson et al., 2008). Aside from the threat to the atmospheric environment such aerosols pose, agricultural burning also causes the loss of forest system carbon, biomass and nutrient stocks due to deforestation, leading to long-term soil infertility despite achieving short-term soil fertility (Kotto-Same et al., 1997; Kanmegne, 2004). Despite the long-term harm to agricultural outcomes, impoverished farmers resort to agricultural burning to secure food and income.299 Following Blankespoor et al. (2021) who examine the effect of conflict on agricultural activity in the Central African Republic, we sum at the grid level the MODIS Burned Area data product (v6), which provides a burned- area estimate per 500m pixel by month (NASA, 2020a). Then, according to the main food crops for each country- crop calendar (FAO, 2020) we define each month into three seasons: (i) land preparation; (ii) sowing and growing; and (iii) harvest. Finally, the controls and other outcomes considered in our analysis are described below. 299 70 percent of deforestation in Africa is attributed to agricultural burning, compared to 50 percent in Asia and 30 percent in Latin America (Nwaga et al., 2010). In Cameroon, about half of the annual rate of deforestation, at 0.6 percent overall, is for agricultural purposes, while the other half is attributed to logging (Gockowski et al., 2005). 4.3 Sample and Main Data for Cameroon, Chad and Niger 159 Lake Chad Regional Economic Memorandum  |  Development for Peace 4.4 Econometric Specification and Issues We examine in a panel-DiD framework the average allow their effects to vary flexibly over time. First, we effect of the Boko Haram (BH) shock in CCN areas control for the log of the Euclidean distances to the largest neighboring the BH area. To do so, we first investigate city and the capital city in the cell’s country.300 We do so the geographical scope of the BH effect, i.e. how “far” because spatial patterns of economic development over into CCC a significant BH effect is observed. Second, time could be affected by proximity to the main economic we verify that this effect only appears in 2009, thus and political centers of the country. We also control for confirming parallel trends and the local exogeneity of the the log of the Euclidean distance to N’Djamena, the foreign BH shock, and also investigating how the effect capital and largest city of Chad. In Map 4.1, N’Djamena varied over time during the 2009–2013 period. is located in the North-West of Chad, close to the border with Cameroon. Since N’Djamena has been growing Model 1. The model examines the geographical scope of rapidly over time, for reasons unrelated to Boko Haram, the effect and can be formalized as follows: we need to avoid conflating the economic impact of Boko 250 Haram with the rapid expansion of N’Djamena per se. NTLs,c,t = α + ∑βd BHs,c,d * Post 2009 Dummyt d=25 + λs + Kc,t + Xs,cBX,t + εs,c,t (1) Due to attacks close to the border areas, Chad and Niger increased border controls as well as military  here s denotes the cell, c the cell’s country, and t the w presence at their borders with the North-East of year. NTL is the log of mean night light intensity (sum Nigeria. Cameroon also increased controls at the border of lights divided by cell area). Since NTL can be zero in with Chad that is close to the BH area. This may have some years, we use log (mean night light intensity + 1). resulted in public expenditure—and thus night lights— As discussed earlier, for our main regressions we focus on in these areas, which would cause an upward bias of the 1,546 cells with some night lights at one point in 2000– effect. In other words, this would make us under-estimate 2013, thus yielding 21,644 observations in total. λs and how negative the effect is. We thus consider a dummy if Kc,t correspond to cell fixed effects and country-year fixed the cell is a border cell and is within 50 km from the BH effects, respectively. The main variables of interest are the area. interactions of the dummies BHs,c,d equal to one if the cell is d kilometers (in terms of simple Euclidean distance) Resource-rich areas may have also seen their NTL away from the BH area in Nigeria (with d ranging from change over time, for example due to commodity price 25 km through 250 km at an increment of 25 km) fluctuations. For example, there is oil production and oil multiplied by a dummy Post 2009 Dummyt equal to one refining in the three countries and Niger is also a major if the Boko Haram conflict has started, hence post-2009. exporter of uranium. We create a dummy equal to one if the cell intersects with oil- or uranium-producing areas or Controls. We include various important time-invariant contains an oil refinery.301 controls Xs,c which we interact with year fixed effects to 300 These two are different in Cameroon where the largest city is Douala, followed closely by the capital city Yaoundé. 301 We use the Petroleum Dataset version 1.0 (Lujala et al., 2007) to identify onshore oil producing areas and we digitize locations of oil refineries from national sources (e.g. Nigeria Department of Petroleum Resources, 2020). Even though both Chad and Niger have a long history with the oil industry, the only refinery in Chad, Djarmaya, opened in 2011. In Niger, the Agadem oilfield and the Soraz refinery near Zinder opened in 2011. Niger exports oil via Chad or Cameroon. U.S. Geological Survey (2006) then capture the locations of uranium producing areas near Arlit, Niger. 160 4.4 Econometric Specification and Issues Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram Finally, due to heightened insecurity in Northeastern 200 km) vary each year (relative to the omitted year Nigeria, areas close to the border in the three countries 2000) instead of only comparing the 2009–2013 period received Nigerian refugees but also Cameroonese, to the pre-2009 period. More formally, we estimate the Chadian or Nigerien returnees. Some of them were following panel model: accommodated by the governments and international 2013 organizations in formal refugee camps. Others moved to NTLs,c,t = α + ∑ Υi × BH 200kms,c,d i=2001 localities in these areas. As such, this may have induced + λs + Kc,t + Xs,cBX,t (2) population increases and public investments, and thus amplified night lights, in these areas. This would cause  here the dummy variable BH 200kms,c,d coded as 1 if the w an upward bias and thus make us under-estimate the cell is within 200km from the BH area is interacted with negative local effect of Boko Haram. We thus add two year dummies ϒi generated for each year between 2001 dummies for whether there is a refugee camp in the cell and 2013. Our expectation is that the effect becomes (ca. 2015) and the estimated log number of (refugees + negative and statistically significant only after 2009. returnees) in each cell (ca. 2015). However, the influx of refugees + returnees could also have negative economic effects, for example if social tensions are increased as a result. One could then argue that there is overcontrolling. We will thus show that results are little sensitive to the omission of these controls.302 Spatial Autocorrelation. To account for spatial autocorrelation, standard errors are clustered at the 3rd level administrative unit, which corresponds to arrondissements in Cameroon (N = 343), sous-prefectures in Chad (N = 336) and communes in Niger (N = 265).303 For our full sample, this corresponds to 12, 33 and 40 cells per unit on average in each country respectively (areas of 1,452, 3,993 and 4,840 sq km, respectively). We use standard errors clustered using administrative units instead of Conley standard errors because, as discussed in Section 4.7, the latter are computationally intensive with many spatial units. However, we will show in the same section that results hold when clustering at a higher level or using Conley standard errors. Model 2. The second model examines the temporal scope of the effect. In particular, we will find that Boko Haram only has a significant effect within 200 km from the BH area. We then slightly modify Equation (1) so as to let the effect of proximity to the Boko Haram area (within 302 Refugee camp locations come from UNHCR (2020). The estimated local numbers of refugees and internally displaced people come from Direction Régionale de l’Etat Civil et des Réfugiés (2016); IOM (2016); UN OCHA (2015). 303 Administrative country boundaries come from GADM version 3.6. 4.4 Econometric Specification and Issues 161 Lake Chad Regional Economic Memorandum  |  Development for Peace 4.5 Average Effects in Cameroon, Chad and Niger 4.5.1. Baseline Results given the rapid intensification of the Boko Haram insurgency after 2009. By 2013, the effect is about -0.20, Results from the panel-DiD model (1) are shown in so cells “close” to the BH area have lost 20 percent of their Figure 4.2. There is a significant effect of Boko Haram level of economic activity on average. in the range between 25 (0–25) and 200 (175–200) km. The average effect within 50 km (across the 25 and  early Effect of Proximity to the Boko Figure 4.3: Y Haram Area (0–200 km) 50 bins) is -0.15 (p < 0:01), implying that the rise of Effect of 200 km from Boko Haram (relative to 2000) Boko Haram reduces NTL by 15 percent. The average 0.1– effect for 50-100 km (across the 75 and 100 bins), 100- 150 km (across the 100 and 125 bins) and 150–200 km (across the 175 and 200 bins) is -15, -11 and -7 percent 0– (p < 0:01), respectively. The average effect within 200 km is then -0.12 (p < 0:01), implying an average decrease of -0.1– 12 percent. For the sake of simplicity, in the rest of the analysis we focus on a simple 0–200 km dummy, thus -0.2– estimating an average effects across all affected bins.  ost-2009 (Incl.) Boko Haram Effect by Figure 4.2: P -0.3– Distance to the Boko Haram Area 2000 2002 2004 2006 2008 2010 2012 2014 Estimated post-2009 effect (incl.) for each bin with 95 percent confidence 0– Notes: The figure shows the yearly effect (relative to the year 2000) of a dummy equal to one if the cell is within 200 km from the Boko Haram area. See Equation 2 for details on the specification. * p<0.10, ** p<0.05, *** p<0.01. -0.05– If we use same model but further separate the BH -0.1– 0-200 km dummy into 0–50, 50–100, 100–150 and 150–200 km dummies, we find that Boko Haram has -0.15– no effect before 2009 in the four groups of cells. The effect by 2013 is then about -15, -20, -20 and -30 percent -0.2– (p < 0:01), respectively (see Figure A4.3). Aggregating some of these effects, cells within 100 km have lost almost -0.25– 25 percent while the cells farther away (but still within 25 50 75 100 125 150 175 200 225 250 Distance bin (km, to Boko Haram area) the 200 km window) have lost about 15 percent. In the Notes: The figure shows the post-2009 (incl.) Boko Haram effect for each distance (to the rest of the analysis, we will also sometimes distinguish Boko Haram area) bin. 25 corresponds to 0–25 km, 50 corresponds to 25–50 km, ..., and 250 corresponds to 225–250 km. See Equation (1) for details on the specification. See Appendix 0–100 and 100–200 km. for data sources. * p<0.10, ** p<0.05, *** p<0.01. Finally, we use the panel-DiD model of eq. (1) and the We find evidence that the assumptions of parallel harmonized NTL data from Li et al. (2020) to study trends and local exogeneity of the BH shock hold. As the long-term effects of the shock. There are several seen in Figure 4.3, when using the model of Eq. (2) no caveats with this analysis. First of all, there may still be effect is observed before 2009, a small effect is observed in comparability issues between DMSP (used for the 2000– 2009, and the effect decreases after that. This is expected 2013 period) and VIIRS (2014–2018) in the data of Li et 162 4.5 Average Effects in Cameroon, Chad and Niger Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram al. (2020). Second, Boko Haram had attacked Cameroon 4.5.2 Foreign Conflict as a Trade Shock by 2014 and Chad and Niger by 2015 and there may have Disproportionately Impacting been local and spill-over effects of these attacks. However, Cities? only 60 cells were ever affected in CCN. To attempt to study the long-term effects of foreign conflict, we exclude In this section, we examine whether the foreign the 60 cells as well as 166 cells within 50 km of these conflict shock disproportionately impacted trade- 60 cells. We also control for the log of the Euclidean reliant cities, mostly due to trade disruptions. To do distance to a CCN Boko Haram event in year t. so, we first show using the night lights data and other data on rural economic development that urban areas As seen in Figure A4.4, the negative effect of Boko were far more impacted than rural areas. Next, we argue Haram increased in magnitude over time, reaching that curfews, the in-migration of refugees and/or the -35 percent by 2015 and -50 percent by 2018. We outmigration of residents were not driving the results. see some recovery effects in 2016 when West African Finally, we do not find evidence for spill-over effects on troops managed in 2015 to regain some of the territory conflict. Thus, incomes did not decrease in the border captured by Boko Haram in Nigeria, another implicit test regions because conflict factors (e.g., armies and weapons) of our identification strategy. We thus see positive spill- moved from the BH area to these regions. Ultimately, we over effects of a successful foreign counter-insurgency believe that conflict in the BH reduced CCN’s trade with campaign. Next, the high standard errors for the VIIRS Nigeria but also trade between the regions of Cameroon- observations likely reflect the fact that the assumptions Chad and Niger that historically used the BH area as a made by Li et al. (2020) to recreate consistent NTL for trade corridor. the whole period also introduced a significant amount of noise. Lastly, we may not be capturing a long-term effect per se as the conflict never ended.304 4.5.2.1 Other Results on Night Lights and Rural Economic Outcomes Overall, we find very strong negative local effects of foreign conflict. The question now is which sectors, Night lights. Our analysis thus far focused on the sample and thus locations, foreign conflict disproportionately of cells that were ever lit at some point between 2000 impacts and why. and 2013. We now consider other samples of cells. Table 4.1:  Post-2009 Effect of Proximity to the Boko Haram Area (0–200 km), Night Lights Col. (1)–(3) and (5): Log (Mean Night Light Intensity + 1) in Year t Dependent Variable: Col. (4): Dummy if Mean Night Light Intensity in Year t > 0 Intensive All Extensive Extensive Pure Intensive Sample: (1) (2) (3) (4) (5) -0.097*** -0.007** -0.001 -0.004 -0.143*** BH 200 Km * Post-2009 [0.027] [0.003] [0.001] [0.003] [0.046] Cell FE, Country-Year FE Y Y Y Y Y Year FE*Controls Y Y Y Y Y Observations 21,644 356,874 348,470 373,227 7,835 Adjusted R-squared 0.89 0.89 0.21 0.27 0.93 Notes: SEs clustered at the 3rd-level administrative unit. * p<0.10, ** p<0.05, *** p<0.01. 