Trade Costs, Conflicts, and Defense Spending

Alexander Tarasov; Roman Zakharenko

This paper provides a formal model of endogenous border formation and choice of defense spending in a world with international conflict. We examine both the case of democratic governments and of dictatorships. The model is consistent with three observations. First, breakup of countries should follow a reduction in the likelihood of international conflicts. Second, the number of regional conflicts between smaller countries may increase as a result of the breakup of larger countries.

The analysis of the effects of international trade on conflict has almost exclusively focused on the volume of trade flows, mostly disregarding any consideration related to the content of trade flows. This paper empirically explores the determinants of bilateral conflict taking into account several measures describing relevant dimensions of trade flows at the product level, as the degree of complementarity between the two countries, the extent of substitutability of the partner as a destination market and an imports provider, and the level of rivalry between the members of each pair as exporters and importers in third markets. Proposing an innovative instrument to address the endogeneity of trade variables, I estimate a directed model which takes advantage of a continuous event-based measure of interstate conflict. Results show that the three considered dimensions of the patterns of trade are relevant to explain interstate conflict. Results also vary when explaining the frequency or the hostility level of conflict events. According to the results, liberal and realist approaches emerge as coexisting explanations of the consequences of trade on political relations between countries.

Trade Costs, Conflicts, and Defense Spending∗ Michael Seitz Alexander Tarasov Roman Zakharenko Boston Consulting Group University of Munich Higher School of Economics January 29, 2014 Abstract This paper develops a quantitative model of trade, military conflicts, and defense spending. Lowering trade costs between two countries reduces probability of an armed conflict between them, causing both to cut defense spending. This in turn causes a domino effect on defense spending by other countries. As a result, both countries and the rest of the world are better off. We estimate the model using data on trade, conflicts, and military spending. We find that, after reduction of costs of trade between a pair of hostile countries, the welfare effect of worldwide defense spending cuts is comparable in magnitude to the direct welfare gains from trade. Keywords: general equilibrium, gains from trade, defense spending JEL Codes: C5, C6, F13, F51, H56 1 Introduction The traditional trade literature formulates a number of channels through which a country can gain from trade: the comparative advantage and love for variety effects, the redistribution of production factors towards more productive firms, the lower markups set by firms are some of them. In this paper, we quantitatively explore a new potential source of gains from trade. Recent empirical studies showed that a rise in trade between two countries reduces the We are thankful to Francesco Caselli, Sergey Izmalkov, Dalia Marin, Thierry Mayer, Igor Muraviev, Dmitry Pervushin for useful discussions. Denis Deryushkin and Sergei Motin provided helpful research assistance. Alexander Tarasov gratefully acknowledges financial support from the Deutsche Forschungsgemeinschaft through SFB/TR 15. ∗ 1 Figure 1: Military Spending and Trade Sources: WorldBank Development Indicators, National Material Capabilities dataset. probability of an armed conflict between them (e.g. Martin et al. (2008); Hegre et al. (2010)). Our reasoning is the following: if trade brings about peace, it should also bring about defense spending cuts across the world, which in turn will bring about even more peace. Figure 1 testifies in favor of this hypothesis: trade volumes have been increasing over the last few decades, whereas the size of defense spending (proxied by share of military personnel in the population) has been decreasing during the same period. While in the modern world military conflicts are quite rare, countries’ defense spending are still substantial and, therefore, this additional effect of trade openness may have considerable welfare implications. In particular, we address the following questions in the paper: what is the magnitude of welfare gains due to reduced conflict probability and defense spending cuts? Is it comparable to the “traditional” gains from trade? The quantitative model we develop is based on the following key assumptions. First, bilateral trade volumes are reduced in case of a military conflict with a certain country, leading to welfare losses. As a result, countries are less likely to be engaged in a conflict when they are connected to each other by stronger trade links. Second, the probability of having a conflict with each of potential opponents affects a country’s decision on how much to spend on defense. Indeed, for years 1993-2001 (our main dataset), a cross-country measure of involvement into conflicts has a 32% correlation with the average (over the years) share 2 of the defense spending in the GDP, with a high level of statistical significance. Finally, the size of defense spending in turn has an impact on the probability of conflict with all other countries. The model thus features both causal links between trade and conflicts (conflicts reduce trade, trade prevents conflicts), as well as both causal links between conflicts and defense spending (anticipation of conflicts causes more defense spending, more defense spending makes conflicts more likely). An important corollary of the above assumptions is that increased trade between a pair of countries will have a multiplicative effect on the global defense spending: it will cut the defense spending of not only the two trading partners, but also of other countries. Furthermore, the reduction of defense spending of other countries will have a further downward effect on the defense spending of the two trading partners. For example, increased trade between Russia and the United States should induce a reduction of defense spending of not only Russia and the United States, but also the defense spending of other potential opponents of Russia (e.g. China or Eastern European countries) and the United States (e.g. China or Venezuela); the reduction of China’s defense spending would reduce the defense spending of other China’s potential opponents (e.g. India), as well as further decrease of defense spending in Russia and the United States. To assess the welfare gains of diminished conflict frequency and defense spending cuts, we need first to clarify the theoretical foundations of such welfare gains. Indeed, in a “perfect” world of full information, zero transaction costs and fully rational players, all players make decisions in such a way that aggregate welfare is maximized. Then, according to the envelope theorem, the overall marginal welfare effects of diminished trade costs are equal to the direct marginal effect, while the indirect marginal effects (via fewer conflicts and less defense spending) are zero. At the same time, there is a widespread belief that in a decentralized equilibrium, there are too many conflicts and too much defense spending, and an effort to reduce both is desirable; such belief is at the core of peace studies.1 To formalize this idea, one has to relax one or more “perfect world” assumptions. Martin et al. (2008), for example, take the asymmetric information approach: every pair of countries bargains over a joint “peace surplus” where the outside options (i.e. welfare in case of disagreement and conflict escalation) are private information. In such a setting, suboptimal conflicts may take place. In this paper, we take another, perhaps more extreme, approach (primarily due to mathematical tractability concerns). We assume prohibitively high transaction costs of negotiation 1 For example, the Peace Research Institute Oslo, a half-century-old think tank, defines its purpose as “to engage in research concerning the conditions for peaceful relations between nations, groups and individuals.” 3 between governments and, as a result, impossibility of welfare transfers among countries. In this setting, country A may attack country B even if the welfare gains of the former are smaller than the welfare losses of the latter, rendering such an attack socially suboptimal. When strengthened trade links cause A to become more peaceful towards B, then, while the additional welfare effect of reduced hostility is zero for A due to the envelope theorem, such effect is positive for B. Moreover, such reduced hostility causes global defense spending cuts, which further improves the welfare of all countries including A. To assess the magnitude of these additional gains from trade, we estimate the model applying what we call the constrained maximum likelihood (i.e. ML with constraints on the parameter space) estimation, and perform a counterfactual analysis. The empirical identification of the interdependencies between trade, military conflicts, and defense spending comes from the structural and functional assumptions in the model. Focusing on some of the most hostile country pairs, we quantitatively examine how unilateral trade liberalization between the two countries affects the exporter, importer, and world welfare. We find that in all experiments both the exporter and the importer gain from unilateral trade liberalization. Moreover, since more peaceful relations between the countries launch a worldwide wave of defense spending cuts, the rest of the world gains as well.2 For instance, a reduction in the cost of exporting from South Korea to North Korea that leads to one dollar rise in the value of exports raises the welfare of North Korea by $0.1678 (in terms of the compensating variation of income), the welfare of South Korea by $0.0551. At the same time, the global welfare gains are $0.7361, including a gain of the United States of $0.2711, and a gain of Japan of $0.1209. These numbers suggest that the world gains from trade due to defense spending cuts can be substantial, especially when the two countries that increase trade have a history of hostility. To the best of our knowledge, this paper is the first one that studies the interplay between trade, conflicts, and