* T A C W HE FTERMATH OF IVIL AR Siyan Chen Norman V. Loayza Marta Reynal-Querol American University The World Bank Universitat Pompeu Fabra , CEPR and CESifo July 2007 Abstract Using an “event-study” methodology, this paper analyzes the aftermath of civil war in a cross-section of countries. It focuses on those experiences where the end of conflict marks the beginning of a relatively lasting peace. The paper considers 41 countries involved in internal wars in the period 1960-2003. In order to provide a comprehensive evaluation of the aftermath of war, the paper considers a host of social areas represented by basic indicators of economic performance, health and education, political development, demographic trends, and conflict and security issues. For each of these indicators, the paper first compares the post- and pre-war situations and then examines their dynamic trends during the post-conflict period. It conducts this analysis both in absolute and relative terms, the latter in relation to control groups of otherwise similar countries. The paper concludes that, even though war has devastating effects and its aftermath can be immensely difficult, when the end of war marks the beginning of lasting peace, recovery and improvement are indeed achieved. JEL Code: 010, 057 Keywords: Civil War * For excellent comments, we are grateful to Ian Bannon, Paul Collier, Ibrahim Elbadawi, Jim de Melo, Nicholas Staines, participants of the Post-Conflict Conferences at Oxford University (November 2005) and the World Bank, and three anonymous referees. Marta Reynal-Querol acknowledges the financial support of the grant SEJ2006-10974/ECON from the Spanish Ministerio de Educación. The views expressed in this paper are those of the authors and do not necessarily reflect those of the World Bank, its Board of Directors, or the countries they represent. The Aftermath of Civil War I. Introduction War has devastating consequences for a country, including death, displacement of people, and destruction of public infrastructure as well as physical and social capital. World Bank (2003), one of the most recent and comprehensive reports, concludes that the economic and social costs of civil wars are not only deep but also persistent, even for years after the end of the conflict. However, when the end of war represents the beginning of lasting peace, there are good reasons to believe that recovery, albeit gradual, is possible. This is what neoclassical models of economic growth and convergence would predict and what the evidence of recovery in Europe (after World War II), Korea, and Vietnam, among others, would seem to indicate. The objective of this paper is to contribute some stylized facts on the evidence regarding the economic, social, and political aftermath of civil wars. The scarce literature that studies the consequences of civil wars has usually focused on the costs during conflict. Few studies analyze the costs of civil war after peace agreements are signed, and we would like to contribute to this literature. Working with a cross-section of countries with well defined pre- and post-war periods, this paper uses an event-study methodology to provide a general evaluation of the aftermath of internal wars along basic economic, social, and political dimensions. Although this paper is mainly descriptive, it gives motivation and evidence on various hypotheses surrounding the consequences of internal wars. It will hopefully induce more specific and analytical research in future work. Brief review of the literature. There is little controversy on the dire effects of civil and international wars. They kill people, destroy infrastructure, weaken institutions, and erode social trust. Moreover, the destruction of infrastructure and institutions leaves the population under conditions that increase the risk of disease, crime, political instability, and further conflict. Collier et al. (2003) provide a review of the literature on the costs of civil war. Collier (1999) finds that during civil war countries tend to grow at 2 2.2 percentage points less than during peace. Using World Health Organization data on 23 major diseases in populations distinguished by gender and age groups, Ghobarah, Huth, and Russett (2003) find that civil war increases substantially the incidence of death and disability produced by contagious diseases. Soares (2005) provides an estimation of the welfare cost of violence in a sample of countries applying a willingness-to-pay approach to account for the health consequences of war. For instance, Soares estimates that the civil conflict in Colombia, by reducing life expectancy at birth by 2.2 years, produces a loss of 9.7% of GDP. Other studies focus on the neighbouring effects of civil war. Murdoch and Sandler (2002 and 2004) show that civil wars reduce growth over an entire region of neighbouring countries. Montalvo and Reynal-Querol (2007) explore the influence of refugees from civil wars on the incidence of malaria in the refugee-receiving countries. They show that for each 1000 refugees there occur between 2000 and 2700 cases of malaria in the refugee-receiving country. The empirical literature on the aftermath of civil and international war is scarcer. It seems to indicate that countries do recover in the post-conflict period to at least their pre-war situations. In a cross-country empirical analysis, Przeworski et al. (2000) finds that post-war economic recovery is rapid. Their results indicate that the average rate of growth during the five years following a war is 5.98 percent. They also find that wars cause more damage under dictatorships than under democracies, but, in contrast, recoveries are faster under dictatorships than under democracies. Barro and Sala-i- Martin (1995) explain post-war recoveries --considering the examples of Japan and Germany following World War II-- arguing that whenever a war destroys a given production factor relatively more than other factors, the rate of return of the latter increases, thus creating the forces of convergence that spur rapid growth. Collier and Hoeffler (2004) provide a systematic empirical analysis of aid and policy reform in the post-conflict growth process. It is based on a comprehensive data set of large civil wars, covering 17 societies during their first decade of post-conflict economic recovery. They find that during the first 3 post-conflict years absorptive capacity is no greater than normal, but that in the rest of the first decade it is approximately double its normal level. They also find that growth is more sensitive to policy in post-conflict societies. 