304 The data from Li et al. (2020) generate more consistent results between the pre- and post-2013 periods for CCN than for Nigeria. Indeed, Li et al. (2020) explain that the harmonization of the DMSP and VIIRS series might work differently for locations with a level of night lights below vs. above 30. Nigeria has locations both below and above 30 whereas there are fewer such values in CCN. As such, harmonization should be less problematical there. 4.5 Average Effects in Cameroon, Chad and Niger 163 Lake Chad Regional Economic Memorandum  |  Development for Peace Table 4.1 shows the results for: i) the “intensive” margin economic activities: greenness, land use, and agricultural sample, i.e. only the cells that were ever lit between 2000 burning. and 2013 (Col. 1); ii) all observations (Col. 2); iii) the “extensive” margin sample, which excludes cell-years with The effects of foreign conflicts on rural economic night light intensity > 0 in t-1 (Col. 3); iv) the “extensive” development are theoretically ambiguous. Rural areas margin sample but with a simple dummy coded 1 if light are possibly isolated from such shocks if they do not trade intensity is higher than 0 in year t (Col. 4); and v) the with foreign areas. However, if they sell their agricultural “pure intensive” margin sample that consist of cell-years products to the foreign area, the level of demand decreases. for which light intensity > 0 in both t-1 and t (Col. 5). Furthermore, if urban areas are negatively impacted by reduced trade with the foreign area, this could in turn The negative effects of foreign conflict are particularly impact the demand for agricultural products in rural pronounced in urban areas (as reflected in cells that areas. In a context of high population growth, the latter are lit between 2000 and 2013). In the “intensive” and mechanisms would lead to slower rates of land expansion. “pure intensive” samples (Cols. 1 and 5), the average effects are -10 percent and -14 percent both significant Alternatively, if urban areas import rural products at the 0.01 level, respectively, whereas these effects are from the foreign area, insecurity may lead urban smaller in the full sample (Col. 2) and insignificant at the areas to demand local rural products instead. Reduced extensive margin (Cols. 3 and 4).305 More generally, the economic opportunities in urban areas trading with intensive margin effect of -0.097*** in Col. 1 represents the foreign area could also lead urban residents to seek about 47 percent of the mean in the sample (which is economic opportunities in the rural sector (in the region, 0.47) whereas the extensive margin effect of -0.004 it is common for urban residents to have farming relatives in Col. 4 represents only 5 percent of the mean in the in surrounding rural areas). In such cases, we might sample (0.07). observe faster land expansion. We thus do not find any effect of foreign conflict on Greenness. The measures of greenness, land use and the likelihood that non-lit cells become lit, a proxy for burned area are available at the grid cell level for 23,945 rural economic development. Villages and small towns cells without night lights at any point between 2000 and close to the BH area are thus not less likely to generate 2013, which correspond to more rural areas.306 In terms enough luminosity picked up by the satellites. These of the Greenness index, data is available on the monthly results could suggest that the geographical scope of the basis. When studying monthly patterns, we find that spill-over effects is limited to more urban areas, likely greenness peaks in August in Niger and Chad—at the because these urban settlements rely more extensively on height of the rainy season—and is high in Cameroon regional trade with, or through, the BH area than their around May (the light rainy season) and September (the rural counterparts (more on this later). heavy rainy season). Once one accounts for country- month effects, greenness could capture land expansion or However, one caveat is that NTL may not measure land abandonment and thus proxy for rural growth. well rural growth or decline, even when focusing on the extensive margin only. Thus, to better examine F  or greenness (available in 2001–2013), the model is the rural effects, we study other reasonable proxies for rural same panel-DiD model as before except the dependent variable is the log of (mean greenness + 1) in the cell s in 305 We find similar non-effects at the extensive margin when separating 0–100 km and 100–200 km (not shown). 306 Results are similar if we keep all cells including those cells that are ever lit between 2000 and 2013 (not shown). 164 4.5 Average Effects in Cameroon, Chad and Niger Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram  ost-2009 Effect of Proximity to the Boko Haram Area (0–200 km), Rural Outcomes Table 4.2: P Share LogMean Col. (3)–(4) and (6)–(9): Log (Agricultural Burning + 1) in t Dep. Var.: Crop+ Green. t Col. (5): Dummy if Agricultural Burning > 0 in t Mos Pure Land All All All Extensive Intensive Sowing Harvest Sample: Prep. Growing (1) (2) (3) (4) (5) (8) (9) (6) (7) -0.000 -0.001 0.042** 0.007* 0.007 0.047 0.001 0.005* 0.070** BH 200 km * Post-09 [0.001] [0.001] [0.017] [0.004] [0.006] [0.035] [0.016] [0.003] [0.027] Cell FE, Cntry-Yr FE Y Y Y Y Y Y Y Y Y Year FE*Controls Y Y Y Y Y Y Y Y Y Observations 311,285 311,285 232,040 232,040 61,496 311,285 311,285 311,285 311,285 Adjusted R-squared 1.00 1.00 0.85 0.40 0.38 0.71 0.77 0.31 0.79 Notes: SEs clustered at the 3rd-level administrative unit. * p<0.10, ** p<0.05, *** p<0.01. year t (mean = 0.31).307 As can be seen in Col. (1) of Table land that was not agricultural before), slash-and-burn 4.2, we find no effect of Boko Haram on greenness (the is common there. We thus investigate how agricultural point estimate only represents 5 percent of the mean in burning (available in 2001–2013) varies with the rise of the sample that is 0.31).308 Boko Haram, depending on the season: “harvest”, “land preparation” and “sowing-growing”.310 Since agricultural Land Use. We know the share of land that can be burning area can be equal to 0, we use log(burning + 1) classified as either “cropland” or “mosaic cropland” (at as the dependent variable. As we did for NTL, we explore least 50 percent cropland) for the full period 2000–2013 different margins (intensive, extensive, etc.) in Table 4.2 (mean = 0.09). As seen in Col. (2), we also find no effect of Col. 3–9. Boko Haram on land intensification or de-intensification (the point estimate only represents 1 percent of the mean In Col. 3, which includes all cell-years, we find a in the sample that is 0.09).309 positive and significant effect of 0.042***. It is however smaller than what was found for night lights. In particular, Agricultural Burning. When studying monthly the point estimate represents 20 percent of the mean in patterns in agricultural burning, we find that it peaks in the sample (20.4) against 47 percent for night lights. November-December when the harvest season ends. This type of agricultural burning corresponds to the practice The burning effect is driven by both extensive margin of stubble burning, where farmers set fire to the straw (Col. 4) and pure intensive margin (Col. 6) effects. stubble that remains after crops have been harvested. More precisely, in Cols. 4 and 5, we restrict the sample Agricultural burning is then observed until April–May, to cell-years whose burning in t-1 is zero. In Col. 4, the at the end of the land preparation season. For land outcome is log(burning + 1) in t whereas in Col. 5 it is a preparation (which includes the preparation of new dummy equal to one if burning > zero. The effect on the 307 Greenness has negative values. To use logs, we first shift all observations by the absolute value of the minimum value in the data (so that the new minimum value is 0) and then add +1. Also, since greenness is not bottom-coded we do not need to distinguish the intensive and extensive margins as we did for NTL. 308 For the sake of simplicity, we use mean greenness averaged across the 12 months of a given year. We obtain the same non-results if we regress for each cell- year-month greenness on country-month dummies and use as our measure the log of the average of the residuals (not shown). There also no effects for 0–100 vs. 100–200 km (ibid.). 309 The coefficients are not significantly different between 0–100 km and 100–200 km (not shown). 310 We rely on crop calendars from FAO GIEWS. “Harvest”: October–November in Cameroon; September–November in Niger; September–December in Chad. “Land preparation”: December–April in Cameroon; December–May in Niger; January–April in Chad. “Sowing-growing”: May–September in Cameroon; June–August in Niger and Chad. 4.5 Average Effects in Cameroon, Chad and Niger 165 Lake Chad Regional Economic Memorandum  |  Development for Peace dummy is small and not significant (Col. 5). However, to Nigeria decreased, there was also less competition the effect on log(burning + 1) is positive and significant from rural products coming from Nigeria. However, (0.007*). Thus, BH resulted in more burning amongst even if land use did not change overall, it could still be those cells without any burning in the previous year. In that the shock very negatively impacted some farming Col. 6, we focus on the pure intensive margin effect for communities. As their members likely live close to the cell-years with burning > 0 in both t-1 and t. The effect subsistence level, they found ways to increase short-run is not significant but seven times higher than for the incomes even if it meant borrowing against the future. extensive margin (0.047, or about 5 percent). Overall, while some rural areas were negatively impacted, the rural sector does not appear to have been driving Burning, while traditionally used, is not a sustainable the economic crisis observed in the region, hence our farming practice as it depletes the nutrients in the characterization of the Boko Haram-led economic shock soil. Results suggest that agriculture is little mechanized as an “urban” shock. (i.e., more traditional) in these areas, and that farmers are willing to increase short-term incomes at the expense of future incomes. Thus, farming households (and their 4.5.2.2 Income Shocks, Migration, and Urban Land possibly more urban-based members) may have become Expansion more present-biased in the face of the shock. Note that these results hold if we exclude border cells within Curfews. First of all, the reduction in night lights was 50 km from the Boko Haram area in case the measures of not due to curfews. While curfews were indeed imposed agricultural burning pick up fires related to destruction in some parts of Northeastern Nigeria, especially around caused by Boko Haram itself (not shown). the city of Maiduguri, there were no curfews occurring in CCN before Boko Haram actually entered these Finally, in Cols. 7–9 which disaggregate the results countries. by different seasons, we show the effects are driven mainly by the end of the harvest period. This finding Refugees and Returnees. Second, we could imagine implies that burning was not a result of preparing new that the inflows of refugees and returnees had negative land that had not been exploited before (Col. 7) but came economic effects on host communities in the border from increasing income as soon as the harvest season was regions. Such inflows could also have had positive effects over (Col. 9). This practice is particularly damaging in the if they generated economic activity and/or led to local long run since soils cannot recover at all. Also, the fact increases in public expenditure. The results reported so that it is at the end of the harvest season indicates that the far are conditional on various controls for the location of observed effects are for parcels that were already exploited refugee camps and the (log) number of returnees in each the year before, not new parcels (in line with the non- cell c. 2015 (all interacted with year fixed effects to allow results for greenness and land use).311 their effects to vary over time). Our baseline intensive margin effect is -0.097*** (Col. 1 of Table 4.1). If we omit Overall, we find no effect on rural lights or land the refugees/returnees controls, we obtain a slightly more expansion. Thus, the positive effects of the shock on rural negative effect of -0.103***. If anything, this suggests growth must have somewhat compensated its negative that the inflows of refugees/returnees had, on net, slightly effects. For example, even if the export of rural products positive, not negative, local economic effects.312 311 Throughout (Cols. (3)–(9)), the effects are stronger for 0–100 km than for 100–200 km (not shown). 312 Results are similar whether we omit the “refugees” controls only or the “returnees” controls only (not shown). As expected, cross-sectional regressions for the 1,546 cells confirm that the border regions had more refugee camps c. 2015 (Ibid.). However, conditional on the baseline controls, they did not receive more returnees c. 2015 (Ibid.). 