Author links open overlay panelAchyuta Adhvaryu ^a1

, James Fenske ^b2

, Gaurav Khanna ^c3

, Anant Nyshadham ^d4Show moreShareCite

https://doi.org/10.1016/j.jdeveco.2020.102598Get rights and content

Abstract

Evidence suggests that natural resources have driven conflict and underdevelopment in modern Africa. We show that this relationship exists primarily when neighboring regions are resource-rich. When neighbors are resource-poor, own resources instead drive economic growth. To motivate the empirical study of this set of facts, we present a simple model of parties engaged in potential conflict over resources, revealing that economic prosperity is a function of equilibrium conflict prevalence, determined not just by a region’s own resources but also by the resources of its neighbors. Structural estimates confirm the model’s predictions, and reveal that conflict equilibria are more prevalent where institutional quality is worse.

Introduction

No understanding of the development of modern Africa would be complete without an appreciation of the importance of natural resources. Rents from natural resources can drive economic growth; yet the countries with the highest resource endowments tend to have the slowest rates of growth (Lowes and Montero, 2019; Sachs and Warner, 2001; Gylfason, 2001). One reason for this is that as the gains from expropriating resources rise, conflict becomes more likely (Buonanno et al., 2015; Fearon, 2005; Dube and Vargas, 2013; Caselli et al., 2015; De La Sierra, 2015; Blattman and Miguel, 2010). Similarly, resources can empower the state and its rivals, and can be used to fuel repressive and destructive activities (Mitra and Ray, 2014; Nunn and Qian, 2014; Caselli and Tesei, 2016; Acemoglu and Robinson, 2001; Dube and Naidu, 2015). Where these motives are salient, and where conflict is destructive enough, resource windfalls may indeed hamper economic development (Bannon and Collier, 2003; Blattman and Annan, 2010).

Critical in these arguments is the idea of strategic interaction between rival factions in the face of economic incentives. Conflict and growth are outcomes of a game in which both a party’s own resources as well as the resources of rivals determine equilibrium choices (Mitra and Ray, 2014; Esteban and Ray, 2011a, 2011b; Esteban et al., 2012; Grossman, 1991; Hodler, 2006). How does the joint distribution of resources across potential rivals (in addition to one’s own resources) work to foment or prevent conflict, and what economic outcomes ensue in equilibrium? Ignoring the role of resources in neighboring regions provides an incomplete characterization of the interdependency of conflict and economic growth. These core questions are the focus of our paper.

As evidence on the importance of resources across potential rivals in the African context, consider the following patterns in the spatial distributions of conflict prevalence, economic development, and resources. In Fig. 1, we plot the log density of nighttime illumination (a proxy for prosperity) against a resource index for 0.5° by 0.5° grid cells across the entirety of sub-Saharan Africa, grouped into percentile bins.¹ Importantly, to capture the idea that resource endowments of potential rivals matter, we split all neighboring grid-cells within a 500 km radius into being either above or below the median value of this resource index.

We see in Fig. 1 a remarkable difference in the relationship between natural resources and prosperity by neighbors’ resources. With resource-rich neighbors, there is a clear inverse U-shaped relationship between natural resources and the density of nighttime illumination – after a certain point, when neighbors are rich, one’s own resources become detrimental for economic development.² In contrast, with resource-poor neighbors, there is an increasing and fairly linear relationship between natural resources and nighttime illumination. That is, when neighbors are relatively poor, resources continue to be good for development at all levels.

Fig. 2 helps to explain part of these divergent relationships. Here, we graph conflict incidence against the resource index for sub-Saharan African countries in the same time period, again splitting the sample by neighbors’ resources. The resulting relationship is substantially more positive for resource-rich neighbors – that is, natural resources fuel conflict to a greater extent when potential rivals also have greater resources. If the impacts of conflict on development are destructive enough, then the increased likelihood of conflict could dominate the positive effects of natural resources to produce the inverse U-shaped relationship we see in Fig. 1.