3 Organski and Kugler (1977, 1980) analyze the economic effects of the two World Wars on a sample of mainly European countries. They find that in the “long run” --15 to 20 years-- the effects of war are dissipated in both losers and winners, occurring typically a return to pre-war growth trends. Miguel and Roland (2005) analyze the impact that U.S. bombing on Vietnam had on the country’s subsequent economic development. They compare the heavily bombed districts with the rest and find that U.S bombing did not have a lasting negative impact on poverty rates, consumption levels, infrastructure, literacy, and population density, as measured around 2002. Inferring to other cases, they conclude that local recovery from the damage of war can be achieved if “certain conditions” are met. As mentioned above, we use an event-study methodology. Regarding its application to the study of conflict, there are some important papers that precede our work. Chen and Siems (2004) use it to examine the effects of terrorism on global capital markets. They examine the U.S capital market’s response to 14 terrorist attacks from 1915, and the response of global capital markets to both the Iraq’s invasion of Kuwait in 1990 and the 9/11 terrorist attacks in New York and Washington. They find that terrorist attacks and military invasions have great potential to affect capital markets around the world in a short period of time. They also find that U.S capital markets recover sooner than other global capital markets. Abadie and Gardeazabal (2003) use an event-study methodology to analyze the impact of terrorism on firms in the Basque Country. They find that firms having a significant presence in the Basque Country improved their performance more than other firms when truce became credible and, correspondingly, a worse relative performance at the end of the ceasefire. Davis and Weinstein (2002) consider the Allied bombing of Japanese cities in WWII as a shock to the relative size of the cities. They find that, in the wake of the destruction there was an extremely powerful recovery. Most cities returned to their relative sizes within about 15 years. The paper’s methodology. In this paper, we use an “event-study” methodology to analyze the aftermath of war in a cross-section of countries. We focus on those experiences where the end of conflict marks the beginning of a relatively lasting peace. The event-study methodology consists of transforming calendar time into “event time” in order to be able to aggregate and extract meaningful statistics from a collection of 4 experiences that have a given event in common. In our case, the “event” is the occurrence of civil war, and the pre- and post-war periods are defined as periods free of war. These considerations guide the selection and preparation of the sample. Since our objective is to provide a comprehensive evaluation of the aftermath of war, we examine a host of social areas. These are represented by basic indicators of economic performance, health and education, political development, demographic trends, and conflict and security issues. For each of these indicators, the paper first compares the post- and pre-war situations and then analyzes their dynamic trends during the post- conflict period. The purpose is to examine the nature of the recovery from war, suggesting evidence on the costs of war and the extent of a peace dividend. The comparative analysis is done controlling for country fixed effects and considering the experience of conflict countries both on their own and in comparison with two control groups of otherwise similar countries. Basic conclusions. As result of war, post-conflict countries find themselves behind otherwise similar developing countries in terms of income per capita, some aspects of health and educational achievement, and key areas of political development. Moreover, longer wars produce larger damage in economic activity and make economic recovery significantly slower. However, when peace is achieved and sustained, recovery is indeed possible. Virtually all aspects of economic, social, and political development experience gradual improvement in the aftermath of civil war. Progress in social areas is accompanied by a continuous reallocation of public resources away from military expenditures and, above all, a steady rise in average income per capita. An important caveat on this paper is that it serves only as a broad overview: Its conclusions refer to the typical or average country afflicted by war and reflect mostly a descriptive and statistical examination. Future research should analyze in greater detail the heterogeneity of post- conflict situations, their causal mechanisms, and the policies that make them successful. The rest of the paper proceeds as follows. Section II describes the data, their sources, and methodology of analysis. Section III presents and discusses the results, first, on the comparison between the pre- and post-war periods and, second, on the trends of change after the war. Section IV offers some concluding remarks. 