166 4.5 Average Effects in Cameroon, Chad and Niger Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram Population Outflows due to Heightened Insecurity. areas decreases. However, housing is durable (Glaeser and We could also imagine that populations afraid of the Gyourko, 2005). Thus, if people outmigrate, housing rise of Boko Haram in Nigeria left the border regions, prices decrease, incentivizing them to stay. Individuals thus causing reductions in luminosity. Indeed, changes more sensitive to lower housing prices are more likely to in night light intensity (sum of night lights divided stay, thereby resulting in a greater proportion of poorer by area) may reflect both changes in nighttime lights individuals. As housing supply is now relatively higher per capita (in other words, per capita incomes) and (compared to demand), there is less construction. Since population changes (net migration). Of course, the two construction takes the form of land expansion in poor subcomponents are mechanically correlated. If incomes countries (Jedwab et al., 2020, 2021), one prediction decrease, local residents will more likely out migrate to could be that there is less land expansion in these areas as other areas and non-local residents will less likely migrate a result of the shock. However, the effect should not be in. We now discuss the respective contributions of each instantaneous since the construction sector often reacts channel, which allows us to discuss the potential role of with some temporal lag. In addition, people may wait for outmigration. a few years before deciding whether to outmigrate and thus just “weather” the shock. In particular, observers Suppose that income (NTL) per capita increases in initially did not expect the BH insurrection to last long a cell relative to other cells. Under this hypothetical as Nigeria was the most developed country in West situation, people migrate in and population density in Africa. The residents of Cameroon, Chad and Niger also settled areas initially increase (built-up area is fixed in the probably expected the BH shock to be temporary. short-run as construction takes time). As a result, housing prices increase. Housing supply eventually responds. In To conclude, with the negative BH-led economic areas where land is relatively cheap and construction shock, we may expect a strong effect on NTL per technology not so advanced, housing supply is likely to capita that is only weakly associated with an effect on respond by using more land, not building taller structures. population density and land expansion. In that case, Hence, the cell’s built-up share should eventually increase. most of the effect on NTL should come from changes in As urban land expands, the population density in settled NTL per capita. areas that initially increased redecreases. As population increases, NTL per capita may also decrease after initially To better assess the plausibility of the previous increasing if increased labor supply reduces wages. hypothesis, in Table 4.3 we focus on urban population However, the levels of income per capita and population outcomes by leveraging data from the Global Human density are likely to remain higher than they were before Settlements (GHS) database. GHS use satellite data to the initial per capita income increase. In this case, cell obtain for each cell built-up land area over time, more growth may be captured by a combination of NTL per precisely c. 1975, 1990, 2000 and 2013/14, which capita, population density (population divided by built- nicely coincides with the end of our period of study.313 up areas) and land expansion (built-up area divided by Furthermore, GHS reconstructs city populations c. 1975, total area). 1990, 2000 and 2015, using urban population levels at a relatively low administrative level circa these years and Now, when income (NTL) per capita decreases in a given then allocating the population within these administrative location to another location, people out-migrate (or areas depending on the distribution of built-up area.314 migrate-in less). As a result, population density in settled However, the population levels reported by GHS may 313 Built-up area is from GHS Builtup (Corbane et al., 2019). See https://ghsl.jrc.ec.europa.eu/ for details. 314 Note that the GHS database focuses on urban agglomerations with more than 50,000 inhabitants c. 2015. 4.5 Average Effects in Cameroon, Chad and Niger 167 Lake Chad Regional Economic Memorandum  |  Development for Peace Table 4.3: Post-2009 Effect of Proximity to the Boko Haram Area (0–200 km), Urban Outcomes LogUrb. (6), (8)–(9): Log (Built-Up Area + 1) t Dependent Variable: Log (Mean Light Intensity + 1) t Pop. t (7): Dummy if Built-Up Area t > 0 All All Niger Niger Niger All Extensive Intensive Sample: (1) (2) (3) (4) (5) (6) (7) (8) (9) -0.18*** -0.17*** -0.24*** -0.24*** 0.07 0.17 -0.05 -0.49 0.11 BH 200 km * Post-2009 [0.06] [0.06] [0.07] [0.07] [0.19] [0.46] [0.10] [0.86] [0.19] Log(BuiltUp Area/ 9.13*** Area+1)t [1.85] 0.023 Log(Urb. Pop./Area+1)t [0.015] Cell FE, Country-Year FE Y Y Y Y Y Y Y Y Y Year FE*Controls Y Y Y Y Y Y Y Y Y Observations 4,638 4,638 1,689 1,689 1,165 4,638 2,237 2,237 2,401 Notes: SEs clustered at the 3rd-level administrative unit. * p<0.10, ** p<0.05, *** p<0.01. not be reliable in our context because of the lack of census In Col. 2, we control for the log of built-up density data. In Niger, there were censuses in 2001 and 2012, (or urban built-up area divided by total area) since so their population level c. 2015 actually reflects 2012. it is available for all cells in 1990–2013. The effect is For Chad, the last two censuses were 1993 and 2009. only slightly lower, at -0.17 (p < 0:01), hence -17 percent. For Cameroon, these were 1988 and 2005. As such, Thus, assuming built-up density captures the effects of the reported population levels for 2015 likely measure both built-up density and population density in settled populations pre-Boko Haram. Thus, in this analysis, we areas—thus, population—almost all of the effect of the report the results based on built-up land area or the results shock on night light intensity must be due to the income based on urban population sizes but for Niger only. shock (i.e., NTL per capita).315 In Table 4.3 Cols. 1–4, we use log(NTL + 1) for the In Cols. 3–4, we focus on Niger, the only country with years 1992—which we call “1990”—2000 and 2013. city population data post-2009. In Col. 3, we run the This is the same panel-DiD regression (eq. (1)) as before same regression as in Col. 1 for Niger only. The estimated but we exclude the years in between. Note that we use effect is -0.24 (p < 0:01). Thus, the negative effect on data from 1990, 2000, and 2013 to mimic the structure NTL appears to have been stronger in Niger than in of the GHS data. First, in Cols. 1–3, we focus on the Cameroon/Chad. However, if we control for log urban 1,546 cells with NTL > 0 at any point in 2000–2013 × 3 population density (total city population divided by area) years, hence N = 4,638. The sample of 1,546 cells is the in the cell, we observe the same effect. Thus, almost all sample where we showed strong negative effects on NTL. of the effect of the shock on night light intensity must More precisely, according to Figure 4.2, we had an effect be due to reductions in income per capita (i.e., NTL per of almost -0.20 (hence -20 percent) by 2013. In Col. 1, capita).316 Relatedly, if we use log(total city population) as we use the same BH 200 Km dummy × post-2009 (in the dependent variable, thus comparing the population this case, the year “2013”, hence 2013/14) and obtain size of existing urban agglomerations over time, we also -0.18 (p < 0:01), hence the same result. find no effect of Boko Haram post-2009 (Col. 5). 315 Note that we use the log of (urban built-up area divided by total area +1). Indeed, some cells with NTL > 0 have an urban built-up area of 0 according to GHS. We thus verify that these cells also have very low levels of NTL. 316 Since some cells have no urban population according to GHS, we use log(total city population area + 1). 168 4.5 Average Effects in Cameroon, Chad and Niger Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram Alternatively, we study if built-up area changed due Overall, we find that the negative effects of foreign to the shock. Since structures are durable, built-up areas conflict on local economic development are driven by did not shrink. However, the shock may slowed down per capita incomes falling, not migration. If anything, urban land expansion. We thus study log(built-up area affected individuals appeared to have stayed in these + 1) for the years 1990, 2000 and 2013.317 In Col. 4, we areas despite the massive income shock, one plausible focus on the same 1,546 cells but study how log built-up explanation being the fact that the shock was seen as area grew slower with the shock. Given that structures temporary (even it was not in the end). are durable, we control for log (built-up area +1) in t-1 interacted with year fixed effects.318 4.5.2.3 Foreign Conflict, Local Conflict, and Local As seen in Col. 6, the coefficient is positive and not Economic Development significant. Thus, the main negative effect on night light intensity is not due to urban land expansion slowing Foreign conflict should have direct economic effects. down. Next, in Cols. 7 and 8, we study the extensive However, foreign conflict can also have a direct impact margin and focus on cell-years whose built-up area in on local conflict, for example by increasing the supply t-1 is zero (we no longer need to control for past built- of weapons and trained mercenaries in the region. up area). In Col. 7, the dependent variable is a dummy Alternatively, foreign conflict, by reducing local incomes, if the cell has some built-up area in t. In Col. 8, it is increases the likelihood of local conflict. In that case, we the log of (built-up area +1). The effect is negative but still capture a direct economic effect of foreign conflict insignificant, which leads us to conclude that while urban but the effect is magnified by a local conflict effect. While land expansion could have slowed down, reductions in possible, we show below that the Boko Haram shock income (NTL) per capita drove the results. did not increase the likelihood of local conflict in CCN. Consequently, the effect estimated so far are the pure Finally, in Col. 9, we focus on the pure intensive direct economic effects of foreign conflict. margin, keeping cell-years whose built-up area in t-1 is higher than 0 (we control for past built-up area). The For the years 2000–2013, we employ the same panel- positive effect suggests accelerated urban land expansion DiD model as before, but we now use measures of in cells where there were already built-up areas. Since the conflict as the dependent variable. In Panel A of Table overall effect (Col. 6) is positive, the intensive margin 4.4, the dependent variable is a dummy equal to one if a effects must have been stronger than the extensive margin conflict event unrelated to Boko Haram occurred in the effects. This may be counter-intuitive since the coefficient cell in year t. In Panel B, it is the number of non-Boko in Col. 9 is lower in absolute value than the coefficient Haram conflict events in the cell in year t (unlogged in Col. 8. However, the coefficient captures percentage because there are few events in a same cell in each year). changes, so the absolute effects depend on the initial Next, for each conflict database, we study the effect for all levels of built-up area in cells with built-up areas in t-1. cells first and then for the intensive sample (where NTL > 0 at any point in 2000–2013) and the extensive sample separately. Finally, ACLED and UCD focus on armed 317 Since most cells have the same area, log built-up area is similar to the log of the share of built-up areas. 318 This allows for the durability effect to vary over time, for example due to changing construction technologies. Adding a lag of the dependent variable in a panel model introduces a dynamic panel bias (Nickell, 1981) so these results should be taken with caution. However, we do not need these controls when studying the extensive margin. 4.5 Average Effects in Cameroon, Chad and Niger 169 Lake Chad Regional Economic Memorandum  |  Development for Peace Table 4.4: Effects of Boko Haram on Domestic Conflict, Various Databases, 2000–2013 Conflict Database: ACLED (Armed Conflict) Uppsala (Armed Conflict) SCAD (Social Conflict) All Intensive Extensive All Intensive Extensive All Intensive Extensive Sample: (1) (2) (3) (4) (5) (6) (7) (8) (9) Panel A:  Dep. Var.: Dummy if Non-Boko Haram Conflict Event in the Cell in Year t 0.0002 0.0047 -0.0002 0.0003 0.0035 0.0000 0.0005 0.0028 0.0002 BH 200Km * Post-09 [0.0004] [0.0042] [0.0003] [0.0002] [0.0025] [0.0001] [0.0005] [0.0046] [0.0002] Mean 0.0011 0.0106 0.0004 0.0004 0.0028 0.0002 0.0005 0.0047 0.0002 Panel B:  Dep. Var.: Number of Non-Boko Haram Conflict Events in the Cell in Year t 0.0024 0.0219 0.0000 0.0009 0.0092 -0.0000 0.0010 0.0055 0.0003 BH 200Km * Post-09 [0.0022] [0.0223] [0.0006] [0.0007] [0.0072] [0.0001] [0.0007] [0.0070] [0.0002] Mean 0.0011 0.0106 0.0004 0.0004 0.0028 0.0002 0.0005 0.0047 0.0002 Cell FE, Cntry-Yr FE Y Y Y Y Y Y Y Y Y Yr FE*Controls Y Y Y Y Y Y Y Y Y Observations 356,874 21,644 335,230 356,874 21,644 335,230 356,874 21,644 335,230 Notes: SEs clustered at the 3rd-level admin. unit. * p<0.10, ** p<0.05, *** p<0.01. conflict (Cols. (1)–(6)) whereas SCAD focuses on (not We then obtain similar non-results if, for the full necessarily armed) social conflict ((7)–(9)).319 (intensive + extensive) sample,  we: (i) study a composite index based on the number of conflict events plus 0.5 × As seen, none of our variables is significant. Therefore, the number of fatalities, thus giving more weight to more the likelihood of domestic conflict did not increase lethal conflict events (note that 0.5 is arbitrary); and (ii) significantly, which suggests that any economic impact of examine specific types of conflict. The results when using BH on neighboring areas in CCN must have been due to conflict data from ACLED, UCD and SCAD can be seen reduced trade, not a direct effect of BH on conflict supply in Tables A4.1, A4.2 and A4.3, respectively.321 We then factors. Likewise, the effect of the BH-led economic find similar non-results if we focus on the intensive shock was not reinforced by an indirect feedback effect sample only (not shown, but available upon request). in which poverty led to conflict, which in turn further caused poverty.320 Next, CCN’s governments increased military presence in the region. As such, domestic conflict might have In addition, the effect of BH on conflict appears been prevented in areas close to BH. Yet, if increased stronger (but is still not significant) in the more military presence came from redeployment, which is urban intensive sample than in the more rural plausible given the time it takes to expand an army, it extensive sample. Indeed, more urban areas have been might have increased conflict in areas farther away disproportionately hit by the BH-led economic shock. from BH. However, we do not observe negative effects. In addition, we find similar non-results as in Table 4.4 319 The total number of conflict events that took place in 2000–2013 is 900 in ACLED, 221 in UCD, and 276 in SCAD. The discrepancy between ACLED, UCD and SCAD could be due to them capturing distinct aspects of conflict or the way they assign the events to specific locations. However, results hold if we combine the three databases (not shown). 