defense spending. At the same time, there exist studies of an interplay of any two of these three. Substantial research, by both economists and political scientists, has been devoted to the analysis of interrelationship between trade and military conflicts. Polachek (1980) argues that mutual dependence between trading partners reduces the probability of a conflict between them. He finds that this hypothesis is consistent with data. Gowa and Mansfield (1993) show in a game-theoretic model that free trade more likely takes place within, rather than across, political-military coalitions. They provide empirical 2 In the paper, we assume away the general equilibrium effects of trade liberalization that work through the adjustment of the cost of labor across countries. As a result, the only effects of unilateral trade liberalization on the rest of the world are due to reduced hostility and associated defense spending cuts. 4 evidence supporting this prediction. A similar hypothesis is tested in Morrow et al. (1998), who find that the effect of alliances on trade volumes is uncertain. Glick and Taylor (2010) study the historic data on trade and conflicts to quantify the impact of conflicts on trade flows and on welfare. They find that conflicts reduce trade of not only the belligerent nations, but also of neutral ones, and that the impact of a conflict on trade is long-lasting and may extend well beyond the conflict itself. Martin et al. (2012) offer a theory and empirics of regional trade agreements (RTA), in which they argue that the primary benefits of RTA’s include not only trade creation but also conflict prevention, and for this reason RTA’s are more likely to be created in places that are more inclined to both trade and conflict, e.g. Europe. In a closely related paper, Vicard (2012) differentiates RTA’s by their “depth” (i.e. extent to which the markets were integrated) and concludes that only deep enough agreements prevent conflicts. Polachek and Seiglie (2007) overviews a general theory and empirical testing methodology of the trade-conflict nexus. There are a number of theoretical studies that consider globalization as an important determinant of the likelihood of conflict. Skaperdas and Syropoulos (2001) explore the impact of trade openness on arming through the terms of trade effects. Alesina and Spolaore (2005) examine the link between the size distribution of countries and international conflict. Alesina and Spolaore (2006) develop a framework with endogenous border formation, defense spending, and international conflict. In particular, they explore how break-up of countries affects the likelihood of international and regional conflicts. Acemoglu et al. (2012) construct a dynamic theory of resource wars and study the interaction between the scarcity of resources and the incentives for war. Our paper makes a step further in this direction of research by quantifying the role of trade in international conflict. The political science literature also explores a link between conflicts and defense spending. One of the first papers examining the impact of defense spending on military conflicts is Wallace (1979), who finds a strong positive correlation between arms races and the probability of escalating a dispute to war. In particular, he reports that 23 out of 28 disputes where the countries were involved in an arms race escalated to war. Controlling for other factors effecting dispute escalation (such as history of disputes, relative defense burdens, etc.), Sample (1998) also finds a strong positive correlation between arms races and the likelihood of dispute escalation. Kamlet and Mowery (1987) analyze the influence of economic and political factors on the budgetary priorities of the U.S. in 1955-81. They find that the level of armed conflict the U.S. was involved in positively affects the level of the budget spending on defense. Collier and Hoeffler (2002) find that the presence of external threat and/or 5 international war has a strong positive impact on country’s military expenditure. Aizenman and Glick (2006) also report a positive cross-country correlation between military spending (as a share of GDP) and external military threat. To our knowledge, the only study that links trade to defense spending is Acemoglu and Yared (2010) who argue that the militarist sentiments, proxied by defense spending, lead to greater isolation of countries and thereby reduce international trade flows. In their paper, militarism and defense spending are viewed as exogenously driven by public preference. Our study makes the exact opposite assumptions: defense spending is a fully rational choice made for a purpose of welfare improvement in case of a conflict; defense spending is cut due to reduced conflict probability, which is in turn reduced by declining costs of trade. The rest of the paper is organized as follows. Section 2 develops a theoretical framework. Section 3 describes the estimation procedure. In Section 4, the description of the data is provided. Section 5 discusses the results and performs counterfactual experiments. Section 6 concludes. 2 The Model We consider a world consisting of N asymmetric countries. Each country has two types of bilateral relations with the other countries: trade relations and political relations, each of which affects the utility of a representative consumer. Political relations, which exist at the government level, are comprised of deciding whether to be in peace or in conflict with every other country. We eliminate the possibility of conflict between more than two countries due to theoretical complexity and low empirical frequency of such conflicts in the past sixty years. A military conflict occurs if at least one of the parties decides to initiate a conflict (to “attack”), which has an impact on countries’ incomes and on the bilateral trade costs between the countries. The impact on country’s income is in turn affected by the country’s defense spending, which is funded through a tax on consumers. The model of trade is based on Armington assumptions that there is perfect competition between producers of a certain country, products are distinguished by a country of origin, and individuals consume both domestic and imported varieties. Trade costs are generally exogenous, but are increased by a factor following an endogenous decision to initiate a military conflict. The timing of events in the model is as follows. First, each country makes a defense spending and a corresponding tax level decision. Second, for every country i and for every 6 potential opponent j 6= i, a pair of stochastic shocks, denoted by ǫij and λij , is realized. The income shock, ǫij , affects the loss of income of i in case of conflict with j, while the trade shock, λij , affects the cost of importing from j in case of the same conflict. Third, each country decides whether to attack each of its potential opponents, and peace/conflict outcomes are realized. Finally, production, trade, and consumption take place. We describe the details of the model in the reverse chronological order. 2.1 Consumption There are Ni different varieties produced in country i. Following the Armington assumptions, Ni is assumed to be exogenous. Each variety is produced in a perfectly competitive environment, resulting in the producer price of a variety in country i being equal to the marginal cost of production in that country. We assume that the marginal cost of production in country i is equal to the wage rate divided by the country-specific labor productivity yi . For analytical tractability of the model, we ignore the trade general equilibrium effects, which require exports to be equal to imports via wage adjustment. Instead, we assume that in each country i the wage rate is equal to the labor productivity yi , thus the domestic price is equal to unity. An alternative way is to introduce a homogenous freely traded good, the presence of which will assume away the general equilibrium effects on wages and, thereby, equalize the marginal cost of production across countries. Transporting a good from country j to country i incurs iceberg transport costs, τij ≥ 1, that are exogenous in general, but rise in case of a military conflict between i and j, as detailed in Section 2.2.1. Thus, the price of a domestically produced variety in country i is equal τii ≡ 1, while the price of a variety imported from j is equal to τij . As varieties produced in each country are symmetric, the utility of a representative consumer in country i can be written as follows: s ! s−1 X s−1 , (1) Nj qijs ui = C i j∈N where qij is consumption of a variety produced in j, s > 1 is the elasticity of substitution between varieties, and Ci is a country-specific constant, detailed in Section 3.1. The budget constraint is then X Nj qij τij ≤ ei , (2) j∈N where ei is the disposable income of a representative consumer in country i. 7 The utility maximization problem implies that the quantity demanded of each variety is qij = P The indirect utility is then ui = C i e i X j∈N Nj τij1−s ei τij−s k∈N Nk τik1−s 1 ! s−1 (3) . = Ci e i X Tij j∈N 1 ! s−1 , (4) where Tij ≡ Nj τij1−s is labeled as the trade propensity. Finally, the trade volumes from j to i are given by Xij ≡ Li Nj qij τij = Ei P Tij k∈N Tik , (5) where Li is the population size of country i and Ei ≡ ei Li is the aggregate disposable income in country i. Assuming that the number of varieties Nj and thus trade propensity Tij are proportional to the exporter economic size, the above model of trade is the simplest one that results in trade flows as predicted by the gravity equation. 2.2 Effects of Conflicts In this section, we model the effects of military conflicts on trade flows and countries’ disposable incomes. 2.2.1 Effect on Trade We assume that a military conflict between i and j increases trade costs between the two countries and, thereby, reduces the trade propensity, Tij , by a stochastic factor. For every pair of countries i 6= j, we define wi:j as the conflict status between i and j, endogenous to the model.3 Specifically, wi:j is equal to unity in case of conflict between country i and j and zero in case of peace. We then define the trade propensity between i and j, endogenous to the model, in the 3 Throughout the paper, colon is inserted between subindex elements if the corresponding link is undirected. For example, wi:j ≡ wj:i . Colon is not inserted for directed links, e.g. generically Tij 6= Tji . 