In this paper, we formalize these linkages and shed light on the underlying mechanisms through the lens of strategic interaction, stressing the role of resources in neighboring regions. We present a simple game-theoretic model in which two groups decide whether or not to engage in conflict. Rather than incorporating mechanisms separately as much of the literature does, we simultaneously integrate several mechanisms into a single model that can be tested empirically. We model both supply-side channels that determine the ease of engaging in conflict, and demand-side channels that capture the benefits of acquiring territory.

In our model, offensive and defensive capabilities for each group increase with resource endowments. This endowment effect of resources on conflict is a feature of early models of conflict (e.g., Hirshleifer (1989); Grossman and Kim (1995)) though it has received limited attention in more recent empirical work, with the exception of a handful of important recent studies (Mitra and Ray, 2014; Caselli et al., 2015; Dube and Vargas, 2013; Bazzi and Blattman, 2014; Morelli and Rohner, 2016; Guariso and Rogall, 2017). In addition to this endowment effect, we explicitly model a “rapacity effect,” by stipulating that each group’s return to fighting is increasing in the neighboring group’s resources.³ Conflict arises as an outcome if either group chooses to fight.⁴ If both fight, the probability of success is determined by the relative strength of each group, which itself depends on the spatial distribution of resource endowments.

Nash equilibria in this model are determined by the resource endowments of each group (along with other fundamentals such as the cost of fighting and the fraction expropriated when winning). When both groups have low levels of resources, peace results. This is because neither group has much strength, and the gains from fighting are also not high for either group, since contestable resources are few. When one group has more resources (loosely speaking) than the other, a one-sided conflict equilibrium results – what one might call an “uncontested attack.” When both groups have abundant resources, both are impelled to conflict. In line with the mechanisms outlined in Acemoglu and Johnson (2005); Acemoglu et al. (2001, 2005), we model the quality of institutions as shifting the cutoffs for conflict onset, by either changing the costs of war or the fraction of appropriable resources, or both.

We test the model’s predictions using disaggregated spatial data on resource endowments, conflict, and satellite data on nighttime lights. Rather than focusing on countries, we partition sub-Saharan Africa into a 0.5° × 0.5° grid. At each point, we match the likelihood of conflict events and the intensity of nighttime lights to a “natural resource” indicator (which equals 1 if any natural resource from the following is present: oil and natural gas reserves, deposits of “lootable” diamonds, gold, zinc, and cobalt) at that point. We also use historical rainfall patterns as an alternative agricultural measure of long-term wealth accumulation and resource abundance. We then match these points i to every neighbor j within a given radius. We use two sources of data on conflict, from (a) the Armed Conflict Location & Event Data Project (ACLED), which allows us to focus specifically on territorial conflicts, and (b) the Conflict Sites version of the UCDP Peace Research Institute Oslo (PRIO) data, which allows us to measure the spatial intensity of conflicts.⁵ Using these data, we ascertain whether two regions were involved in joint conflict over the past 10 years, after restricting neighboring regions to be across ethnic borders. This restriction does not qualitatively affect our results.

The model’s main predictions amount to a partitioning of the ij “resource space” into Nash equilibria regions defined by resource thresholds. Both a region’s own resources and importantly, the resources in surrounding areas jointly determine the likelihood of conflict. In the empirical analysis, we begin by drawing a heat map of the raw data on the involvement of shared conflict for points i and j over the resource index for these points. In this simple plot, we find striking confirmation of the model’s implications regarding equilibria regions over the ij resource space. Groups represented by our disaggregated points i and j behave in a manner markedly consistent with the predictions of our simple model.

In a regression framework, we test these implications by estimating the relationships between i– and j-specific natural resource endowments with whether or not regions i and j were involved in the same conflict. In keeping with recent work that finds time-invariant characteristics are often better at predicting conflict than shocks (Bazzi et al., 2018), we focus on the longer-run “endowment” of resources in the cross section. Yet, we show that our results are similar when leveraging the panel dimension of our data, where we interact the presence of the resource with temporal changes to world prices of the resources.