320 We also do not find stronger effects for 0–100 km than for 100–200 km (not shown, but available upon request). 321 For ACLED, we consider battles, violence against civilians, protests/riots, non-violent strategic developments, and explosions/remote violence. For UCD, we consider state violence (government forces are involved), non-state violence (none of the warring parties is a government), and one-sided violence (armed force is used against civilians). For SCAD, we consider demonstrations, riots, strikes, and violence. Note that the significant effect for UCD and one-sided violence (A2) is due to conflict ending in Eastern Chad in 2008, so not Boko Haram in Western Chad. 170 4.5 Average Effects in Cameroon, Chad and Niger Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram if we drop cells located within 50 km from a military result as confirming that the very negative economic or gendarmerie headquarter c. 2020 (Table A4.4). In impact of Boko Haram on neighboring areas in CCN francophone countries the gendarmerie is a paramilitary was driven by reduced trade in the region. organization with law enforcement duties among the civilian population and gendarmes often intervene where there is a national emergency crisis.322 Lastly, one way to interpret these non-results is that reduced urban incomes (especially related to a trade shock) does not automatically lead to more conflict. Otherwise, the average effects would have been significant. Thus, foreign conflict does not always beget domestic conflict. A body of literature has shown that negative income shocks, most often related to weather related shocks, lead to increased instances of conflicts ( Berman and Couttenier, 2015; Harari and Ferrara, 2018). Hegre and Sambanis (2006) also show that conflict begets more conflict. Lower incomes are often one of the main channels explaining spillover effects. Indeed, with lower incomes, the cost of hiring soldiers is lower (i.e. the opportunity cost of conflict labor is lower) (Harari and Ferrara, 2018). The existing literature relies on shocks that disproportionately affect the agricultural sector and thus rural areas. However, our shock disproportionately impacts urban areas, and urban areas might be more negatively impacted by conflict than agriculture. Indeed, urban production relies more on trade and thus security whereas rural production relies more on fixed factors of production such as land. Subsequently, rural production should be less affected by conflict than urban production. As such, there could be reduced economic incentives to engage in conflict when the income shock originates in urban areas. To summarize, while it is possible that the shock led to increased conflict in some areas of CCN, on average we do not find significant effects of Boko Haram activities in Nigeria on domestic conflict. We interpret this non- 322 Military headquarters include the headquarters of military regions (5–8 depending on the country). Gendarmerie headquarters include the headquarters of “compagnies” or “legions de gendarmerie” (15–23). Sources used include administrative sources, security reports, newspaper articles, and Wikipedia. There is no data for the pre-2009 period. 4.5 Average Effects in Cameroon, Chad and Niger 171 Lake Chad Regional Economic Memorandum  |  Development for Peace 4.6 Heterogeneous Effects for Cameroon, Chad and Niger Now that we have identified the “nature” of the Boko interaction between the 200 km Boko Haram dummy Haram shock for neighboring areas in CCN, we can and cell-specific characteristics defined c. 2009 or before. investigate the factors that accentuated or mitigated Lastly, given the country-year fixed effects we compare these spillovers of foreign conflict. cells within the same country. As seen in Figure 4.3, the 95 percent confidence interval values of the estimated effects vary 4.6.1 Heterogeneity with Respect to Initial significantly, from -0.10 to -0.30 percent in 2013. Economic Conditions For the year 2018, the effects varied by between -30 and -80 percent (Figure A4.4). However, given issues when We first explore how the effects vary depending on harmonizing the DMSP and VIIRS series, these values initial economic conditions,  i.e. night light intensity in respectively represent upper- and lower-bound values of 2008 (Boko Haram rose in 2009). For each cell, we create the 95 percent confidence intervals. a dummy equal to one if the cell’s night light value in 2008 is below the 10th or 25th percentile (i.e., the cell Likewise, the effect varies across the three countries. is “less” developed) or above the 75th or 90th percentile In particular, we use the same panel-DiD model as (i.e., the cell is “more” developed) in the cell’s country. In before but interact the “200 km Boko Haram x post- a triple-difference framework, we then interact the “200 2009” dummy with three dummies for whether the cell’s km Boko Haram x post-2009” dummy with the dummy country is Cameroon, Chad or Niger. For the year 2013 to see if the effect is stronger, or weaker, for less, or more, and relative to the year 2008, we find an effect of about developed areas. -5 percent (n.s.), -20 percent (***) and -25 percent (**), respectively (not shown, but available upon request). Our analysis reveals that those places that were initially Thus, in Cameroon, no significant effect is found on more developed than other areas were relatively less average. In the three countries, we then observe marked affected by the rise of Boko Haram. As seen in Cols. (1)– heterogeneity in the effects, as suggested by the wide (2) of Table 4.5, places that were relatively less developed confidence intervals (-0.13/0.03, -0.36/-0.10 and -0.47/- are the places where the effect was most negative, with 0.04, respectively).323 the overall effect about -0.14 (***). The overall effect in the third row corresponds to the combined effect Thus, the disruption effects of Boko Haram were of the effect of the BH 200 km x Post-09 dummy and very heterogeneous. However, for a given shock and its interaction with the chosen pre-2009 characteristic. country, it does not answer the question of which When we examine the effect for places that were initially locations “suffered” more vis-á-vis others. Conversely, more developed (Cols. (3)–(4)), then we find that the which locations were ultimately more resilient to the interaction is strongly positive, enough to make the negative effects of the shock? To answer these questions, observed negative effect of BH—about -14 percent— we use the same panel-DiD model as before but add the 323 The stronger effects in Chad and Niger might be explained by the heterogeneous effects shown below or the fact that Chad’s and Niger’s regions close to Boko Haram historically disproportionately relied on their trade links with Northeastern Nigeria. In contrast, Cameroon’s North was also trading with Southern Nigeria via Southern Cameroon (see Map 4.1). In particular, Niger’s Southeast is poorly connected to the more developed Western areas of Niger and its Northeast correspond to the Sahara, hence its Southeastern areas’ over-reliance on Northeastern Nigeria. 172 4.6 Heterogeneous Effects for Cameroon, Chad and Niger Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram disappear (Col. (3)) or even turn positive (Col. (4); 4.6.2 Factors of Resilience to Foreign 0.11**). Conflict Shocks Overall, while we found stronger negative effects for We now examine the heterogeneous effects of other the more urban intensive sample than for the more categories of initial (pre-2009) conditions. However, rural extensive sample, within the intensive sample due to lack of data, we sometimes use post-2009 cell we actually find stronger negative effects for less data. Next, in order to capture an interacted effect developed areas (which may for example include small that is different from the interacted effect with initial towns). If anything, the most developed areas relatively development (or “explain” some of the interacted effect gained from (or lost relatively less) from the presence of with initial development), we simultaneously control for BH. The relative gain in the most developed areas suggests the interaction of the BH 200 km x Post-09 dummy and that their sectors were more resilient to the BH shock, for the dummy equal to one if the cell’s night light intensity example because they trade more with other places within is above the 75th percentile value in 2008. When doing their respective country, with other regions of Nigeria, or so, we found that the average residual decline due to Boko with neighboring countries. Likewise, these places may Haram was 14 percent (see Col. (3) of Table 4.5). Finally, have attracted more economic outmigrants coming from we study the interacted effect of each characteristic one at negatively impacted areas. a time, mostly due to power issues. To improve our understanding of the factors of Note that using the 10th and 90th percentile values resilience in the face of an economic shock brought captures a more local, possibly stronger, effect, that about by foreign conflict, we next study heterogeneous could be better identified as a result. At the same time, effects related to trade diversification, agricultural if the studied characteristic has an effect above the 10th development, infrastructure, human capital, and percentile value of distance or below the 90th percentile institutions. value of distance, the effect may be mis-estimated because places above the 10th percentile or places below the 90th percentile are also directly affected by the characteristic. Table 4.5: Baseline Heterogeneous Effects of Boko Haram Dep. Var.: Log (Mean Night Light Intensity + 1) in Year t Interaction of BH 200km * Post-09 Interaction: with Dummy if Night Light Intensity in 2008 is ... Below 10th Below 25th Above 75th Above 90th Percentile: (1) (2) (3) (4) -0.001 -0.001 -0.135*** -0.110*** BH 200 km * Post-09 [0.030] [0.030] [0.026] [0.026] -0.135*** -0.135*** 0.166*** 0.216*** Interaction [0.026] [0.026] [0.026] [0.039] -0.14*** -0.14*** 0.03 0.11** Overall Effect [0.03] [0.03] [0.03] [0.04] Cell FE, Cntry-Yr FE Y Y Y Y Yr FE*Controls Y Y Y Y Observations 21,644 21,644 21,644 21,644 Notes: The dummy used for the interaction withBH200km * Post-09 is constructed using the 10th, 25th, 75th and 90th percentile values of night light intensity in the cell’s country in 2008. SEs clustered at the 3rd-level admin. unit. * p<0.10, ** p<0.05, *** p<0.01. 4.6 Heterogeneous Effects for Cameroon, Chad and Niger 173 Lake Chad Regional Economic Memorandum  |  Development for Peace In that case, using the 25th or 75th percentile could Trade Diversification. We first present our findings help better estimate the effect. With the 25th or 75th on heterogeneity based on market potential (MP). For percentile, more cells are included in the “relatively more each cell i and other cells j, MP of cell i is the weighted treated” group, which may also improve precision. There sum of the sum of night lights of other cells j, using as is thus a trade-off. As a result, we report the effects for the weights the driving time (in hours) circa 2008 between 10th and 25th percentiles as well as the 75th and 90th cell i and cell j to the power α. To begin, we assign to a percentiles. Figure 4.4 shows the interacted effects and cell the maximum speed between the speed(s) based on their confidence intervals. Two vertical lines are added, road categories applied in Jedwab and Storeygard (2020) one at 0 and one at 0.14. Indeed, a resilience effect of and the speed of travel across off-road cells from a hiking 0.14 is needed to offset the average residual decline due function (Tobler, 1993) that incorporates slopes from to Boko Haram (14 percent).324 Verdin et al. (2007). Then, we use the least-cost path algorithm to calculate the minimum travel time between  eterogeneous Resilience Effects Figure 4.4: H each cell and each other cell.325 Next, we assume α = 3 Depending on Initial Local Conditions in our baseline specification.326 Finally, when crossing 1. >75th Market Potential CCN+NGA 2. >90th Market Potential CCN+NGA borders, we impose that drivers have to go through 3. >75th Market Potential CCN Only 4. >90th Market Potential CCN Only border crossings (whose locations we know for the year 5. <10th Livestock Market 6. <25th Livestock Market 2008). The cost to cross the border is then assumed to be 7. >75th Sh. Ethnicity in BH 8. >90th Sh. Ethnicity in BH 4 hours. 9. <10th Airport 10. <25th Airport 11. <10th Paved Road 12. <25th Paved Road MP can first be defined using the cells of CCN and 13. <10th Paved/Improved Road 14. <25th Paved/Improved Road Nigeria but excluding the BH area itself since we aim 15. <10th Main Power Line 16. <25th Main Power Line to capture how the cell can trade with other areas than 17. <10th Cotton Ginning Fact. 18. <25th Cotton Ginning Fact. the BH area. As seen in rows 1–2 of Figure 4.4, we find 19. >75th Cotton Suit. 20. >90th Cotton Suit. a positive (but not significant) resilience effect if the cell 21. >75th Groundnut Suit. 22. >90th Groundnut Suit. is in the top 10th percentile in market potential in 2008 23. >75th Food Crop Suit. 24. >90th Food Crop Suit. (no effect is observed for the top 25th percentile). The 25. <10th Hospital 26. <25th Hospital point estimate is relatively high, at 0.11, enough to almost 27. <10th Hospital/Health Center 28. <25th Hospital/Health Center offset the negative independent effect of the BH shock. 29. <10th Higher Educ. Institution 30. <25th Higher Educ. Institution Standard errors are high, with the 95 percent confidence 31. <10th Military Forces HQ 32. <25th Military Forces HQ interval values ranging from -0.05 to 0.26. We thus 33. <10th Paramilitary Forces HQ 34. <25th Paramilitary Forces HQ observe heterogeneous effects of the heterogeneous effect 35. <10th 1st Lev. Admin. Capital 36. <25th 1st Lev. Admin. Capital itself. 37. <10th 2nd Lev. Admin. Capital 38. <25th 2nd Lev. Admin. Capital 39. <10th Border Crossing 40. <25th Border Crossing Nigeria’s economy dramatically suffered as a result of Boko Haram and a cleaner test of the trade 0 00 0 20 30 5 05 5 25 35 .0 0.1 .1 0.1 0. 0. 0. 0. -0 0. 