Trade and Thy Neighbor’s War†

Contributor Notes

Author’s E-Mail Address: mqureshi@imf.org

This paper examines the spatial dispersion effects of regional conflicts, defined as internal or external armed conflicts in contiguous states, on international trade. Our empirical findings-based on different measures of conflict constructed using alternate definitions of contiguity and conflict-reveal a significant collateral damage in terms of foregone trade as a result of spillovers from conflict in neighboring countries. The magnitude of this negative externality is somewhat larger for international conflicts than intrastate warfare, but about one-third of conflict in the host economies. Further, the impact is persistent-on average, it takes bilateral trade three years to recover from the end of intrastate conflicts in neighboring states, and five years from international conflicts. These findings are robust to alternate definitions of conflict, estimation methods, and specifications, and underscore the importance of taking into account spillover effects when estimating the economic costs of warfare.

Abstract

This paper examines the spatial dispersion effects of regional conflicts, defined as internal or external armed conflicts in contiguous states, on international trade. Our empirical findings-based on different measures of conflict constructed using alternate definitions of contiguity and conflict-reveal a significant collateral damage in terms of foregone trade as a result of spillovers from conflict in neighboring countries. The magnitude of this negative externality is somewhat larger for international conflicts than intrastate warfare, but about one-third of conflict in the host economies. Further, the impact is persistent-on average, it takes bilateral trade three years to recover from the end of intrastate conflicts in neighboring states, and five years from international conflicts. These findings are robust to alternate definitions of conflict, estimation methods, and specifications, and underscore the importance of taking into account spillover effects when estimating the economic costs of warfare.

I. Introduction

“All wars are follies, very expensive and very mischievous ones.”

- Benjamin Franklin, 1783.

Political violence and war—both internal and international—have plagued the world for long. The incidence of international conflicts exhibits a declining trend in recent years, but interstate conflicts of different intensity, ideological origins, and dynamics have increasingly prevailed, particularly in less developed countries. In 2008, for example, there were 25 major conflicts raging in Africa, Asia, and other parts of the world, 20 of which represented interstate warfare.1 The economic costs of such conflicts resulting from physical and human capital losses, infrastructure destruction, lower investments, and market disruption are substantial, leading some scholars to call regional conflicts as “the greatest risk to the world today”.2

From a global perspective, the challenges posed by conflicts are compounded further because seldom are their negative impacts confined within the national boundaries. In fact, the few existing studies investigating the economic consequences of spatial diffusion of conflicts, notably, Murdoch and Sandler (2002, 2004) and Collier and Hoeffler (2004), find evidence of profound negative spillover effects of intrastate conflict on the economic growth rates of neighboring states.3 The effect is strong in both the short and long runs, and large in cumulative terms—estimated to cost a typical neighbor about 43 percent of initial GDP (Collier and Hoeffler, 2004).

The pertinent question therefore is how do such large spatial effects set in? A variety of possibilities exist, at least theoretically, but empirical evidence does not support the direct spillover effects from migration, human capital dilution or a decline in long term investment ratios.4 In this paper, we investigate another potential but unexplored channel through which conflict in neighboring countries could have important consequences for economic growth: the reduction in international trade. Warfare in neighboring states could disrupt trade flows directly through, for example, a rise in transport costs, or, an interruption of input supplies and exports, if borders are hardened due to security concerns and external trade routes are damaged or blocked. There could also be indirect effects if the poor security situation and uncertainty surrounding the spreading out of the conflict deters traders from demanding the products originating in the region, and investors from making productive investments—thereby reducing trade and restricting growth of the tradable sector.5

The available anecdotal evidence supports these conjectures. For example, Malawi—a landlocked country in sub-Saharan Africa—had to rechannel its trade through South Africa when civil unrest erupted in neighboring Mozambique in the 1980s. Transport costs from Burundi to the nearest ports in Kenya and Tanzania were high during the war between Uganda and Tanzania in the late 1970s and early 1980s, while passage through conflict zones was unreliable and trade flows were frequently disrupted (World Bank, 1992). Similarly, the war in Azerbaijan at the end of 1980s interfered with the oil deliveries from the Caspian sea, and the wars in Afghanistan and Tajikistan disrupted central Asia’s trade flows (Brown, 1996). More recently, the war in Iraq was accompanied by a reduction in total trade (in real terms) of neighboring Syria by 10 percent in 2003.6 A better understanding of the implications of warfare, not only on host economy’s trade but also on its neighbors’, is therefore essential to assess the devastating economic effects of conflicts and prepare an adequate response at both the national and international levels.

This paper attempts to systematically investigate the spatial dispersion of warfare consequences on international trade flows by addressing three key questions. First, are bilateral trade flows directly affected by conflict in contiguous states. Second, do countries successfully insulate themselves from neighborhood warfare as it continues over time. Third, are regional warfare effects persistent in nature. To address these questions, we develop a simple theoretical framework of bilateral trade flows that explicitly incorporates the effect of neighborhood conflict through both demand and supply side mechanisms, and motivates the empirical analysis. On the demand side, foreign consumers differentiate between products by country of origin—expressing ideological and political preferences for products from nonconflict regions. On the supply side, regional conflict increases trade costs, affecting exporter competitiveness and reducing trade. The derived reduced form equation specifies a causal relationship, predicting that conflict in contiguous states depresses bilateral trade.7

Our empirical assessment is based on an extended armed conflict dataset that covers 145 countries over the period 1948–2006. Using this dataset, we construct several composite measures—with alternate definitions of contiguity—to reflect internal and international conflict presence and intensity in the immediate region. Our wide range of measures for regional conflict reveal interesting spatial and temporal trends. First, we find that while the prevalence and intensity of international conflicts in neighboring states has declined over the sample period, intrastate conflicts have become more widespread and increased in intensity. Second, regional conflict is the most prevalent in Africa and parts of Asia—for example, no country in South Asia and only a handful in Africa had no neighbor involved in some form of conflict in 2006. Furthermore, the correlation between self-involvement in warfare and regional conflicts is low, indicating that the economic costs of warfare are unlikely to be fully accounted for by studying domestic conflict only.

