Scarcity Without Leviathan - The Violent Effects Pt 1

Scarcity without Leviathan: The Violent Effects of Cocaine Supply Shortages in the Mexican Drug War Abstract This paper asks whether scarcity increases violence in markets that lack a centralized authority. We construct a model in which, by raising prices, scarcity fosters violence. Guided by our model, we examine the link between scarcity and violence in the Mexican cocaine trade. At a monthly frequency, scarcity created by cocaine seizures in Colombia—Mexico’s main cocaine supplier—increases violence in Mexico. The effects are larger in municipalities near the US, with multiple cartels, and with strong PAN support. Between 2006 and 2009 the decline in cocaine supply from Colombia could account for 10%-14% of the increase in violence in Mexico. 1 Introduction Agents in illegal markets cannot rely on the rule of law. The state does not solve disputes, protect property rights, or enforce contracts. As in other environments without a centralized authority, agents use violence as a substitute for the institutions that a well-functioning state provides.1 Illegal drug markets in countries like Colombia, Mexico, and Afghanistan constitute major examples of environments where armed groups resort to ruthless violence in order to run the illegal trade, extract rents, and protect their revenue and assets. We study the role of scarcity—or supply shortages—as a factor that exacerbates the use of violence in markets without a centralized authority. We do so in the context of the cocaine trade in Mexico. Starting in the late 1990s, Mexico became the main point of entry of drugs into the US. The industry transformed into a vertical market with little integration, where powerful Mexican cartels purchase cocaine from producers in Colombia (and to a lesser extent Peru and Bolivia) and smuggle it across the US border.2 As Figure 1 shows, since 2006, the expansion of the Mexican cocaine trade coincided with a vast increase in violence, driven by clashes between cartels (drug trafficking organizations), and between cartels and Mexican authorities. In 2007, there were 8,686 homicides in Mexico, out of which 2,760 were drug-related; in 2010, there were 25,329 homicides, out of which 15,258 were drug-related. Between 2006 and 2009 there were 60,000 drug-related killings in Mexico, twice the death 1 In developing countries, local strongmen contest areas in which the state does not central- ize authority (see Acemoglu et al., 2013; Fearon and Laitin, 2003; Sánchez de la Sierra, 2017) and warlords fight over the extraction and control of valuable resources (see Skaperdas, 2002). 2 Cocaine produced in Andean countries used to be shipped to US markets through the Caribbean. Following intensified enforcement in the Caribbean by US authorities, Colombian and Mexican traffickers began to smuggle drugs through Mexico. Although Mexico also produces drugs such as cannabis, heroin and amphetamines, close to 60% of the profits obtained by Mexican cartels is generated via drug-trafficking activities, especially those related to cocaine, and not via drug production (see Kilmer et al., 2010). Grillo (2011) and Valdés (2013) provide thorough reviews of the history of drug trafficking in Mexico and its nexus with crime and violence. toll for Afghanistan between 2001 and 2011 (see Crawford, 2011). We ask whether the surge in drug-related violence in Mexico that started in 2006 is related to shortages in the supply of cocaine caused by larger and more frequent cocaine seizures in Colombia. In 2006, the Colombian government redefined its anti-drug strategy, emphasizing the interdiction of drug shipments and the detection and destruction of cocaine processing labs over the eradication of coca crops.3 Cocaine seizures in Colombia went from 127 metric tons per year in 2006 to 203 metric tons in 2009. As Figure 1 shows, the net supply of Colombian cocaine fell from 522 to 200 metric tons in this period—at the same time that the homicide rate in Mexico roughly tripled. This large negative supply shock was noticeable in downstream markets. The retail price per pure gram of cocaine in the US went from about $114 in 2006 to about $180 in 2009—a 45% increase—and from $40 to $68 for wholesale purchases between 10 and 50 grams—a 53% increase. To motivate our analysis, we present a model of conflict in illegal markets where scarcity— driven by supply disruptions in upstream markets—fuels violence along trafficking countries. In our model, cartels use violence to dispute revenue and control over factors of production (e.g., routes, land). If demand for cocaine is inelastic, as evidence for illegal drugs suggests (Becker et al., 2006), exogenous supply reductions in upstream markets increase revenue. Rapacity over the additional money flowing into the cocaine trade triggers violence. Our model predicts that scarcity has a greater effect on violence in Mexican areas that are more important for trafficking, like those near the US border, where there are at least two cartels, and where more frequent arrests and crackdowns against cartel leaders increase turnover, which precludes informal cooperation.4 3 The change in strategy followed intense criticism of the Colombian government’s focus on disrupting the early stages of the cocaine production chain via the eradication of coca crops. The new focus on interdiction greatly increased the efficiency of supply reduction efforts (see Mejı́a and Restrepo, 2016). 