0. -0 Q Coef. of triple interaction J Significant at 5 percent ‹ 10 percent diversification hypothesis could be to define MP using Notes: The figure shows the interacted effect of the 200 km BH*Post-2009 dummy with the variable shown at left. Each row represents a separate regression. The 2nd vertical line is only the cells of CCN, thus excluding Nigeria (rows for x = 0.14 because 14 percent is the average residual decline due to Boko Haram (= the independent effect of the 200 km BH dummy). * p<0.10, ** p<0.05, *** p<0.01. 3–4). The effect with the 10th percentile value is now even 324 In the 40 specifications described below, the effect of BH 200 km x Post-09 is almost always equal to 0.14. 325 The road data come from Jedwab and Storeygard (2020). The data include information on the surface of each road in 2008, i.e. whether the road is a highway, a paved road, an improved (gravel or laterite) road, or a dirt road. 326 Results are generally not sensitive to the alpha used (not shown). A high α implies a high trade cost of distance, making cells farther away from cell i matter less. α is not known in our context. Jedwab and Storeygard (2020) use 3.8 but they study the effect of market potential for the whole continent, thus focusing on long-distance trade. 174 4.6 Heterogeneous Effects for Cameroon, Chad and Niger Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram stronger (and significant). The point estimate—0.16—is Rohner et al. (2013b) discuss how a lack of inter-ethnic enough to fully offset the Boko Haram shock and the trust hampers trade. Therefore, ethnically connected 95 percent confidence interval values now range from areas should trade more. Amarasinghe et al. (2020) 0.00 to 0.33. Therefore, some locations among the most then find that ethnic connectivity, among other factors, connected locations might have even (relatively) gained is particularly important for the diffusion of economic with the BH shock. spillovers. We use the Murdock (1959) map to obtain for each cell the main ethnic group in terms of area in the While such locations perhaps trade more—which cell. For each cell/group, we then obtain the share of the means that they could have been affected relatively group’s total homeland area that is within the BH area. more by the income shocks experienced in the BH By interacting this share with the BH 200 km x Post-09 areas—the positive effects may indicate that their dummy, this allows us to test if cells that were historically economy is more diversified  (i.e. they trade more in more “connected” to other cells in the BH area are more general, not just with Northeastern Nigeria). As a result, directly affected, likely because of stronger trade links they may be on net less susceptible to foreign conflict with the BH area (via ethnicity-based trade networks). shocks. More generally, taking the simple average across More precisely, we use dummies if the share is above the rows 3 and 4, we obtain +10 percent. 75th or 90th percentile value in the country. As seen in row 8 of Figure 4.4, we find a negative significant effect of Next, we use the location of major livestock markets about -4 percent for the most connected cells (i.e., when as a proxy for general markets. In the region, markets using the 90th percentile). Thus, ethnic connectivity are used for agricultural products, cattle that is eventually plausibly helped the diffusion of the economic shock exported to urban markets in Southern Nigeria or caused by Boko Haram. Southern Cameroon, and manufactured products bought with income from the sale of agricultural products and Infrastructure. We now investigate how infrastructure cattle. Given the lower demand from Nigeria, we could factors related to trade or not may have mattered for the expect a negative interacted effect for the cells closest diffusion of the economic shock as well as local economic to the markets. At the same time, as livestock markets resilience. We examine how proximity to airports mediates proxy for markets more generally, the negative effects the impact of Boko Haram. We calculate the distance of could be (more than) compensated by positive effects for each locality to all airports in the same country.328 We locations trading more in general. In addition, cattle can find a positive and significant effect for the 10th and 25th travel to Nigeria through other routes not impacted by percentiles (rows 9–10; +7 percent on average) but the the “closure” of the BH region.327 We find positive effects effect is, as expected, higher for the 10th percentile. It of livestock markets (rows 5–6 of Figure 4.4). However, could be that cities close to airports have specific sectors the effect is weaker for the 10th percentile value than that are more resilient to land-based economic shocks for the 25th percentile value, possibly due to a more (i.e. overland trade with Northeastern Nigeria). negative impact for locations specialized in cattle export. The positive and significant effect for the 25th percentile Amarasinghe et al. (2020) show that road connectivity, (+5 percent) may then capture a more general resilience along with ethnic connectivity, is a critical factor in effect for trade-oriented regions. the diffusion of economic spillovers. We use the road network database of Jedwab and Storeygard (2020) 327 The location of 81 livestock markets in Chad and 10 livestock markets in Cameroon (c. 2004–2005) is obtained from République du Tchad (2010). The location of 66 livestock markets in Niger (in the 2010s) is obtained from USGS FEWS.NET (2017). There are fewer markets in Cameroon as most of the cattle is produced in Chad or Niger. 328 The locations of airports (circa 2003) come from U. S. Geological Survey (2003). 4.6 Heterogeneous Effects for Cameroon, Chad and Niger 175 Lake Chad Regional Economic Memorandum  |  Development for Peace to obtain for each cell and the year 2008 the minimal trade do not appear as important as the ones related to distance to a paved road (incl. highways), the minimal trade in our context. distance to a paved or improved road, and the minimal Euclidean distance to all roads (i.e., paved, improved, and Agricultural Development. We turn to heterogeneity dirt roads). We then create dummies based on whether with regard to agricultural development. Two main cash the cell’s distance to a paved road, a paved/improved road crops are grown in the area, cotton and groundnut. With or any road is below the 10th or 25th percentile value the shock, the demand from Nigeria likely decreased. in the country. As seen in rows 11–14, we find stronger At the same time, the supply of cotton and groundnut effects for paved roads (+5 percent) than for other roads. from Nigeria was also reduced, which may have increased The only significant effect is for the most connected cells, prices for local producers. The effect of the shock on i.e. cells whose distance to a paved road is below the 10th producing areas is thus ambiguous. In addition, if cash percentile value in the country (+8 percent). crop production is “fixed” in space, because of land suitability being an unsubstitutable factor of production Other types of infrastructure that are not related or because of past sunk investments in transformation to trade but possibly important include access to factories, then these locations remain valuable even in electricity and mobile networks. A reliable access times of crisis. In that case, we might expect these areas to to electricity is particularly important in countries be affected relatively less. where power failures are frequent. We thus investigate heterogeneity with respect to proximity to a major We estimate mean cotton and groundnut production electricity transmission line, assuming that such locations within 50 km from the cell’s centroid. We then create are more protected against regional power outages. In dummies based on whether cotton suitability is higher rows 15–16 of Figure 4.4, we interact the BH 200 km x than the 75th or 90th percentile value in the country.331 Post-09 dummy with a dummy if the cell’s distance to a Next, for cotton ginning factories, we use proximity to power line (c. 2008) is below the 10th or 25th percentile a factory, and thus create dummies if the cell’s distance value in the country. We find a positive but not significant to a factory is below the 25th or 10th percentile value effect of +5 percent for the 10th percentile and no effect in the country. Finally, note that there was no formal for the 25th percentile. The average effect is +2 percent.329 groundnut oil extraction plant in the area during the period. Groundnut oil was instead extracted artisanally Next, we examine heterogeneity with respect to GSM by local producers.332 coverage. More precisely, for each cell we obtain the area share that is covered by 2G mobile phone coverage c. As can be seen in Figure 4.4, we see positive interacted 2009.330 We then create dummies if coverage is above the effects for cotton ( rows 17–20; average affect of 75th or 90th percentile value in the country. However, +4 percent), which are only significant in two out of the we do not find any effect (not shown, but available upon four cases. No effect is observed for groundnut (rows 21– request). Therefore, infrastructure factors not related to 22), possibly because it is considered a less profitable cash crop in the area. 329 Data is obtained from the Africa Infrastructure Country Diagnostic (AICD) database of the World Bank. 330 The source of the data on 2G mobile phone geographic coverage is the Global System for Mobile Communications (GSMA) c. 2009, who summarizes submissions of mobile operators data that provide representation of network coverage with roaming detail based on strong and variable signal strength. 331 The distance threshold of 50 km is arbitrary. Results hold with 100 km (not shown, but available upon request). 332 Cotton and groundnut suitability-based measures of production c. 2010 are from SPAM 2010 (IFPRI, 2019). According to their website: “SPAM relies on a collection of relevant spatially explicit input data, including crop production statistics, cropland data, biophysical crop ‘suitability’ assessments, population density, as well as any prior knowledge about the spatial distribution of specific crops or crop systems.” The locations of cotton ginning factories are digitized from a map on Cotton Zones, Ginning Factories and Exports of West Africa in OECD (2006). 176 4.6 Heterogeneous Effects for Cameroon, Chad and Niger Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram The interacted effect with overall food crop suitability part public universities in CCN. For each cell we obtain then merits particular attention. Access to food crops the Euclidean distance to a higher education institution is important because, in time of (urban) crisis, urban and create dummies based on the 10th and 25th percentile areas surrounded by land that is relatively more suitable values in the country.335 As seen in rows 29–30, we find for food production may be more resilient to the shock. significant positive effects for both percentile values. The People are more likely to stay in these locations to weather effects are on average twice higher than the effects found an economic shock. We interact the BH 200 km x Post- for health (+6 percent vs. +3 percent). 09 dummy with dummies based on food crop suitability (averaged across 12 major food crops in Sub-Saharan Government Expenditure. We examine more broadly Africa).333 We see positive interacted effects (rows 23–24; if the effect of Boko Haram depends on government average effect of +5 percent, close to what we found for expenditure. Indeed, locations supported by the cotton). These are only significant for the 75th percentile. presence of government services may be more resilient due to the fact a larger share of their economy does not Overall, the resilience effects appear weaker for depend on local economic conditions but government agricultural development. However, if we focus on the budget allocations most often made at the national cotton industry or food suitability, we find resilience level. In addition, the presence of government services effects that are about 5 percent on average. may also positively, or negatively, impact the ability of local economies to bounce back in the face of a massive Human Capital. Health infrastructure proxies for both economic shock. human capital and government social expenditure as the health sector is mostly public in CCN. We construct We first examine heterogeneity with respect to major measures of the distance to hospitals or health centers military and paramilitary headquarters  (c. 2020 (2013–17) and create dummies based on whether it is as information is not available for earlier years). For below the 10th or 25th percentile value in the country.334 each cell we obtain the minimal Euclidean distance to We do not see any effect for hospitals (rows 25–26). When a major military headquarter or a major paramilitary considering hospitals and health centers simultaneously, headquarter and create dummies based on the 10th and we then see positive significant effects (rows 27–28; 25th percentile values in the country.336 As seen in rows average effect of +3 percent). The non-effects for hospitals 31–34, the interaction effects are strong and significant in suggests that these effects are not driven by health supply three out of the four cases (average of about +5 percent). per se. Instead, locations with health centers might have The effect is larger for military headquarters than for higher levels of social services and offer higher levels of paramilitary headquarters. social protection in times of crisis. We then study if the effect of Boko Haram depends We then examine heterogeneity with respect to higher on proximity to “regional” capitals (for 1st level education institutions (c. 2020), which are for the most administrative units) or “district” capitals (2nd-level 333 FAO (2013) provides for the period 1981–2010 a measure of food crop suitability that is based on both soils and the climate and the following 12 crops: manioc (cassava), maize, rice paddy (Japonica), rice paddy (Indica), common wheat, sorghum (low alt.), common millet, potato, potato yam, sugar beet, cowpea and common bean. 334 We rely on Maina et al. (2019). Cameroon (2014–17), Chad (2013–16) and Niger (2013–17) have 183 (2,836), 41 (824) and 78 (1,151) hospitals (health centers), respectively. Data does not exist for the pre-2019 period. 