To address concerns regarding identification that arise from the use of the cross-sectional dimension of the data rather than time-varying shocks, we show that our results are robust to controlling for local geographic, agricultural, and climatological characteristics, as well as spatial fixed effects of varying size. Standard errors are clustered using conservatively-defined geographic levels to account for potential spatial correlation in the error term. When testing the model using rainfall data, we find a (two-dimensional) structural break in the relationship between region i and j‘s historical rainfall patterns on the one hand and conflict on the other.⁶

The results of this analysis are in line with the heat map evidence and in strong support of the model’s predictions. Own and neighbor resource endowments are both statistically and economically significant determinants of the spatial distribution of conflict in sub-Saharan Africa. These patterns are consistent when analyzing many different resources and both sources of conflict data, further alleviating concerns of omitted confounders driving the results. We complement this evidence with procedures relying on optimal bandwidth regression discontinuity (RD) methods (Calonico et al., 2014) to measure the rise in the likelihood of conflict when crossing the resource threshold from a peace to a conflict equilibrium region.

We also investigate contexts in which resources do not lead to conflict. In keeping with the importance of institutions as mediators of conflict and development (Acemoglu and Johnson, 2005; Acemoglu et al., 2001, 2005; Caselli and Tesei, 2016; Mehlum et al., 2006; Barro, 1996; Carreri and Dube, 2018), we show that for ij pairs where baseline institutional quality (measured in alternative specifications by property rights, risk of expropriation, political stability, and voice and accountability) is lower the estimated resource value at which conflict ensues is lower (i.e., conflict is more likely to break out for smaller resource endowments) and the explanatory power of our model is substantially higher. These results suggest that good institutions at baseline potentially raise the costs of conflict and/or lower the gains from expropriation, and weaken the link between resources and conflict overall. In particular, the roles played by rapacity and relative strength in conflict may become less important in the presence of stronger baseline property rights and lower risks of expropriation.

Finally, we estimate analogous regression equations for satellite data on nighttime lights (Henderson et al., 2012; Michalopoulos and Papaioannou, 2013; Pinkovskiy and Sala-i Martin, 2015) to highlight how the resource-conflict dependence results in a non-monotonic reduced form relationship, seen in Fig. 1, between resource abundance and development (Sala-i Martin and Subramanian, 2013). Additional evidence using regression discontinuity methods and two-stage least squares analyses supports this story, showing that as we move across regions on either side of the optimally determined threshold the rise in conflict correspondingly leads to a sharp drop in light density.

Our main contribution is to the literature on natural resources and conflict. The closest papers to ours in this large literature are Besley and Persson (2010); Caselli et al. (2015); Harari and La Ferrara (2018), and Berman et al. (2017). We build on these studies in several ways. First, we highlight the importance of strategic interaction in determining conflict prevalence and levels of development across space, simultaneously incorporating several mechanisms from this literature into a single theoretical model with clear testable predictions. Taking seriously the role played by neighboring regions allows us a nuanced understanding of this relationship. Resources lead to conflict particularly when neighboring regions are resource-rich. When a rival is poor, more resources may no longer drive conflict, and instead fuel development.

Second, like Harari and La Ferrara (2018) and Berman et al. (2017), our analysis is on a granular (grid point) level and spans the entirety of sub-Saharan Africa. What is different is that we analyze point pairs in potential joint conflict, drawing focus on strategic interaction effects in the determination of equilibrium conflict.⁷ Importantly, this allows us to focus on within-country conflicts (which are far more prevalent in our data), whereas other work studying the impacts of a neighbors’ resources (Caselli et al., 2015) look at cross-country conflicts.⁸ Country and ethnic group boundaries are endogenous and give pause when studying territorial conflicts. Additionally, ethnic group borders may be measured with error, and as such we find the grid level analysis in recent work more suited for our study. Yet, we show that our results are strong when we restrict grid cells to opposing sides of ethnic boundaries. Unlike much other work, we also show consistent results for a broad set of natural resources, allowing for greater generalizability.