8 following way: Tij = T (Tij0 ; wi:j ; β; λij ) ≡ Tij0 wi:j (λλ̄ij − 1) + 1 ∈ [0, Tij0 ], (6) where Tij0 is the trade propensity in case of peace, λij ∈ [0, 1] is a realization of the trade shock drawn from the standard uniform distribution, λ̄ > 0 is a parameter to be estimated, and β is the vector of the parameters in the model to be estimated, to be specified throughout section 2, such that λ̄ ∈ β.4 As can be seen, a military conflict between i and j reduces the trade propensity by a country-pair specific factor that depends on λ̄: a higher λ̄ means a greater reduction of trade due to conflict. To the best of our knowledge, we are the first ones to assume that the effect of conflict on trade is stochastic. Other studies (see, for example, Martin et al. (2008)) assume deterministic effects where trade is reduced by a constant factor. The potential heterogeneity of the effects of conflicts on trade leads to the selection problem: a conflict is more likely to happen when its effect on trade is small. Thus, if such heterogeneity is not accounted for, the model will underestimate the effects of conflicts on trade. The above specification implies that λ̄ governs both the mean and the variance of trade shocks. Ideally, we would like to estimate them separately. However, since we do not observe how much trade there would have been without a conflict, identification of both mean and variance seems complicated if not impossible. Therefore, the stochastic nature of the effect of conflict on trade is assumed rather than tested. 2.2.2 Effect on Income In the paper, we assume that the disposable income, ei , depends on conflict statuses as well. Specifically, we define the disposable income in the following way: ei = ei (wi , mi ; β; ǫi ) ≡ yi 1 − mi + σ i X j6=i ! wi:j Yj (θ log(mi + α) + zij γ − ǫij ) , (7) where wi ≡ {wi:j }j6=i is the vector of conflict statuses with all opponents of i; mi is i’s defense spending as a share of its GDP; ǫi ≡ {ǫij }j6=i is the vector of income shocks drawn from the standard logistic distribution.5 Yi ≡ yi Li is the GDP in the absence of conflicts, zij 4 In the subsequent analysis, we divide all function arguments into four categories: peace-level trade propensities, country decisions (defense spending and conflicts), parameters of the model to be estimated, and realizations of shocks. The arguments of different types are separated by semicolon. ex 5 By the “standard” logistic distribution, we imply a distribution with c.d.f. equal to F (x) = 1+e x. 9 is the vector of control variables affecting conflict incidence (such as the history of conflicts between the two countries, geographic and cultural distance between them, etc.). Finally, {θ, α, γ, σi }i=1..N ∈ β are the parameters to be estimated such that θ > 0, α > 0, σi > 0. Notice that if wi:j is equal to zero for all j 6= i (that is, country i does not have conflicts), the disposable income is yi (1 − mi ), i.e. productivity minus the defense tax. The intuition behind (7) is as follows. Parameters {σi }i=1..N stand for the country-specific effects that govern the magnitude of income gains or losses of a country in case of conflicts. The assumption that countries choose their defense spending optimally allows us to identify {σi }i=1..N from the data. The income gains or losses are assumed to be proportional to the opponent’s size, Yj : for Cuba, a conflict with the United States may be more costly than a conflict with the Bahamas.6 The logarithmic dependence of income on defense spending, mi , is assumed to make ei concave enough with respect to mi and to avoid the problem of multiple local utility maxima. The parameter α is introduced to justify zero defense spending by some countries; it also affects the curvature of income with respect to defense spending. The logistic distribution of income shocks is chosen because (i) since income gains from conflicts range from positive values (otherwise no one would initiate a conflict) to negative ones (because most of the time, conflicts do not happen), it is desirable to have the whole real line as a support of income shocks and (ii) the logistic distribution has an analytically computable c.d.f. and the integral of c.d.f., both of which are used in the model. All shocks in the model are assumed to be independent from each other. Identification of correlation between shocks for the same dyad (ǫij , ǫji , λij , λji ) is impossible due to data availability; the potential correlation between shocks on different dyads sharing the same country, e.g. between i : j and i : k, is captured, at least partially, by country fixed effects σi . 2.3 Transformation of Utility With all above definitions, we can rewrite the consumer utility (4) as a function of all model ingredients: ui = u(Ti ; wi , mi ; β; ǫi , λi ) = Ci e(wi , mi ; β; ǫi ) X j∈N 6 T (Tij0 ; wi:j ; β; λij ) 1 ! s−1 , (8) It is also natural to assume that welfare gains or losses of a country depend on defense spending of the opponent. However, since defense spending is endogenous in the model, this assumption would greatly complicate the analysis. The opponent’s economic size, Yj , among other things, serves as a proxy for the opponent’s military potential. 10 where Ti ≡ {Tij0 }j∈N and λi ≡ {λij }j6=i . Country i makes a decision on its conflict status with its opponent j, wij , by comparing its utility from peace and from conflict with j. A conflict happens if at least one party finds it optimal to initiate a conflict. Typically, it is the case that at most one member of a dyad is willing to initiate a conflict. In other words, wij is not necessarily equal to wji , implying that wi:j ≡ max{wij , wji }. From the expression for the consumer utility in (8) it can be seen that the decision to initiate a conflict, wij , is interdependent with conflict statuses on other dyads {wi:k }k6=i,j . This in turn makes the analysis of the model very complex. To make the model more tractable, we transform (“simplify” ) the utility in two steps, believing that the results will not be strongly affected. First, we linearize the utility with respect to the conflict status vector wi , as follows: ui ≈ u(Ti ; 0, mi ) + " X = Ci yi Si (Ti ) + j6=i X j6=i " wi:j [u(Ti ; 1j , mi ; β; ǫij , λij ) − u(Ti ; 0, mi )] − Ci mi yi Si (Ti ) + # wi:j {(1 + σi Yj (θ log(mi + α) + zij γ − ǫij )) Sij (Ti ; β; λij ) − Si (Ti )} X j6=i # wi:j {Sij (Ti ; β; λij ) − Si (Ti )} , (9) where 0 is a N × 1 vector of zeros, 1j is a vector of same size with unity corresponding 1 P 0 s−1 is the love-of-variety component to country j, and zeros elsewhere; Si (Ti ) ≡ T j ij 1 P 0 0 λ̄ s−1 of utility in case of total peace, while Sij (Ti ; β; λij ) ≡ is the same k6=j Tik + Tij λij component in case of conflict with j only. Note that if country i has no conflicts or has a conflict with only one other country,7 the linearized utility is equal to the original utility. In the second step, we simplify (9) further by assuming that the total cost of defense spending, as presented by the last line of (9), does not depend on the conflict incidence and is equal to −Ci mi yi Si (·). The defense spending mi is typically a small fraction of GDP, while conflict incidence wi:j is a rare occurrence, hence the interaction between the two seems to be a negligible issue. At the same time, such simplification greatly improves the analytical tractability of the model. 7 . . . which is true for the vast majority of country-year observations: no conflicts in 1038 out of 1629 country-year observations, one conflict in 379 cases, and more than one conflict in the remaining 212 cases. 11 Thus, we end up with the following “linearized” utility: uLi (Ti ; wi , mi ; β; ǫi , λi ) ≡ Ci yi (1 − mi )Si (Ti ) X + C i yi wi:j {(1 + σi Yj (θ log(mi + α) + zij γ − ǫij )) Sij (Ti ; β; λij ) − Si (Ti )} (10) j6=i 2.4 Attack Decisions In this section, we model attack decisions made by each country with respect to its potential opponents. Country i decides to initiate a conflict with j 6= i (in this case, wij is equal to one and, therefore, wi:j is equal to one as well) if it is better off from doing so. In other words, the utilities under having a conflict with j and without are compared. From (10), it is straightforward to see that the difference between the welfare in case of a conflict and in case of no conflict is given by (note that, due to linearization, all terms that are not relevant to the pair {i, j} are cancelled out) uLi (·; 1j , mi ; ·) − uLi (·; 0, mi ; ·) = Ci yi {(1 + σi Yj (θ log(mi + α) + zij γ − ǫij )) Sij (·) − Si (·)} . Thus, country i initiates a conflict if 8 (1 + σi Yj (θ log(mi + α) + zij γ − ǫij )) Sij (Ti ; β; λij ) ≥ Si (Ti ). As can be seen, the attack decision depends on the realizations of the income shock, ǫij , and the trade shock, λij . Specifically, the conflict takes place iff ǫij ≤ ǫ⋆ij , where ǫ⋆ij is the income shock cutoff given by ǫ⋆ij (Ti ; mi ; β; λij ) Si (Ti ) 1 ≡ θ log(mi + α) + zij γ + 1− . σ i Yj Sij (Ti ; β; λij ) (11) The income shock cutoff increases with country i’s defense spending. The greater is the defense spending, the higher the probability of initiating a conflict. The income shock cutoff decreases with the peace-level trade propensity with j, Tij0 , as the losses from a conflict with j are greater when there are more trade opportunities. It increases with the peace-level trade 8 From a game-theoretic standpoint, there exists another equilibrium: if country i is certain that it will be attacked by j (wji = 1 with probability one), its own decision wij does not affect anything, so i may choose to attack j with certainty, as well, regardless of the value of ǫij . This equilibrium however is not trembling-hand-proof: if j attacks i with probability slightly less than one, this equilibrium falls apart. We ignore this equilibrium throughout the analysis. We thank Sergei Izmalkov for raising the issue. 