335 The location of higher education institutions comes from Wikipedia, reports, and newspaper articles. Cameroon, Chad and Niger have 31, 21 and 11 such institutions, respectively. Data does not exist for the pre-2019 period. 336 Cameroon, Chad and Niger have 5 (15), 8 (23) and 10 (23) military (paramilitary) headquarters, respectively. 4.6 Heterogeneous Effects for Cameroon, Chad and Niger 177 Lake Chad Regional Economic Memorandum  |  Development for Peace administrative units).337 For each cell we obtain the (especially when related to security for which the Euclidean distance to a regional capital or a district capital resilience effect is about +5/+6 percent). We do not and create dummies based on the 10th and 25th percentile find significant effects for access to electricity or mobile values in the country. The effects are not significant (rows networks, technologies that might only produce resilience 35–38; +2 percent on average). The effect is larger for if more resilient sectors are already present in the local regional capitals than for district ones. economy. More broadly, one can see that the interacted effect is While our results could have straightforward policy higher for military headquarters (about +6 percent) implications, one important caveat is that we only than for paramilitary headquarters (+5 percent), measure population, not real wages or welfare more regional capitals (+3 percent) or district capitals generally. Some “better endowed” locations may have (0 percent).338 Thus, security might have been a more experienced a slower relative decline in their population important concern than government employment. Given possibly because they were also attracting economic that Boko Haram had not entered CCN then, one refugees from equally affected neighboring locations. Our interpretation could be that firms reduced investments as analysis only captures relative population growth patterns a result of increased uncertainty in the region, especially and suggests that initially (pre-shock) better endowed in potentially more unsafe areas located farther away from locations, by being more resilient, grow faster than less military and paramilitary headquarters. well endowed locations. As such, economic shocks due to foreign conflict may accentuate spatial inequality. Finally, we examine heterogeneity with respect to border crossings/posts. A negative effect could be In addition, mostly due to power issues, we estimate expected in such areas due to reduced trade. However, each interacted effect one by one rather than such areas likely received more public investments and saw simultaneously. Some of the heterogeneity variables are an increase in military and police presence. For each cell also correlated with each other and may as such capture we compute the minimal Euclidean distance to a border similar dimensions. crossing circa 2008 and then create dummies based on the 10th and 25th percentile values in the country.339 As seen in rows 3–40, we find a negative, but insignificant, effect for the 10th percentile and no effect for the 25th percentile. As such, any negative effect due to reduced trade must have been offset by government expenditure. To summarize, factors of resilience in the face of an economic shock brought about by foreign conflict include trade diversification and infrastructure related to trade (resilience effect of about +5–10 percent), agricultural development (+5 percent), human capital (+3–6 percent), and government expenditure 337 For each country, we obtain a list of 1st-administrative level capitals—regional capitals (9 in Cameroon c. 2005, 22 in Chad c. 2020 and 7 in Niger c. 2014, respectively)—and a list of 2nd-administrative level capitals—departments capitals (48, 68 and 57, respectively). Sources used include the Humanitarian Data Exchange. While for Chad and Niger we use capitals defined post-2009, the total number of capitals barely changed there in the 2010s. 338 The coefficient of correlation between the 10th percentile dummies for these four types of government expenditure is between 0.16 and 0.65 (mean = 0.43). The dummies thus do not necessarily capture the same locations. 339 The locations of border crossings are obtained from Jedwab and Storeygard (2020). 178 4.6 Heterogeneous Effects for Cameroon, Chad and Niger Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram 4.7 Robustness and Other Considerations Spatial autocorrelation. To account for spatial for Borno or Maiduguri, likely because these were more autocorrelation, we cluster standard errors at the 3rd level affected; (iii) use a dummy for whether the cell is within administrative unit (N = 343, 336, and 265 in Cameroon, a 6.5 hours driving distance from the BH area. 6.5 hours Chad and Niger, respectively). We verify that the baseline corresponds to the 20th percentile in driving time to the negative effect of Col. (1) in Table 4.1 remains strongly BH area. We use the 20th percentile because 200 km significant when (Table A4.5): (i) clustering standard corresponds to the 20th percentile in Euclidean distance errors at the 2nd (36; 58; 53) or even 1st (8; 10; 23) to the BH area; and (iv) use the negative of the log administrative level; and (ii) using Conley standard distance to the BH area. The last two regressions are less errors using a distance cut-off of 100, 200 or even comparable to our baseline regression. The coefficients, 300 km. However, given how computationally intensive while different, remain strongly negative. computing Conley standard errors are when the number of spatial units is high, we first residualize the data, thus removing any variation due to the fixed effects and the controls. Using Conley standard errors is not feasible for regressions involving the full/extensive sample of cells, which we use for our analysis on the extensive margin of night lights, rural outcomes, and conflict. We also verify that these regressions and other regressions return similar results if we cluster standard errors at the 2nd or 1st administrative level (not shown, but available upon request). More generally, for the analysis on the extensive margin of night lights, greenness, land use and local conflict, we already find no effects. Thus, more conservative standard errors would not change our conclusions. Other Definitions of the Treatment. For the sake of simplicity, proximity to BH is constructed using Euclidean distance to the BH area, which we define as the area of the states of Borno, Yobe and Adamawa that is between the Yobe river in the North (in Yobe) and the Benue river in the South (Adamawa). Table A4.6 shows that the results hold if we: (i) define the BH area as the state of Borno (where 60 percent of conflict events took place) or the full area of the Borno, Yobe and Adamawa; (ii) use a dummy for whether the cell is within 300 km from Maiduguri, Northeastern Nigeria’s main city, which was particularly impacted by Boko Haram activities. We use 300 km instead of 200 km because Maiduguri is about 100 km from the border. The effects are stronger 4.7 Robustness and Other Considerations 179 Lake Chad Regional Economic Memorandum  |  Development for Peace 4.8 Conclusion What are the spillover effects of foreign terrorism and conflict on regional economies? Adopting a difference- in-difference framework leveraging the unexpected rise of the Boko Haram insurgency in Northeastern Nigeria in 2009, we studied its effects in neighboring areas in Cameroon, Chad and Niger. We found strong negative effects on regional economic activities— proxied by reductions in nighttime lights—particularly amongst areas within 200 km from the Boko Haram area. Our findings suggested that this negative impact was concentrated in urban areas and was particularly pronounced among those areas that were initially less developed and connected, which highlights the role of trade diversification and infrastructure in mitigating the effects of economic shocks brought about by foreign conflict. We also found that the rise of Boko Haram resulted in more agricultural burning—an agricultural practice that is profitable in the short-term but typically leads to long-term environmental and economic losses. Overall, these findings attest to both the short-term and long-term negative impacts of foreign conflicts on regional economies. More generally, we believe our findings might have important policy implications. First, conflicts have spillover effects that significantly impact regional economies as a whole, not only in the short run but also in the long run as well. For example, foreign conflicts push individuals in the urban sector to seek opportunities in the rural sector and engage in agricultural practices—namely, agricultural burning— that potentially jeopardizes long-run economic gains. Peace interventions can have positive effects “beyond” the country or countries in which they take place. Second, certain types of mitigation measures are perhaps more effective than others at alleviating the negative spillover economic effects of foreign conflict. In our context, initially more developed, connected, infrastructure- endowed, and government-protected areas were better able to “weather” the impact of the shock. 180 4.8 Conclusion Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram References Abadie, Alberto and Javier Gardeazabal, “The Economic Costs of Conflict: A Case Study of the Basque Country,” American Economic Review, March 2003, 93 (1), 113–132. Adelaja, Adesoji and Justin George, “Effects of conflict on agriculture: Evidence from the Boko Haram insurgency,” World Development, 2019, 117, 184 – 195. Adhvaryu, Achyuta, James Fenske, Gaurav Khanna, and Anant Nyshadham, “Resources, conflict, and economic development in Africa,” Journal of Development Economics, 2021, 149 (C). Akresh, Richard, Sonia Bhalotra, Marinella Leone, and Una Okonkwo Osili, “War and Stature: Growing Up during the Nigerian Civil War,” American Economic Review, May 2012, 102 (3), 273–277. Alix-Garcia, Jennifer, Sarah Walker, Anne Bartlett, Harun Onder, and Apurva Sanghi, “Do refugee camps help or hurt hosts? The case of Kakuma, Kenya,” Journal of Development Economics, 2018, 130 (C), 66–83. Amarasinghe, Ashani, Roland Hodler, Paul Raschky, and Yves Zenou, “Key Players in Economic Development,” SSRN Scholarly Paper ID 3562861, Social Science Research Network March 2020. Angrist, Joshua D. and Adriana D. Kugler, “Rural Windfall or a New Resource Curse? Coca, Income, and Civil Conflict in Colombia,” The Review of Economics and Statistics, May 2008, 90 (2), 191–215. Annan, Jeannie, Christopher Blattman, Dyan Mazurana, and Khristopher Carlson, “Civil War, Reintegration, and Gender in Northern Uganda,” Journal of Conflict Resolution, December 2011, 55 (6), 877–908. Barrios, Salvador, Luisito Bertinelli, and Eric Strobl, “Climatic change and rural–urban migration: The case of sub- Saharan Africa,” Journal of Urban Economics, 2006, 60 (3), 357–371. Bauer, Michal, Christopher Blattman, Julie Chytilová, Joseph Henrich, Edward Miguel, and Tamar Mitts, “Can War Foster Cooperation?,” Journal of Economic Perspectives, Summer 2016, 30 (3), 249–274. Bazzi, Samuel and Christopher Blattman, “Economic Shocks and Conflict: Evidence from Commodity Prices,” American Economic Journal: Macroeconomics, October 2014, 6 (4), 1–38. Bellows, John and Edward Miguel, “War and local collective action in Sierra Leone,” Journal of Public Economics, 2009, 93 (11), 1144 – 1157. Berman, Eli, Jacob N. Shapiro, and Joseph H. Felter, “Can Hearts and Minds Be Bought? The Economics of Counterinsurgency in Iraq,” Journal of Political Economy, 2011, 119 (4), 766–819. Berman, Nicolas and Mathieu Couttenier, “External Shocks, Internal Shots: The Geography of Civil Conflicts,” The Review of Economics and Statistics, 2015, 97 (4), 758–776. __ , __ , and Raphael Soubeyran, “Fertile Ground for Conflict,” Journal of the European Economic Association, 12 2019, 19 (1), 82–127. __ , __ , Dominic Rohner, and Mathias Thoenig, “This Mine Is Mine! How Minerals Fuel Conflicts in Africa,” American Economic Review, June 2017, 107 (6), 1564–1610. Bertoni, Eleonora, Michele Di Maio, Vasco Molini, and Roberto Nisticò, “Education is forbidden: The effect of the Boko Haram conflict on education in North-East Nigeria,” Journal of Development Economics, 2019, 141, 102249. Besley, Timothy and Marta Reynal-Querol, “The Legacy of Historical Conflict: Evidence from Africa,” American Political Science Review, 2014, 108 (2), 319–336. __ and Torsten Persson, “The Logic of Political Violence*,” The Quarterly Journal of Economics, 08 2011, 126 (3), 1411–1445. References 181 Lake Chad Regional Economic Memorandum  |  Development for Peace Blankespoor, Brian, “Building rural development in the Lake Chad Region,” 2021. __ , Sering Touray, and Roy Katayama, “Estimating the effect of conflict on agricultural activity in the Central African Republic with remotely sensed data,” 2021. Blattman, Christopher and Edward Miguel, “Civil War,” Journal of Economic Literature, 2010, 48 (1), 3–57. __ and Jeannie Annan, “THE CONSEQUENCES OF CHILD SOLDIERING,” The Review of Economics and Statistics, 2010, 92 (4), 882–898. __ and __ , “Can Employment Reduce Lawlessness and Rebellion? A Field Experiment with High-Risk Men in a Fragile State,” NBER Working Papers 21289, National Bureau of Economic Research, Inc June 2015. __ , Julian C. Jamison, and Margaret Sheridan, “Reducing crime and violence: Experimental evidence from cognitive behavioral therapy in Liberia,” NBER Working Papers 21204, National Bureau of Economic Research, Inc May 2015. Blomberg, S. Brock and Gregory D Hess, “How Much Does Violence Tax Trade?,” The Review of Economics and Statistics, 2006, 88 (4), 599–612. Bosker, Maarten and Joppe de Ree, “Ethnicity and the spread of civil war,” Journal of Development Economics, 2014, 108 (C), 206–221. Brodeur, Abel, “The Effect of Terrorism on Employment and Consumer Sentiment: Evidence from Successful and Failed Terror Attacks,” American Economic Journal: Applied Economics, October 2018, 10 (4), 246–82. Bruederle, Anna and Roland Hodler, “Nighttime lights as a proxy for human development at the local level,” PloS one, 2018, 13 (9), e0202231. Brückner, Markus and Antonio Ciccone, “International Commodity Prices, Growth and the Outbreak of Civil War in Sub-Saharan Africa,” Economic Journal, May 2010, 120 (544), 519–534. Burger, Martijn, Elena Ianchovichina, and Bob Rijkers, “Risky Business: Political Instability and Sectoral Greenfield Foreign Direct Investment in the Arab World,” The World Bank Economic Review, 06 2015, 30 (2), 306–331. Burke, Marshall, Solomon M. Hsiang, and Edward Miguel, “Climate and Conflict,” Annual Review of Economics, August 2015, 7 (1), 577–617. Bó, Ernesto Dal and Pedro Dal Bó, “Workers, Warriors, and Criminals: Social Conflict in General Equilibrium,” Journal of the European Economic Association, 08 2011, 9 (4), 646–677. Caselli, Francesco, Massimo Morelli, and Dominic Rohner, “The Geography of Interstate Resource Wars *,” The Quarterly Journal of Economics, 02 2015, 130 (1), 267–315. Castillo, Juan Camilo, Daniel Mejía, and Pascual Restrepo, “Scarcity without Leviathan: The Violent Effects of Cocaine Supply Shortages in the Mexican Drug War,” The Review of Economics and Statistics, May 2020, 102 (2), 269–286. CCAPS, “Social Conflict Analysis Database,” 2020. Chiovelli, Giorgio, Stelios Michalopoulos, and Elias Papaioannou, “Landmines and Spatial Development,” Working Paper 24758, National Bureau of Economic Research June 2018. Ciccone, Antonio, “Economic Shocks and Civil Conflict: A Comment,” American Economic Journal: Applied Economics, October 2011, 3 (4), 215–27. Cook, David, “The Rise of Boko Haram in Nigeria,” 2011. Corbane, Christina, Martino Pesaresi, Thomas Kemper, Panagiotis Politis, Aneta J Florczyk, Vasileios Syrris, Michele Melchiorri, Filip Sabo, and Pierre Soille, “Automated global delineation of human settlements from 40 years of Landsat satellite data archives,” Big Earth Data, 2019, 3 (2), 140–169. Couttenier, Mathieu and Raphael Soubeyran, “Drought and Civil War In Sub-Saharan Africa,” Economic Journal, March 2014, 124 (575), 201–244. 