Third, of the studies mentioned, only Besley and Persson (2010) explicitly model and examine equilibrium impacts on economic development. We are able to study this outcome at a highly granular level using satellite data on nighttime illumination, and highlight a negative externality to having richer neighbors.

We also contribute to the literature on the determinants of growth (Barro, 1991; Jones, 2016). While both natural resources and conflict have long been regarded as playing key roles in the economic growth of developing nations (Rodrik, 1999; Sachs and Warner, 2001), few papers have disciplined the interrelationships between resources, conflict, and growth through a model of strategic interaction, and tested its predictions across a large set of regions incorporating data on resource endowments of many types. We are also able to directly incorporate the role of property-related institutions in the model to study the potential for institutional quality to modulate the aforementioned relationships (Mehlum et al., 2006).

The remainder of the paper is structured as follows. Section 2 sets up our model and delivers its main predictions through a set of lemmas and propositions. Section 3 describes our data. Section 4 details our empirical strategy, and section 5 presents and interprets our results. Finally, section 6 is a concluding discussion.

Section snippets

Model

To motivate our empirical study, we model the interaction of two parties, i and j, who play a symmetric, simultaneous game that determines peace or conflict between them. Our static model generates testable empirical predictions, and extensions to the basic model, including heterogeniety in institutional structures, provide additional refinements to our predictions. The model is simple to better communicate the primary testable implications.⁹

Data

To test the implications of the model, we combine spatial data on rainfall, oil and gas reserves, diamond deposits, gold mines, zinc deposits, cobalt mines, conflicts, and nighttime lights. We begin with a data set at the 0.5° by 0.5° latitude/longitude grid level covering the whole of Sub Saharan Africa.¹⁴

Estimation strategy

The theoretical model allows us to divide the conflict-resources space into four distinct Nash Equilibrium regions. When there are low resources for both parties, there is a lower probability of conflict as neither party has resources to build an army and there is little wealth to expropriate from one’s neighbor. On the other hand, having a large amount of resources for either party leads to more conflict, and this is especially true when both parties have high levels of resources.

In order to

Results

In this section, we present and discuss empirical evidence in support of the model developed in section 2. The analysis is carried out in multiple stages as discussed in section 4 above.

Conclusion

We present a model of resource-driven conflict and the corresponding impacts on development, and test its implications in sub-Saharan Africa. We stress the role played by a neighboring region’s resources. In our model, natural resources impel societies to conflict by increasing both the capacity for aggression as well as the gains from expropriation. We extend the model to account for 1) the idea that resources raise the opportunity costs of conflict, and 2) the possibility that neighboringRecommended articles

References (77)

• D. Acemoglu et al.Institutions as a fundamental cause of long-run growthHandb. Econ. Growth(2005)
• J. BaiCommon breaks in means and variances for panel dataJ. Econom.(2010)
• J. Gonzalo et al.Estimation and model selection based inference in single and multiple threshold modelsJ. Econom.(2002)
• J. Gonzalo et al.Subsampling inference in threshold autoregressive modelsJ. Econom.(2005)
• T. GylfasonNatural resources, education, and economic developmentEur. Econ. Rev.(2001)
• R. HodlerThe curse of natural resources in fractionalized countriesEur. Econ. Rev.(2006)
• C.I. JonesThe facts of economic growthHandb. Macroecon.(2016)
• J.D. Sachs et al.The curse of natural resourcesEur. Econ. Rev.(2001)
• D. Acemoglu et al.Unbundling institutionsJ. Polit. Econ.(2005)
• D. Acemoglu et al.The colonial origins of comparative development: an empirical investigationAm. Econ. Rev.(2001)

View more references