GLOBALIZATION, ECONOMIC SHOCKS, AND INTERNAL ARMED CONFLICT*

*The authors are grateful for financial support from the World Bank project ‘Economics of Civil War, Crime and Violence’ and from the Carnegie Corporation of New York. We are also grateful for comments on earlier versions of the article from DPE’s reviewers and from Aysegul Aydin, Jon Hovi, Ragnar Torvik, Erich Weede, members of Working Group 5 of the Centre for the Study of Civil War, PRIO, and the Conference on Globalization, Territoriality, and Conflict, San Diego, January 2004.

Abstract

Critics of trade liberalization argue that globalization increases countries’ vulnerability to economic shocks and hence may exacerbate domestic social conflict. Such social conflict may also be transformed into armed conflict. Others argue that globalization promotes economic growth and reduces poverty, which leads to a reduction in the risk of internal conflict. Several studies find trade to reduce the risk of interstate conflict. This article investigates the impact of trade and trade shocks on the risk of intrastate conflict. A set of operationalizations of economic shock is developed and used to analyze the risk of conflicts that involve at least 25 battle deaths per year. The analysis finds no robust evidence for a direct relationship between trade openness, trade shocks, and the risk of armed conflict. There is somewhat more basis for concluding that globalization affects the risk indirectly through its effect on long‐ and short‐term growth. In the long run, trade‐induced growth reduces the risk of domestic conflict.

*The authors are grateful for financial support from the World Bank project ‘Economics of Civil War, Crime and Violence’ and from the Carnegie Corporation of New York. We are also grateful for comments on earlier versions of the article from DPE’s reviewers and from Aysegul Aydin, Jon Hovi, Ragnar Torvik, Erich Weede, members of Working Group 5 of the Centre for the Study of Civil War, PRIO, and the Conference on Globalization, Territoriality, and Conflict, San Diego, January 2004.

Keywords:

• Globalization
• Economic Shocks
• Armed Conflict
• Civil War

Notes

*The authors are grateful for financial support from the World Bank project ‘Economics of Civil War, Crime and Violence’ and from the Carnegie Corporation of New York. We are also grateful for comments on earlier versions of the article from DPE’s reviewers and from Aysegul Aydin, Jon Hovi, Ragnar Torvik, Erich Weede, members of Working Group 5 of the Centre for the Study of Civil War, PRIO, and the Conference on Globalization, Territoriality, and Conflict, San Diego, January 2004.

¹In the last decade, it has also often been noted that globalization increases the ease with which rebel organizations and terrorists may purchase and move around arms, illicit goods such as drugs or diamonds, and even the rebels or terrorists themselves. Li and Schaub (2004), however, do not find evidence for these claims.

²Signatory bonuses paid by international oil companies to members of the Angolan government are of a magnitude no domestic actor would be able to pay. According to MacMillan (2004: 158), these lie between 100 and 800 million US dollars per drilling tract.

³Rodrik is not very explicit as to the nature of these distortions, beyond referring to ‘foreign exchange bottlenecks, import compression, debt crises, and bouts of high inflation’ (Rodrik, 1999: 386). These economic ills are particularly severe where institutions are relatively non‐developed, and policies are characterized by myopicness, macrofinancial expansion that is unsustainable in the long run, and where Dutch‐disease effects may follow.

⁴Only data constraints inhibit specifying even more lags. With seven lags, however, this component of the model requires 15 degrees of freedom, and the first 7 years of each country’s data cannot be used. Given the low number of armed conflicts in our dataset, estimating a model with more than seven lags would be impracticable. As shown below, even three lags put severe strains on the data available.

⁵Again, data constraints are the only reasons for stopping at 7 years. We lose about 25% of the observations when we go 7 years back in time. Beyond this, we consider the chance of obtaining significant results to be very slim.

⁶The last aspect has an indeterminate effect on the openness measure, since it also reduces GDP – the denominator of the measure.

⁷The remoteness variable in Wei’s measure is , where the weight w_j is country j’s share of world trade; , and D_ij is the great‐circle distance between i and j’s capitals. W_i for the first year in each four‐year period is used. Wei’s remoteness concept is a bit strange, since it penalizes countries that are far from a large trading block even if they are close to another. If US trade were to collapse dramatically, this does not make Mauritius any less remote just because the w_j,t associated with the USA decreases. Wei’s measure reflects such a change as a decrease in remoteness, however. But, in reality, a US collapse would serve to render Mauritius even more remote when compared to a pre‐collapse world. Our closeness term avoids this problem, since D_ij discounts distant countries’ impact on natural openness rather than highlighting them.

⁸In practice, there is much more variation in countries’ total production G_j and distance D_ij between countries than in actual openness O_j . This simplification therefore introduces only marginal error into the measure, and simplifies data construction considerably.

⁹The World Development Indicators do not include data for West Germany for 1970 and 1971. The exports and imports data for 1972 were used for Germany as a proxy. This is necessary, since the existence of West Germany strongly affects the remoteness/closeness of all European countries.

¹⁰See Hegre and Raleigh (2006) for an extensive discussion of the relationship between population size and the risk of conflict.

¹¹See Hegre (2003) for a version of the model with more control variables. Since the economic shock variables are missing values for several observations, the variables ethnic dominance, mountainous terrain, proximity of regime change, and in conflict were removed from the analysis to maximize sample size. None of these variables are significant in the sample used here nor show any evidence of biasing the results.

¹²For a fuller discussion of these variables, see Hegre et al. (2001), Sambanis (2002), Fearon and Laitin (2003), Collier and Hoeffler (2004), and Hegre and Sambanis (2006).

¹³The natural‐openness measure is based on log(trade/gdp) and therefore has a much smaller range than the trade‐openness measure. Hence, the large estimate for openness in Model 2‐4 does not reflect any substantial difference between the original and the instrumented variable.

¹⁴We excluded the observations for Russia 1991–2000. The terms‐of‐trade dataset reports a four‐fold increase in terms of trade from 1991 to 1992, followed by a 35% reduction from 1992 to 1993. We believe there are good reasons to be skeptical towards data for Russia in the year following the fall of the Soviet Union for reasons that have nothing to do with conflict. Since Russia experienced two Uppsala conflicts in the following years (the bombardment of the Duma in 1993 and the onset of the Chechnyan war in 1994), this anomalous observation is sufficient to render the shock variable positive and significant when included in some models. The estimates for the shock variables in Table II are not significant even with Russia included.

¹⁵That is, the variance in the three log differences from t−3 to t−2 from t−2 to t−1, and from t−1 to t.

¹⁶We also estimated a model where we included seven years of lagged change variables. No estimate beyond t−3 were statistically significant, however.

¹⁷The likelihood ratio test is based on a re‐estimation of the model without the Huber–White standard errors.

¹⁸The risk in the year of such a shock relative to the year before for a country with openness of 100% is calculated as exp(−0.3(1.64+100(−0.038)))=1.91.

¹⁹This discussion ignores that the onset of conflict itself also alters the risk of (renewed) conflict (cf. Collier et al., 2003, ch. 4).

²⁰The growth variable is measured as the difference in log income from year t−2 to year t−1. For values close to zero, this is roughly similar to growth measured as proportional (percent/100) change from year t−2 to year t−1.