The results from our empirical estimations provide strong evidence of systemic effects of regional conflicts. Specifically, while controlling for conflict in the host economies as well as for other regional characteristics, we find that conflict in neighboring states has a significant but modest adverse effect on bilateral trade flows. The magnitude of this effect is somewhat larger for international wars as compared to intrastate wars, and about one-third of that for host country conflict. For example, bilateral trade flows are estimated to be, on average, about 12 percent lower if at least one of the trading partners is involved in a conflict, but the estimated change as a result of an additional intrastate and international conflict in the dyad’s neighborhood is about 2–4 percent. An increase in the measure of regional conflict intensity also imposes higher trade costs—for example, a shock of one standard deviation to the measures of intrastate and international conflicts reduces bilateral trade by about 7 and 3 percent, respectively.

The results for different subsamples based on income classification show that the negative impact is significant for both high and low income countries, and relatively more pronounced for their trade with each other. In addition, regional warfare effects appear to be dynamic in nature— increasing with the duration of the conflict, and, on average, persisting for 3-5 years after its end. The post-regional conflict trade recovery period is, however, shorter (between 2-4 years) for high income countries. These results are robust to alternate datasets, proxies of conflicts, estimation methods, and model specifications.

This paper, to our best knowledge, is the first to analyze the spatial dimensions of conflicts for international trade.8 In doing so, it combines two strands of literature—regional contagion and spillover effects, and the nexus between international trade and conflict. The first body of literature has mostly focused on financial contagion across regions, and spillovers of technology and knowledge in the economics domain (see, for example, Claessens and Forbes, 2001; Holod and Reed, 2004), while studies in political science have been preoccupied with the likelihood of international spillovers of regional conflicts. Notable exceptions to this include Ades and Chua (1997), Murdoch and Sandler (2002, 2004), Collier and Hoeffler (2004), Chauvet, Collier and Hoeffler (2007), and De Groot (2009) who examine the spillover effects of neighborhood conflict and political instability on economic growth.

The second strand of literature—both in political science and more recently in economics—has been limited to analyzing the impact of conflict on bilateral trade when the host economy is itself involved in warfare, either with the trading partner or with another country.9 These studies almost unanimously show significant and large adverse effects of self-involvement in conflict on trade. For example, Glick and Taylor (2004), study the effects of World Wars I and II when at least one country in the trading pair is engaged in war. They find strong contemporaneous and persistent negative effects of warfare on bilateral trade flows even after controlling for possible endogeneity. Similarly, Blomberg and Hess (2006) investigate the effects of different types of violence on bilateral trade. They find a significantly negative impact of external and internal conflict, as well as of terrorism, that is larger than the estimated tariff-equivalent costs of border and language barriers. Martin, Mayer, and Thoenig (2008) analyze the two-way relationship between military conflicts and trade. They show that military conflicts substantially reduce trade, while the probability of war escalation is lower for countries that engage in greater bilateral trade. Multilateral trade openness has the opposite effect in their model as it decreases bilateral trade dependence, and escalates the probability of bilateral war.

The rest of the paper is organized as follows. Section II develops the analytical framework, and describes the construction of composite measures to reflect neighborhood conflict. Section III discusses the data. Section IV presents the empirical results and the sensitivity analysis. Section V concludes.

II. Methodology

A. Analytical framework

To examine the impact of conflict in contiguous states on trade, we develop a simple theoretical framework that augments the traditional gravity model of bilateral trade flows. Existing research tends to postulate the impact of conflict on international goods flows from the supply side only— specifically, by assuming an increase in trading costs. We introduce both demand and supply side mechanisms through which regional warfare could affect trade by: (i) the explicit modeling of Armington’s (1969) assumption that consumers differentiate between products based on the country of origin, expressing preferences for products originating from nonconflict regions; and (ii) a rise in trade costs as security is increased on the borders and external trade routes are blocked or damaged, which affects the supply of exports.

Consider a world economy with i=1,…, N countries, where each country specializes in the production of one good and all goods are differentiated by the country of origin. Consumers in country j have identical and homothetic preferences, approximated by a constant elasticity of substitution (CES) utility function, given by

Uj=i=1N(bijcij)σ1σ,(1)

where Cij is the consumption in country j of the good imported from country i, σ is the elasticity of substitution between goods (σ >1), and bij reflects the (relative) evaluation or preference of the consumers in country j of the good imported from country i. As hypothesized by Pollins (1989), we assume that consumers in country j employ a common logic, which incorporates both economic as well as socio-political and ideological concerns into their import decisions. In some cases, these concerns may be revealed by straightforward boycotts or import bans—such as that in the context of “conflict diamonds” originating from central and western Africa—while in others, they may reflect a general foreign policy orientation and purposeful attempt to exert pressure on the exporter to change the status quo by manipulation of economic ties. The exporter in question could itself be directly involved in the conflict or perceived to be politically or economically involved in a neighboring conflict. The foreign consumers therefore have a preference for goods originating from nonconflict regions such that

bij=exp(β1Wari+β2Nwari),(2)

where Wari and Nwari—assumed to lie within a continuous range [0, ∞]—denote the status of conflict in country i and its neighborhood, respectively, and β1 and β2 are (negative) parameters that reflect the responsiveness of consumers’ preferences to the degree of conflict in country i and its neighborhood, respectively. A value of zero for Wari and Nwari indicates peace while strictly positive values indicate conflict, so that bij approaches one for peace and zero for a high level of conflict.10 Hence, the aggregate utility function of country j is strictly decreasing with the warfare exposure of exporter i.