12 propensities with other countries, Tik0 . The idea behind is that if a country trades intensively with countries other than j, it is less dependent on trade with j and thus more likely to engage in a conflict. This relationship between trade with the rest of the world and conflict incidence has a strong empirical support (see Martin et al. (2008)). How does the income shock cutoff depend on the opponent’s size, Yj ? One can show P that, if j is a “small country” for i, i.e. Tij0 constitutes a small fraction of k Tik0 , then the term in square brackets in (11) is approximately proportional to Tij0 , i.e. can be presented as kTij0 + o(Tij0 ) for some k < 0. Assuming further that Tij0 is proportional to the exporter’s economic size Yj , we have that ǫ⋆ij (·) (marginally) does not depend on Yj , and the opponent size (marginally) does not matter for the decision to attack them. Note that ǫ⋆ij is a realization of a random variable, as it depends on the realization of λij . We can calculate the probability that i initiates a conflict with j as follows: Pij (Ti ; mi ; β) = Z F (ǫ⋆ij (Ti ; mi ; β; λ))dλ = Eλ F (ǫ⋆ij (Ti ; mi ; β; λ)), (12) λ∈[0,1] where F (·) is the c.d.f. of the income shock, and Eλ is the expectation over appropriate shocks, not to be confused with the aggregate disposable income Ei . Finally, the utility (10) can be rewritten as uLi (Ti ; wi , mi ; β; ǫi , λi ) = Ci yi (1 − mi )Si (Ti ) + X wi:j uLij (Ti ; mi ; β; ǫij , λij ), (13) j6=i where uLij (Ti ; mi ; β; ǫij , λij ) = Ci yi σi Yj Sij (Ti ; β; λij ) ǫ⋆ij (Ti ; mi ; β; λij ) − ǫij is the additional utility of i in case of conflict with j. 2.5 Defense Spending Decisions The government in each country chooses its military spending, mi , by maximizing the expected utility, before the realizations of ǫij and λij are known, and before conflict statuses are realized. The latter is equal to (irrelevant function arguments are dropped for brevity): uE i (T; m; β) = Ci yi (1 − mi )Si (Ti ) " Z X L + Pji (·)Eǫ,λ uij (·; ǫ, λ) + (1 − Pji (·)) j6=i 13 (14) 1 λ=0 Z ǫ⋆ij (·;λ) ǫ=−∞ # uLij (·; ǫ, λ)dǫdλ . Here T = {Ti }i=1..N and m = {mi }i=1..N . The first term in the second line of (14) describes the expected additional utility due to possible attack by j. The second term in the second line of (14) represents the expected utility due to possible attack of j by i. Thanks to the assumption that the mean of income shock ǫij is zero, we have the following: Eǫ,λ uLij (·; ǫ, λ) = Ci yi σi Yj Eλ Sij (·; λ)ǫ⋆ij (·; λ) , while Z 1 λ=0 R Z ǫ⋆ij (·;λ) ǫ=−∞ uLij (·; ǫ, λ)dǫdλ = Ci yi σi Yj Eλ Sij (·; λ)F F (ǫ⋆ij (·; λ)) , where F F (x) ≡ y∈(−∞,x) F (y)dy. To determine the optimal defense spending, the government in country i maximizes (14) ∂ǫ⋆ (Ti ;mi ;β;λi ) with respect to mi taking {mj }j6=i as given. Given the fact that ij ∂mi = miθ+α , the first derivative of (14) with respect to mi is given by ∂uE i (T; m; β) = Ci yi ∂mi ! P θ Y P (·)E S (·; λ) −Si (Ti ) + mσii+α λ ij j6=i j ji . P σi θ + mi +α j6=i Yj (1 − Pji (·))Eλ Sij (·; λ)F (ǫ⋆ij (·; λ)) (15) We arrive at the following set of the first-order conditions: ∂uE i (T; m; β) ∂mi ( = 0, mi > 0, ≤ 0, mi = 0. (16) Note that, while the marginal effectiveness of defense spending, miθ+α , decreases with mi , the probability of a conflict increases via a rise in ǫ⋆ij (Ti ; mi ; β; λij ) and, therefore, the concavity of the utility with respect to mi is not guaranteed. While estimating the parameters, we verify the second-order conditions of optimal defense spending for all countries. To prevent the problem of overidentification, we assume that for countries that spent zero on defense, the parameter σi is equal to a pre-defined value, low enough such that the constraints (16) for these countries hold. For better numerical results, we modify the remaining constraints on the parameter space (16) by dividing both sides by Ci yi S(Ti ) > 0, adding unity, and taking logs, to arrive at the 14 following, ∀i : mi > 0: Gi (β) ≡ log σi + log θ − log(mi + α) − log Si (·) (17) ! X + log Yj Pji (·; β)Eλ Sij (·; β; λ) + (1 − Pji (·; β))Eλ Sij (·; β; λ)F (ǫ⋆ij (·; β; λ)) = 0. j6=i By G(β) we denote the vector of all constraints described by (17). In the above analysis, we have formulated the probability of country j being attacked by country i and examined its dependence on the parameters in the model (see (12)). We have also established the conditions that determine the choice of military spending made by countries, (17). The proposed framework has several empirically plausible properties that are at the core of the analyzed model: • Countries spend more on defense if a conflict is more likely. This follows from the fact that the optimal mi solving (17) rises with Pji and with ǫ⋆ij . Conversely, a conflict is more likely if countries spend more on defense. That is, Pij rises with mi . • A rise in the defense spending of a certain country increases the probability of a conflict with its opponents and, therefore, increases the defense spending of all other countries. In other words, an arms race takes place. • Greater trade volumes result in larger welfare losses in case of conflict making military conflicts less likely. That is, Pij decreases with Tij . In the next section, we fit the model to the data on bilateral military conflicts, countries’ military spending, and bilateral trade volumes. 3 3.1 Fitting Model to Data Calibration We borrow the value for the elasticity of substitution, s, from the existing trade literature. Specifically, we choose s = 4, which is the mean value of the elasticities estimated in Broda and Weinstein (2006). The peace-level trade propensities Tij0 were calibrated as follows. First, from (5), it follows that T̂ij T̂ik = X̂ij , X̂ik where X̂ij is the empirically observed trade volume while T̂ij is the trade 15 propensity under the empirically observed conflict status. In other words, the ratio of trade propensities equals the ratio of trade volumes, ∀i, j, k. Second, the level of trade propensities is immaterial: multiplying T̂ij , ∀j by some importer-specific constant ki will not change any results in the model. For this reason, we choose the benchmark level of empirical trade propensities such that they are numerically equal to empirically observed trade volumes X̂ij , i.e. T̂ij = X̂ij , ∀i, j. Second, the relationship between T̂ij and the peace-level trade propensity Tij0 can be determined by (6). One problem with this relationship is that the magnitude of the trade shock λij is unknown, therefore there is no way to back out Tij0 precisely even if the parameter λ̄ is known. Our method of handling the problem is to find Tij0 by calculating the expected (with respect to λij ) trade propensity, conditional on empirically observed conflict status and on parameter values, and equating it to the empirically observed trade volumes: ! λ̄ ⋆ F (ǫ (T ; m ; β; λ))dλ λ 1 i i ij . + ŵji X̂ij = Eλ (T̂ij |ŵij , ŵji , β) = Tij0 1 − ŵij − ŵji + ŵij λR ⋆ F (ǫij (Ti ; mi ; β; λ))dλ λ̄ + 1 λ (18) λ̄ In (18), the ratio of the two integrals is the expectation of λij , conditional on λij being low 1 enough so that i chooses to attack j. The last term λ̄+1 is the unconditional expectation λ̄ of λij : the decision of j to strike i does not depend on λij . The system of equations (18), for every trade link ij, is used to find peace trade propensities Tij0 , ∀i, ∀j 6= i. Note that the parameter vector β is also unknown and is estimated as described in section 3.2; the estimate of β itself depends on the values of Tij0 . We re-estimate Tij0 and β iteratively until convergence. P To meet the budget-balancedness condition j Xij0 = Ei = (1 − mi )Yi , that the total expenditure on goods imported from all countries (including domestically produced goods) equals the total disposable income, we set the “self-trade” propensity equal to the difference between the disposable income and the sum of imports. By assumption, a country cannot have conflicts with itself, hence the peace-level self-trade propensity equals its empirical P counterpart: Tii0 = T̂ii = Êi − j6=i X̂ij . We proxy Êi by (1 − mi )Ŷi , which is the most simple (but not the most accurate, cf.(7) approximation. The productivity of a country Yi is set equal to its observed GDP Ŷi . For countries that actually had conflicts, such calibration has an obvious shortcoming, since throughout the model we assume that conflicts move the actual realization of income away from the peace-level productivity Yj . Thus, we sacrifice some realism for the sake of mathematical tractability of the model. R 16 Finally, the utility scale parameter Ci is chosen such Ci Si (Ti ) = 1. This way, the utility in case of peace among all countries is numerically equal to disposable income ei , thus a utility change due to changing model parameters is numerically equal to the compensating variation of income. For this reason, in the policy experiment section of this paper, all marginal changes in utility due to changing trade costs are nominated in dollars. 3.2 Estimation The set of parameters in the model we estimate is β = θ, α, λ̄, γ, {σi }i=1..N . We estimate the unknown parameters using the maximum likelihood estimator with constraints on the parameter space. In particular, we fit the predicted probabilities of conflict initiation (12) to actually observed conflict initiation status. In doing so, we account for a discrepancy between the theoretical model and the actually available data. Specifically, the theoretical model predicts that a pair of countries i and j decide independently whether or not to attack each other, meaning that it is possible that both may decide to attack (wij = wji = 1). In the data however, as described in detail in section 4, only one side may be a conflict initiator, thus ŵij + ŵji ≤ 1, where ŵij and ŵji are observed in the data conflict initiation statuses. To reconcile the discrepancy, we assume that, if both countries decide to attack, each of them becomes the “observed” attacker with probability 12 . With this assumption, the theoretical probabilities of observed outcomes are as follows (dropping some function arguments): Pr(ŵij = 0, ŵji = 0) = (1 − Pij (·; β))(1 − Pji (·; β)), 1 Pr(ŵij = 1, ŵji = 0) = Pij (·; β)(1 − Pji (·; β)) + Pij (·; β)Pji (·; β), 2 1 Pr(ŵij = 0, ŵji = 1) = (1 − Pij (·; β))Pji (·; β) + Pij (·; β)Pji (·; β). 2 The corresponding loglikelihood function can be then written as follows: L(β) = XX i j6=i 1 ŵij log(Pij (·; β)) + ŵji log 1 − Pij (·; β) + (1 − ŵij − ŵji ) log(1 − Pij (·; β)) . 