4 As we discuss in our concluding remarks, there may be other mechanisms linking scarcity and violence. The purpose of our model is to show that a standard theory of conflict in which Using monthly data for Mexico from December 2006 to December 2010, we then explore the relationship between scarcity and violence in the context of the Mexican cocaine trade. We use cocaine seizures in Colombia as an external source of scarcity. Although Figure 1 shows that the rise of violence in Mexico since 2006 coincides with the sharp decline of the cocaine supply in 2006, this low-frequency correlation could simply reflect a change in policies in both countries, such as President Calderón’s war on drugs in Mexico and the emphasis on interdiction policies in Colombia. To overcome this difficulty, we exploit monthly deviations in cocaine seizures in Colombia, obtained after flexibly removing time trends. Our empirical strategy investigates whether the scarcity generated by high-frequency changes in cocaine seizures in Colombia generates violence in Mexico, and then asks which municipalities are affected the most. The key identifying assumption is that the monthly variation in seizures is exogenous to changes in enforcement in Mexico or unobserved changes in cocaine markets. Our main finding is that, during months with supply shortages caused by large cocaine seizures by Colombian authorities, violence increases in Mexico, especially in municipalities close to the US border. The effect is stronger in municipalities with a cartel presence, but only when two or more cartels are contesting each other. The increase in violence is also larger in municipalities that have historically supported PAN, the governing party that spearheaded the crackdown on the cocaine trade between 2006 and 2012. Building on Dell (2015), our interpretation is that these municipalities were more supportive of federal government efforts to defeat cartels. As a consequence, cartel leaders were arrested, killed or removed more frequently in these municipalities, precluding non-aggression treaties between cartels and exacerbating the effect of scarcity on violence. According to our preferred estimates, a 10% increase in monthly cocaine seizures in Colombia raises homicides in Mexico by 0.59%-0.91%. Under additional assumptions mentioned in the text, this means that the sharp decline in the cocaine supply from Colombia between 2006 and 2009 (from 520 mt to 200 mt per year) accounts for 10%-14% of the incartels contest the cocaine trade revenue—a description we believe captures the main features of this market—can to explain all of our empirical findings. crease in the homicide rate in Mexico during this period, and 25% of the differential increase in violence observed in the North of Mexico relative to the rest of the country. Our paper relates to three strands of literature. The first one is the literature on conflict over resources. Our model builds on the economic theory of conflict (Grossman, 1991, 2001; Hirshleifer, 2001). An increase in the value of the contested resources has an ambiguous effect on violence. On the one hand, rapacity over the additional resources—the temptation to prey on others and the necessity to defend against predators —leads to more violence (Skaperdas, 2002; Collier and Hoeffler, 2004). On the other hand, resource booms may also increase wages and, hence, the opportunity cost of engaging in violence (see Becker, 1968; Grossman, 1991; Dal Bó and Dal Bó, 2011).5 We contribute to this literature by showing that rapacity dominates in the Mexican cocaine trade: higher revenues from cocaine increase violence, especially in areas contested by multiple cartels. This adds to the empirical literature that finds a link between resource booms and violence (Collier and Hoeffler, 2004; Buonanno et al., 2015; Couttenier et al., 2017; Dube and Vargas, 2013) as well as the theoretical literature that emphasizes the role of the number of contenders (Kalyvas, 2006). Our paper also contributes to the literature on violence in illegal markets (Goldstein, 1985). Out model relates to works that analyze the motivations behind violence between the state and cartels (Lessing, 2015) and between competing cartels (Castillo and Kronick, 2018). On the empirical side, our finding that higher revenues lead to more drug-related violence in the Mexican cocaine market is consistent with results by Angrist and Kugler (2008) and Mejı́a and Restrepo (2013), who find that the rise of the cocaine trade spurred violence in Colombia. (Lind et al., 2014, show evidence for a causal link in the opposite direction for opium cultivation in Afghanistan.) Similar results hold for other illegal markets (Garcı́a-Jimeno, 2016; Owens, 2014; Chimeli and Soares, 2017). Finally, our work contributes to the debate about the causes of the upsurge in violence 5 In line with this channel, other authors find that national income and commodity prices reduce the likelihood of conflict in Africa (Miguel et al., 2004; Brückner and Ciccone, 2010) and that booms in labor-intensive commodities like coffee may actually reduce violence (see Dube and Vargas, 2013). in Mexico since 2006. Some observers blame the upsurge on the struggle for power created by the policies implemented by Calderón, who in 2006 started an aggressive strategy of beheading cartels by killing or arresting their leaders (Guerrero, 2010; Merino, 2011).6 Using a regression discontinuity design, Dell (2015) finds a significant increase in anti-drug crackdowns in municipalities in which