Shock Persistence and the Study of Armed Conflict: Empirical Biases and Some Remedies

AbstractPoor employment prospects for potential insurgents are often thought to increase the intensity of armed conflict. A large empirical literature tries to identify the strength of this “opportunity cost” channel, in part by regressing conflict on commodity price shocks that affect the demand for workers. In this research note we develop a theoretical framework to interpret these empirical results. We argue that because commodity price shocks are usually persistent, the estimated strength of the opportunity cost mechanism will be biased upwards (towards zero)—even for labor-intensive commodities whose price shocks are not permanent. We define this bias analytically and, using regressions on simulated data, show that it is quantitatively important for commodities studied in the literature. The bias occurs because persistent shocks that reduce employment prospects today are correlated with unobserved dynamic motivations to fight, such as the expected value of an oil field or a fighter’s subjective value of a grievance. We conclude that the opportunity cost mechanism may be even stronger than has been estimated, and that researchers should use transient, seasonal or anticipated shocks to identify its magnitude.Con frecuencia, se tiene la tendencia a pensar que el hecho de que los potenciales insurgentes tengan unas expectativas pobres en materia de empleabilidad contribuye a aumentar la intensidad de los conflictos armados. Existe una amplia bibliografía, de carácter empírico, que trata de identificar la fortaleza de este canal de «coste de oportunidad». Esto, lo hace, en parte, mediante la regresión del conflicto con respecto a las perturbaciones en materia de precios de los productos básicos, las cuales afectan la demanda de trabajadores. En esta nota de investigación, desarrollamos un marco teórico para interpretar estos resultados empíricos. Argumentamos que, debido a que las perturbaciones en materia de precios de los productos básicos suelen ser persistentes, la fortaleza estimada del mecanismo de coste de oportunidad estará sesgada al alza (hacia cero), incluso para aquellos productos básicos intensivos en mano de obra para los cuales las perturbaciones en materia de precios no resultan permanentes. Definimos este sesgo de forma analítica y demostramos, mediante el uso de regresiones en datos simulados, que este resulta cuantitativamente importante para los productos estudiados en la bibliografía. El sesgo se produce porque los cambios persistentes que contribuyen, actualmente, a reducir las perspectivas de empleo se correlacionan con motivaciones dinámicas para luchar que no han sido tenidas en cuenta, como el valor esperado de un campo petrolero o el valor subjetivo que un combatiente otorga a un agravio. Concluimos que el mecanismo de coste de oportunidad puede ser incluso más fuerte de lo que se había estimado, y que los investigadores deben usar perturbaciones transitorias o estacionales para poder identificar su magnitud.Des perspectives d’emploi médiocres pour les rebelles potentiels intensifieraient souvent les conflits armés. Une abondante littérature empirique tente d’identifier la force de ce canal de « coût d’opportunité », en partie en remontant le conflit jusqu’à une crise des prix des matières premières qui a une incidence sur la demande de travailleurs. Dans cette note de recherche, nous développons un cadre théorique pour interpréter ces résultats empiriques. Nous affirmons que puisque les crises des prix des matières premières sont généralement de longue durée, l’estimation de la force du mécanisme coût-opportunité sera haussière (vers zéro), même pour les marchandises intensives en travail dont les crises de prix ne sont pas permanentes. Nous définissons ce biais sur le plan analytique et, à l’aide de régressions à partir de données simulées, nous montrons son importance quantitative pour les matières premières étudiées dans la littérature. Le biais intervient quand des crises persistantes qui réduisent les perspectives d’emploi aujourd’hui sont corrélées à des motifs de conflit dynamiques, comme la valeur attendue d’un champ pétrolifère ou la valeur subjective qu’un combattant accorde à un grief. La conclusion indique que l’importance du mécanisme coût-opportunité pourrait bien être plus élevée que nos estimations, et suggère aux chercheurs d’utiliser des crises passagères ou saisonnières pour identifier son ampleur.Keywords: Commodity price shocksconflictopportunity cost mechanismJEL: D74O13Q34 AcknowledgmentWe thank the editor, two anonymous referees, and participants at various conferences and seminars for helpful comments.Disclosure statementThe views expressed here are the authors’ and do not necessarily reflect those of the World Bank, its executive directors, or the countries they represent.Notes1 For example, Dube and Vargas (Citation2013) find that higher coffee prices reduce conflict intensity in Colombia and, as coffee price shocks are typically fairly persistent, their results might be even stronger than estimated. Exactly how much stronger is difficult to calculate as they instrument coffee price shocks using other country’s exports (which have an unknown persistence), rather than using the international price series. Bazzi and Blattman (Citation2014) find that price shocks for all commodity types reduce conflict intensity (their Tables 7 and 8). As these are bundles of different commodities, its persistence is also difficult to calculate.2 Motivated by the high fraction of part-time fighters in insurgencies (Appendix Figure 3.2), our model assumes insurgent fighters can transition freely between working and fighting. However, in contexts where there are substantial fixed costs of moving from fighting to working (or vice-versa), it could be that the opportunity cost mechanism itself is weaker in response to temporary/seasonal shocks than persistent shocks, and in those contexts our theory may be less applicable.3 In terms of practical application, researchers need to be careful that (i) the transient/seasonal shocks used actually affect labor incomes and the opportunity cost of fighting (rainfall shocks outside the growing season may not), and (ii) results are not confounded by other seasonal variables, such as temperature and religious