2 (19) Since in the model country’s military spending is endogenous and chosen to maximize its expected utility function, we use the conditions implied by (17), for countries with positive defense spending, as restrictions on the parameter space in the estimation procedure. In this 17 case, the estimate of β is β̂ = arg max L(β̃) β̃ subject to (17) and to zero lower bounds on θ, α, and σi . Appendix A details the computational algorithm and the asymptotic variance of the estimate. The identification of θ and α comes from the relationship between conflict incidence and military spending in the data. Parameter λ̄ is identified from the relationship between military conflict incidence and trade volumes found in the data (see, for instance, Martin et al. (2008)). Finally, {σi } , ∀i : mi > 0 can be identified due to the presence of the constraints in the estimation procedure. Note also that the above estimation strategy allows us to disentangle all the interdependencies between trade volumes, military conflicts, and defense spending. Indeed, the structure of the model implies that the probability that i initiates a conflict with j, Pij (Ti ; mi ; β), depends on the exogenous controls, the parameters in the model, the exogenous trade propensities in case of peace, and the endogenous defense spending mi . That is, the only source of endogeneity is the presence of mi . However, from (17), mi can be implicitly expressed in terms of the exogenous variables (and parameters) and the probabilities of initiating a conflict. This in turn allows us to control for endogeneity (arising from the endogenous nature of conflicts and defense spending) in the estimation procedure. 3.3 Additional Adjustments The availability of data suggests some adjustments to the theoretical model and to the estimation method described above. First, given the availability of data on multiple years, we use them all while estimating the parameters in the model. We treat observations from different years as independent ones and ignore potential autocorrelation of shocks for the following reasons: (i) while ignored autocorrelation leads to inconsistent short-term predictions, it does not compromise longterm effects that are our primary goal, (ii) autocorrelation of shocks would complicate the model beyond reason, especially given the discrete nature of the dependent variable. Second, given the availability of data on conflict dates, we can expand the support of ŵij from {0, 1} (conflict, no conflict) to the entire unit interval (fraction of the year with active conflict). Intuitively, a conflict that lasted one day has a smaller effect on trade than a conflict that lasted the entire year. Empirically, there is indeed a strong correlation between conflict length and trade (for those observations with wi:j > 0). While such adjustment 18 does not require any changes in the estimation procedure, it does require some theoretical justification. To justify intermediate values of ŵij , we assume that shocks in the model are realized every day rather than once per year, and every country i updates every day its decision of whether to be in conflict with every other country j. To make sure that the first-order conditions of optimal defense (16) hold for any country in any given year, we allowed the parameters σi to vary across years, effectively making them country-year fixed effects. Most expectations in the model over λ are not computable analytically and had to be approximated numerically. To do so, we define a grid of points for λ, such that the change of c.d.f. between any two consecutive points is the same, and approximate the integral by the average value of the integrand across the grid points. The number of grid points was driven by computer memory constraints and set equal to 25. 4 Data In this section, we describe the data we use to estimate the model. We have dropped several countries due to data availability concerns; the full list is provided in Appendix B.1. The remaining sample consists of 181 countries, observed in 9 years as detailed below. 4.1 Conflicts Our primary source of data is the dyadic Militarized Interstate Dispute dataset, version 3.10 (Ghosen et al., 2004). The dataset provides the list of all “militarized interstate disputes”: i.e., “conflicts in which one or more states threaten, display, or use force against one or more other states” for years 1993-2001. For each conflict, the dataset includes the list of all country pairs which had a hostility between each other. For every such country pair, we use information on the start and end dates of the hostility, the level of hostility on a scale from 1 to 5, and a flag for the initiator of the conflict. Although data on earlier conflicts (from 1951 onward) is also available, we do not use it because of lower quality of the data: unlike MID v3.10, it does not provide information for every possible pair of countries for conflicts with multiple participants. Additionally, the pre-1993 world was characterized by quite a different political map; some auxiliary data (e.g. GDP for Socialist countries) is not easily available for those years. Nevertheless, the data on early conflicts is used as a control variable as described in section 4.5.1. 19 Following Martin et al. (2008), by a “conflict” we consider only conflicts with hostility level of 3 (Display of Force) and above (Use of Force, War) – overall, 483 out of 512 conflict observations meet this criterion. The empirical conflict status ŵij is computed as fraction of the year that i and j were in a conflict, such that i was the conflict originator (in the data, one and only one member of a dyad may be an originator). 4.2 GDP The gross domestic product is measured in current U.S. dollars. The data is from the World Development Indicators of the World Bank. The missing observations were filled by data from IMF’s World Economic Outlook. For North Korea which had no GDP data in any year, we employed the 1991 estimate of 22.9 billion USD, given in the Wikipedia article “Economy of North Korea.” The observations that were still missing were replaced by values predicted by the regression log GDPit Lit = δ0 + δ1 log PECit Lit + δi + δt + νit , where PECit is Primary Energy Consumption as reported by National Material Capabilities dataset, version 4.0 (Singer et al., 1972); δi is the country fixed effect, δt is the year fixed effect, νit is the residual, and year t ranges from 1981 to 2007. 4.3 Defense Spending The primary source of information on defense spending is Stockholm International Peace Research Institute (SIPRI), which reports such spending as a share of GDP for a wide range of countries and years. The missing data was filled by the calculated ratio of defense expenditure to GDP, where the former is taken from above mentioned National Material Capabilities dataset. The correlation between the two sources is approximately 66%. For some country-year observations, e.g. Cuba in 1993, defense spending observations were missing. Out of 181 × 9 = 1629 country-year observations, 45 had such missing defense data. We illustrate the handling of such observations by continuing the Cuba example. To calculate the probability that Cuba attacks other countries in 1993, zero defense spending was assumed; such probability however was only used to fit the constraint (16) for countries other than Cuba; it was excluded from the loglikelihood function (19). As mentioned earlier, we also do not verity the first-order condition (16) for countries 20 that spend zero on defense (as well as for countries with missing defense data), and a corresponding σi is assumed equal to a default value. In all, we have 1490 out of 1629 country-year observations with positive defense spending, each accompanied by an unknown value of σi and by a constraint (17). 4.4 Trade The primary source of trade data is Bilateral Trade dataset, version 3.0, by Barbieri and Pollins (2009). The missing data was filled using Feenstra et al. (2005), with correlation between the two sources exceeding 99%. While calculating the attack probability, the missing observations were replaced by zeros. However, if a trade data X̂ij was missing, the corresponding attack probability Pij was dropped from the loglikelihood function (19). In all, 46280 country-country-year observations out of 181 × 180 × 9 = 293220 had missing data on trade. After excluding missing trade and defense data, 239528 country-country-year observations out of 293220 were included into the loglikelihood function. 4.5 Control Variables Besides the constant, there are four sets of control variables: conflict history, contiguity, other distance variables, time trend. Below we describe the first three. 4.5.1 Conflict History A major predictor of conflicts between a pair of countries is preceding history of conflicts between them. The history of conflicts between countries i and j is measured as the total length of conflicts of hostility levels 3-5 between years 1951 and 1992, as reported by Militarized Interstate Dispute dataset, version 2.0 (Zeev, 2005). Since the political map of the world has considerably changed in the early 1990s, a number of adjustments had to be made. Conflicts between countries that later merged into one (e.g. between Federal Republic of Germany and German Democratic Republic) were dropped. Conflicts between such countries and third countries were added to the history of conflicts of their successors (e.g. a conflict between GDR and USA contributed to a history of conflicts between modern Germany and USA). For countries that split into several countries, their conflict history was carried over to the main successor country (e.g. a conflict between USSR and USA contributed to a history of conflicts between modern Russia and USA). 21 To account for possible nonlinear effects of past conflicts on modern conflicts, we add conflict history squared to the set of control variables. 4.5.2 Contiguity The data on geographic contiguity in taken from Direct Contiguity dataset, version 3.10 (Stinnett et al., 2002). The dataset identifies five types of contiguity, from most proximate type-1 (countries separated by a land or river border) to most distant type-5 (countries separated by 150-400 miles of water); we include a dummy for each of the five contiguity types. 4.5.3 Other Distance Variables An additional source of information on distance, both geographic and cultural, is the GeoDist database (Mayer and Zignago, 2011). We use the following variables from the dataset: log of (population-weighted) geographic distance between countries, a dummy for common official language, a dummy for colonial link ever between the two countries, and a dummy for a pair of countries having a common colonizer after 1945. 5 Results In this section, we present the results of the estimation procedure and perform counterfactual analysis. 