I. Introduction

Civil conflicts have catastrophic consequences, including death, displacement of people and destruction of institutional, physical, and social capital. Since the Second World War, intra-state armed conflicts have become the most common type of aggression across the world, according the UCDP/PRIO Armed Conflict Dataset (Figure 1).¹ It is therefore unsurprising to witness the emergence of interdisciplinary research looking into the causes and consequences of civil wars. The theoretical literature has focused on the pivotal question of why conflicts occur in spite of the prohibitively high cost of rebellion, and draws attention to explanations like information asymmetries and the absence of the rule of law (Grossman, 1991; Skaperdas, 1992; Hirshleifer, 1995; Sandler, 2000; Sambanis, 2001; Fearon and Laitin, 2003; Collier and Hoeffler, 2004a). On the empirical side, a burgeoning range of studies has aimed to decipher the causes of conflicts that have persistent and damaging effects on economic development, mainly through the destruction of institutional and physical infrastructure, the loss of human capital, and heightened uncertainty making investment riskier (Collier, 1999; Nafzinger, Stewart, and Vayrynen, 2000; Addison, Chowdhury, and Murshed, 2002; Collier and Hoeffler, 2002; Elbadawi, Hegre, and Milante, 2008). Despite the econometric problems plaguing much of the empirical work, studies tend to link the incidence of civil conflict to low levels of per capita income, slow economic growth, and geographic conditions assisting insurgents.² Additionally, while Fearon and Laitin (2003) and Collier and Hoeffler (2004a) argue that vertical income inequality within a country does not increase the risk of conflict, others including Robinson (2001) and Esteban and Ray (2008) show that greater horizontal inequality in the distribution of income could lead to the onset of intra-state wars, particularly in ethnically fragmented societies.

View Full Size

Civil Conflicts across the World, 1960 – 2010

Citation: IMF Working Papers 2013, 002; 10.5089/9781475531152.001.A001

Source: UCDP/PRIO.

• Download Figure
• Download figure as PowerPoint slide

This paper provides a broad empirical analysis of the determinants of post-conflict economic transitions during the period 1960–2010. Countries emerging from a conflict struggle to break the curse of Sisyphus, with 20 percent relapsing into a subsequent conflict in the first year and 40 percent within five years after the end of conflict, primarily because of the economic legacy of conflict and because the factors that trigger violence in the first place—such as dysfunctional institutions and social grievances—remain unresolved.³

Accordingly, a better understanding of the causes and consequences of intra-state wars remains an important objective in the quest to achieve development and reconciliation, especially in view of the recent wave of popular uprisings across the Middle East and North Africa and long-lasting civil conflicts elsewhere. Though we explore the causes of intra-state wars in modeling the risk of conflict recurrence, our main focus in this paper is on the empirical patterns of post-conflict economic performance by analyzing 146 cases of civil conflict in 94 countries across Africa, the Americas, Asia, Europe, and the Middle East during the period 1960–2010.

Our empirical model captures post-conflict dynamics through an array of variables including an estimated risk of conflict recurrence. From a methodological perspective, we adopt a dynamic panel estimation approach based on the system-generalized method of moments (GMM) that uses pooled cross-country and time-series data and corrects for biases associated with the joint endogeneity of the explanatory variables as well as the problems caused by unobserved country-specific effects. We also use annual data, instead of the five-year averages commonly used in the literature, to fully exploit the time-series dimension of the data. This additional layer of information, provided by the over-time variation in dependent variables and underlying determinants, allows us to obtain more precise estimates. Our empirical model captures the dynamics of post-conflict economic transitions through a core set of explanatory variables including the initial level of per capita income, duration and intensity of the conflict, demographic conditions, measures of human capital, geographic variables and natural resource endowments, macroeconomic conditions, external factors, and measures of political regime and governance quality. Furthermore, because post-conflict countries have a tendency to relapse into subsequent conflicts even years after the cessation of violence, we introduce an estimated measure of the risk of conflict recurrence as an explanatory variable in our econometric analysis of post-conflict growth dynamics.

An important contribution of this paper is the estimation of a forward-looking measure of conflict recurrence risk. The strength and sustainability of post-conflict recovery depends considerably on economic agents’ perception of the stability of peace, because post-conflict countries have a tendency to relapse into subsequent conflicts and the risk of a new bout of conflict inhibits economic growth. Therefore, it is necessary to estimate a measure of conflict recurrence risk and explore its explanatory power in the growth regression. Using a logistic regression approach, we model the probability of conflict recurrence after the end of conflict with the latest available information in each period. First, we find that there is a high degree of persistence in conflict recurrence risk in countries with a history of civil conflict. Second, we identify that the estimated probability of conflict recurrence has a statistically significant and economically large negative effect on the post-conflict growth experience across all specifications of the model.

The empirical results show that domestic as well as international factors contribute to post-conflict economic performance. According to our dataset of 146 cases of internal armed conflict in 94 countries during the period 1960–2010, real GDP per capita growth barely accelerates after the end of a conflict, with an average of 1.5 percent a year over the subsequent five years, and shows significant cross-country variation, ranging from -10 percent in Liberia to 20 percent in Bosnia and Herzegovina. We find a statistically significant negative coefficient on the initial level of real GDP per capita, confirming conditional convergence after the end of conflict. Although the duration of conflict is statistically significant with a negative impact on post-conflict per capita income growth, we could not identify the effects of conflict intensity or the depth of economic contraction during the conflict. Human capital and institutional factors, however, appear to contribute positively to post-conflict growth, while resource-dependent countries suffer from a double curse of higher risk of conflict recurrence and lower post-conflict growth. We also ascertain landlocked status as a statistically and economically significant factor; however, other geographical variables such as the variability in climatic conditions, do not appear to be important. Ethnic fractionalization seems to have a statistically significant negative effect on post-conflict growth, even when we control for the risk of conflict recurrence. We find that global real GDP growth and net foreign direct investment (FDI) flows are statistically significant factors, with economically large effects on the post-conflict growth process. Contrary to popular expectations, however, the results show that UN peacekeeping operations, foreign aid, and IMF-supported economic programs do not appear to have a significant influence on post-conflict growth in our sample of countries.