The representative consumer in country j maximizes the utility function subject to the following budget constraint

Yj=i=1Npijcij,(3)

where Yj is the aggregate income of all consumers in country j, and pij is the exported product’s price in country j. The demand for each product Cij, derived from maximizing (1) subject to (3), is given by

cij=bijσ1(pijPj)σYjPj,(4)

where Pj is the consumer price index of country j, defined as

Pj=i=1N(bijσ1pij1σ)1/1σ.(5)

From (4), we can see that the demand for country i’s good is an increasing (decreasing) function of peace (conflict) in country i and its neighborhood. The total nominal value of exports, X, from country i to j can hence be written as

Xij=pijcij=bijσ1(pijPj)1σYj,(6)

where the first equality holds by definition, and the second is obtained from (4).

We now assume that prices between countries differ due to trade costs such that the price at destination is given by pij=piTij, where pi reflects net of transport (or free on board) domestic prices, and Tij represents “iceberg trade costs”.11 In general, Tij is considered to include all factors that limit trade between i and j, and is expressed as a function of natural trade costs (D), for example, distance, common border, and access to sea, and artificial trade costs (τ), for example, tariffs, institutional quality, infrastructure, and currency conversion costs. That is,

Tij=Dijexp(τij).(7)

Substituting (7) in (6), total exports can be expressed as

Xij=pijcij=bijσ1(pijPj)1σYj=bijσ1(piTijPj)1σYj.(8)

If conflict erupts in the neighborhood of country i, then from (7) the resultant tightening of security at the borders, or any blockages and damage to the external trading routes, would manifest as a rise in τ and Tij, which for a given preference parameter bij would imply lower export supply.

Following Anderson and van Wincoop (2003), we can simplify (8) by assuming symmetric transportation costs (Tij=Tji), and making use of the market clearing condition given by12

yi=j=1NcijTij.(9)

Multiplying both sides of (9) by pi and using pij=piTij, we have

piyi=j=1N(piTijPj)1σbijσ1Yj.(10)

Further, denoting country i’s aggregate output by Yi=piyi, world output by Yw, and country i’s share in world output by si= Yi/Yw, we obtain from (10)

p˜i=si1/1σP˜i,(11)

where

P˜i=(j=1Nsjbijσ1(TijPj)1σ)1/1σ.(12)

Substituting (12) into (5) and (8), we obtain

Xij=YiYjYwbijσ1(TijP˜iP˜j)1σ,(13)

where

P˜j=(i=1Nsibijσ1(TijP˜i)1σ)1/1σ.

Equation (13) is the familiar gravity model, whereby bilateral trade between countries depends on their outputs, trade costs, and the implicit price measures or “multilateral resistance” terms Pi and Pj In addition, the model includes the term, bij, that reflects consumers’ political preference for trading partners. Taking logs of equation (13) and adding scripts for the time dimension, we obtain the following equation for estimation purposes

log(Xijt)=β0+β1Warit+β2Nwarit+β3log(YitYjt)+β4log(Dij)+β5τijlog(Pit)σ1log(Pjt)σ1+μijt.(14)

Equation (14) gives us the impact of regional warfare on bilateral trade through both the supply (τ) and demand (Nwar) sides. However, relevant data on changes in transport costs as a result of regional conflict that would allow us to separately identify the effect of τ is not available. Hence, in the estimated model, the coefficient for Nwar captures both the demand as well as supply side effects of conflicts in neighboring states—and is expected to be negative—while τ includes other variables typically used to proxy artificial trading costs such as currency union, free trade agreements, common language, and historical ties.

B. Estimation issues

An important issue in estimating equation (14) is that of measuring the unobserved multilateral price terms.13 Including them in the error term and estimating the model with the pooled Ordinary Least Squares (OLS) approach could lead to a serious omitted variables bias problem, originating from the correlation of these terms with any of the right-hand side variables. Research following Rose (2000) attempts to control for this bias by introducing country-specific idiosyncrasies (or country fixed-effects (CFE)) in the gravity model—both for cross-sectional and panel estimations. However, given that there is a time-series element to the potential bias that is not eliminated with this procedure; Anderson and van Wincoop (2004) propose that separate country fixed-effects should be included for each year (CYFE) to take into account changes in multilateral resistance over time.

Glick and Rose (2002) argue that including the CFE or CYFE may still not resolve the omitted variables problem. This is because the unobserved variables could be correlated with the bilateral characteristics of the dyads, which may bias the CFE/CYFE estimates. They therefore propose using the panel data fixed-effects estimator that adds country-pair specific effects (CPFE) to the gravity equation, thereby controlling for any strong bilateral likelihood to trade. In a recent paper, Baier and Bergstand (2007) note that in a panel setting, the multilateral price variables (Pi and Pj) are likely to be time varying. Thus, controlling for them through CPFE may not fully account for the omitted variable problem. They therefore propose including both the CPFE as well as the CYFE to control for possible correlation between the unobserved omitted and time invariant bilateral variables, and between the omitted and time varying variables, respectively. We denote this estimation approach as country-year and county-pair fixed effects (CYPFE).

As discussed later in Section IV, our estimation results for equation (14) are broadly similar for both the CPFE and CYPFE approaches when we use the world sample. Considering the large N and T dimensions of our dataset, and the associated computational difficulties in estimating CYPFE, we retain the CPFE as our preferred estimation approach for other specifications involving, for example, different subsamples. The final estimating equation therefore takes the form

log(Xijt)=β0+β1Warit+β2Nwarit+β3log(YitYjt)+β4log(Dij)+β5τij+k=6KβkZk,ijt+uit+λt+μijt,(15)

where Z is a vector consisting of k control variables that reflect regional characteristics such as the (median) regional polity score and real income per capita; uij reflect the country-pair specific effects, λt are time-specific factors common across all countries, and μijt is the normally distributed error term. The reason for taking into account other neighborhood features in the model is to ensure that our regional conflict variable does not capture spillovers from other potential socio-economic channels.