5.1 The Parameter Estimates The parameter estimates are reported in Table 1. Note that we have estimated the values of logs of θ, α, and σi , because these parameters must be positive. The following control variables are significant: history of conflicts and its square, contiguity of types 1 and 2, distance between countries, linguistic and colonial links. Almost all significant parameters have the expected sign: the effect of conflict history on conflict incidence is positive and concave, the effect of geographic proximity is positive, the effect of linguistic proximity is negative, while almost all measures of past political ties have a positive effect on conflict incidence. The estimate of λ̄ implies an unconditional reduction of trade, following a conflict, by a 22 Table 1: Estimates of the model parameters Parameter log θ log α λ̄ constant conflict history history squared type-1 contiguity type-2 contiguity type-3 contiguity type-4 contiguity type-5 contiguity log weighted distance common official language colonial link ever common colonizer after 1945 time trend σi # of observations loglikelihood Estimate (Std.Err.) -0.5899 (0.1772) -6.6904 (1.4722) 1.9221 (1.9411) 1.0064 (0.8060) 0.4110 (0.0329) -0.0109 (0.0012) 1.7532 (0.2427) 1.9168 (0.4296) -0.3651 (1.8584) 0.8865 (0.4715) 0.0639 (0.5247) -0.8117 (0.0906) 0.2990 (0.1649) 1.3872 (0.2306) -0.7814 (0.2395) -0.0145 (0.0279) yes 239528 -1330.48 factor of 3.9 The estimate of α implies that the marginal effectiveness of defense spending by a country that spends nothing on defense is about 40 times higher than that of a country that spends on defense 5% of its GDP; such sharp concavity of welfare with respect to defense spending ensured the second-order condition of optimal defense spending for every country-year observation. Given the estimates, for any country we can calculate the probability of being attacked by any other country in any given year. The highest such probability was 65.35%, that Israel attacks Jordan in 1993. The lowest probability was 10−59 , that Tuvalu attacks Fiji in 2001. Overall, the predicted probabilities are consistent with subjective expectations. For example, for the United States in 2001, the greatest threats of attack came from Russia (2.47%),10 North Korea (1.77%), and China (0.67%). For Israel in 2001, the top threats were Jordan (53.88%), Lebanon (29.42%), and Egypt (8.55%).11 9 Given the fact that the prospect of trade reduction prevents some conflicts, the actual trade reduction is smaller than the unconditional reduction. 10 The high estimate is based on a long history of conflicts between Russia and the USA. 11 Palestine was not included into the data, thus the Palestinian threat for Israel was not accounted for. 23 5.2 Counterfactuals With these results, we ask the primary question of the paper – how increased trade affects conflicts and defense spending. For that purpose, we define a counterfactual experiment as follows: reduced trade costs τij and a corresponding increase of trade propensity Tij0 such that the total-peace level of trade Xij0 increases by one dollar, assuming constant defense spending and constant trade propensities on other dyads. Note that, due to the substitution effect, a rise of trade propensity Tij0 causes imports to i from countries other than j to drop. Using the assumption that the initial peace-level trade propensity Tij0 is numerically equal to Xij0 , and taking into accountP(5), we can show that the new peace-level trade propensity is T0 Tij1 = Tij0 + ∆T , where ∆T ≡ P k Tik0 = EiE−Ti 0 is considered to be a marginal change in trade k6=j ik ij volume. We can also show that the new peace-level trade volumes are Xij1 = Xij0 +1+o(∆T ), 0 Xik 12 1 0 as defined, and that Xik = Xik − Ei −X 0 + o(∆T ), ∀k 6= j. ij In the present paper, we report the results of three counterfactual experiments. In experiment #1, the importer (country i) is Jordan and the exporter (country j) is Israel. In experiment #2, the importer is North Korea and the exporter is South Korea. In experiment #3, the importer is Russia and the exporter is the United States of America. In all three cases, the experiment year was 2001, the last year of available data. Next, we outline the effects of such trade cost changes. 5.2.1 Effects on defense spending The change in defense spending by each country can be calculated by totally differentiating the first-order condition of optimal defense spending given by (16) with respect to m and k : Tij and by solving for the vector of ∂m ∂Tij   dm1     ...  = −  dmN  ∂ 2 uE 1 ∂m21 ... ∂ 2 uE 1 ∂m1 ∂mN ... ... ... ∂ 2 uE N ∂mN ∂m1 ... ∂ 2 uE N ∂m2N −1        ∂ 2 uE 1 ∂m1 ∂Tij ... ∂ 2 uE N ∂mN ∂Tij    ∆T.  Since the typical response to reduced trade costs is reduction (rather than increase) of defense spending worldwide, and since the natural lower bound on defense spending is zero, countries that already spent zero on defense were not allowed to adjust their spending. T 0 +∆T 1 0 1 1 − Xij = To show the result for Xij , observe that Xij = Ei Eiji +∆T , thus Xij 1 1 + o(∆T ). Likewise, we can show the result for Xik , k 6= j. 12 24 0 (Ei −Tij )∆T Ei +∆T = Ei Ei +∆T = Table 2: Defense spending cuts by selected countries following reduction of costs of imports to Jordan from Israel in 2001. Country United States of America United Kingdom Egypt Jordan Israel Saudi Arabia WORLD Defense spending change, $ (Std.Err.) -0.0031 (0.0025) -0.0019 (0.0013) -0.0026 (0.0016) -0.0280 (0.0106) -0.0477 (0.0217) -0.0061 (0.0037) -0.0965 (0.0424) Table 3: Defense spending cuts by selected countries following reduction of costs of imports to North Korea from South Korea in 2001. Country United States of America United Kingdom China North Korea South Korea Japan WORLD Defense spending change, $ (Std.Err.) -0.0442 (0.0261) -0.0015 (0.0011) -0.0028 (0.0021) -0.0481 (0.0231) -0.0238 (0.0142) -0.0196 (0.0144) -0.1503 (0.0771) Tables 2, 3, 4 report defense spending cuts by various countries in the two specified experiments. The standard errors of the estimates are due to the standard errors of the estimated model parameters. In all three experiments, the United States and the United Kingdom were among the top respondents to changing political arena, which is a consequence of high involvement of both into global political affairs. In all three experiments, the defense spending cuts by third countries (i.e. countries other than the importer and the exporter) constitute at least a quarter of the global cuts. 5.2.2 Welfare Implications Below we propose a classification of all welfare effects of increasing Tij0 on the “aggregate” expected utility Lk uE k , ∀k. In formulas below, we show the marginal effects, to be multiplied by a factor of ∆T . The expected welfare is affected both directly and via changing defense spending; moreover, the direct welfare effect for the importer i is qualitatively different from that for all other countries. For this reason, we divide welfare effects into three groups: direct effect for importer, direct effect for other countries, and indirect effect via changing 25 Table 4: Defense spending cuts by selected countries following reduction of costs of imports to Russia from the United States of America in 2001. Country United States of America United Kingdom France Russia China Japan WORLD Defense spending change, $ (Std.Err.) -0.0816 (0.0408) -0.0020 (0.0013) -0.0010 (0.0008) -0.0133 (0.0080) -0.0016 (0.0010) -0.0067 (0.0040) -0.1152 (0.0537) defense spending for all countries. • Direct effect on the importer. Following a decrease in trade costs and associated rise in trade propensity Tij0 , the importer i experiences, among other things, an expansion of the opportunity set, changed expected welfare due to possible attack by neighbors, and changed expected welfare due to possible own attack on neighbors. In math, ∂Si (Ti ) Li ∂uE i (T; m; β) = Ci Yi (1 − mi ) 0 ∂Tij ∂Tij0 X ∂ǫ⋆ik (Ti ; ·; λ) ∂Sik (Ti ; β; λ) ⋆ ǫik (·; λ) + Sik (·; λ) Yk Pki (·)Eλ + C i Yi σ i 0 ∂T ∂Tij0 ij k6=i X ∂Sik (Ti ; β; λ) ∂ǫ⋆ik (Ti ; ·; λ) ⋆ ⋆ + C i Yi σ i Yk (1 − Pki (·))Eλ F F (ǫik (·; λ)) + Sik (·; λ) F (ǫik (·; λ)) ∂Tij0 ∂Tij0 k6=i (20) 1 Due to the assumptions made in Section 3.1, the first line in (20) is merely s−1 , and is equal to the love of variety welfare effect equivalent to that of conventional trade models. It is the only effect that exists even in the absence of conflicts. The second line of (20) is the change in utility associated with conflicts initiated by neighbors of i. Finally, the third line of (20) is the change in utility associated with conflicts initiated by i itself. Note that the second and the third lines are typically negative for k = j, as they reflect by how much the gains from trade are reduced by possible conflict incidence. • Peaceful importer effect. Changing trade patterns make the importer i more dependent on the exporter j, which reduces the probability that i attacks j. Thereby, j’s 26 expected welfare is increasing, which is labeled as peaceful importer effect. Note that substitution of trade flows causes i to import less from other exporters k 6= i, j and thereby makes i more hostile towards k. Thus, the peaceful importer effect for k 6= i, j is negative. The peaceful importer effect does not apply to the importer i itself. In math, the peaceful importer effect is equal to (dropping function arguments) ∂Pik (Ti ; ·) ∂Pik (Ti ; ·) ∂uE Lk k = C k Yk σ k Yi Eλ Ski (·; λ) (ǫ⋆ki (·; λ) − F F (ǫ⋆ki (·; λ))) . 