festivals. Guardado and Pennings (Citation2020) get around the second problem by utilizing subnational variation in the timing and intensity of harvest, which allows them to include month fixed effects and temperature controls in their specification.4 We focus on conflict intensity for simplicity, because that is where commodity price shocks are most likely to have an effect, but also because most insurgencies are long-lasting with low intensity (Fearon Citation2008). The connection between shocks and conflict incidence or onset is weaker (and is close to zero for most countries). The relation is also more complex, in part because shocks can affect bargaining which staves off conflict onset/incidence (Chassang and Padró i Miquel Citation2009). A recent empirical literature has emphasized the complexity of modeling conflict onset (rather than intensity): Malone (Citation2022) finds the opportunity cost mechanism tends to affect militant formation rather than conflict onset and Buhaug et al. (Citation2021) find that droughts are more likely to start a conflict in areas with a recently downgraded ethnic group.5 Alternatively, she might instrument the wage with a relevant shock.6 If shocks are anticipated, they will also leave the expected value of the prize (EtΠt+1Prize) unchanged—because the value of the changes in future profits have already been “priced in” (as in the stock market). Hence the estimated size of the opportunity cost mechanism is close to being unbiased using anticipated shocks, even when those shocks are persistent and the value of the prize is unobserved (results available upon request). Seasonal shocks are both transient and anticipated.7 Bias in some dynamic models of conflict might be due to omitted lags; see Beck and Katz (Citation2011). But in the greed models here, the estimated coefficient on lagged wages or lagged violence is zero, and does not affect the contemporaneous elasticity.8 This applies to types of commodities and production where insurgent groups are be able to tax production in the event their campaign is successful. Commodities with a benchmark world price studied in the literature are often easier to tax, given exports must pass through ports or through border crossings. But this is not universal. For example, Crost and Felter (Citation2020) find that higher world banana prices increase conflict and extortion by rebels in the Philippines only for large plantations, but not for small farmers.9 This is a good approximation of the “power form” of the contest function (Garfinkel and Skaperdas Citation2007, Equation 3) in a low-level insurgency where the strength of the government is large (and constant)–see Appendix 1. ϕ>1 ensures positive but diminishing marginal returns to fighting p′(Vt)>0,p″(Vt)<0.10 When ϕ→1 (weak opportunity cost mechanism), p(Vt) is very curved so when wages increase, violence barely needs to fall in order to increase p′(Vt) and make fighting more attractive. Conversely, when ϕ is very large (strong opportunity cost mechanism) p(Vt) is very flat, so when wages increase, violence has to fall substantially in order to increase p′(Vt) to make fighting more attractive to regain equilibrium.11 In this former case, let the price of domestic consumption goods be the numeraire and the international price of the commodity be θt, and assume that the volume of cash crops produced for export is Ltα . Then the amount of consumption goods that can be purchased is θtLtα.12 We abstract from any other costs of losing, like death, exile or imprisonment, which make the model more complicated. One could consider a more general setup with a fixed cost of losing, or where a losing insurgent is not able to work in the second period. The fixed cost from losing has little qualitative effect on our argument. Moreover, the inability to work in the second period actually strengthens the bias from persistent shocks, as a positive persistent shock that raises wages in t makes the value of winning (and hence working in t + 1) more valuable.13 The constant κ1=ϕlnψ(1−ϕ−1)+ϕlnδ. In Equation (5), κ2= ϕlnψ(1−ϕ−1)+ϕlnδ and κ3=(1−ρ)lnθ¯ and κ4=κ2+κ3−ϕlnα14 Fearon (Citation2008) and Chassang and Padró i Miquel (Citation2009) both argue that permanent changes in the level of economic development, or income, increase both the opportunity cost of violence and the spoils of war, leaving the level of violence unchanged, which we also find as ρ→1.15 This result (analytically and quantitatively) requires ψ, the efficiency of the fighting technology, to be adjusted to maintain the insurgency at only a low level so L≈1 (this is also the empirically relevant case, Fearon Citation2008).16 One might be concerned that seasonal or transient shocks might only have a small effect on wages or employment opportunities, as they are smoothed by employers. This seems unlikely in most cases, as employment in conflict-affected countries is typically informal, and informal employment lacks the binding long-term contracts that could isolate workers from changes in labor demand. One possible exception is “labor tying” arrangements in South Asia, but Mukherjee and Ray (Citation1995, 208) argue “the incidence of tying has undergone a steep secular decline to low current levels.” Drawing on de Janvry, Duquennois, and Sadoulet (Citation2022) and Guardado and Pennings (Citation2020), we report empirical evidence from Malawi and Iraq (respectively) that rural employment varies strongly with the seasons. See Appendix 3.1 for further discussion.17 As ρ→1 (more permanent), βOppMeas.→0, as in Proposition 1 above.18 i.e. βOppMeas.=−1.8 rather than βOppTrue=−3.19 The PIH states that only permanent changes in income should affect consumption (and transient shocks should be saved). Using data from Thailand, Paxson (Citation1993, 70) estimates how consumption varies with incomes across the seasons, and concludes that “the timing of income flows has little to do with the timing of [consumption] expenditure across seasons.”20 Appendix 2.3 includes a counterinsurgency information model of violence (motivated by Berman, Shapiro, and Felter Citation2011), which generates similar results through the same income-effect mechanism: a persistent increase in wages reduces violence directly, but also makes the agent richer, which makes passing-on counterinsurgency information for payment less attractive.