Growth-enhancing policies and institutional reforms are necessary to improve economic growth prospects after the end of conflict. The empirical findings presented in this paper suggest a strategy for reconciliation and recovery centered on three main pillars—implementing growth-enhancing policies, reforming dysfunctional institutions, and addressing urgent social needs—to reduce the risk of conflict recurrence and pave the way to broad-based, sustainable growth in real GDP per capita after the end of a conflict. Sound macroeconomic management helps to sustain the pace of economic recovery and prevent a relapse into conflict; strengthening institutional capacity is necessary for addressing longstanding grievances, reducing the risk of future civil conflicts, and raising the economy’s growth potential over the medium term.

The remainder of this paper is organized as follows. Section II provides a brief survey of the growth literature, with a view to identifying the key factors that contribute to the strength of post-conflict performance. Section III covers data-related issues, followed in Section IV by an explanation of the theoretical model and econometric methodologies used in this paper. Section V presents the empirical findings, while the robustness of the results is discussed in Section VI. Concluding remarks are offered in Section VII.

II. A Brief Survey of Growth and Conflict Literature

Deciphering the determinants of cross-country differences in growth remains one of the most important pursuits in economics. The neoclassical growth models, developed by Ramsey (1928), Solow (1956), Swan (1956), Cass (1965), and Koopmans (1965), predict that the steady-state level of income is determined by the existing technology and the exogenous rates of saving, population growth, and technological progress. Accordingly, low-income countries with lower stocks of physical capital are expected to build up a capital base at a faster rate and converge towards the income level of high-income countries. Although the absolute convergence and the steady-state equilibrium envisaged in the neoclassical growth theory hinge on the principle that countries are comparable in every respect except for the initial level of physical capital and income, empirical studies indicate that convergence is actually conditional upon other steady-state parameters including saving and investment rates, human capital, government policies, population growth, fertility preferences, and the state of technology.⁴ Therefore, the hypothesis of conditional convergence implies that countries with different steady-state parameters are highly likely to converge to different steady-state levels.

Empirical dissatisfaction with the neoclassical growth theory led to the emergence of alternative models of economic growth. Challenging the neoclassical assumption of exogenously determined technology, the new growth models—an augmented version of the standard neoclassical approach—imply that technological progress is endogenous and that physical capital is not the only kind of capital important for economic growth; they hold that the definition of capital should also include human capital as well as intangible capital such as knowledge. Accordingly, the “endogenous” growth theory explains cross-country differences in growth with differences in resources devoted to research and development (Grossman and Helpman, 1989) and the initial stocks of human capital (Lucas, 1988; Becker, Murphy, and Tamura, 1990; Rebelo, 1991) and human capital specifically devoted to research and development (Romer, 1986). In fact, Barro (1991) shows that the growth rate of real GDP per capita is positively related to the initial level of human capital, while Levine and Zervos (1993) and Sachs and Warner (1997) find that an increase in human capital results in faster transitional growth, mainly through an improvement in productivity.

Institutional characteristics contribute to the variations in cross-country economic performance. While differences in technology and physical and human capital certainly matter for economic performance, cross-country income disparities can only partially be explained by the variation in factor endowments (Lucas, 1990). Over the past several decades, alternative theories on potential sources of cross-country income differences have started incorporating factors such as history and institutions that have been ignored by the standard neoclassical growth model. Institutions signify various characteristics of the socioeconomic and political organization of a country and form an underlying infrastructure of incentives—the “rules of the game” in economic, political and social interactions—that may hinder or facilitate economic activity. As argued by Coase in a seminal paper (1937) and expanded by North (1990; 1991) and Olson (1993), institutions and the way they are structured provide the basic guidelines for economic interactions and help explain why some countries are more developed than others. Several empirical studies have shown a significant relationship between the quality of institutions and overall economic performance, and have argued that weak institutions interfere with economic growth by encouraging economic agents to participate in redistributive endeavors rather than productive economic activity.⁵

Political institutions influence economic outcomes, mainly through the process of institutional change and economic reforms. One of the most important institutions that underscore every transaction is the political regime. Although the empirical evidence on the economic impact of political institutions remains mixed, the democratic transition appears to be associated with industrialization and modernization over the long run. Barro (1996) finds a nonlinear effect of democracy, with an initial increase in growth and a negative relation once some level of democracy has been reached. Tavares and Wacziarg (2001) uncover a positive effect of democracy on human capital accumulation and a negative effect on physical capital accumulation, while Rodrik and Wacziarg (2005) identify a positive short-term effect of democratization on economic growth, but no effect over a longer horizon and no effect when democracy is sustained for long periods. On the other hand, Murtin and Wacziarg (2011) suggest that the level of per capita income is more likely to be a determinant of the level of democracy than the other way around, once income persistence is controlled for in the empirical estimation, and that the level of democracy has not been a significant determinant of economic growth over the course of the past century. Nonetheless, democracy may still influence economic performance through the process of institutional change and economic reforms. Using a dataset on economic reforms for 150 countries over the period 1960–2004, Giuliano, Mishra, and Spilimbergo (2012) show that democracy has a positive and significant impact on the adoption of economic reforms in the financial and banking sectors, and in product markets, agriculture, and trade. There is also a large literature on the relationship between political institutions and the incidence of civil conflict, with findings similar to the relationship between political institutions and economic growth. Among others, Sambanis (2001), Hegre et al. (2001) and Reynal-Querol (2002) find that partial democracies are more prone to civil conflicts than full democracies and autocracies.⁶