5 II. Data and Methodology In exploring the patterns of behavior of various economic, social, and political variables in post-war countries, this study focuses on internal (or civil) conflicts. The information on conflicts comes from the Armed Conflict Dataset of International Peace Research Institute, Oslo (PRIO). We group internal and internationalized internal armed conflict as internal wars.1 In order to focus on major conflicts, we limit our analysis to those with the highest intensity level in the PRIO dataset, i.e., more than 1000 battle- related deaths per year during the war. In an attempt to provide a comprehensive set of stylized facts on post-war transitions, this paper examines the following dimensions: economic performance, including the level and growth rate of GDP per capita, the share of domestic investment in GDP, the share of government expenditure in GDP, the share of military expenditure in government expenditure, and the inflation rate; health and education, represented by the rates of infant mortality, adult female mortality, adult male mortality, primary school enrollment, and secondary school enrollment; political development, including indices of democracy and autocracy, civil liberties and political rights, and law and order; demographic development, such as the old dependency ratio, young dependency ratio, and female-male ratio; and other forms of conflict, specifically the incidence of terrorist attacks. Detailed description of these variables, including definitions and sources, is provided in Appendix 2. Given its wide-ranging coverage of themes and variables, the paper uses an “event-study” methodology that can produce clear and succinct results. This methodology consists of reorganizing the data by converting calendar time into event time.2 In this particular application, the occurrence of a war is the event that serves to anchor the data. For instance, we define the last year before the start of a war as event year -1, the next-to-the-last year as event year -2, and so on. Similarly, the first year after the end of a war is defined as event year 1, the second year as event year 2, etc. 1 According to PRIO’s definitions, internal armed conflict occurs between the government of a state and internal opposition groups without intervention from other states; internationalized internal armed conflict occurs when such conflict involves intervention from other states. 2 For other presentations of this methodology, see Bruno and Easterly (1998) and Wacziarg and Welch (2003). 6 The definition of the war event is crucial in our empirical evaluation. We define it such that its pre and post periods can be characterized as relatively free of war. In particular, in order to ensure that we analyze the aftermath following the true resolution of an armed conflict, we require at least 10 years of peace after the war. This means that in cases of elongated conflicts with temporary ceasefire periods, our “war event” includes initial war, (short) interwar peace, and resumption of war. In case a country undergoes two wars with more than ten years of peace in between, the two wars are treated as independent events.3 In order to reach a favorable compromise between sample size and period extension, we measure the pre-conflict period as the 7 years before the war, and similarly the post-conflict period as the 7 years after the war. These are the “event” years under consideration. One difficulty in applying the event-study methodology has to do with sample changes across event years. Ideally, we should have a constant sample comprised of the same countries for all event years. Unfortunately, a preliminary assessment of the data reveals that for each individual variable, quite a few countries have data only for a subset of years under consideration. For instance, a country may have GDP per capita growth rate data in the first three years after the war, but no more thereafter. In addition, since our sample period is from 1960 through 2003 and we look at seven years both before and after the war, a country could have started the conflict “too early” (e.g., 1962) or ended the conflict “too late” (e.g., 2000), in the sense that it would not have a well- defined pre-war period in the former case or a well-defined post-war period in the latter case. On the other hand, however, if we do restrict to a perfectly constant sample, we might end up with too few countries included. In order to achieve a balance between the two extremes, we set our criterion in the following way. For the comparison of pre- and post-war periods, a country will be included in the sample for a particular variable, if for this variable it has at least 5 years of observations in the 7-year window before the war and likewise after the war. (Naturally, to be considered in the sample, the country would still have to meet the criterion of being war-free 10 years before and 10 years after the war). For the analysis of the aftermath of conflicts, the data availability restriction is 3 A concern arises when some countries experience external war during the pre- or post-internal conflict periods. In such cases, the periods around the war event cannot be characterized as peaceful. To eliminate this contamination, we exclude these countries from our samples for all variables. 