In addition to the omitted variable bias issue, a potential estimation concern that has preoccupied researchers investigating the effect of the host economy’s conflict on trade is to address the possible reverse causality from trade to conflict. Two hypothesis have been postulated on this front: the first argues that trade between two countries reduces the likelihood of a conflict between them (see, for example, Polachek, 1992; and Oneal and Russett 1999), while in the second view, trade causes conflict as countries compete over economic and political goals (for example, Gasiorowski, 1986; Gowa 1994; and Barbeiri, 2002).

Endogeneity concerns are, however, mitigated in equation (15) as our main variable of interest is warfare in dyad’s neighboring countries. From a theoretical point of view, it is easier to believe in the exogeneity of regional conflicts in bilateral trade models since trade between i and j is unlikely to affect interstate conflict in their neighboring states. As for international conflicts in the region, trade may have an effect to the extent that both i and j are neighbors, in which case, as hypothesized in the political science literature, their bilateral trade could affect the probability of going to war with each other. Although, even then, the impact of bilateral trade on the regional conflict variable is likely to be small in our case as the latter variable includes international conflict involvement of all neighboring states (and not only of the neighboring country j).

Similar argument holds for our variable depicting conflict within the trading pair—as the measure includes all types of interstate and international conflict that a pair is involved in— which is unlikely to be strongly affected by their bilateral trade. Nevertheless, we address any endogeneity concerns that may arise for this variable through the inclusion of CPFE and time effects; and by applying the instrumental variable (IV) methodology discussed in the sensitivity analysis.

C. Defining neighbor at war

The most important issue in the estimation of equation (15) is to measure conflict in the host and neighboring economies in an appropriate way. The definition and measurement of armed conflict is a subject of ongoing debate in political science literature, which has led to a proliferation of conflict datasets in recent years. Differences among datasets exist in the type of event coded, the spatial and temporal dimensions, the violence threshold for event inclusion, the timing and duration of the conflict, and the treatment of interstate, intrastate and extra state wars. Some of the prominent conflict datasets often used in academic literature include those of the Correlates of War (COW) project, the Uppsala Conflict Data Program (UCDP), the Center for International Development and Conflict Management and the Center for Systematic Peace (CIDCM-CSP), Fearon and Laitin (2003), and Sambanis (2004).14

In this paper, we define conflict in a contiguous country using the Major Episodes of Political Violence (MEPV) dataset compiled by the CIDCM-CSP.15 The MEPV defines major armed conflicts as those episodes of political violence, which “involve at least 500 ‘directly-related’ fatalities and reach a level of intensity in which political violence is both systematic and sustained (a base rate of 100 ‘directly-related deaths per annum’)”. There are three important advantages of using this dataset: first, it provides information on different types of armed conflict: (i) civil—intrastate involving rival political groups; (ii) ethnic—intrastate involving the state agent and a distinct ethnic group; and (iii) international—involving at least two states.16 Second, it assigns magnitude scores—ranging from 1 (lowest) to 10 (highest)—to each episode based on the scale of the conflict and the available estimates of fatalities. The scores are considered to be comparable across time, countries, and typologies of warfare, and provide readily usable information on the existence as well as the intensity of major armed conflicts.17 Finally, the dataset provides information on conflicts in a large number of countries—about 162 in recent years—over a long time horizon (1948–2006).

To construct our measures for warfare in neighboring countries, we use six different variables of conflict from the MEPV dataset—international violence (intviol), international war (intwar), civil violence (civviol), civil war (civwar), ethnic violence (ethviol), and ethnic war (ethwar)— and proceed in two main steps. First, for each country-year observation, we create variables to indicate conflict intensity and presence. To this end, we combine the magnitude scores for the four measures of intrastate conflict—civviol, civwar, ethviol and ethwar—to reflect societal conflict intensity (civconf), and the two measures for extrastate conflict—intviol and intwar—to indicate international conflict intensity (intconf). We add these two measures to create a measure for overall conflict intensity (conflict), and then use the three composite measures—civconf, intconf, and conflict—to create binary variables that indicate conflict presence. Hence, the variables civconfd, intconfd, and conflictd take the value of one if the country experienced a civil war, international war, or any type of conflict, respectively, in the observation period, and zero otherwise.

Next, we combine these six conflict measures with the geographical contiguity data of Stinnett et al. (2002) to obtain measures for conflict presence and intensity in the neighborhood.18 While neighbors for a specific state could be defined on the basis of linguistic, religious, historical and economic linkages, our theoretical framework rests on the importance of geographical proximity for conflict externalities. In addition, using contiguous states to constitute the neighborhood prevents us from making any subjective judgment errors about what countries to include in a certain region. Thus, following Murdoch and Sandler (2004), we employ two definitions of neighboring states. The first simply defines countries as neighbors if they share a common border. The second builds on the first definition but takes into account the fact that the length of a common border may influence the extent of the spatial transmission of negative externalities from a neighbor’s conflict. It therefore determines the geographical importance of each neighbor based on the length of their common border.19

Using the above definitions, we compute weighted averages of the warfare measures for the neighboring states. Specifically, based on the first definition which assigns equal weights to the n contiguous states, our measures for neighborhood conflict presence and intensity for country j (ji) in time period t are, respectively, given by:

nwardjt=1ni=1n(wardit),and(16)
nwarjt=1ni=1n(warit),(17)

where war = civconf, intconf, and conflict.