Geographical circumstances can affect economic growth, both directly and indirectly, through various channels. While the quality of institutional arrangements is among the fundamental determinants of cross-country differences in macroeconomic performance, institutions may well be shaped by a broad range of factors including geographic conditions, demographic composition, and the structure of factor endowments. A number of studies led by Hall and Jones (1997), Gallup, Sachs, and Mellinger (1998), and Sachs and Warner (1998) take into account geographic variables and find that countries located in the tropics tend to grow at a slower pace than those in more temperate climates. The variability of climatic conditions such as temperature and precipitation appears to have significant effects on the growth rate of real GDP per capita and thereby on the incidence of intra-state conflicts (Miguel, Satyanath, and Sergenti, 2004; Hendrix and Glaser, 2007). On the other hand, other scholars argue that the impact of geographic factors on economic development comes indirectly through historical conditions (Diamond, 1997; Olsson and Hibbs, 2005) or through the legacy of institutions (Engerman and Sokoloff, 1997; Acemoglu, Johnson, and Robinson, 2001; Easterly and Levine, 2003).

Demographic conditions, including ethnic fractionalization, may well affect economic growth through various channels. The size and composition of population are significant factors in determining human capital formation and overall productivity by influencing savings, investments, and technological change, while population density and the extent of urbanization potentially support greater economies of scale. As a result, changes in the demographic structure tend to have substantial effects on economic performance over time. While a young and growing population provides a demographic dividend with far-reaching spillovers throughout the economy, a youth bulge could also become a threat to stability and a source of civil unrest (Urdal, 2004). Ethnic fractionalization is also considered to be an important factor in explaining the cross-country differences in economic growth as well as the variation in social tensions that tend to heighten the risk of civil conflict. While Fearon and Laitin (2003) argue that ethnic—or religious—heterogeneity has no bearing on the incidence of civil conflict, after controlling for per capita income, Easterly and Levine (1997) find that, at least in the case of sub-Saharan Africa, a higher degree of ethnic fractionalization explains a significant amount of the variation in growth rates through political instability, underdeveloped financial systems, and distorted foreign exchange markets. Likewise, utilizing a dataset of 138 countries over the period 1960–2008, Esteban, Mayoral, and Ray (2012) show that measures of ethnic and social fragmentation are significant correlates of the occurrence of civil conflict. As shown by Easterly (2001), however, institutions hold a critical role in determining the impact of ethnic polarization on economic growth and the risk of intra-state conflict. On the whole, we reason that the growth experience, regardless of conflict incidence, depends on a complex interaction of geography, institutions, policies, factor endowments, and developments beyond the control of a country.

External factors such as terms of trade, global demand, remittances, and foreign capital and aid flows can also be influential. Global growth is expected to have a considerable effect on domestic economic activity, especially in natural resource-dependent countries.⁷ Another important factor is the terms of trade, which measure the price of a country’s exports relative to its imports. As noted by Mendoza (1995), changes in the terms of trade affect export revenues and overall economic growth. Their impact could be particularly magnified in resource-dependent economies, resulting in greater volatility in output and the real exchange rate. Likewise, workers’ remittances and foreign capital flows can be a significant source of foreign exchange and influence the evolution of economic growth; the extent will depend on the size of the population that has migrated abroad and the magnitude of inflows relative to the domestic economy. While workers’ remittances are typically stable over time and do not result in claims on assets or other contractual obligations, certain types of capital flows, such as FDI, can set the stage for knowledge and technology transfer and, in so doing, help expand the economy’s production frontier. The impact of foreign aid on growth, on the other hand, is more controversial. While some studies, such as Hansen and Tarp (2000) and Dalgaard, Hansen, and Tarp (2004), identify a positive effect, others find a negative impact of foreign aid on growth, once the quality of institutions and policies is controlled for (Fayissa and El-Kaissy, 1999; Burnside and Dollar, 2000).⁸

The structure of factor endowments matters for the risk of civil conflict and post-conflict growth dynamics. Hausmann, Hwang, and Rodrik (2007), among others, show that countries dependent on natural resource exports experience, on average, slower grow than more diversified economies do. This so-called “resource curse” acts through the volatility of commodity prices and the “Dutch disease” that tends to occur when a commodity boom leads to appreciation of the real exchange rate and undermines non-resource sectors (Bruno and Sachs, 1982; Corden and Neary, 1982; Sachs and Warner, 1995). Resource dependence, particularly on hydrocarbons and precious stones, has also been linked to the risk of intrastate conflict through the opportunity cost of rebellion and the value of the state prize (Collier and Hoeffler (2002; 2004)), for example, argue that civil conflict is associated with low opportunity cost of rebellion.⁹ The most convincing evidence in favor of the opportunity cost argument comes from the analysis of the interaction between economic shocks and the incidence of civil conflict. Miguel, Satyanath, and Sergenti (2004) show that rainfall variability in sub-Saharan Africa affects income growth and the opportunity cost of conflict, while Sambanis (2005) suggests that high unemployment contributes to the onset and duration of civil conflict by lowering the opportunity cost of rebellion. The state prize theory, on the other hand, predicts that the conflict risk rises with the value of the prize (Besley and Persson, 2011), with the strongest evidence in favor of the “state prize” argument coming from the historical analysis of Africa (Reno, 1998; Bates, 2008) and from studies showing a link between natural resource wealth and civil conflicts (Fearon, 2005; Ross, 2006).¹⁰

Post-conflict rebound in economic growth shows great variation and is not typically a sustainable phenomenon. According to our dataset of 146 cases of internal armed conflict in 94 countries during the period 1960–2010, real GDP per capita growth accelerates after the end of a conflict, with an average of 1.5 percent a year over the subsequent five years. This is a very weak performance relative to the rest of the world economy and especially considering the low level of initial income and the depth of contraction during the conflict. Furthermore, as illustrated in Figure 2, the distribution of post-conflict growth rates is spread out, ranging from a five-year average of -10 percent per year in Liberia to 20 percent per year in Bosnia and Herzegovina. The empirical regularities of post-conflict economic transitions identified in this paper are broadly consistent with other studies. For example, analyzing growth performance in 62 post-conflict countries from 1974 to 1997, Collier and Hoeffler (2004b) find that the economic recovery tends to be marked by a rebound in growth in the first three years followed by an “above normal” average growth rate over the next four to seven years if peace is maintained and the security situation improved. However, since the initial recovery stage is simply a spurt of growth, with the onset of peace initiating a mean reversion in economic activity supported by foreign aid flows and reconstruction projects, this growth rebound is not typically a sustainable phenomenon. Likewise, studying sub-Saharan African countries, David, Bastos, and Mills (2011) uncover that changes in the terms of trade and the quality of institutions are the most important determinants of post-conflict economic growth, while trade openness and international aid do not have statistically meaningful effects.