PAN—Calderón’s party—won local elections. Crackdowns are followed by a spike in drug-related violence. Similarly, Calderón et al. (2015), who compare the area of influence of a cartel with a synthetic control group, find that the killing of a cartel leader raises violence.7 We emphasize the role of policy changes in Colombia. The dramatic increase in cocaine seizures in Colombia since 2006 created large and frequent shortages in downstream cocaine markets, which translated into more frequent and larger surges of violence in Mexico. This is not to say that PAN and Calderón’s policies had no effect; indeed, our evidence suggests that Colombian policies and the Mexican policies interacted to cause large increases in violence. The paper is organized as follows. Section 2 presents our model and discusses its testable predictions. Section 3 describes our data, empirical strategy and results. Section 4 concludes. 2 A model of scarcity and violence Our model features cartels that operate in a transit country (Mexico) and buy cocaine in upstreams markets (Colombia) to traffic to a downstream market (the US). The key feature of our model is that cartels fight over the control of the cocaine trade and the revenue it commands. In particular, cartels fight for the control of municipalities, some of which are strategically more important and, thus, account for a larger share of total cocaine revenue. When total revenue increases—as a consequence of scarcity spurred by cocaine seizures 6 Other observers have defended this strategy, arguing that by reducing cartel entry it could decrease violence in the long term (Poiré and Martı́nez, 2011; Villalobos, 2012). 7 Two recent papers shift the focus away from Calderón’s policies. Dube et al. (2016) show that negative agricultural shocks facilitated the rise of the Mexican drug sector, thus, increased violence. Dube et al. (2013) find that violence in Mexico increased as a consequence of the 2004 expiration of the US Federal Assault Weapons Ban. in Colombia—rapacity over the additional resources fuels violence. Rapacity is more severe in municipalities that are important for the cocaine trade and that command a larger share of the trade revenue, and in municipalities contested by multiple cartels. We then introduce a dynamic version of the model where cartels take into account the threat of retaliation when they are fighting over revenue. This extension shows that transitory increases in revenue have a strong effect on violence because they increase the benefits of one-shot deviation while having a limited impact on the future value of cooperation. Policies that make cartel leaders more impatient, such as an increase in arrest rates, make deviations more attractive and cause violence to be more responsive to spikes in revenue. We discuss the connection between this model and our empirics at the end of this section. Static setting: Let Qs be the quantity of cocaine supplied by upstream countries like Colombia, which we assume is exogenous and determined by supply-reduction policies in these countries. There are N Mexican cartels, indexed by c ∈ C = {1, 2, . . . , N}. Each cartel buys an amount Qsc of cocaine at a price P s and traffics it to downstream markets at a cost C(Qsc ), which is increasing, convex, and satisfies limQ→0 C ′ (Q) = ∞ and limQ→∞ C ′ (Q) = 0. We abstract from seizures by Mexican authorities, which implies that cocaine trafficked is P given by c Qsc = Qs . Cartels then sell cocaine in the downstream market at a price P d (Qs ), which they take as given.8 When deriving our results, we allow demand, P d (Qs ), to have any price elasticity. However, as we discuss in Subsection 2.1, evidence from related markets suggests that demand for cocaine traded through Mexican is inelastic. The Mexican cocaine trade generates a total revenue R(Qs ) = Qs P d (Qs ). Our key assumption is that, because there are no formal property rights in the cocaine market, revenue is unsecured and must be disputed with other cartels. In particular, a cartel is only entitled to a share ηQsc P d (Qs ) of the revenue it generates, leaving a share of total revenue 8 We assume cartels are price takers. The term cartel is, therefore, misleading. The powerful Colombian drug trafficking organizations of the 1980s and 1990s called themselves carteles to emphasize that a few organizations controlled the whole trade, even though they did not collude to increase prices. The name stuck, and present-day Mexican organizations use the same term. equal to (1 − η)R(Qs ) to be contested among cartels. Thus, η measures the extent to which a centralized authority can credibly protect property rights. We formalize the dispute over the unsecured revenue as a dispute over municipalities, the control of which allows cartels to capture more of the cocaine revenue. In particular, we assume there are I municipalities, indexed by i ∈ I = {1, 2, . . . , I}. We denote by Ic ⊂ I the subset of municipalities where a cartel operates, and let Ni ⊂ C be the set of cartels that operate in municipality i. Municipalities are heterogeneous in how relevant they are for the total cocaine trade. For instance, municipalities close to the US or located along strategic routes have a greater importance, which allows cartels that control these areas to obtain a larger share of the disputed cocaine trade revenue. We model this heterogeneity by assuming that in municipality i a fixed share si ≥ 0 of the total revenue from the cocaine trade is P disputed, with the