With the second definition, which determines the weights of each neighbor by the length of the common border relative to the total length of country j’s border, (16) and (17) are modified and given as

w_nwardjt=i=1nwardit×(borderlength)ijttotalborderjt,and(18)
w_nwarjt=i=1nwarit×(borderlength)ijttotalborderjt.(19)

Thus, (16)(19) give us 12 different measures to describe neighborhood or regional conflict. As the magnitude scores assigned in the MEPV dataset range from 0 to 10, all measures can only take nonnegative values by construction.20

III. Data

A. Neighbor at war

The MEPV dataset documents 1,448 observations for some form of societal conflict, and 330 for major armed international conflict (excluding wars of independence) in about 160 countries for the period 1948–2006. Based on the first definition of contiguity, this gives us 3,589 and 1,070 country-year observations where one or more neighboring states was involved in societal and international conflict, respectively.21 Figure 1 summarizes this information and shows the percentage of countries with at least one neighbor involved in any type of conflict during 1948–2006. It also gives the maximum number of neighbors in conflict at a given point in time, and the distribution of countries according to the total duration of regional conflict. Clearly, the percentage of countries experiencing regional conflict increased sharply in the early 1980s, but declined toward the end of 1990s.22 For countries in the sample which had some form of conflict in a neighbor, about one-half endured it for a total of at least 20 years, and one-fourth for 40 years or more, suggesting a high degree of persistence (and recurrence) in regional conflicts.

Figure 1.
Figure 1.

Conflict in neighboring countries, 1948-2006

Citation: IMF Working Papers 2009, 283; 10.5089/9781451874280.001.A001

Source: Author’s calculations based on CIDCM-CSP MEPV and COW Direct Contiguity databases.* Total number of years a country had conflict in the region during the sample period.

Figures 2 and 3 present the neighborhood conflict intensities as measured by the ncivconf and nintconf measures, respectively, and the total number of neighboring countries involved in a conflict, at different points in time (1950, 1980, and 2006).23 The charts indicate that the prevalence of international warfare in neighboring states has declined since 1950, but armed intrastate conflicts in contiguous regions have increased—both in terms of intensity and the countries involved—particularly in Africa, the Middle East, and South Asia. For example, in 2006, the ten countries with the highest values for societal conflict (ncivconf) are in Africa and Asia; and the lowest in Europe.

Figure 2.
Figure 2.

Societal (intrastate) armed conflict in neighboring states, 1950-2006

Citation: IMF Working Papers 2009, 283; 10.5089/9781451874280.001.A001

Source: Author’s calculations based on CIDCM-CSP Major Episodes of Political Violence and COW Direct Contiguity datasets.
Figure 3.
Figure 3.

International armed conflict in neighboring states, 1950-2006

Citation: IMF Working Papers 2009, 283; 10.5089/9781451874280.001.A001

Source: Author’s calculations based on CIDCM-CSP Major Episodes of Political Violence and COW Direct Contiguity datasets.
Figure 4.
Figure 4.

Conflict and bilateral trade, 1948-2006*

Citation: IMF Working Papers 2009, 283; 10.5089/9781451874280.001.A001

Source: Author’s calculations*Unconditional relationship with the period average of conflict measures and log of bilateral trade for all dyads in the sample.

It is also interesting to note that a large number of otherwise peaceful countries rank relatively high in terms of regional instability due to their geographical closeness to states with some form of conflict. For example, both Finland and Norway appear shaded in the maps due to their proximity with Russia. This observation is supported by Table 2, which shows that the correlation between the presence and intensity of domestic conflict—as measured by civconf, intconf, and conflict—and regional conflicts is low. Thus, in estimating the trade losses emerging from conflict, it may not be sufficient to study domestic warfare only as regional conflict may also have a role to play thereby providing additional information.

B. Dyadic data description

The regional conflict data described above is in country-year form, which is converted to dyad-year format to correspond with the annual bilateral trade data obtained from the IMF’s Direction of Trade Statistics used for estimation purposes. For each dyadic observation, neighborhood conflict is defined as the sum of the conflict measures for both trading partners. Data on other variables used in the analysis—which are also converted into dyadic form—such as real GDP (in 2000 US dollars), real GDP per capita (in 2000 US dollars), population and geographical size are compiled from the World Bank’s World Development Indicators 2007. The polity index, scaled between -10 (lowest) and 10 (highest), has been obtained from the Polity IV Project. The source of information on geographical attributes including length of the borders, colonial ties, and language is the CIA World Factbook 2004 and Rose (2000), while currency unions and free trade agreements have been obtained from Tsangarides et al. (2008) and the Regional Trade Agreements database of the World Trade Organization, respectively.24

The estimated sample is an unbalanced panel dataset covering 145 countries over the period 1948-2006, yielding 9,024 individual country pairs (rather than 145×144/2=10,440 because of missing observations), and 199,912 observations.25 We estimate the gravity model for the world sample as well as for various subsamples to explore whether the role of trade and regional conflicts differs across subsets of countries. We do so by dividing the sample into two income groups—low and lower middle income (low); and upper middle and high income (high)—based on the World Bank’s income classification, and subsequently report the results for four subsamples (world, high-high, high-low, and low-low). Specifically, the first subsample covers all countries for which the required data are available; the second comprises those observations where both trading partners belong to the high income group; the third includes dyads where one partner belongs to the high income and the other to the low income group; and the fourth constitutes those pairs where both countries are in the low income group.26

Table 1 gives the distribution of observations across the different subsamples. About 27 percent of the observations in the sample belong to the upper middle and high income group, but they represent about 80 percent of world trade conducted in the sample period. The low and lower middle income dyads constitute 20 percent of the sample but account for only a fraction of total trade. Trade conducted between high and low income groups is also almost one-fifth of the trade conducted between the high income countries. Conflict appears to be prevalent in the neighborhood of both high and low income dyads—but whereas 64 percent of the high income dyads had some form of societal conflict in the neighboring countries, this percentage is much higher (about 90 percent) for the low income dyads. The neighborhood societal conflict intensity measures of the low income trading pairs also appear to be twice as high as for the high income dyads. While the share of trade conducted between pairs with regional conflict is higher than those without, this likely reflects the widespread existence of conflict rather than a proof of causality.

Table 1.

Correlation between domestic and regional conflicts, 1948-2006

article image
Source: Author’s calculations based on CIDCM-CSP MEPV and COW Direct Contiguity datasets.
Table 2.

Distribution of societal and international conflict in neighbors, 1948-2006

article image
Source: Author’s calculations.