View Full Size

Distribution of Post-Conflict Growth Rates, 1960 – 2010

Citation: IMF Working Papers 2013, 002; 10.5089/9781475531152.001.A001

Source: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

III. Data Overview

We assemble a panel dataset of 146 cases of civil conflict in 94 countries spanning the years from 1960 to 2010. To investigate the empirical patterns of post-conflict economic transitions, we gauge the strength of growth by the annual rate of growth in real GDP per capita and use a comprehensive set of conflict data—encompassing 146 cases of civil conflict in 94 countries across Africa, the Americas, Asia, Europe, and the Middle East—extracted from the latest version of the UCDP/PRIO Armed Conflict Dataset (Appendix Table 1).¹¹ The UCDP/PRIO dataset characterizes four different types of conflicts according to location and participants: (1) extra-systemic armed conflicts, (2) inter-state armed conflicts, (3) intra-state armed conflicts, and (4) internationalized intra-state armed conflicts. In this paper, we concentrate on episodes of internal armed conflict including all three categories based on the level of intensity: low-intensity conflict with at least 25 battle-related deaths a year and fewer than 1,000 battle-related deaths during the course of the conflict; cumulatively high-intensity conflict with an accumulated total of at least 1,000 battle-related deaths, but fewer than 1,000 in any given year; and high intensity conflict with at least 1,000 battle-related deaths a year. Although low-intensity conflicts may not necessarily have a measurable impact on economic activity during and after the conflict, we still include such episodes, excluding terrorist attacks, in the sample to develop a broader understanding.

We use annual observations, instead of five-year averages, to avoid a loss of valuable information. Most of the empirical literature uses five-year averages, with the aim of removing business cycle fluctuations. Although year-to-year variations in data may include cyclical components, averaging does not necessarily eliminate the cyclical effect. Moreover, as argued by Aguiar and Gopinath (2007), cycles tend to be the trend in developing countries. That is, in contrast to business cycles in advanced economies, some shocks tend to have a permanent impact in developing countries and especially in post-conflict episodes during which an economy is subject to various structural breaks. Moreover, uncertainty is better measured over the business cycle and accordingly, using five-year averages could underestimate the importance of volatility, particularly in resource-dependent countries. Additionally, because of the frequent occurrence of conflicts, five-year averages may include developments during the conflict period and may consequently fail to represent the true extent and magnitude of post-conflict performance. In contrast, if averaging only applies to non-conflict periods, then spell durations become endogenous because the post-conflict risk can potentially decrease the non-conflict years. In other words, these averages over different spells are not comparable; and averaging growth rates over time would lead to misleading results or, at least, reduces the dimension of data. Accordingly, we use a set of annual observations that allows for a more complete empirical analysis with more robust results.

In testing our model of economic transitions, we include in the sample only countries with a history of conflict. In the construction of our dataset, we adopt a conservative strategy and limit the sample to conflict episodes, since post-conflict economic dynamics reflect specific factors including an unprecedented destruction of human and physical capital that tends to lead to an unbalanced growth path. Similarly, institutions after a conflict behave differently than they did before the conflict (or in countries with no history of intra-state conflict) with a corresponding effect on growth dynamics. Another important reason to focus on countries with a history of conflict is related to certain variables, such as the duration and intensity of conflict and the extent of destruction, which must be included in the regression as explanatory variables, and which do not exist for a country with no history of an internal conflict. Accordingly, we construct an unbalanced panel over the period 1960–2010, using annual data, in such a way that the cross-sectional dimension is defined by country and that the time dimension is defined by years.¹²

We make use of a host of variables, including the risk of conflict recurrence, to explain the strength of post-conflict economic growth. Economic performance depends on a multitude of interlinked factors ranging from the stocks of human and physical capital, natural resources and geography to institutions, politics, and developmental stage, but we need to keep the construction of the dataset in accordance with the growth literature as well as our objective of working with a panel data set covering 94 countries over the period 1960–2010. Accordingly, our empirical model considers a broad range of variables: these include the initial level of real GDP per capita at the end of the conflict; duration and intensity of the conflict as defined by the UCDP/PRIO dataset, but scaled to account for the population size; human capital as measured by educational attainments; demographic variables such as population density, the size of the youth bulge, and ethnic fractionalization; geographic variables and climatic conditions; natural resource dependence; terms of trade; global real GDP growth; degree of trade openness; international factors such as foreign aid, IMF-supported economic programs, and UN peacekeeping operations; and institutional variables.¹³ In addition, because post-conflict countries have a tendency to relapse into subsequent conflicts, we introduce an estimated measure of the forward-looking risk of conflict recurrence as an explanatory variable in our econometric estimations.

We test the unit root or stationarity properties of our panel dataset and find no evidence of non-stationarity. The majority of the tests for unit root or stationarity in panel data that are widely used in the empirical literature assume a balanced panel dataset (Levin, Lin, and Chu, 2002; and Im, Pesaran, and Shin, 2003). In our study, however, the panel dataset of post-conflict growth episodes is unbalanced because of frequent occurrences of civil conflict, making most of these tests for unit root in panel data inapplicable. One prominent exception is the approach proposed by Choi (2001) that introduce a Fisher-type unit root test for even unbalanced panel datasets. The results, presented in Appendix Table 6, show that there is no evidence for non-stationarity in our panel dataset.