7 imposed only on event years after the war (i.e., a country does not need to have sufficient pre-war data). Our samples are variable specific --it is quite likely, then, that a country meets the requirement for one variable but fails for another. Our empirical analysis studies the typical patterns of countries that experienced civil war, examining, first, the average difference between the post- and pre-war periods and, then, the average rates of change in the years after the war. The analysis is made considering the experience of conflict countries both on their own and with respect to two control groups of countries. The control groups are the full sample of non-conflict developing countries and the subset located in the geographic region of the corresponding conflict country (see below for details). Since some of the variables under consideration may follow world trends (e.g., the wave of democratization in the case of political development variables or the discovery of new vaccines in the case of health indicators), the comparison with the full sample of developing countries is necessary to separate these world trends from the real costs of war and the merits of pacification. The comparison with respect to regional countries is relevant because it can also capture some of these world trends while matching more closely the level of development of corresponding conflict countries. The main disadvantage of the regional control group is that its geographic proximity to conflict countries may make them susceptible to the effects of war. The potential disadvantages of both control groups are reduced by the way we implement the comparisons with respect to them. Considering a given indicator variable, for each conflict country and event year, we measure the control value as the median for the control group in the calendar year corresponding to the event year. Then, we take the difference between the conflict-country value and the control value in a given event year for each variable under consideration. Two series of differences are generated, corresponding, respectively, to the two control groups. Clearly, the control values (and the sample of countries from which they are computed) are specific to each indicator variable under study. Appendix 1 provides summary information on the various country samples. A country is marked with double asterisk if it is included in the samples for both pre- and post-war comparison and post-war analysis. A single asterisk indicates that this country is 8 used only for post-war evaluation. For example, 17 countries are considered in the internal war comparison of GDP per capita growth rate before and after the war; these countries together with other 7 that lack pre-war information are used for evaluation of post-war only. Three variables, i.e., military expenditures, law and order, and terrorist attacks are examined only in the event years after the war due to their lack of available data in the pre-war period.4 Altogether, we work with 41 countries involved in internal wars (15 from Africa, 17 from Asia, 3 from Europe, and 6 from Latin America) during the period 1960-2003. Among these countries, six (Burma, Cambodia, Iraq, Liberia, Sri Lanka and Sudan) were entangled in two internal conflicts. III. Results As mentioned above, we carry out two complementary exercises. In the first, we evaluate and compare the central tendency of each variable before and after the war, both by itself and with respect to two control groups (Table 1). In addition, for the level and growth rate of GDP per capita only, we examine to what extent the duration of the war affects the difference between the pre- and post-war periods (Table 2). In the second exercise, we estimate the average slope (or rate of change) of each variable during the post-conflict period, also by itself and with respect to the two control groups (Table 3). Likewise, we assess whether the duration of the war has an impact on the rate of change of per capita GDP, both in levels and growth rates, in the post-war period (Table 4). To be precise, the following regression equations represent the exercises just described. For the pre-post war comparison, y =α +α *Post +µ +ε (1) i,t 1 2 i,t i i,t y =α +α *Post +α *Dur *Post +µ +ε (2) i,t 1 2 i,t 3 i i,t i i,t 4 For example, WDI started to collect military expenditures data (% as central government expenditures) in 1990; and ICRG provides ratings on law and order after 1984. 9 where the subscripts i and t represent country and event year, respectively; y is the variable under consideration; Post is a dummy variable that takes the value of 1 for the years of the post-war period, and 0 otherwise; µ is a country-specific effect (modeled as a country dummy); α is the main parameter of interest and represents the average 2 difference in the variable y between the post- and pre-war periods; α represents the 3 effect of each additional war year on the post-pre difference; Dur is the duration of the war in number of years; and ε is the regression residual. For the post-war exercise, we use the following regression equations, y =β +β *Year +µ +ε (3) i,t 1 2 t i i,t y =β +β *Year +β *Dur *Year +µ +ε (4) i,t 1 2 t 3 i t i i,t where, Year indicates the event year after the war (1 through 7), β is the main 2 coefficient of interest and represents the average change in the variable of interest from year to year in the post-war period, and β represents the effect of each additional year of 3 war on the post-war average change. The dependent variable, y, is measured by itself and in deviation from the median of each control group. It can then take three values, according to, ⎛ y ⎞ ⎜ i,t ⎟ y =⎜y − y ⎟ i,t i,t i,t ⎜ ~ ⎟ y − y ⎝ i,t i,t ⎠ where y represents the median of the non-conflict developing-country control group i,t ~ associated with country i in year t; and, similarly, y denotes the median of the non- i,t conflict regional-country control group for the same country and year. Given the large number of variables under consideration, Tables 1 and 3 report, respectively, only the estimated α and β coefficients, and associated standard errors, for the three versions 2 3 of each dependent variable. 10