At least one country in the trading pair has neighbor(s) involved in a societal conflict. Measure is defined as the average of the neighborhood magnitude scores for civil violence, civil war, ethnic violence, and ethnic war.

At least one country in the trading pair has neighbor(s) involved in an international conflict. Measure is defined as the average of the neighborhood magnitude scores for international violence, and international war.

Overall conflict measure is the average magnitude score for neighborhood international and societal conflicts.

At least one country in the pair has neighbor(s) involved in some type of conflict (societal, international or both).

Figure 3 plots the (unconditional) relationship between regional conflict measures and (log of) bilateral trade for all dyads averaged over the sample period. The negative relationship between bilateral trade and the intrastate and overall conflict measures is apparent from the plots, but the relationship between international conflicts in neighbor states and trade appears positive from the fitted line. These plots are, however, impressionistic only as they do not take into account other factors that affect trade such as income and geographical attributes. In what follows, we control for the other determinants of trade to formally investigate its link with regional warfare.

IV. Empirical Results

A. World sample

The estimation results for equation (15) for the world sample are presented in Tables 3-5. We investigate the impact of conflict in the neighborhood on bilateral trade by sequentially including the measures for intrastate, international and overall conflict presence and intensity. For completeness and comparative purposes, we estimate the gravity model using all the estimators discussed earlier, namely, pooled OLS, CFE, CPFE, and CYPFE.27 As our dataset pools a large number of country pairs over almost 50 years, the error term is likely to exhibit correlation patterns for given country pairs. To take this into account, we cluster the robust standard errors at the country pair level in all estimations.28

Table 3.

Estimation results for conflict presence (world sample, 1948-2006)

article image
Source: Authors’ calculations.Dependent variable is log of real bilateral trade.Robust clustered standard errors in parentheses; Time effects included in all specifications.***, ** and * indicate significance at the 1%, 5% and 10% significance levels, respectively.

CFE= country-fixed effects included; CPFE= country-pair fixed effects included; CYPFE=country-year and country-pair fixed effects included.

b Dummy variable equal to one if at least one country in the trading pair is involved in any type of conflict.

Sum of average number of neighbors involved in civil and ethnic violence and war (ncivconfd) for the trading pair.

Sum of average number of neighbors in international violence and war (nintconfd) for the trading pair.

Sum of average number of neighbors in any type of conflict (nconflictd) for the trading pair.

Table 4.

Estimation results for conflict intensity (world sample, 1948-2006)

article image
Source: Authors’ calculations.Dependent variable is log of real bilateral trade.Robust clustered standard errors in parentheses; Time effects included in all specifications.***, ** and * indicate significance at the 1%, 5% and 10% significance levels, respectively.

CFE= country-fixed effects included; CPFE= country-pair fixed effects included; CYPFE=country-year and country-pair fixed effects included.

bSum of magnitude scores for all types of conflict (conflict) in the trading pair.

Sum of average of neighborhood magnitude scores for civil and ethnic violence and war (ncivconf) for the trading pair.

Sum of average of neighborhood magnitude scores for international violence and war (nintconf) for the trading pair.

Sum of average of neighborhood magnitude scores for all types of conflict (nconflict) for the trading pair.

Table 5.

Estimation results for border weighted conflict (world sample, 1948-2006)

article image
Source: Author’s calculations.Dependent variable is log of real bilateral trade.Robust clustered standard errors in parentheses; country-pair and time effects included in all specifications.***, ** and * indicate significance at the 1%, 5% and 10% significance levels, respectively.

Sum of magnitude scores for all types of conflict (conflict) in the trading pair and neighbor at war is conflict intensity.

Dummy variable equal to one if at least one country in the pair is involved in any conflict and neighbor at war is conflict presence.

Sum of border length weighted average of neighbor magnitude scores for civil and ethnic violence and war (w_ncivconf) for the pair.

Sum of border length weighted average number of neighbors in civil and ethnic violence and war (w_ncivconfd) for the pair.

Sum of border length weighted average of neighbor magnitude scores for international violence and war (w_nintconf) for the pair.

Sum of border length weighted average number of neighbors in international violence and war (w_nintconfd) for the pair.

Sum of border length weighted average of neighbor magnitude scores for all types of conflict (w_nconflict) for the pair.

Sum of border length weighted average number of neighbors in any type of conflict (w_nconflictd) for the pair.

Table 3 presents the results for conflict presence when all neighboring states are assigned the same average weight in the regional conflict measure as in (16). In the OLS estimation, when only time effects are included along with the traditional gravity variables, the estimated coefficients of the control variables are plausible and in line with earlier studies. For example, larger economies trade more, as do countries that share a common currency, border, language, and free trade agreement, but distance reduces bilateral trade.29 The war variable in this specification is a dummy variable that takes the value of one if at least one of the countries in the dyad is involved in any type of conflict (civil, international or both). As expected, the estimated coefficient for this variable is significantly negative, indicating that conflict involvement of the trading partners disrupts their trade flows.

Importantly, the estimated coefficient of our main variable of interest, neighbor at war—as measured by the societal (ncivconfd) and overall conflict (nconflictd) indicators—supports the main predictions of the model and shows that the presence of regional conflict affects bilateral trade flows negatively. The impact of international conflict (nintconfd) in contiguous states is negative but insignificant in the OLS specification. In addition, we find that good institutions in neighboring countries—proxied by the median polity index score—have a positive, albeit small, impact on bilateral trade. We also estimate equation (15) with the average regional income and income per capita for the trading partners instead of the polity score, but the results remain essentially the same and are not reported here for brevity.