IV. Theoretical Framework and Empirical Methodology

We adopt an augmented version of the neoclassical growth model to capture the impact of human capital, institutions, and the risk of conflict recurrence. As a basis for estimating the determinants of economic growth, this paper follows the theoretical framework extended by Mankiw, Romer, and Weil (1992) to incorporate human capital into the neoclassical growth model:

where an economy’s output (Y) is produced by the stock of physical capital (K), the stock of human capital (H), and the stock of labor (L), with labor-augmenting technology (A) growing exogenously at the rate of g, and α, β > 0 and α + β < 1. The effect of institutions and other potential factors on economic growth can also be incorporated by expressing the production model as a function of:

where Z denotes the influence of a vector of potential determinants such as institutions, geography, and policy variables on the productivity of inputs.

B. Estimating the Risk of Conflict Recurrence

We estimate the probability of conflict recurrence and use it as an explanatory variable in our growth regressions. The strength of post-conflict recovery depends considerably on economic agents’ perception of the stability of peace after the end of violence. Therefore, since the risk of plunging into a new bout of conflict deters investment spending and inhibits overall growth in the economy, it is necessary to include a measure of forward-looking risk assessment as an explanatory variable in estimating the determinants of post-conflict macroeconomic performance. However, quantifying the relationship between the probability of conflict recurrence and the strength of income growth remains a challenge. There are different modeling approaches in the early warning/leading indicators literature; these include linear regression and limited dependent variable probit/logit techniques developed to test the statistical significance of various indicators in determining the incidence of an event risk.¹⁵ While these statistical tools have been extensively applied in the empirical literature to estimate the probability of currency/financial crisis, debt/loan default or economic recession, we adopt a similar approach to estimate the risk of conflict recurrence.

The risk of an event at time t + 1 can be measured by estimating its distribution given the available information at time t. Measuring the risk of conflict recurrence over the subsequent years by using proxy variables would yield biased estimates mainly because there is no good proxy to convey forward-looking risk expectations, especially in many developing and low-income countries that have experienced episodes of civil conflict. Therefore, we model the perceived risk of conflict recurrence with the latest available information in each period by imposing more structure on a general model of conflict and defining risk as the probability of conflict recurrence in the next period. Assuming conflict is a latent variable, the model of hostility in country i at time t (h_it) becomes as a function of:

$\begin{array}{cc}\left(6\right)& h_{\mathit{it}}=ɸX_{\mathit{it}}+v_i+{ϵ}_{\mathit{it}}\end{array}$

where X_it is a set of independent variables contributing to the level of enmity at time t in country i and v_i represents fixed effects for country i. Although the level of hostility is unobservable, we can still estimate Equation (7) since conflict incidence is an observable variable, assuming that it happens when the level of hostility passes a certain threshold:¹⁶

$\begin{array}{cc}\left(7\right)& {\mathit{cr}}_{\mathit{it}}=1\mathit{if}{\mathit{h}}_{\mathit{it}}>0\mathrm{and}{\mathit{cr}}_{\mathit{it}}=0\mathit{if}{\mathit{h}}_{\mathit{it}}<0\end{array}$

If ϵ has a distribution of Δ, a conflict occurs with the probability of 1 – Δ(ɸX_it + v_i). These binary observations provide enough variation to estimate β in Equation (6)—a well-established exercise in the literature for characterizing the causes of civil conflicts. At the same time, the forward-looking probability of conflict recurrence can be defined as:

$\begin{array}{cc}\left(8\right)& \mathrm{E}\left[r_{\mathit{it}}\right]=\mathrm{probability}({\mathit{cr}}_{i,t+1}=1|X_{\mathit{it}};ɸ,v_i)\end{array}$

Based on this framework, the perceived risk of conflict recurrence can be calculated as:

$\begin{array}{cc}\left(9\right)& \begin{array}{ccc}\mathrm{E}\left[r_{\mathit{it}}\right]& =& \int P({\mathit{cr}}_{i,t+1}=1|X_{i,t+1},{ϵ}_{i,t+1};v_i)P\left(X_{i,t+1}\right|X_{i,t})\mathrm{\Delta }\left({ϵ}_{i,t+1}\right)d{X}_{i,t+1}d{ϵ}_{i,t+1}\hfill \\ & =& \int 1(ɸX_{i,t+1}+v_i+{ϵ}_{i,t+1}>0)P\left(X_{i,t+1}\right|X_{\mathrm{it}})\mathrm{\Delta }\left({ϵ}_{i,t+1}\right)d{X}_{i,t+1}d{ϵ}_{i,t+1}\hfill \\ & =& \overline{\mathrm{\Delta }}(X_{\mathrm{it}};v_i)\hfill \end{array}\end{array}$

Therefore, risk can be measured by the probability of a conflict at the next period given the information set available at time t in country i (X_it), and the above equation can be estimated by using the following likelihood estimator:

$\begin{array}{cc}\left(10\right)& L_{\mathit{it}}={\left(\overline{\mathrm{\Delta }}\left(ɸX_{\mathrm{it}}+v_i\right)\right)}^{1\left({\mathit{cr}}_{i,t+1}==0\right)}{\left(1-\overline{\mathrm{\Delta }}\left({\mathit{x}}_{\mathit{it}}+v_i\right)\right)}^{1\left({\mathit{cr}}_{i,t+1}==1\right)}\end{array}$

One major advantage of our method, compared to earlier studies, is that we control for unobserved fixed effects (v_i) and therefore deal with unobserved variables that may contribute to the incidence of relapsing into subsequent conflicts. Following Wooldridge (2005), assuming that the unobserved fixed effect v_i is distributed by a distribution Ω(.;π), the likelihood estimator becomes a function of:

$\begin{array}{cc}\left(11\right)& L={\mathrm{\Sigma }}_{i,t}\left[{\int }_{-\infty }^{\infty }{\left(\overline{\mathrm{\Delta }}\left(ɸX_{\mathit{it}}+v_i\right)\right)}^{1\left({\mathit{cr}}_{i,t+1}==0\right)}{\left(1-\overline{\mathrm{\Delta }}\left(ɸX_{\mathit{it}}+v_i\right)\right)}^{1\left({\mathit{cr}}_{i,t+1}==1\right)}d\Omega \left(v_i;\pi \right)\right]\end{array}$

For this exercise, we assume $\overline{\mathrm{\Delta }}$ has the logistic functional form as $\frac{exp\left(ɸX_{\mathit{it}}+v_i\right)}{1+exp\left(ɸX_{\mathit{it}}+v_i\right)}$ and v_i is distributed by $N\left(0,{\sigma }_v^2\right)$. It is worth mentioning that a significant and economically large σ_v shows that ignoring these unobserved fixed effects may lead to biased results.

This is the first attempt in the conflict literature to control for unobserved fixed effects in a nonlinear probability model. We measure the risk of conflict recurrence with the logit technique, using the level and growth rate of real GDP per capita and whether the country had a civil conflict before.¹⁷ The estimated baseline coefficients for Equation (12), based on a set of 195 countries, with or without a history of conflict, across the world, are presented in the first column of Table 1. The results show that the variance of unobserved fixed effects (σ_v) is significant in all cases, illustrating that country-specific fixed effects are crucial in measuring conflict recurrence risk (or the initial occurrence of civil conflict, for that matter). Another important conceptual distinction between this paper and earlier studies is the timing of the model, as we infer the risk of conflict recurrence at time t + 1 by using the available information set at time t, which is an abstraction and not burdened by the problem of endogeneity often present in conflict regressions. Various models of conflict recurrence risk, presented in Table 1, exhibit a high degree of cross-correlation, with an average of 80 percent (Appendix Table 7). Furthermore, to check the robustness of our estimations, we develop an alternative measure of conflict recurrence risk, presented in the second column of Table 1 and discussed thoroughly in Section VI, using an array of variables such as a country’s history of conflict, the level and growth rate of global real GDP, the frequency of cabinet changes, and the degree of ethnic fractionalization, but without relying on the level and growth of domestic per capita income.

Table 1.

Determinants of the Risk of Conflict Recurrence: Random-Effect Logistic Panel Model

Notes: The dependent variable is a dummy for the recurrence of a conflict at the next period; the coefficients are estimated with a random-effect logit panel model for the sample period 1960 – 2010. z-statistics are reported in parenthesis, with ***, **, and * indicating statistical significance at the 1 percent, 5 percent, and 10 percent levels, respectively. Source: Authors’ estimations.

Table 1.

Determinants of the Risk of Conflict Recurrence: Random-Effect Logistic Panel Model

Variables	Models
	1	2	3	4	5	6	7	8	9
	Baseline	Alternative
Log (real GDP per capita)t	-0.363***		-0.275**	-0.087	-0.507***	-0.399***	-0.457***	-0.260**	-0.282**
	(-3.114)		(-2.245)	(-0.701)	(-3.204)	(-3.078)	(-3.031)	(-2.025)	(-2.392)
Real GDP per capita growtht	-1.639**		-1.667**	-1.834**	-2.687**	-1.988**	-3.397***	-2.305***	-1.766**
	(-2.223)		(-2.243)	(-2.226)	(-2.565)	(-2.441)	(-3.301)	(-2.869)	(-2.379)
Had conflictt	5.383***	2.619***	5.238***	4.932***	5.627***	5.407***	5.606***	5.292***	5.428***
	(9.567)	(3.494)	(9.346)	(8.572)	(7.698)	(8.495)	(7.536)	(9.296)	(9.562)
Log (global real GDP)t		1.443			1.724
		(0.873)			(1.053)
Global real GDP growtht		-4.703			-3.813
		(-0.903)			(-0.679)
Duration of conflictt		0.044***
		(3.327)
Terms of tradet					-4.630**		-4.890***
					(-2.546)		(-2.921)
Natural resource rentt						4.294***	6.847***
						(2.724)	(3.693)
Natural resource rentt2						-3.134	-5.235**
						(-1.492)	(-1.974)
Secondary school enrollment ratet								-11.822*
								(-1.817)
Secondary school enrollment ratet2								76.206**
								(2.257)
Ethnic fractionalization		0.675	1.677**
		(0.866)	(2.178)
Youth bulget		0.121***
		(3.930)
Frequency of cabinet changest		0.352***		0.190**	0.293***
		(3.619)		(2.166)	(2.889)
Type of political regimet				0.013
				(1.050)
Type of political regimet2				-0.016***
				(-6.631)
UN peacekeeping operationst									0.931***
									(3.549)
1960s	-0.920***		-0.866***	-0.625***				-0.877***	-0.824***
	(-4.619)		(-4.327)	(-2.834)				(-3.314)	(-4.101)
1970s	-0.196	0.103	-0.159	0.167	0.708	-0.060	0.066	-0.153	-0.103
	(-1.205)	(0.163)	(-0.975)	(0.869)	(1.028)	(-0.346)	(0.345)	(-0.748)	(-0.624)
1980s	0.423***	0.551	0.452***	0.691***	1.215**	0.525***	0.779***	0.521***	0.494***
	(2.835)	(1.242)	(3.008)	(3.938)	(2.553)	(3.371)	(4.520)	(3.154)	(3.263)
1990s	0.531***	0.309	0.515***	0.626***	0.866***	0.621***	0.752***	0.633***	0.558***
	(3.827)	(1.130)	(3.676)	(4.079)	(3.272)	(4.264)	(4.756)	(4.242)	(3.988)
Constant	-4.782***	-21.707	-6.038***	-5.734***	-18.662	-4.996***	-4.559***	-5.190***	-5.473***
	(-4.627)	(-1.531)	(-5.003)	(-5.279)	(-1.331)	(-4.388)	(-3.403)	(-4.848)	(-5.110)
σν2	1.202***	1.086***	1.156***	1.163***	1.443***	1.292***	1.440***	1.182***	1.210***
	(6.790)	(5.250)	(6.450)	(6.404)	(7.255)	(7.007)	(7.447)	(6.577)	(6.846)
Log Likelihood	-1469	-1060	-1449	-1323	-1028	-1254	-1098	-1387	-1463
χ2	184.9	86.17	179.7	205.5	133.2	134.4	138.1	177.0	197.5
Number of observations	7,087	2,848	6,795	5,409	4,564	5,340	4,646	6,303	7,087
Number of countries	195	105	184	159	165	168	153	184	195