0 0 ∂Pik ∂Tij ∂Tij • Defense spending cuts by neighbors. Reduced hostility of i towards j causes both i and j to cut their defense spending, which makes both less hostile towards each other and towards third countries, which causes further defense spending cuts worldwide. While the welfare effect of own defense spending cuts is zero due to the envelope theorem, the effect of neighbors’ cuts is positive. In math, the welfare effect of defense spending cuts by neighbors for country k is equal to Lk X ∂uE (T; m; β) ∂mn k ∂mn n ∂Tij0 , where ∂uE ∂Pnk (·; mn ; ·) k = C k Yk σ k Y n Eλ Skn (·; λ) (ǫ⋆kn (·; λ) − F F (ǫ⋆kn (·; λ))) . ∂mn ∂mn Tables 5, 6, 7 report the welfare effects for the three experiments of this paper. The defense cuts by neighbors effect is truly global. In all three experiments, Brazil is thousands of kilometers away from both exporters and importers, but is one of top beneficiaries, primarily due to defense cuts by the United States which is highly involved in global affairs. The three experiments show that declining defense spending may account for a half or more of all welfare effects of declining trade costs, so we can conclude that the defense spending consequences of changing trade costs cannot be ignored. We can also split the defense cuts by neighbors effect by every particular neighbor. For example, in the Korean experiment (Table 6), the American welfare gain of $0.2711 includes $ 0.0040 due to defense spending cuts by Canada, $0.0019 due to Mexico, $0.0026 due to the United Kingdom, $0.0011 due to France, $0.0101 due to Russia, $0.0235 due to China, $0.2083 due to defense spending cuts by North Korea, $0.0070 due to South Korea, and 27 Table 5: Welfare effects for selected countries following reduction of costs of imports to Jordan from Israel in 2001. Country Welfare effect, $ (Std.Err.) Direct effect on importer Jordan 0.1285 (0.1350) Peaceful importer effect Israel 0.0714 (0.0373) Saudi Arabia -0.0011 (0.0004) Defense cuts by neighbors United States of America 0.0292 (0.0172) Brazil 0.0034 (0.0018) United Kingdom 0.0151 (0.0075) France 0.0042 (0.0023) Spain 0.0016 (0.0010) Germany 0.0045 (0.0027) Italy 0.0056 (0.0030) Iran 0.0015 (0.0008) Turkey 0.0020 (0.0011) Egypt 0.0109 (0.0052) Syria 0.0021 (0.0009) Lebanon 0.0018 (0.0009) Israel 0.0100 (0.0053) Saudi Arabia 0.0294 (0.0123) United Arab Emirates 0.0028 (0.0014) Japan 0.0041 (0.0028) Australia 0.0014 (0.0008) WORLD 0.1556 (0.0771) All effects combined WORLD 0.3524 (0.1070) 28 Table 6: Welfare effects for selected countries following reduction of costs of imports to North Korea from South Korea in 2001. Country Welfare effect, $ (Std.Err.) Direct effect on importer North Korea 0.1671 (0.1043) Peaceful importer effect South Korea 0.0428 (0.0237) Japan -0.0013 (0.0008) Defense cuts by neighbors United States of America 0.2711 (0.1324) Canada 0.0047 (0.0031) Mexico 0.0030 (0.0021) Colombia 0.0015 (0.0008) Venezuela 0.0011 (0.0007) Brazil 0.0059 (0.0036) Argentina 0.0015 (0.0010) United Kingdom 0.0131 (0.0077) Netherlands 0.0012 (0.0008) France 0.0075 (0.0044) Spain 0.0031 (0.0020) Germany 0.0053 (0.0035) Italy 0.0055 (0.0033) Russia 0.0038 (0.0024) Sweden 0.0012 (0.0008) Norway 0.0011 (0.0007) South Africa 0.0011 (0.0007) Iran 0.0015 (0.0008) Turkey 0.0012 (0.0007) Israel 0.0013 (0.0007) Saudi Arabia 0.0037 (0.0019) United Arab Emirates 0.0025 (0.0012) China 0.0167 (0.0099) Taiwan 0.0013 (0.0010) South Korea 0.0123 (0.0082) Japan 0.1221 (0.0776) India 0.0029 (0.0019) Thailand 0.0012 (0.0007) Malaysia 0.0011 (0.0007) Singapore 0.0022 (0.0012) Australia 0.0049 (0.0030) WORLD 0.5280 (0.2792) All effects combined WORLD 0.7361 (0.2829) 29 Table 7: Welfare effects for selected countries following reduction of costs of imports to Russia from the United States of America in 2001. Country Welfare effect, $ (Std.Err.) Direct effect on importer Russia 0.3123 (0.1436) Peaceful importer effect United States of America 0.2071 (0.1517) Ukraine -0.0019 (0.0010) Defense cuts by neighbors United States of America 0.2047 (0.1008) Canada 0.0075 (0.0042) Mexico 0.0049 (0.0030) Colombia 0.0015 (0.0007) Venezuela 0.0012 (0.0007) Brazil 0.0062 (0.0031) Argentina 0.0016 (0.0009) United Kingdom 0.0173 (0.0083) Netherlands 0.0011 (0.0006) France 0.0089 (0.0045) Spain 0.0041 (0.0023) Germany 0.0056 (0.0031) Italy 0.0054 (0.0027) Russia 0.0041 (0.0019) Finland 0.0015 (0.0008) Sweden 0.0018 (0.0009) Norway 0.0019 (0.0009) South Africa 0.0010 (0.0005) Iran 0.0018 (0.0008) Israel 0.0012 (0.0005) Saudi Arabia 0.0031 (0.0012) United Arab Emirates 0.0020 (0.0008) China 0.0110 (0.0051) South Korea 0.0024 (0.0014) Japan 0.0436 (0.0243) India 0.0019 (0.0010) Singapore 0.0014 (0.0006) Australia 0.0030 (0.0016) WORLD 0.3745 (0.1693) All effects combined WORLD 0.8893 (0.2397) 30 $0.0088 due to Japan. Furthermore, the American welfare gain of $0.2083 due to defense cuts by North Korea is the product of $4.3300 of American welfare gain per dollar of North Korean defense spending cut, times the $0.0481 of the magnitude of such cut (as shown in Table 3). Furthermore, the marginal American welfare gain of $4.33 due to North Korean defense cuts is the product of (i) the marginal effect of such defense cuts on the probability of North Korean aggression against the Unites States, (ii) economic size (proxy for military potential) of North Korea, and (iii) the overall sensitivity of the American welfare to military conflicts. 6 Conclusion The principal message of this paper is that increases in international trade, especially those between belligerent nations, may lead to much larger positive welfare effects than estimated by existing models of trade. In the three counterfactual experiments presented in this paper, the magnitude of the additional welfare effects due to defense spending cuts worldwide is comparable to that of direct welfare effects of increased trade. We also show that the welfare effects of rising trade apply not only to the two trading partners, but also to other nations, often on other continents, due to interdependence of global political relations and of national defense spending. To assess these effects, we develop a novel structural estimation technique that allows to slice the total welfare effect of rising trade between two nations into several components. First, rising trade makes the importer more dependent on the exporter and lowers the probability of conflict between the two, also makes both better off. Second, more peaceful relations between the two launch a wave of defense spending cuts worldwide, which makes the world even more peaceful. The model allows to show the welfare gains of every country due to defense spending cuts by every other country. The proposed model has a number of limitations which remain to be addressed in future research. One key limitation is that our model ignores the problem of “trade zeros”: the model assumes that a conflict reduces trade by a factor, therefore a positive trade volume in case of peace remains positive in conflict; zero trade in conflict implies zero trade in peace as well. Empirically, however, it is quite plausible that trade may be positive in case of peace and zero in conflict, something that our methodology cannot replicate. Another limitation is that we implement a static model and thus do not account for intertemporal effects of conflicts and defense spending, possibly underestimating the true 31 effects. We also do not allow for military coalitions and, thereby, exclude a substitution effect (if the U.S. reduce their defense spending as a result of increased trade with Brazil, the U.S. allies like Japan and South Korea may increase rather than decrease their defense spending to make up for the reduced protection from the U.S.) from the consideration. At the same time, introducing military coalitions into the model may complicate the analysis beyond the limits of tractability. References Acemoglu, D., A. Golosov, Mikhai Tsyvinski, and P. Yared (2012). A dynamic theory of resource wars. Quarterly Journal of Economics 127 (1), 283–331. Acemoglu, D. and P. Yared (2010). Political limits to globalization. American Economic Review: Papers and Proceedings 100 (2), 83–88. Aizenman, J. and R. Glick (2006). Military expenditure, threats, and growth. Journal of International Trade and Economic Development 15 (2), 129–155. Alesina, A. and E. Spolaore (2005). War, peace and the size of nations. Journal of Public Economics 89 (7), 1333–1354. Alesina, A. and E. Spolaore (2006). Conflict, defense spending and the number of nations. European Economic Review 50 (1), 91–120. Barbieri, Katherine, O. M. G. K. and B. Pollins (2009). Trading data: Evaluating our assumptions and coding rules. Conflict Management and Peace Science 26 (5), 471–479. Broda, C. and D. E. Weinstein (2006). Globalization and the gains from variety. Quarterly Journal of Economics 121 (2), 541–585. Collier, P. and A. Hoeffler (2002, November). Military expenditure: Threats, aid and arms races. Working Paper 2927. Feenstra, R. C., R. E. Lipsey, H. Deng, A. C. Ma, and H. Mo (2005, January). World trade flows: 1962-2000. Working Paper 11040, National Bureau of Economic Research. Ghosen, F., G. Palmer, and S. Bennett (2004). The mid 3 data set 1993 2001: Procedures, coding rules, and description. Conflict Management and Peace Science 21, 133–154. 32 Glick, R. and A. M. Taylor (2010). Collateral damage: Trade disruption and the economic impact of war. The Review of Economics and Statistics 92 (1), 102–127. Gowa, J. and E. D. Mansfield (1993). Power politics and international trade. American Political Science Review 87 (2), 408–420. Hegre, H., J. R. Oneal, and B. Russett (2010). Trade does promote peace: New simultaneous estimates of the reciprocal effects of trade and conflict. Journal of Peace Research 47 (6), 763–774. Kamlet, M. and D. Mowery (1987). Influences on executive and congressional budgetary priorities, 1955-1981. American Political Science Review 81 (1), 155–178. Martin, P., T. Mayer, and M. Thoenig (2008). Make trade not war? Review of Economic Studies 75 (3), 865–900. Martin, P., T. Mayer, and M. Thoenig (2012). The geography of conflicts and regional trade agreements. American Economic Journal: Macroeconomics 4 (4), 1–35. Mayer, T. and S. Zignago (2011). Notes on cepiis distances measures: The geodist database. Morrow, J., R. Siverson, and T. Tabares (1998). The political determinants of international trade: The major powers, 1907-90. American Political Science Review 92 (3), 649–661. Polachek, S. W. (1980). Conflict and trade. Journal of Conflict Resolution 24 (1), 55–78. Polachek, S. W. and C. Seiglie (2007). Trade, peace and democracy: an analysis of dyadic dispute. Handbook of Defense Economics 2, 1017–1073. Sample, S. (1998). Military buildups, war, and realpolitik: A multivariate model. Journal of Conflict Resolution 42 (2), 156–175. Singer, J. D., S. Bremer, and J. Stuckey (1972). Capability distribution, uncertainty, and major power war, 1820-1965. in bruce russett (ed). Peace, War, and Numbers, 19–48. Beverly Hills: Sage. Skaperdas, S. and C. Syropoulos (2001). Guns, butter, and openness: On the relationship between security and trade. American Economic Review: Papers and Proceedings 91 (2), 353–357. 