When the CFE are included to the model, the fit of the estimated equation improves. The size of the estimated coefficients for ncivconfd and nconflictd drops slightly but holds significance at the one percent level, and the effect of neighborhood involvement in international conflicts also becomes significantly negative. Controlling for the CPFE and CYPFE, as in the last two columns for each type of conflict measure, we lose the cross-sectional information of the data and all time invariant variables drop from the estimation. However, the estimated coefficients of war and neighbor at war variables remain broadly the same. Bilateral trade flows are estimated to be, on average, about 12 percent (e−0.13-1=0.12) lower if at least one trading partner is involved in conflict. The effect of conflict in the neighbors can be computed by predicting the impact from conflict in an additional neighbor. By definition, the effect of an additional conflict in the neighborhood depends on the estimated coefficient for Nwar and the number of neighbors. The average number of neighbors for a country in our sample is four, implying that the average weight for each neighbor in the bilateral measure is 1/8=0.125. Therefore, the estimated change in bilateral trade flows as a result of an additional conflict in the neighborhood is a reduction of about 2 and 3 percent for intrastate and international conflicts, respectively, as implied by the CPFE and CYPFE models.30

The results for conflict intensity—as measured by the average of the conflict magnitude scores— further support the finding that conflict in contiguous states reduces bilateral trade flows. Specifically, the estimates presented in Table 4 suggest that a one standard deviation increase in the sum of the neighborhood conflict intensity scores for the dyad reduces bilateral trade by about 7 and 3 percent for intrastate and international conflicts, respectively.31 In comparison, a one standard deviation increase in the intensity of domestic conflict in a dyad is associated with about a 9 percent decline in bilateral trade. This estimate is close to that of Blomberg and Hess (2006), who find that a one standard deviation shock to their indicator for violence in dyads—a composite measure for external war, inter-ethnic conflict and genocide, revolution and coups, and terrorism—reduces bilateral trade by 8.4 percent.

In Table 5, we present the results for the border length weighted conflict presence and intensity measures.32 The estimated impact of an additional neighbor at war is computed along the same lines as above, with the only difference that we use the average border length of a neighbor as a weight instead of the number of neighbors. The predicted change in bilateral trade flows as a result of an additional societal and international conflict in the region is the same as before (a drop in bilateral trade flows of 2-3 percent, while the estimates for societal and international conflict intensity imply that increasing the joint score of the dyad by one standard deviation would reduce their trade by about 6 and 4 percent, respectively.

Taken together, the results from our four different types of measures constructed for societal, international and both types of conflict suggest moderate but strong negative spillovers from nearby conflicts. These results hold when controlling for domestic warfare and other regional characteristics such as polity and income, indicating that our findings with respect to conflicts in neighbors are not simply picking up effects of other related phenomenon, but are important in their own right. Further, these estimates present the direct impact of regional warfare only, and in this respect are likely to present a lower bound of its total effect on trade. If, for example, conflict in neighboring countries increases the likelihood of conflict in host economy—as shown by Hegre and Sambanis (2006)—then an indirect effect on trade through domestic conflict would also set in. To get some idea of the magnitude of this additional effect, we use the estimate of Hegre and Sambanis (2006)—which suggests that a neighbor at war increases the host country’s likelihood of civil war onset by about 32 percent. For an economy with the existing probability of civil war at say, for example, 10 percent, this estimate implies an increase in the probability of civil conflict to about 13 percent with a neighbor in conflict—the rise in probability when combined with our estimate of the effect of domestic conflict on bilateral trade (12 percent), suggests an additional reduction in trade by about half a percentage points as a result of the spillover of neighborhood societal warfare to the host economy.

B. Subsamples

To investigate the trade effects of regional conflicts across various income groups, we estimate equation (15) using the different measures of regional conflict for the high-high, high-low and low-low dyads. The results presented in Tables 6 through 9 provide strong evidence that bilateral trade between high-high and high-low dyads is adversely affected by regional as well as domestic conflict. Adding one additional neighbor with some type of conflict to a trading partner in the high-high and high-low dyads is estimated to reduce trade between them by about 4 and 5 percent, respectively. However, the pair is estimated to trade, on average, about 10 percent less if at least one of the trading partners is itself involved in some form of conflict.

Table 6.

Estimation results for conflict presence (subsamples, 1948-2006)

article image
Source: Author’s calculations.Dependent variable is log of real bilateral trade.Robust clustered standard errors in parentheses; country-pair and time effects included in all specifications.***, ** and * indicate significance at the 1%, 5% and 10% significance levels, respectively.

Dummy variable equal to one if at least one country in the trading pair is involved in any type of conflict.

Sum of average number of neighbors involved in civil and ethnic violence and war (ncivconfd) for the trading pair.

Sum of average number of neighbors in international violence and war (nintconfd) for the trading pair.

Sum of average number of neighbors in any type of conflict (nconflictd) for the trading pair.

Table 7.

Estimation results for conflict intensity (subsamples, 1948-2006)

article image
Source: Author’s calculations.Dependent variable is log of real bilateral trade.Robust clustered standard errors in parentheses; country-pair and time effects included in all specifications.***, ** and * indicate significance at the 1%, 5% and 10% significance levels, respectively.

Sum of magnitude scores for all types of conflict (conflict) in the trading pair.

Sum of average of neighborhood magnitude scores for civil and ethnic violence and war (ncivconf) for the trading pair.

Sum of average of neighborhood magnitude scores for international violence and war (nintconf) for the trading pair.

Sum of average of neighborhood magnitude scores for all types of conflict (nconflict) for the trading pair.

Table 8.

Estimation results for border weighted conflict presence (subsamples, 1948-2006)

article image
Source: Author’s calculations.Dependent variable is log of real bilateral trade.Robust clustered standard errors in parentheses; country-pair and time effects included in all specifications.***, ** and * indicate significance at the 1%, 5% and 10% significance levels, respectively.

Dummy variable equal to one if at least one country in the trading pair is involved in any type of conflict.

Sum of border length weighted average number of neighbors involved in civil and ethnic violence and war (w_ncivconfd) for the trading pair.

Sum of border length weighted average number of neighbors in international violence and war (w_nintconfd) for the trading pair.

Sum of border length weighted average number of neighbors in any type of conflict (w_nconflictd) for the trading pair.

Table 9.