understanding that i∈I si = 1. An alternative interpretation of si is that trafficking requires inputs, such as routes and land, that cannot be traded in regular markets and that, therefore, must be contested. Parameter si then captures the share of revenue accruing to the cartel that controls the inputs available in municipality i. To appropriate or defend the unsecured revenue (1 − η)si R(Qs ) in municipality i, the cartels that operate in it engage in violent conflict. We model the dispute over revenues with a contest success function (Skaperdas, 2002). Cartels in Ni simultaneously decide amounts gc,i ≥ 0 to invest in weapons or soldiers in municipality i. As a result, cartel c is able to obtain a fraction qc,i of the total revenue contested in municipality i, where gc,i qc,i = P . c′ ∈Ni gc′ ,i This contest success function represents all the strategies by which cartels appropriate and extract drug rents. For instance, they can act as toll collectors and extort payments from other cartels operating in municipality i; they can fight for turf in order to use routes and labor for their trafficking activities; they can prey on one another by violently stealing cash, weapons, assets or other resources; or they may eliminate competition in key municipalities to increase their market share. The aggregate level of drug-related violence in municipality i is therefore linked to the total expense in the local conflict by all cartels, vi = P c∈Ni gc,i . Our assumptions imply that a Cartel’s profits are then given by X (qc,i (1 − η)si R(Qs ) − gc,i ) . πc = −P s Qsc − C(Qsc ) + ηP d (Qs )Qsc + i∈Ic The Online Appendix characterizes the full equilibrium of this market (including cartels’ choice of quantities). In order to characterize the outcome of conflict, we note that by the time cartels dispute the control of municipalities, the amount paid for drugs, −P s Qsc , the cost of smuggling, C(Qsc ), and the revenues not subject to appropriation, ηP d (Qs )Qsc , are bygones. Thus, cartels simply care about the disputed revenue si R(Qs ).9 Also, because the outcomes qc,i are independent across municipalities, it suffices to solve the conflict in each municipality separately.10 Cartel c’s maximization problem in municipality i is thus given by max qc,i (1 − η)si R(Qs ) − gc,i . gc,i The Online Appendix shows that the unique Nash equilibrium of the conflict among cartels 9 One implicit assumption behind this feature is that there is no risk sharing between Mexican cartels and suppliers in Colombia. We believe that the absence of a centralized authority precludes the possibility of this kind of vertical relations. If cartels and suppliers were able to enforce complex contracts, cartel payments to suppliers could be conditioned on the municipalities that they control, which implies that profits (revenue net of payments to Colombian suppliers) and not revenue would determine the incentives to fight over a municipality. Even if this were the case, profits and violence would still increase for a sufficiently inelastic demand, as shown by Castillo and Kronick (2018). 10 The key simplifying assumption in our model is that conflict over municipalities only determines who owns the disputed revenue, but does not affect production and trafficking. In a more realistic and complicated model, each municipality could be endowed with inputs needed for drug trafficking, and cartels would only be able to use the inputs in the municipalities they control. This would result in more complicated strategic considerations because now the value of controlling a municipality would depend on who controls all other municipalities, and investments to control different municipalities could be strategic substitutes or complements. Modeling these considerations is beyond the scope of our paper. operating in municipality i results in a level of violence |Ni | − 1 vi = (1 − η)si R(Qs ). |Ni | (1) This expression leads to our main proposition.11 Proposition 1 If demand is inelastic (R(Qs ) decreasing in Qs ), a decrease in the supply of cocaine from downstream markets increases revenue and leads to more violence, vi . The increase in violence is greater in municipality i under the following circumstances: 1. If third party enforcement is weaker or absent (smaller η). 2. If municipality i is more important for cocaine trafficking (greater si ). 3. If municipality i has more cartels (greater |Ni |). Also, violence only changes if there are at least two rival cartels operating in the same municipality. If demand is elastic (R(Qs ) increasing in Qs ), a decrease in the supply of cocaine from downstream markets reduces revenue and violence, vi . Proof. The proof follows by differentiating equation (1). The main result in Proposition 1 echoes a general finding in the conflict literature: when the value of contested resources increases, agents are encouraged to invest more in conflict, which ends up fueling violence. This prediction is quite general and applies to the unique symmetric equilibrium obtained when the contest success function is symmetric, homogeneous of degree zero, and has decreasing marginal returns (e.g., all ratio contest success functions described in Hirshleifer, 1989). The same comparative statics apply to the equilibrium with the least and the largest levels of violence when we use a generic contest success function such that investments in the conflict by different cartels are strategic complements.12 11 This expression also helps us clarify which type of scarcity generates violence. Anti-drug poli- cies in municipality i displace the cocaine trade and reduce si . From the perspective of municipality i, such policy creates a scarcity of cocaine in that municipality and could reduce violence. As this example illustrates, it is the aggregate level of scarcity, which raises prices and revenue, that generates violence. The displacement of production from one municipality to another only reallocates violence. 