Notes: The dependent variable is a dummy for the recurrence of a conflict at the next period; the coefficients are estimated with a random-effect logit panel model for the sample period 1960 – 2010. z-statistics are reported in parenthesis, with ***, **, and * indicating statistical significance at the 1 percent, 5 percent, and 10 percent levels, respectively. Source: Authors’ estimations.

View Table

Table 1.

Determinants of the Risk of Conflict Recurrence: Random-Effect Logistic Panel Model

Variables	Models
	1	2	3	4	5	6	7	8	9
	Baseline	Alternative
Log (real GDP per capita)t	-0.363***		-0.275**	-0.087	-0.507***	-0.399***	-0.457***	-0.260**	-0.282**
	(-3.114)		(-2.245)	(-0.701)	(-3.204)	(-3.078)	(-3.031)	(-2.025)	(-2.392)
Real GDP per capita growtht	-1.639**		-1.667**	-1.834**	-2.687**	-1.988**	-3.397***	-2.305***	-1.766**
	(-2.223)		(-2.243)	(-2.226)	(-2.565)	(-2.441)	(-3.301)	(-2.869)	(-2.379)
Had conflictt	5.383***	2.619***	5.238***	4.932***	5.627***	5.407***	5.606***	5.292***	5.428***
	(9.567)	(3.494)	(9.346)	(8.572)	(7.698)	(8.495)	(7.536)	(9.296)	(9.562)
Log (global real GDP)t		1.443			1.724
		(0.873)			(1.053)
Global real GDP growtht		-4.703			-3.813
		(-0.903)			(-0.679)
Duration of conflictt		0.044***
		(3.327)
Terms of tradet					-4.630**		-4.890***
					(-2.546)		(-2.921)
Natural resource rentt						4.294***	6.847***
						(2.724)	(3.693)
Natural resource rentt2						-3.134	-5.235**
						(-1.492)	(-1.974)
Secondary school enrollment ratet								-11.822*
								(-1.817)
Secondary school enrollment ratet2								76.206**
								(2.257)
Ethnic fractionalization		0.675	1.677**
		(0.866)	(2.178)
Youth bulget		0.121***
		(3.930)
Frequency of cabinet changest		0.352***		0.190**	0.293***
		(3.619)		(2.166)	(2.889)
Type of political regimet				0.013
				(1.050)
Type of political regimet2				-0.016***
				(-6.631)
UN peacekeeping operationst									0.931***
									(3.549)
1960s	-0.920***		-0.866***	-0.625***				-0.877***	-0.824***
	(-4.619)		(-4.327)	(-2.834)				(-3.314)	(-4.101)
1970s	-0.196	0.103	-0.159	0.167	0.708	-0.060	0.066	-0.153	-0.103
	(-1.205)	(0.163)	(-0.975)	(0.869)	(1.028)	(-0.346)	(0.345)	(-0.748)	(-0.624)
1980s	0.423***	0.551	0.452***	0.691***	1.215**	0.525***	0.779***	0.521***	0.494***
	(2.835)	(1.242)	(3.008)	(3.938)	(2.553)	(3.371)	(4.520)	(3.154)	(3.263)
1990s	0.531***	0.309	0.515***	0.626***	0.866***	0.621***	0.752***	0.633***	0.558***
	(3.827)	(1.130)	(3.676)	(4.079)	(3.272)	(4.264)	(4.756)	(4.242)	(3.988)
Constant	-4.782***	-21.707	-6.038***	-5.734***	-18.662	-4.996***	-4.559***	-5.190***	-5.473***
	(-4.627)	(-1.531)	(-5.003)	(-5.279)	(-1.331)	(-4.388)	(-3.403)	(-4.848)	(-5.110)
σν2	1.202***	1.086***	1.156***	1.163***	1.443***	1.292***	1.440***	1.182***	1.210***
	(6.790)	(5.250)	(6.450)	(6.404)	(7.255)	(7.007)	(7.447)	(6.577)	(6.846)
Log Likelihood	-1469	-1060	-1449	-1323	-1028	-1254	-1098	-1387	-1463
χ2	184.9	86.17	179.7	205.5	133.2	134.4	138.1	177.0	197.5
Number of observations	7,087	2,848	6,795	5,409	4,564	5,340	4,646	6,303	7,087
Number of countries	195	105	184	159	165	168	153	184	195

Notes: The dependent variable is a dummy for the recurrence of a conflict at the next period; the coefficients are estimated with a random-effect logit panel model for the sample period 1960 – 2010. z-statistics are reported in parenthesis, with ***, **, and * indicating statistical significance at the 1 percent, 5 percent, and 10 percent levels, respectively. Source: Authors’ estimations.

The higher the level and growth rate of real GDP per capita, the lower the risk of conflict recurrence over the next period in that country. Regression results, presented in Table 1, show that the impact of the level and growth rate of real GDP per capita on the perceived risk of conflict recurrence over the subsequent period after the end of aggression is negative and highly statistically significant across all specifications of the model. According to our baseline model, the estimated coefficients on real GDP per capita and its annual growth rate are -0.32 and -1.72, respectively, which are equivalent to the marginal effects of -0.02 and -0.11. In other words, a one percent increase in the level of real GDP per capita or in its rate of growth would lower the probability of conflict recurrence in the subsequent period by 2 percent and 11 percent, respectively, in that country.