182 References Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram __ and __ , “A Survey of the Causes of Civil Conflicts: Natural Factors and Economic Conditions,” Revue d’économie politique, 2015, 125 (6), 787–810. Crost, Benjamin, Joseph Felter, and Patrick Johnston, “Aid Under Fire: Development Projects and Civil Conflict,” The American Economic Review, 2014, 104 (6), 1833–1856. __ , Joseph H. Felter, and Patrick B. Johnston, “Conditional cash transfers, civil conflict and insurgent influence: Experimental evidence from the Philippines,” Journal of Development Economics, 2016, 118 (C), 171–182. Daouda, Hamadou Y., “Poverty and living conditions with Boko Haram in the Lake Chad Basin: the case of southeastern Niger,” Review of African Political Economy, 2020, 47 (163), 126–134. de la Sierra, Raúl Sánchez, “On the Origins of the State: Stationary Bandits and Taxation in Eastern Congo,” Journal of Political Economy, 2020, 128 (1), 32–74. de Ree, Joppe and Eleonora Nillesen, “Aiding violence or peace? The impact of foreign aid on the risk of civil conflict in sub-Saharan Africa,” Journal of Development Economics, March 2009, 88 (2), 301–313. Dincecco, Mark and Massimiliano Gaetano Onorato, “Military conflict and the rise of urban Europe,” Journal of Economic Growth, September 2016, 21 (3), 259–282. Direction Régionale de l’Etat Civil et des Réfugiés, “Région de Diffa Situation Pop Deplacées,” 2016. Djankov, Simeon and Marta Reynal-Querol, “Poverty and Civil War: Revisiting the Evidence,” Review of Economics and Statistics, 2010, 92 (4), 1035–1041. Dube, Oeindrila and Juan F. Vargas, “Commodity Price Shocks and Civil Conflict: Evidence from Colombia,” The Review of Economic Studies, 03 2013, 80 (4), 1384–1421. Eberle, Ulrich J., Dominic Rohner, and Mathias Thoenig, “Heat and Hate: Climate Security and Farmer-Herder Conflicts in Africa,” CEPR Discussion Papers 15542, C.E.P.R. Discussion Papers Dec 2020. Elvidge, Christopher D, Kimberly Baugh, Mikhail Zhizhin, Feng Chi Hsu, and Tilottama Ghosh, “VIIRS nighttime lights,” International Journal of Remote Sensing, 2017, 38 (21), 5860–5879. Emran, M Shahe, Forhad Shilpi, Harold Coulombe, and Brian Blankespoor, “Temporary Trade Shocks, Spatial Reallocation, and Persistence in Developing Countries: Evidence from a Natural Experiment in West Africa,” World Bank Policy Research Paper 8962, World Bank 2019. Enders, Walter and Gary A. Hoover, “The Nonlinear Relationship between Terrorism and Poverty,” American Economic Review, May 2012, 102 (3), 267–72. Enobi, Akepe Linus and Samuel F. Johnson-Rokosu, “Terrorism financing. The socio-economic and political implications of boko haram insurgency in lake Chad Basin,” Academic Journal of Economic Studies, 2016, 2 (1), 25–41. Publisher: Faculty of Finance, Banking and Accountancy Bucharest,” Dimitrie Cantemir . . . . European Space Agency, “Land Cover CCI Product User Guide Version,” 2017. __ , “C3S Global Land Cover products 2016-2018,” 2019. Fafchamps, Marcel, “Ethnicity and Networks in African Trade,” The B.E. Journal of Economic Analysis & Policy, December 2003, 2 (1), 1–53. Fallah, Belal, Caroline Krafft, and Jackline Wahba, “The impact of refugees on employment and wages in Jordan,” Journal of Development Economics, 2019, 139 (C), 203–216. FAO, “Climate change predictions in Sub-Saharan Africa: impacts and adaptations (ClimAfrica),” 2013. __ , “GIEWS - Global Information and Early Warning System,” 2020. Fearon, James D., Macartan Humphreys, and Jeremy M. Weinstein, “Can Development Aid Contribute to Social Cohesion after Civil War? Evidence from a Field Experiment in Post-conflict Liberia,” American Economic Review, May 2009, 99 (2), 287–91. References 183 Lake Chad Regional Economic Memorandum  |  Development for Peace __ , __ , and , “How Does Development Assistance Affect Collective Action Capacity? Results from a Field Experiment in Post-Conflict Liberia,” American Political Science Review, 2015, 109 (3), 450–469. Fenske, James and Namrata Kala, “1807: Economic shocks, conflict and the slave trade,” Journal of Development Economics, 2017, 126 (C), 66–76. Foyou, Viviane E., Peter Ngwafu, Maribel Santoyo, and Andrea Ortiz, “The Boko Haram Insurgency and its Impact on Border Security, Trade and Economic Collaboration Between Nigeria and Cameroon: An Exploratory Study,” African Social Science Review, 2018, 9 (1), 7. Glaeser, Edward L., “A World Of Cities: The Causes And Consequences Of Urbanization In Poorer Countries,” Journal of the European Economic Association, October 2014, 12 (5), 1154–1199. __ and Jesse M. Shapiro, “Cities and Warfare: The Impact of Terrorism on Urban Form,” Journal of Urban Economics, 2002, 51 (2), 205 – 224. Glaeser, Edward L. and Joseph Gyourko, “Urban Decline and Durable Housing,” Journal of Political Economy, 2005, 113 (2), 345–375. Glick, Reuven and Alan M. Taylor, “COLLATERAL DAMAGE: TRADE DISRUPTION AND THE ECONOMIC IMPACT OF WAR,” The Review of Economics and Statistics, 2010, 92 (1), 102–127. Gockowski, James, Jean Tonyé, Chimere Diaw, Stefan Hauser, Jean Kotto-Same, Rosaline Njomgang, Appolinaire Moukam, Dieudonné Nwaga, Téophile Tiki-Manga, Jerome Tondoh, Zac Tschondeau, Stephan Weise, and Louis Zapfack, “The Forest Margins of Cameroon,” in Cheryl Palm, Stephen A. Vosti, Pedro A. Sanchez, and Polly J. Ericksen, eds., Slash-and-Burn Agriculture: The Search for Alternatives, Columbia University Press, 2005. Gollin, Douglas, Remi Jedwab, and Dietrich Vollrath, “Urbanization with and without industrialization,” Journal of Economic Growth, 2016, 21 (1), 35–70. Harari, Mariaflavia and Eliana La Ferrara, “Conflict, Climate, and Cells: A Disaggregated Analysis,” The Review of Economics and Statistics, 2018, 100 (4), 594–608. Hartman, Alexandra C., Robert A. Blair, and Christopher Blattman, “Engineering Informal Institutions: Longrun Impacts of Alternative Dispute Resolution on Violence and Property Rights in Liberia,” NBER Working Papers 24482, National Bureau of Economic Research, Inc April 2018. Haslop, Federico, Remi Jedwab, and Elena Ianchovichina, “Consumption Cities vs. Production Cities: New Considerations and Evidence,” 2021. __ , __ , Takaaki Masaki, Carlos Rodriguez-Castelan, and Roman Zarate, “Climate Change, Rural Livelihoods and Urbanization: Evidence from the Permanent Shrinking of Lake Chad,” 2021. Hegre, Håvard and Nicholas Sambanis, “Sensitivity Analysis of Empirical Results on Civil War Onset,” Journal of Conflict Resolution, 2006, 50 (4), 508–535. Henderson, J. Vernon, Adam Storeygard, and Uwe Deichmann, “Has climate change driven urbanization in Africa?,” Journal of Development Economics, 2017, 124 (C), 60–82. Henderson, Vernon, Adam Storeygard, and David N Weil, “A bright idea for measuring economic growth,” American Economic Review, 2011, 101 (3), 194–99. Hodler, Roland, “The curse of natural resources in fractionalized countries,” European Economic Review, August 2006, 50 (6), 1367–1386. IFPRI, “Global Spatially-Disaggregated Crop Production Statistics Data for 2010, Version 1.0,” 2019. Institute for Economics and Peace, “Global Terrorism Database,” 2012-2020. International Monetary Fund, 2. The Economic Consequences of Conflicts, USA: INTERNATIONAL MONETARY FUND, 2019. 184 References Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram IOM, “Cameroon: Displacement Tracking Matrix - Round 2,” 2016. Jedwab, Remi and Adam Storeygard, “The Average and Heterogeneous Effects of Transportation Investments: Evidence from Sub-Saharan Africa 1960- 2010,” Working Paper 27670, National Bureau of Economic Research August 2020. Jedwab, Rémi and Alexander Moradi, “The Permanent Economic Effects of Transportation Revolutions in Poor Countries: Evidence from Africa,” Review of Economics & Statistics, 2016, 98 (2), 268–284. Jedwab, Remi and Dietrich Vollrath, “The Urban Mortality Transition and Poor-Country Urbanization,” American Economic Journal: Macroeconomics, January 2019, 11 (1), 223–75. Jedwab, Rémi, Edward Kerby, and Alexander Moradi, “History, Path Dependence and Development: Evidence from Colonial Railways, Settlers and Cities in Kenya,” The Economic Journal, 2017. Jedwab, Remi, Jason Barr, and Jan K. Brueckner, “Cities without Skylines: Worldwide Building-Height Gaps and Their Implications,” Technical Report 2020. Jedwab, Rémi, Luc Christiaensen, and Marina Gindelsky, “Demography, Urbanization and Development: Rural Push, Urban Pull and... Urban Push?,” Journal of Urban Economics, 2017, 98, 6–16. Jedwab, Remi, Prakash Loungani, and Anthony Yezer, “Comparing cities in developed and developing countries: Population, land area, building height and crowding,” Regional Science and Urban Economics, 2021, 86, 103609. Jia, Ruixue, “Weather Shocks, Sweet Potatoes and Peasant Revolts in Historical China,” Economic Journal, March 2014, 124 (575), 92–118. Johnson, B. T., S. R. Osborne, J. M. Haywood, and M. A. J. Harrison, “Aircraft measurements of biomass burning aerosol over West Africa during DABEX,” Journal of Geophysical Research: Atmospheres, 2008, 113 (D23). Kanmegne, J., Slash and Burn Agriculture in the Humid Forest Zone of Southern Cameroon: Soil Quality Dynamics, Improved Fallow Management and Farmers’ Perceptions Tropenbos-Cameroon series, Wageningen University, 2004. Kilic, Talip, Umar Serajuddin, Hiroki Uematsu, and Nobuo Yoshida, “Costing household surveys for monitoring progress toward ending extreme poverty and boosting shared prosperity,” World Bank Policy Research Paper 7951, World Bank 2017. Kimenyi, Mwangi, Jideofor Adibe, Moussa Djiré, Abigail J. Jirgi, Alpha Kergna, Temesgen T. Deressa, Jessica E. Pugliese, and Andrew Westbury, “The impact of conflict and political instability on agricultural investments in Mali and Nigeria,” Working Paper 17, The Brookings Institute August 2014. Kocornik-Mina, Adriana, Thomas K. J. McDermott, Guy Michaels, and Ferdinand Rauch, “Flooded Cities,” American Economic Journal: Applied Economics, April 2020, 12 (2), 35–66. Kotto-Same, Jean, Paul L. Woomer, Moukam Appolinaire, and Zapfack Louis, “Carbon dynamics in slash-and- burn agriculture and land use alternatives of the humid forest zone in Cameroon,” Agriculture, Ecosystems & Environment, 1997, 65 (3), 245 – 256. Krueger, Alan B. and Jitka Malěcková, “Education, Poverty and Terrorism: Is There a Causal Connection?,” Journal of Economic Perspectives, December 2003, 17 (4), 119–144. Kull, Christian A and Paul Laris, “Fire ecology and fire politics in Mali and Madagascar,” in “Tropical fire ecology,” Springer, 2009, pp. 171–226. König, Michael D., Dominic Rohner, Mathias Thoenig, and Fabrizio Zilibotti, “Networks in Conflict: Theory and Evidence From the Great War of Africa,” Econometrica, 2017, 85 (4), 1093–1132. Lei, Yu-Hsiang and Guy Michaels, “Do giant oilfield discoveries fuel internal armed conflicts?,” Journal of Development Economics, 2014, 110 (C), 139–157. References 185 Lake Chad Regional Economic Memorandum  |  Development for Peace Lewis, Ethan and Giovanni Peri, “Immigration and the Economy of Cities and Regions,” in Gilles Duranton, J. V. Henderson, and William C. Strange, eds., Handbook of Regional and Urban Economics, Vol. 5 of Handbook of Regional and Urban Economics, Elsevier, 2015, chapter 0, pp. 625–685. Li, Xuecao, Yuyu Zhou, Min Zhao, and Xia Zhao, “A harmonized global nighttime light dataset 1992-2018,” Scientific Data, June 2020, 7 (1), 168. Lujala, P¨aivi, Jan Ketil Rod, and Nadja Thieme, “Fighting over oil: Introducing a new dataset,” Conflict Management and Peace Science, 2007, 24 (3), 239–256. Maina, Joseph, Paul O Ouma, Peter M Macharia, Victor A Alegana, Benard Mitto, Ibrahima Socé Fall, Abdisalan M Noor, Robert W Snow, and Emelda A Okiro, “A spatial database of health facilities managed by the public health sector in sub Saharan Africa,” Scientific Data, 2019, 6 (1), 1–8. Martin, Philippe, Mathias Thoenig, and Thierry Mayer, “Civil Wars and International Trade,” Journal of the European Economic Association, 2008, 6 (2/3), 541–550. __ , Thierry Mayer, and Mathias Thoenig, “Make Trade Not War?,” The Review of Economic Studies, 07 2008, 75 (3), 865–900. __ , __ , and , “The Geography of Conflicts and Regional Trade Agreements,” American Economic Journal: Macroeconomics, October 2012, 4 (4), 1–35. McGuirk, Eoin and Marshall Burke, “The Economic Origins of Conflict in Africa,” Journal of Political Economy, 2020, 128 (10), 3940–3997. McGuirk, Eoin F. and Nathan Nunn, “Transhumant Pastoralism, Climate Change, and Conflict in Africa,” NBER Working Papers 28243, National Bureau of Economic Research, Inc Dec 2020. Melnikov, Nikita, Carlos Schmidt-Padilla, and Maria Micaela Sviatschi, “Gangs, Labor Mobility and Development,” NBER Working Papers 27832, National Bureau of Economic Research, Inc September 2020. Michalopoulos, Stelios and Elias Papaioannou, “Pre-Colonial Ethnic Institutions and Contemporary African Development,” Econometrica, 2013, 81 (1), 113–152. __ and __ , “National Institutions and Subnational Development in Africa,” Quarterly Journal