12 One can also construct examples where the logic of parts 1 and 2 of Proposition 1 fail. For Dynamic setting: We now move to a dynamic setup based on Castillo and Kronick (2018).13 The dynamic analysis highlights two key points. First, supply shocks have an effect on violence, particularly if they are short-lived shocks like the ones we focus on in our empirical analysis. Second, beheading cartels by arresting or killing their leaders reduces the importance of retaliation, which makes violence more responsive to current scarcity. The game described above is played out repeatedly, with an independent draw of Qst in each period. Subscript t denotes time. At the end of each period, a cartel leader in municipality i is arrested or killed with probability ai , and is then replaced by a new one. Leaders maximize cartel profits at their own discount rate (1 − ai )βand not at the market rate β. Arrests and killings thus make cartel leaders more impatient.14 Just as in the static problem, the amount paid for drugs, the cost of smuggling drugs, and the revenues not subject to appropriation are bygones. Also, revenue R(Qst ) is fully determined by the realizations of Qst . Thus, in each period, cartel c operating in municipality example, investment by one party could dissuade others from investing in the conflict, making investments by different cartels strategic substitutes. In this case, an increase in the value of the contested resources could lead one cartel to increase its investment, which might then induce a larger decrease in other cartels’ investment. These forces are stronger in asymmetric equilibria where one party has a greater ex-ante chance of winning the conflict and can effectively dissuade the other parties from investing. Thus, this dissuasion mechanism could be at play in municipalities that are under the effective control of a single cartel, but is less likely to apply to municipalities contested by multiple cartels—where we expect conflict to increase with scarcity. We explore this possibility, which relates to part 3 of Proposition 1, in our empirical exercise. 13 Castillo and Kronick (2018) build a richer model of violence between criminal groups and analyze the effects of different government policies on supply and violence. Dynamic interaction opens the possibility of more favorable equilibria under the threat of retaliation (see Abreu et al., 1990). 14 This assumes that once a cartel leader is arrested or killed, the next leader cannot credibly compensate him for long-term investments. This could stem from the lack of enforceable contracts in environments without a centralized authority. i chooses gc,i,t in order to maximize expected payoffs (excluding bygones): X Et0 β t−t0 (1 − ai )t−t0 qc,i,t (1 − η)si R(Qst ) − gc,i,t. (2) t≥t0 N s Let g (Q ) be investment in the conflict in the symmetric static Nash equilibrium in municipality i when supply is Qs , which is the solution to the static model above. We focus on a recursive equilibrium where all cartels in municipality i spend g C (ai , Qs ) ∈ [0, g N (Qs )) and revert to g N (Qs ) as punishment if anyone deviates. This is not the most cooperative subgame perfect equilibrium (SPE), but it provides a simple way to illustrate how informal cooperation arrangements work. The following proposition characterizes this equilibrium: Proposition 2 If demand is inelastic (R(Qst ) decreases in Qst ), the equilibrium that sustains the most cooperation under Nash reversion is recursive and has gc,i = g C (ai , Qs ) ∈ [0, g N (Qs )). If R(Qst ) is unbounded above and converges to zero as Qst increases,15 then the function g C satisfies the following: 1. There exists some finite Q(ai ) > 0 such that g C (ai , Qs ) = 0 for Qs ≥ Q(ai ). 2. Wherever g C (ai , Qs ) is positive, it is increasing in ai , decreasing in Qs , and the cross partial derivative is negative. If demand is elastic (R(Qst ) increases in Qst ), 1 and 2 hold with opposite signs for Qs .16 Proof. See the Online Appendix. We now explain the results for the empirically relevant case, when demand is inelastic. The proposition shows that in the absence of a centralized authority there is room for some de facto protection of property rights, sustained by the threat of future retaliation. Part 1 shows that if ai is low and β is high, an equilibrium with no violence can be sustained—a folk theorem. As the effective discount rate decreases, participants value the future less and some level of violence must be allowed, especially when revenues are large. 15 Our proof, in the online appendix, shows that similar results hold if the assumptions that R(Qst ) is unbounded above and converges to zero as Qst increases are relaxed. 