33 Stinnett, D. M., J. Tir, P. F. Diehl, P. Schafer, and C. Gochman (2002). The correlates of war (cow) project direct contiguity data, version 3.0. Conflict Management and Peace Science 19 (2), 59–67. Stoica, P. and B. C. Ng (1998). On the cramér-rao bound under parametric constraints. Signal Processing Letters, IEEE 5 (7), 177–179. Vicard, V. (2012). Trade, conflict, and political integration: Explaining the heterogeneity of regional trade agreements. European Economic Review 56 (1), 54–71. Wallace, M. (1979). Arms races and escalation: Some new evidence. Journal of Conflict Resolution 23 (1), 3–16. Zeev, M. (2005). Dyadic mid http://psfaculty.ucdavis.edu/zmaoz/dyadmid.html . A dataset (version 2.0). Further details of the estimation procedure We calculate the parameter estimates β̂ by maximizing (19) over β subject to (17). The corresponding Lagrangian is Λ(β̃, µ) = L(β̃) + G(β̃)T µ → max where µ = {µi , ∀i} is the vector of Lagrange multipliers, and where superscript T denotes the transposition operator, not to be confused with trade propensities. Assuming an interior solution, the vector of parameter estimates β̂ and Lagrange multiplier estimates µ then satisfies ∇β L(β̂) + ∇β G(β̂)T µ = 0, (21) G(β̂) = 0, (22) where ∇ denotes the Jacobian operator, i.e. ∇x f (x) is the K1 × K2 matrix of first partial derivatives of a 1 × K2 vector-valued function f of K1 × 1 vector of variables x. Our estimation setup seemingly deviates from that of the standard maximum likelihood due to the fact that increased sample size (country-country-year observations) is likely to be accompanied by an increased number of variables (specifically, country-year fixed effects σi ). 34 In fact, however, the number of degrees of freedom does not change: each additional variable is complemented by an additional constraint (17) that uniquely determines the value of that variable. For this reason, the asymptotic properties of our estimates are standard. The asymptotic variance of the estimates is calculated using a standard technique (e.g., see Wikipedia article on Maximum Likelihood), with some augmentation to account for constraints (17). First, we Taylor-expand the system (21)-(22) around the true value of β̂, i.e. around β, and around the zero value of µ: ∇β (L(β)) + ∇ββ T (L(β)) (β̂ − β) + ∇β GT (β) µ + o({β̂ − β; µ}) = 0 G(β) + ∇β T G(β)(β̂ − β) + o(β̂ − β) = 0 (23) (24) where ∇ββ T is the Hessian (matrix of second derivatives) of its argument. The above system may be solved for the unknowns as follows: β̂ − β µ ! = " −∇ββ T (L(β)) −∇β T (G(β)) 0 −∇β GT (β) #−1 ∇β (L(β)) G(β) ! + o({β̂ − β; µ}) (25) The off-diagonal elements of the block matrix in square brackets in (25) are deterministic; we denote the lower-left matrix by B(β) ≡ −∇β GT (β) , thus the upper-right matrix is B T (β). The upper-left element of the block matrix in square brackets in (25), −∇ββ T (L(β)), is random (since loglikelihood is a function of empirical conflict statuses, which are in turn functions of income and trade shocks), and converges in probability to a square positivedefinite matrix nH(β), where H(β) ≡ −Eǫ,λ ∇ββ T (L(β)), and where n is the number of country-country-year observations that enter the loglikelihood. In the block vector in brackets on the right-hand side of (25), the lower element, G(β), is deterministic and is equal to zero, while the upper element, ∇β (L(β)), is random and has an expectation of zero. Thus, our estimates β̂ and µ have expectations of β and zero, respectively. To find the asymptotic variance of β̂, we apply the Central Limit Theorem to obtain √ n(β̂ − β) = →d 1 nJ(β) √ ∇β (L(β)) + o({β̂ − β; µ}) n N (0, Eǫ,λ nJ(β)Eǫ,λ (∇β (L(β)))2 Eǫ,λ nJ(β)) where J(β) is the upper-left block of the inverse block matrix shown in (25). 35 (26) One can show that (dropping some function arguments) Eǫ,λ J(β) = −1 1 −1 1 −1 T H − H B BH −1 B T BH −1 , n n thus nEǫ,λ J(β) does not depend on n. We can also show that Eǫ,λ (∇β (L(β)))2 = H. Thus, the asymptotic variance of β̂ is equal to (cf.(26)) −1 −1 1 −1 BH −1 BH −1 H H −1 − H −1 B T BH −1 B T H − H −1 B T BH −1 B T n −1 1 −1 1 −1 T BH −1 (27) = H − H B BH −1 B T n n Eǫ,λ (β̂ − β)2 = Stoica and Ng (1998) prove that (27) is equal to the Cramer-Rao variance lower bound, hence the proposed estimator of unknown model parameters is asymptotically efficient. In the numerical procedure, we can approximate nH(β) by −∇ββ T L(β̂) , and B(β) by B(β̂). B B.1 Data details List of countries dropped from sample A country was dropped from the sample if (i) it had no defense spending data in any year, or (ii) it had data on imports from less than 5 trading partners in 2001; we also drop Nauru, the second-smallest nation in the world, for which we could not construct the GDP data for some years. The list of dropped countries is as follows: St. Kitts and Nevis, Monaco, Liechtenstein, Andorra, San Marino, Comoros, Tonga, Nauru, Marshall Islands, Federated States of Micronesia. Each of these had a GDP of under 3 billion dollars and no recorded history of defense spending. B.2 Country statistics 36 Table 8: Some country statistics, year 2001 Country Defense spending, % of GDP Conflict length aggregated across neighbors, countryyears Total imports, % of GDP United States of America Canada Bahamas Cuba Haiti Dominican Republic Jamaica Trinidad and Tobago Barbados Dominica Grenada St. Lucia St. Vincent and the Grenadines Antigua and Barbuda Mexico Belize Guatemala Honduras El Salvador Nicaragua Costa Rica Panama Colombia Venezuela Guyana Suriname Ecuador Peru Brazil Bolivia Paraguay Chile Argentina Uruguay 3.00 1.10 0.44 2.31 0 0.62 0.50 0.75 0.50 1.00 0 0 0 0.50 0.60 0.90 0.80 0.70 1.30 0.80 0 0 3.00 1.50 1.50 3.14 1.60 1.80 1.80 2.10 1.10 3.80 1.10 2.80 1.5081 0.1048 0 0 0 0 0 0 0 0 0 0 0 0 0 0.3656 0.3656 0.2043 0 0.3065 0 0 0.1022 0 0 0 0 0 0 0 0 0 0 0 11.7180 32.8808 49.1025 10.6750 18.0337 38.1901 36.1070 40.3019 40.8193 38.3043 34.0402 77.3815 39.6814 65.6241 29.2754 50.6530 29.2522 37.6782 35.1261 39.1568 26.5591 22.0094 12.6043 13.1988 82.3861 84.0103 26.7463 14.4915 10.9602 20.8488 34.1903 23.1375 7.3384 14.7288 Continued on next page 37 Country Defense spending, % of GDP Conflict length aggregated across neighbors, countryyears Total imports, % of GDP United Kingdom Ireland Netherlands Belgium Luxembourg France Switzerland Spain Portugal Germany Poland Austria Hungary Czech Republic Slovakia Italy Malta Albania Macedonia Croatia Yugoslavia Bosnia and Herzegovina Slovenia Greece Cyprus Bulgaria Moldova Romania Russia Estonia Latvia Lithuania Ukraine Belarus Armenia Georgia 2.40 0.70 1.50 1.40 0.60 2.50 1.10 1.20 1.90 1.50 1.80 1.00 1.70 2.00 1.70 2.00 0.70 1.20 1.90 3.10 5.50 2.45 1.10 3.60 3.00 2.70 0.40 2.50 3.70 1.40 0.90 1.70 3.60 1.30 3.60 0.60 0.9220 0 0.0591 0 0 0.1667 0 0.0591 0.0591 0.0591 0 0 0 0 0 0 0 0.1022 0 0 0.1022 0 0 0.3871 0.5511 0 0 0 0.7823 0 0 0.0860 0 0.0860 0 0.2016 20.7991 43.1469 50.4285 77.5283 57.2586 24.8987 32.2164 23.3166 31.5640 25.5818 26.2372 39.2347 62.6157 63.6119 53.3359 21.1190 121.5685 32.2704 44.8176 38.1870 35.8826 49.8555 50.4138 21.6080 40.8713 51.2197 59.2851 37.6833 11.8982 83.9315 41.9360 51.8912 41.3333 67.6823 41.0939 23.4704 Continued on next page 38 Country Defense spending, % of GDP Conflict length aggregated across neighbors, countryyears Total imports, % of GDP Azerbaijan Finland Sweden Norway Denmark Iceland Cape Verde Sao Tome and Principe Guinea-Bissau Equatorial Guinea Gambia Mali Senegal Benin Mauritania Niger Cote DIvoire Guinea Burkina Faso Liberia Sierra Leone Ghana Togo Cameroon Nigeria Gabon Central African Republic Chad Congo Democratic Republic of Congo Uganda Kenya Tanzania Burundi Rwanda Somalia 2.30 1.30 2.00 1.70 1.50 0 1.30 no data 4.40 0.23 1.00 2.20 1.30 0.60 3.50 1.20 0.75 1.50 1.20 4.41 3.70 0.70 2.33 1.30 0.80 1.80 1.55 1.90 2.86 8.35 2.50 1.30 1.50 6.00 3.50 2.56 0.4382 0 0 0.0134 0 0 0 0 0 0 0 0 0 0 0 0 0 0.7554 0 0.4785 0.4382 0 0 0.0134 0 0 0.0134 0 0 0.8441 2.2823 0 0 0 2.2823 0 25.0072 22.6418 27.5693 18.5413 27.3757 29.6218 42.9252 70.5393 46.9711 19.3426 94.9333 35.3301 35.2381 25.9257 49.8529 17.0250 23.0957 18.0433 16.9399 815.1370 50.4026 49.9575 26.5074 18.9304 16.5178 30.2574 10.5887 21.5112 21.2783 14.4494 17.3810 27.2178 17.4247 19.5625 11.7796 19.8681 Continued on next page 39 Country Defense spending, % of GDP Conflict length aggregated across neighbors, countryyears Total imports, % of GDP Djibouti Ethiopia Eritrea Angola Mozambique Zambia Zimbabwe Malawi South Africa Namibia Lesotho Botswana Swaziland Madagascar Mauritius Seychelles Morocco Algeria Tunisia Libya Sudan Iran Turkey Iraq Egypt Syria Lebanon Jordan Israel Saudi Arabia Yemen Kuwait Bahrain Qatar United Arab Emirates Oman 5.10 7.50 32.70 6.40 1.30 0.74 4.70 0.70 1.50 2.70 4.00 3.30 1.50 1.20 0.20 1.70 2.30 3.40 1.80 3.10 4.50 3.70 3.70 7.24 3.20 5.50 5.40 6.30 8.00 10.60 4.40 7.20 4.00 7.08 9.40 10.80 0 0 0 1.8253 0 0.9812 0.8441 0 0 0.8441 0 0 0 0 0 0 0 0 0 0 0 0.4409 1.0699 2.5054 0 0.5000 0.9973 0 1.4973 0.6452 0 0.2473 0 0 0 0 121.4357 19.4002 27.2792 38.5701 21.9129 29.9271 22.2083 25.1134 22.9668 14.9584 14.1348 4.8122 6.3726 14.7020 44.2940 79.3677 29.2162 17.8380 42.6296 15.7767 14.3243 14.6389 20.1403 29.5683 11.9293 25.2845 34.8417 51.6011 25.0551 20.3195 23.9812 22.0456 46.4132 20.4877 30.5749 28.9443 Continued on next page 40 Country Defense spending, % of GDP Conflict length aggregated across neighbors, countryyears Total imports, % of GDP Afghanistan Turkmenistan Tajikistan Kyrgyzstan Uzbekistan Kazakhstan China Mongolia Taiwan North Korea South Korea Japan India Bhutan Pakistan Bangladesh Myanmar Sri Lanka Maldives Nepal Thailand Cambodia Laos Vietnam Malaysia Singapore Brunei Philippines Indonesia Australia Papua New Guinea New Zealand Vanuatu Solomon Islands Kiribati Tuvalu 9.95 6.28 1.20 2.90 1.20 0.80 1.90 2.10 2.40 9.85 2.60 1.00 3.10 4.16 3.70 1.30 2.30 5.00 4.56 0.80 1.50 2.20 0.80 7.19 1.60 4.60 5.70 1.60 0.53 1.80 0.90 1.20 0 0 0 0 1.5995 0 0.0081 0 0.0672 0 1.0753 0 0.4274 0.4839 0.4839 0.1586 0.4677 0 0.4704 0.0027 0.2177 0 0 0 0.2177 0 0 0 0 0 0 0.0806 0 0.1048 0 0 0 0 0 0 22.6432 62.9105 63.5987 30.9817 20.8336 29.2810 23.3518 49.2217 34.1775 15.6948 27.3303 8.4593 9.5603 5.7107 14.1162 16.5070 40.6860 33.7569 49.8122 14.6684 49.8994 34.6157 39.3159 47.8934 77.3222 126.1993 23.4061 43.5333 19.4895 17.4795 35.6907 24.9090 63.8031 24.2162 69.0734 87.8851 Continued on next page 41 Country Defense spending, % of GDP Conflict length aggregated across neighbors, countryyears Total imports, % of GDP Fiji Palau Samoa 1.90 0 0 0 0 0 43.7271 21.0989 109.0048 42

Log In

Trade Costs, Conflicts, and Defense Spending

Trade Costs, Conflicts, and Defense Spending

Free Related PDFs

FREE RELATED PAPERS

RELATED TOPICS