of Economics, 2014, 129 (1), 151–213. Miguel, Edward and Shanker Satyanath, “Re-examining Economic Shocks and Civil Conflict,” American Economic Journal: Applied Economics, October 2011, 3 (4), 228–32. __ , __ , and Ernest Sergenti, “Economic Shocks and Civil Conflict: An Instrumental Variables Approach,” Journal of Political Economy, 2004, 112 (4), 725–753. Murdock, George Peter, “Africa its peoples and their culture history,” 1959. NASA, “MCD64A1.006 MODIS Burned Area Monthly Global,” 2020. __ , “MOD13A2.006 Terra Vegetation Indices 16-Day Global 1km ,” 2020. Nickell, Stephen, “Biases in Dynamic Models with Fixed Effects,” Econometrica, 1981, 49 (6), 1417–1426. Nigeria Department of Petroleum Resources, “Petroleum Refineries and Petrochemicals,” 2020. Novta, Natalija, “Ethnic Diversity and the Spread of Civil War,” Journal of the European Economic Association, 08 2016, 14 (5), 1074–1100. Nunn, Nathan and Leonard Wantchekon, “The Slave Trade and the Origins of Mistrust in Africa,” American Economic Review, December 2011, 101 (7), 3221–3252. __ and Nancy Qian, “US Food Aid and Civil Conflict,” American Economic Review, June 2014, 104 (6), 1630–66. Nwaga, Dieudonné, Jan Jansa, Monique Abossolo Angue, and Emmanuel Frossard, “The Potential of Soil Beneficial Micro-Organisms for Slash-and-Burn Agriculture in the Humid Forest Zone of Sub-Saharan Africa,” in 186 References Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram Patrice Dion, ed., Soil Biology and Agriculture in the Tropics, Berlin, Heidelberg: Springer Berlin Heidelberg, 2010, pp. 81–107. OECD, “Atlas on Regional Integration inWest Africa,” 2006. [Map 8; accessed September 20, 2020]. OECD/SWAC, “Africapolis (database),” 2020. Omenma, Tochukwu J., Ike E. Onyishi, and Alyious-Michaels Okolie, “A decade of Boko Haram activities: the attacks, responses and challenges ahead,” Security Journal, 2020, 33, 337 – 356. Opoku, Maxwell Peprah, Bernard Nsaidzedze Sakah, and Beatrice Atim Alupo, “The impact of Boko Haram threat on economic activities in Cameroon: perceptions of people in Yaoundé,” Behavioral Sciences of Terrorism and Political Aggression, 2017, 9 (3), 222–237. Qureshi, Mahvash Saeed, “Trade and thy neighbor’s war,” Journal of Development Economics, 2013. Raleigh, Clionadh, Andrew Linke, Håvard Hegre, and Joakim Karlsen, “Introducing ACLED: an armed conflict location and event dataset: special data feature,” Journal of peace research, 2010, 47 (5), 651–660. Rohner, Dominic, Mathias Thoenig, and Fabrizio Zilibotti, “War Signals: A Theory of Trade, Trust, and Conflict,” Review of Economic Studies, 2013, 80 (3), 1114–1147. __ , __ , and , “War Signals: A Theory of Trade, Trust, and Conflict,” The Review of Economic Studies, 02 2013, 80 (3), 1114–1147. Rozo, Sandra V. and Micaela Sviatschi, “Is a refugee crisis a housing crisis? Only if housing supply is unresponsive,” Journal of Development Economics, 2021, 148, 102563. République du Tchad, “Etude de faisabilité pour la définition d’une zone pilote de transit et de services aux acteurs de la filiére d’exportation de bétail et d’un cadre général de certification et de traçabilité,” Technical Report 2010. Seitz, Michael, Alexander Tarasov, and Roman Zakharenko, “Trade costs, conflicts, and defense spending,” Journal of International Economics, 2015, 95 (2), 305–318. Sousa, José De, Daniel Mirza, and Thierry Verdier, “Terror networks and trade: Does the neighbor hurt?,” European Economic Review, 2018, 107 (C), 27–56. Storeygard, Adam, “Farther on Down the Road: Transport Costs, Trade and Urban Growth in sub-Saharan Africa,” Review of Economic Studies, 2016, 83, 1263–1295. Sviatschi, Maria Micaela, “Making a Narco: Childhood Exposure to Illegal Labor Markets and Criminal Life Paths,” Working Papers, Princeton University, Woodrow Wilson School of Public and International Affairs, Research Program in Development Studies. March 2018. Tayimlong, Robert A., “Fragility and Insurgency as Outcomes of Underdevelopment of Public Infrastructure and Socio-Economic Deprivation: The Case of Boko Haram in the Lake Chad Basin,” Journal of Peacebuilding & Development, 2020, 0 (0), 1542316620950188. Tobler, Waldo, “Three presentations on geographical analysis and modeling,” 1993. Trebbi, Francesco and Eric Weese, “Insurgency and Small Wars: Estimation of Unobserved Coalition Structures,” Econometrica, 2019, 87 (2), 463–496. U. S. Geological Survey, “Global GIS DVD,” 2003. __ , “Mineral Operations of Africa and the Middle East,” 2006. UN OCHA, “Chad: situation on the Lake region and the impact of the Nigerian crisis – Situation Report n°6 (20/10/2015),” 2015. UNHCR, “Refugees,” 2020. UNHCR and World Bank, “Forced Displacement by the Boko Haram Conflict in the Lake Chad Region,” 2016. References 187 Lake Chad Regional Economic Memorandum  |  Development for Peace University, Uppsala, “Uppsala Conflict Data Program,” 2020. USGS FEWS.NET, “Production and Market Flow Map : Niger Livestock,” 2017. Vanda Felbab-Brown, In Nigeria, we don’t want them back: Amnesty, defectors’ programs, leniency measures, informal reconcilitation, and punitive responses to Boko Haram, Tokyo: United Nations, 2018. Verdin, Kristine L, Jonathan W Godt, Chris Funk, Diego Pedreros, Bruce Worstell, and James Verdin, “Development of a global slope dataset for estimation of landslide occurrence resulting from earthquakes,” Technical Report 2007. Voigtländer, Nico and Hans-Joachim Voth, “The Three Horsemen of Riches: Plague, War, and Urbanization in Early Modern Europe,” The Review of Economic Studies, 11 2012, 80 (2), 774–811. World Food Program, “Assessing Gender and Markets in the Lake Chad Basin Region. VAM Gender and Markets Study, 2016-2017,” 2016. 188 References Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram Appendix Figure A4.1:  Boko Haram Area Effect in Nigeria, 2000–  oko Haram Area Effect in Nigeria, Figure A4.2: B 2013 2000–2018 Effect of Boko Haram area dummy (relative to 2000) Effect of Boko Haram area dummy (omitted = 2000) 0.05– 0.2– Insurgency starts Counter-insurgency in 2015 0.1– 0– 0– -0.1– -0.2– -0.05– -0.3– -0.4– -0.5– -0.10– -0.6– -0.7– -0.15– -0.8– 2000 2002 2004 2006 2008 2010 2012 2014 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 with 95 percent confidence interval with 95 percent confidence interval Q DMSP-based lights ‹ VIIRS-based lights Notes: The figure shows for Nigeria the yearly effect (relative to the year 2000) of a dummy Notes: We use the same panel-DiD model as for Fig. A4.1 except that we now consider the equal to one if the cell is within the Boko Haram area (the area of Borno, Yobe and full period 2000–2018. For this analysis we rely on the harmonized NTL data (1992–2018) Adamawa that is between the Yobe river in the North (in Yobe) and the Benue river in the from Li et al. (2020) who combine night light data from OLS-DMSP (used until 2013) and South (Adamawa)). More precisely, we use data for 7,761 0.1×0.1 degree grid cells (≈ 11x11km SNPP-VIIRS (use for the period 2014–2018). Note that the high standard errors for the at the equator) in Nigeria for the years 2000–2013 (hence N = 108,654). The dependent VIIRS observations in 2014–2018 likely reflect the fact that the assumptions made by Li et variable is the log of mean light intensity (sum of lights divided by area + 1) in cell c in al. (2020) to recreate harmonized NTL for the whole period 2000–2018 also introduced a year t. We include cell c fixed effects, year t fixed effects, and interact the Boko Haram significant amount of noise. area dummy c (equal to one if the cell is within the Boko Haram area or if its centroid is within 10 km from the area’s border) with a dummy for each year t in 2001–2013. Standard errors are clustered at the Local Government Area (LGA; N = 721). With 7,761 cells, there are 11 cells per LGA.  oko Haram Area Effect in Cameroon, Figure A4.3: B  oko Haram Area Effect in Cameroon, Figure A4.4: B Chad and Tchad, 50 Km Bins, 2000–2013 Chad and Niger, 2000–2018 Effect of X km from Boko Haram (relative to 2000) Effect of 200 km from Boko Haram (omitted = 2010) 0– 0.1– Insurgency starts Counter-insurgency in 2015 0– -0.1– -0.2– -0.1– -0.3– -0.4– -0.5– -0.2– -0.6– -0.7– -0.8– -0.3– -1.0– 2000 2002 2004 2006 2008 2010 2012 2014 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 with 95 percent confidence interval ▬ 0–50 km ▬ 50–100 km ▬ 100–150 km ▬ 150–200 km Q DMSP-based lights ‹ VIIRS-based lights Notes: For 1,546 cells in Cameroon, Chad and Niger, we use the same panel-DiD model Notes: For 1,320 cells in Cameroon, Chad and Niger (CCN), we use the same panel-DiD as eq. (1) except that we now separate the 0–200 km Boko Haram Area Dummy into four model as eq. (1) except that we now consider the full period 2000–2018 (1,320 cells x 19 dummies for whether the cell is within 0–50 km, 50–100 km, 100–150 km or 150–200 km years = 25,080 obs.). We start with the sample of 1,546 cells but exclude cells having ever from the Boko Haram area (1,546 cells x 14 years = 21,644 obs.). To avoid the figure being too experienced a Boko Haram event during the period of study as well as cells within 50 km cluttered, we do not report confidence intervals. See text for details on the specification. from these cells. We also control for the log of the Euclidean distance to any CCN cell with a Boko Haram event in the same year t. For this analysis we rely on the harmonized NTL data (1992–2018) from Li et al. (2020) who combine night light data from OLS-DMSP (used until 2013) and SNPP-VIIRS (use for the period 2014–2018). Note that the high standard errors for the VIIRS observations in 2014–2018 likely reflect the fact that the assumptions made by Li et al. (2020) to recreate harmonized NTL for the whole period 2000–2018 also introduced a significant amount of noise. Appendix 189 Lake Chad Regional Economic Memorandum  |  Development for Peace  ffects of Boko Haram on Domestic Conflict, ACLED Database, 2000–2013 Table A4.1: E Combined Violence Non-Violent Explosions Protests or Conflict Measure: All Events (Including Battles Against Strategic & Remote Riots Fatalities) Civilians Dev. Violence Panel A. Dep. Var.: Dummy if Non-Boko Haram Conflict Event in the Cell in Year t 0.0002 – -0.0003 0.0004 0.0002 0.0000 0.0002 BH 200Km* Post-09 [0.0004] – [0.0003] [0.0003] [0.0003] [0.0001] [0.0002] Panel B. Dep. Var.: Number of Non-Boko Haram Conflict Events in the Cell in Year t 0.0024 0.0212 0.0016 0.0005 0.0001 -0.0001 0.0002 BH 200Km* Post-09 [0.0022] [0.0129] [0.0018] [0.0004] [0.0004] [0.0002] [0.0002] Cell FE, Cntry-Yr FE Y Y Y Y Y Y Y Yr FE*Controls Y Y Y Y Y Y Y Observations 356,874 356,874 356,874 356,874 356,874 356,874 356,874 Notes: SEs clustered at the 3rd-level admin. unit. * p<0.10, ** p<0.05, *** p<0.01.  ffects of Boko Haram on Domestic Conflict, UCD Database, 2000–2013 Table A4.2: E Type of Organized Violence Combined Conflict Measure: All Events (Incl. Fatalities) State Non-State One-Sided Panel A. Dummy if Non-Boko Haram Conflict Event in the Cell in Year t 0.0003 – 0.0000 0.0000 0.0003** BH 200Km * Post-09 [0.0002] – [0.0001] [0.0000] [0.0002] Panel B. Dep. Var.: Number of Non-Boko Haram Conflict Events in the Cell in Year t 0.0009 0.0299 0.0004 0.0000 0.0005* BH 200Km * Post-09 [0.0007] [0.0236] [0.0004] [0.0000] [0.0003] Cell FE, Cntry-Yr FE Y Y Y Y Y Yr FE*Controls Y Y Y Y Y Observations 356,874 356,874 356,874 356,874 356,874 Notes: SEs clustered at the 3rd-level admin. unit. * p<0.10, ** p<0.05, *** p<0.01. Table A4.3: Effects of Boko Haram on Domestic Conflict, SCAD Database, 2000–2013 Type of Social Conflict Combined Conflict Measure: All Events (Incl. Fatalities) Demonstration Riot Strike Violence Panel A. Dummy if Non-Boko Haram Conflict Event in the Cell in Year t 0.0005 – 0.0003 0.0003 0.0002 -0.0001 BH 200Km * Post-09 [0.0005] – [0.0003] [0.0002] [0.0002] [0.0003] Panel B. Dep. Var.: Number of Non-Boko Haram Conflict Events in the Cell in Year t 0.0010 0.0024 0.0007 0.0003 0.0001 -0.0001 BH 200Km * Post-09 [0.0007] [0.0020] [0.0004] [0.0002] [0.0003] [0.0003] Cell FE, Cntry-Yr FE Y Y Y Y Y Y Yr FE*Controls Y Y Y Y Y Y Observations 356,874 356,874 356,874 356,874 356,874 356,874 Notes: SEs clustered at the 3rd-level admin. unit. * p<0.10, ** p<0.05, *** p<0.01. 190 Appendix Technical Paper 3. Estimating the Spillover Economic Effects of Foreign Conflict: Evidence from Boko Haram Table A4.4: Effects on Domestic Conflict, 2000–2013, Excluding Military Headquarters Cells Conflict Database: ACLED (Armed Conflict) Uppsala (Armed Conflict) SCAD (Social Conflict) All Intensive Extensive All Intensive Extensive All Intensive Extensive Sample: (1) (2) (3) (4) (5) (6) (7) (8) (9) Panel A. Dep. Var.: Dummy if Non-Boko Haram Conflict Event in the Cell in Year t BH 200Km * Post- -0.0005 -0.0056* -0.0004 0.0001 0.0011 0.0001 -0.0004 -0.0052 0.0001 09 [0.0004] [0.0031] [0.0003] [0.0001] [0.0020] [0.0001] [0.0004] [0.0032] [0.0002] Panel B. Dep. Var.: Number of Non-Boko Haram Conflict Events in the Cell in Year t BH 200Km * Post- -0.0016* -0.0337** -0.0004 0.0000 0.0014 0.0000 -0.0004 -0.0063* 0.0002 09 [0.0009] [0.0162] [0.0006] [0.0002] [0.0034] [0.0001] [0.0004] [0.0036] [0.0002] Cell FE, Cntry-Yr FE Y Y Y Y Y Y Y Y Y Yr FE*Controls Y Y Y Y Y Y Y Y Y Observations 302,344 12,796 289,548 302,344 12,796 289,548 302,344 12,796 289,548 Notes: SEs clustered at the 3rd-level admin. unit. * p<0.10, ** p<0.05, *** p<0.01. Table A4.5: Post-2009 Effect of Proximity to the Boko Haram Area, Lights, Alternative SEs Dependent Variable: Log (Mean Night Light Intensity + 1) in Year t SEs Clustered using Admin. units Conley SEs - Distance Cut-Off = Standard errors: Level 3 Level 2 Level 1 50 km 100 km 200 km (1) (2) (3) (4) (5) (6) -0.097*** -0.097*** -0.097** -0.097*** -0.097*** -0.097*** BH 200Km * Post-09 [0.027] [0.031] [0.038] [0.029] [0.034] [0.036] Cell FE, Cntry-Year FE Y Y Y Y Y Y Year FE*Controls Y Y Y Y Y Y Notes: Obs.: 21,644. * p<0.10, ** p<0.05, *** p<0.01. Table A4.6: Post-2009 Effect of Proximity to Boko Haram, Alternative Measures of the Shock Dependent Variable: Log (Mean Night Light Intensity + 1) in Year t Borno Only Borno + Yobe City of Driving Log Dist. to Measure: Baseline + Adamawa Maiduguri Time BH Area (1) (2) (3) (4) (5) (6) -0.097*** -0.123*** -0.057** BH 200 Km * Post-09 [0.027] [0.038] [0.026] -0.119*** Maiduguri 300Km * Post-09 [0.039] -0.044** BH 6.5 Hrs * Post-09 [0.018] -0.033*** (-) Log Dist. BH * Post-09 [0.012] Cell FE, Cntry-Year FE Y Y Y Y Y Y Year FE*Controls Y Y Y Y Y Y Notes: Obs.: 21,644. * p<0.10, ** p<0.05, *** p<0.01. Appendix 191