16 For 1, there exists some finite Q(a ) > 0 such that g C (a , Qs ) = 0 for Qs ≤ Q(a ). For 2, g C is i i i increasing in ai and Qs , and the cross partial derivative is positive. Part 2 shows that when violence occurs, temporary reductions in supply Qs increase violence for two reasons. First, they increase current revenues and, hence, the incentives to prey. Second, because of reversion to the mean, expected revenues in the future are lower, which means that the treat of retaliation is a less effective means of preventing violence (see Rotemberg and Saloner, 1986). The second effect implies that short-lived supply shocks are particularly prone to generating violence. Also, the second effect becomes stronger when the arrest rate is high, which explains why the function g C (ai , Qs ) has a negative cross partial derivative: not only is expected revenue in the future low, but cartel leaders heavily discount the threat of future retaliation when they are likely to get arrested. The effect of scarcity (in particular, short-lived shocks) on violence is thus stronger when arrest rates are high. 2.1 Empirical predictions In our empirical exercises we test the implications of our model when the demand for cocaine is inelastic, as suggested by the existing empirical evidence. Becker et al. (2006) summarize these results and conclude that it is consistent with an elasticity of demand of less than 1 for most drugs, with a central tendency towards 1/2, especially over the short term.17 The basic story that emerges from the model is that scarcity generated by temporary and large cocaine seizures in Colombia—of the type that we will explore in our empirical analysis—fosters violence in Mexico. Also, violence increases more in places that are more relevant for the drug trade (where si is higher). In our empirical application, we focus on distance from the US as a measure of si . Since the goal of cartels is to take drugs across the border, places that are close to the US are of strategic importance and, thus, command a 17 The relevant value in our model is the short-run elasticity of demand, which is smaller than that over the long run (see Becker and Murphy (1988)). Early papers found inelastic demands for marijuana and heroin (Nisbet and Vakil, 1972; Roumasset and Hadreas, 1977; Silverman and Spruill, 1977). Wasserman et al. (1991) and Becker et al. (1994) find demand elasticities between -0.4 and -0.75 for cigarette consumption over the short term. In the case of cocaine, the results of more recent studies are consistent with an inelastic demand as well. See, for instance, Bachman et al. (1990), DiNardo (1993) and Saffer and Chaloupka (1999). larger share of the drug trade. Consequently, the violence spikes caused by cocaine seizures in Colombia should be stronger close to the US border. Our model also predicts differential effects depending on the number of cartels present in a municipality. Because violence results from disputes among cartels, our model suggests that scarcity only raises violence in municipalities contested by two or more cartels; there should be no effect in municipalities controlled by a single cartel or where no cartels operate. Finally, our model suggests that supply shortages generate more violence in municipalities where cartel leaders are killed or arrested more frequently. As shown by Dell (2015), crackdowns against the cocaine trade were more recurrent in municipalities that elected PAN mayors because the policies of Calderón and PAN, beheading cartels by killing or arresting their leaders, required the cooperation of local authorities. Thus, we expect supply shortages to have a larger impact on cartel violence in municipalities in which there is a strong PAN support, which are more likely to elect PAN mayors and support their federal anti-drug policies.18 Proposition 2 also suggests that higher levels of violence should be observed when the government’s main strategy is to behead or arrest cartel leaders, as was the case during Calderón’s term. We do not test this prediction because we focus on the role of scarcity and how it interacted with Mexican policies. However, Calderón et al. (2015) find empirical support for this prediction. 3 Data and evidence on the role of scarcity We focus on the period from December 2006 to December 2010, which is the period for which drug-related homicide data are available. This coincides with the period of violence and intense cartel activity in Mexico known as the Mexican Drug War. 18 A simple cross-sectional regression shows that a 10% increase in the historical PAN vote share increases the likelihood of electing a PAN mayor during our period of analysis by 7.4%. Also, incumbents in these places may be under more pressure from their electorate to cooperate with PAN’s national policies. We use a monthly panel of all 2,457 Mexican municipalities (including the 16 boroughs of Mexico City). Our dependent variables are different measures of the monthly homicide rate in each municipality. Our main variable of interest is the monthly homicide rate per 100,000 inhabitants from INEGI (Mexico’s Instituto Nacional de Estadı́stica y Geografı́a), which includes all reported homicides. We also use data on drug-related homicides published by the Mexican Presidency, which only counts homicides related to the illegal drug trade. This dataset further classifies homicides into (1) assassinations of cartel members, which involve targeted killings of cartel members by competing cartels, (2) confrontation deaths, which are the result of battles either between competing cartels or between cartels and government authorities, and (3) assassinations of government officials, which are the result of cartels attacking government forces.19 We explore the effect of scarcity on violence and how it is mediated by the relative importance of trafficking, the number of cartels, and the political power of the PAN party. We use proximity to the US border as our proxy for trafficking importance. We compute it as the geodesic distance from the centroid of each municipality to the closest US entry point. Using the data from Coscia and Rı́os (2012), we construct dummies for the presence of cartels in all municipalities and for the presence of at least two cartels.20 Finally, in order to explore heterogeneity with respect to PAN political power, we compute the average voting share for PAN candidates in municipal elections from 1980 to 2000. For institutional reasons, this measure is only available for 1,998 of the municipalities in our sample. 