Estimation results for border weighted conflict intensity (subsamples, 1948-2006)

article image
Source: Author’s calculations.Dependent variable is log of real bilateral trade.Robust clustered standard errors in parentheses; country-pair and time effects included in all specifications.***, ** and * indicate significance at the 1%, 5% and 10% significance levels, respectively.

Sum of magnitude scores for all types of conflict (conflict) in the trading pair.

Sum of border length weighted average of neighborhood magnitude scores for civil and ethnic violence and war (w_ncivconf) for the trading pair.

Sum of border length weighted average of neighborhood magnitude scores for international violence and war (w_nintconf) for the trading pair.

Sum of border length weighted average of neighborhood magnitude scores for all types of conflict (w_nconflict) for the trading pair.

The impact of neighbors’ and host economy war is much less clear for the low-low dyads. For example, for the measures indicating conflict presence, the estimated coefficients suggest a negligible effect of regional conflicts on bilateral trade, but a strong adverse effect of host country war. This finding is reversed for the measures of conflict intensity where neighbor at war has a much stronger impact than host economy war. The apparent weak relationship is possibly the result of much lower variation in the conflict presence and intensity variables for the low-low subsample. As shown in Table 2, about 90 percent of the observations in the low-low sample reflect regional conflict, making it difficult to draw meaningful conclusions from the estimations. For this reason, we define another sample for the low income group, which includes all trading partners (both high and low) and reestimate equation (15). The results (not reported here for brevity) indicate a strong negative effect of both regional and domestic conflict on trade of about 5 and 12 percent, respectively.

While the results estimated from the world and subsamples consistently indicate a strong negative impact of regional conflicts on bilateral trade, it is also important to assess what our results imply for the change in conflict prevalence and trade between countries. For this purpose, using the CPFE estimates for the world, high income and low income samples, we simulate the effect of changes in societal and international conflict measures—holding everything else constant at the 2006 level—on bilateral trade. Specifically, we answer the counterfactual question that what would be the change in (predicted) trade between countries in 2006, if there are no regional societal and international conflicts, or if they were at the 1950 and 1980 levels. The results of this exercise, graphically shown in Figure 5, indicate that total trade between countries would be about 6 percent higher had there been no regional societal conflict in 2006, and about 2 percent higher had it been at the 1950 level. For international conflicts in neighboring states, the estimated trade would be about 2 and 10 percent higher than if there was no conflict, or if conflict is brought back to the 1950 level when international conflicts were much more widely prevalent than today, respectively.

Figure 5.
Figure 5.

Regional conflict and trade (in percent)*

Citation: IMF Working Papers 2009, 283; 10.5089/9781451874280.001.A001

Source: Author’s calculations.

C. Dynamic effects of neighbors at war

Most conflicts, particularly, intrastate civil and ethnic, tend to last over a long time period. For example, a dyad experiences continuous intrastate conflict in the neighborhood for an average duration of 15 years in our sample. This average is higher—about 21 years—for low-low dyads, and 9 years for high-high dyads. Protracted conflicts are likely to cause greater damage to the transportation infrastructure, lead to further thickening of borders as the risk of conflict spillover increases, and increase the uncertainty faced by traders. These factors imply that the effects of regional conflicts could be dynamic in nature with the duration having important consequences for a dyad’s trading activity. In addition, the effects of conflict could be persistent such that it may take some time for trading activity to resume even after a conflict ends as confidence rebuilds and economic links are reestablished. In order to get a clear picture of the dynamic effects of regional warfare, it is therefore important to investigate whether the effects of a change in the conflict status manifest immediately, or only with time.

To explore the effect of conflict duration on bilateral trade, we include a variable in equation (15) that reflects the number of years that a trading pair has experienced conflict in at least one of its neighboring states. Since the effect of duration may be nonlinear, we also add the quadratic term for this variable to the model. The estimated results for the augmented specification reported in Table 10 indicate that the length a dyad experiences neighborhood conflict matters, particularly for the trade of low income countries. On average, an additional year of warfare in the neighborhood decreases bilateral trade by about 1 percent. The quadratic term for duration is negative and significant, albeit small, suggesting some diminishing returns of regional conflict duration. To see the effect of individual years since the onset of regional conflict we estimate another augmented specification, which includes dummy variables for each year into the conflict. Figure 6 graphically plots the estimated coefficients for the dummy variables, which show that the negative effect of successive years of regional conflict is significant, and increases gradually up to 40 years before exhibiting a declining trend.

Figure 6:
Figure 6:

Impact of regional conflicts on bilateral trade, 1948-2006

Citation: IMF Working Papers 2009, 283; 10.5089/9781451874280.001.A001

Source: Author’s calculations.
Table 10.

Estimation results for regional conflict duration, 1948-2006

article image
Source: Author’s calculations.Robust clustered standard errors in parentheses; CPFE and time effects included in all specifications.***, ** and * indicate significance at the 1%, 5% and 10% significance levels, respectively.

Sum of average of neighborhood magnitude scores of conflict for the trading pair.

Years a dyad has conflict in any contiguous state.

To examine persistence in the trade destructing effects of conflicts, we examine the lag in recovery after the regional conflicts end. For this purpose, we include a dummy variable, nwarend, to equation (15) which is equal to one for the first year after a given type of conflict ends. The results presented in Table 11 reveal interesting differences in recovery from societal and international conflicts in the region. For example, columns (1)-(6) show that for societal conflict, the estimated coefficient for nwarend is positive but not statistically significant. However, the impact of nwarend is significantly negative for international conflicts, suggesting that the adverse effects of such conflicts in the neighborhood remain strong for at least a short while even after the end of the war.

Table 11.

Persistence in regional warfare effects for the world sample, 1948-2006

article image
Source: Author’s calculations.Robust clustered standard errors in parentheses; CPFE and time effects included in all specifications.***, ** and * indicate significance at the 1%, 5% and 10% significance levels, respectively.

Dummy variable equal to one the first year after warfare ended in the neighborhood.