19 Although it comes from an official source, the classification of homicides as drug-related and into additional categories has been criticized (Hope, 2012). Because of these concerns, we use the INEGI homicide rate as our main variable of interest. 20 Their data codes whether every major cartel operated in a given municipality. The cartels included are: Cartel de los Hermanos Beltrán Leyva, Familia Michoacana, Cartel del Golfo, Cartel de Juárez, Cartel de Sinaloa, Cartel de Tijuana and Los Zetas. We also obtained similar results if we focus on municipalities contested by rival cartels (instead of simply looking at municipalities that have multiple cartels), which are defined using the alliances described by Guerrero (2011). Table 1 summarizes our data. It presents statistics for all of Mexico (where the average distance from the US is 763 km), for municipalities in the two quintiles closest to the US (average distance 507 km), and for municipalities in the closest quintile (average distance 333 km). The Mexican population is concentrated in the north: the first quintile includes 25% of the national population and the second quintile includes 35%. While homicide rates were similar across Mexico in 2006, the north of Mexico (quintiles 1 and 2) became more violent subsequently than other parts of the country subsequently—a difference driven by a large increase in drug-related homicides. Consistent with the intuition that the north of Mexico is more important for cocaine trafficking, the data shows that the vast majority of homicides in the north are drug-related, whereas homicides in other areas of the country are less so. There is also more cartel activity in quintiles 1 and 2 than in the country as a whole. In 2010 cartels were active in 40% of the municipalities in the first two distance quintiles compared to 29% of all Mexican municipalities. Finally, northern municipalities are slightly more supportive of PAN, but these differences are not significant. Our variation in scarcity comes from the vast monthly changes in cocaine seizures by Colombian authorities.21 The data on cocaine seizures comes from the Colombian Ministry of Defense. Figure A1 in the Online Appendix presents the monthly series for seizures of cocaine in Colombia and the homicide rate in Mexico. 3.1 Time series evidence We start by providing evidence consistent with the main prediction in Proposition 1, namely that violence increases in Mexico when cocaine becomes scarce due to cocaine seizures in 21 We also constructed a series for the monthly cocaine supply net of seizures in Colombia, and we use it directly as a measure of the supply of drugs reaching Mexico. However, this methodology requires interpolating annual production figures (there is no information on monthly production). Because our estimates only exploit high-frequency changes, the variation that we exploit in this alternative exercise is fully driven by changes in seizures, and the obtained results are very similar to the ones reported here. Colombia. We study the time series effect of supply shocks by estimating the model ln ht = β ln St + Ft (γ) + εt . (3) Time runs at the monthly frequency from December 2006 (t = 0) to December 2010 (t = 48). ht is the homicide rate in Mexico, and St are cocaine seizures in Colombia. The term P3 n Ft (γ) = n=0 γn t + ηy includes both a cubic polynomial in time and year dummies to control flexibly for time trends; it absorbs low-frequency relations or secular trends that could confound our estimates, such as the upward trend in seizures in Colombia and crackdowns in Mexico that started in 2006 with President Calderón. Finally, εt is an error term. Our standard errors are robust to heteroskedasticity, but we assume there is no serial correlation.22 We interpret β as the effect of supply reductions in Colombia on violence in Mexico, which, according to our model, should be positive if demand is inelastic. The effect of cocaine seizures is identified from its high-frequency variation, which we treat as an exogenous shock to supply, Qs . The variation is considerable. The standard deviation in monthly seizures is 8 metric tons, which is high relative to the 40 metric tons of cocaine that Colombia produced per month during the period we analyze. Thus, changes in Colombian seizures produce significant shortages of cocaine in downstream markets. We believe that this monthly variation is determined mostly by chance—for instance, by whether an interdiction operation is successful or unsuccessful, and by factors like politics or funding in Colombia that are exogenous to Mexico. We maintain this assumption throughout and discuss some potential challenges to our interpretation after presenting our results. We present estimates of equation (3) in Table 2. Each panel has a different dependent variable. In columns 1 to 3, we focus on the homicide rate in all Mexican municipalities. In columns 4 to 6, we focus on the homicide rate in municipalities in quintiles 1 and 2 of proximity to the US, and in columns 7 to 9 we focus on the homicide rate in municipalities in the first quintile of proximity to the US. 22 The Online Appendix shows that there is no serial correlation in the high frequency variation in homicides or seizures and that both series are stationary. This justifies the use of standard inference and allows us to abstract from more complicated dynamics in our models. Our estimates in columns 1, 4, and 7 of the top panel suggest that a 10% increase in monthly seizures in Colombia raises homicides in Mexico by 0.59% (standard error=0.27%), by 0.82% in municipalities in quintiles 1 and 2 (s.e=0.33%), and by 1.19% in municipalities in the first quintile (s.e=0.33%). In the remaining panels we document a larger increase in drug-related homicides, which is driven by assassinations of cartel members and confrontation deaths (though the effect on confrontation deaths is not precisely estimated). We do not find any effect on assassinations of government officials (the effect is small but imprecise), which we view as supportive of the emphasis on inter-cartel violence in our model. In columns 2, 5, and 8 we include two monthly controls for all of Mexico, the unemployment rate and the Índice Global de Actividad Económica (IGAE), which is a measure of economic activity. Although these controls might be endogenous, and we prefer our estimates without them, it is reassuring to see that our estimates do not change with their inclusion. Finally, in columns 3, 6, and 9 we control for Mexican weather seasons, which include a dummy for months in the rainy season and dummies for months in the hurricane season. Although our estimates are less precisely estimated in these cases, the point estimates remain roughly unchanged. These exercises suggest that the estimated effects are not driven by the Mexican business cycle or by seasonality.23 We also illustrate the response of violence to supply shocks using event-study figures. We estimate equation (3) including six lags and six leads of ln St . Figure 2 plots the estimated coefficients for the homicide rate in all of Mexico, for the homicide rate in the first two quintiles of proximity to the US, and for the first quintile. The homicide rate spikes during a cocaine shortage and then reverts back to its pre-shock level with some small persistence 23 In Table A2 of the Online Appendix, we explore variants of equation (3) with lags of the homicide rate as explanatory variables. These lags are not significant and they do not affect the estimate of our coefficient of interest. We then explore the timing of the effects by adding lags and leads of seizures in Colombia. Contemporary seizures in Colombia have the strongest effect on violence. The effect vanishes quickly, lasting only an additional month in some specifications. We do not find any significant effect of future Cocaine seizures in Colombia on Mexican violence. (although individual coefficients are less precisely estimated in this exercise). As in Table 2, this pattern becomes more stark in the North of Mexico. To illustrate the quantitative implications of our estimates, we compute the effect of the sharp reduction in the Colombian supply of cocaine from 2006 to 2009 on violence in Mexico. Our calculation requires two assumptions. First, we assume that short-run shocks and persistent supply shocks have the same effect on violence.24 Second, we assume that seizures in Colombia only affect violence through their effect on the net Colombian supply of drugs—an exclusion restriction. Both assumptions imply that β is the effect of a persistent reduction in the supply of Colombian cocaine by 1%. Between 2006 and 2009, the cocaine supply from Colombia fell by 95.6 log points (from 520 mt to 200 mt). Most of the reduction was triggered by an increase in seizures, although coca cultivation and potential cocaine production also fell. We use the results in columns 1, 4 and 7 in the top panel of Table 2, as well as the estimates in columns 4, 8 and 12 of Table A2 in the Online Appendix, where we control for the dynamics of the homicide rate. Our estimates imply that the reduction in cocaine supply increased the homicide rate by 5.6-8.7 log points in all of Mexico, which corresponds to about 10%-15% of the increase experienced during this period. They also imply that scarcity in Colombia increased the homicide rate in quintiles 1 and 2 of Mexico by 7.8-10.1 log points, and by 11.37-15.4 log points in the first quintile of proximity to the US, which also accounts for 10% to 15% of the increase in violence during this period in both regions. 24 This is only a suggestive calculation, since in practice short-run and permanent shocks may have different effects. On the one hand, the possibility of storage implies larger effects from persistent shocks. On the other hand, the possibility of substituting cocaine from other source countries implies smaller effects from persistent shocks. Likewise, our theoretical discussion highlights the fact that temporary shocks have a larger effect than permanent ones. This is so because a temporal shortage that is soon reversed implies large revenues today and low revenues tomorrow, which makes future punishment less severe. In contrast, permanent shocks also raise the deterrence power of future punishment, and so they raise violence less.