One of the most important factors in determining the risk of conflict recurrence is whether the country had a conflict before. The coefficient on the “had conflict” variable is statistically significant at the 1 percent level across all specifications. Accordingly, the marginal effect of having a history of civil conflict increases the probability of conflict recurrence in the subsequent period by 35 percent, which alone highlights the importance of efforts to avoid an incidence of conflict in the first place. While the “had conflict” variable may arguably capture country characteristics, there is no test available to address this question directly. It is also worth noting that country fixed effects are modeled separately in our estimation, based on the logit random-effect framework. Furthermore, we include a range of other explanatory variables in the second column and find that the marginal impact of having a conflict before never drops below 25 percent.¹⁸

The impact of human capital on the risk of conflict recurrence is statistically significant and has a U-shape. The coefficient on human capital, as measured by secondary school enrollment rates, is statistically significant at the 5 percent level, which implies that the risk of conflict recurrence declines with the stock of human capital. However, this relationship is not linear. Including the squared term of secondary school enrollment rate in the regression indicates that the risk of conflict recurrence starts to increase once a country reaches a certain threshold in human capital.¹⁹ In other words, raising educational attainments from a low base has a stabilizing effect, as it helps to increase the productivity of labor and lower the opportunity cost of rebellion. But when a country is relatively well endowed with human capital, further gains would add to labor market pressures in a fragile post-conflict economy and consequently increase the risk of conflict recurrence.

The youth bulge is a significant factor in explaining the risk of conflict recurrence, while ethnic fractionalization does not appear to be robust. In order to identify the dynamic interaction between human capital, demographic conditions, and the risk of conflict, we also include a measure of the youth bulge, along with educational attainments, in the regression. Consistent with the findings of Urdal (2004), we find that the size of the youth bulge is also a momentous contributor to the risk of conflict recurrence, with a positive coefficient at the 1 percent level of significance. Furthermore, we also estimate the role of ethnic fractionalization, as measured by the ethnolinguistic fractionalization (ELF) index, which measures the probability that two randomly drawn individuals from a given country do not speak the same language. In contrast to other studies including Esteban, Mayoral, and Ray (2012) that highlight a relationship between ethnic fractionalization and the incidence of civil conflict, our analysis does not yield robust results for the relationship between ethnic heterogeneity and the forward-looking risk of conflict recurrence.²⁰ Depending on the model specification, the statistical significance of ethnic fractionalization varies, but it always has the same sign—that is, a higher degree of ethnic fragmentation appears to increase the risk of conflict recurrence over the subsequent period.

The magnitude of natural resource rents as a share of GDP has a significant effect on the risk of conflict recurrence. Empirical studies offer mixed results on the link between natural resource wealth and civil conflicts. Collier and Hoeffler (2004a) suggest that the share of primary goods in total exports has a quadratic and statistically significant effect on the incidence of conflict, while Ross (2006) argues along the same lines, but using data on petroleum and diamond production, instead the share of primary exports. Nevertheless, data limitations, especially in countries with a history of conflict, make the identification of the impact of resource dependence empirically challenging. We decided to use natural resource rents as a share of GDP—as a consistent metric for all countries—in the regressions, and find that its coefficient is statistically significant at the 1 percent level in different specifications of the model. In other words, the higher the extent of natural resource rents in a country, the higher the risk of conflict recurrence after the cessation of violence.²¹ We also include the squared term of natural resource rent, which is weakly significant at the 10 percent level only in one specification but still provides evidence that an increase in natural resource rents reduces the probability of future conflicts in higher-income countries.

Institutional factors have statistically and economically large effects on the probability of conflict recurrence. The literature on the institutions-growth nexus has become more diverse and multilayered, thanks to the development of a wide range of institutional measures. For example, the International Country Risk Guide (ICRG) dataset and the World Bank’s Country Policy and Institutional Assessment (CPIA) dataset provide survey-based variables ranging from the rule of law and the risk of expropriation to the quality of governance. The problem is that these series cover a relatively short period—starting from the 1980s in the case of the ICRG and from 1996 in the case of the CPIA dataset. Given that our sample period spans from 1960 to 2010, we need alternative measures to estimate the impact of institutions on the risk of conflict recurrence as well as economic growth. The Polity IV dataset, going back as far as 1800, provides a measure of political regime type; the Cross-Nations Time Series (CNTS) dataset reports the frequency of changes in government.²² Both measures have been used extensively in the literature and should be a reasonable proxy for the quality of institutional arrangements. Our estimation results show that although the type of political regime has a weakly significant effect that tends to increase the risk of conflict recurrence in democratic countries, the relationship is not linear. The negative squared term in political regime type indicates that a competitive election process and executive constraints reduce the probability of conflict recurrence in more democratic societies, while a high degree of institutionalized autocracy would also lower the risk of conflict recurrence. In other words, countries in the middle of the political regime spectrum appear to have a higher perceived risk of conflict recurrence. On the other hand, the coefficient on the number of cabinet changes is negative and highly significant at the 1 percent level in various specifications of the model, implying that frequent changes in government contribute to a higher risk of conflict recurrence.

External factors, especially changes in the terms of trade, have a significant influence on the risk of conflict recurrence. Dube and Vargas (2007) study the impact of commodity prices on intra-state armed conflicts in Colombia and find that a sharp fall in the price of coffee lowers wages, decreases the opportunity cost of violence, and, consequently, increases the risk of conflict outbreaks in coffee-intensive areas. However, they also show that the commodity price link to conflict incidence changes in hydrocarbon-rich areas, where an increase in the price of oil leads to an increase in conflict outbreaks. Likewise, using crosscountry data, Fearon (2005) and Ross (2006) find that an increase in the price of crude oil adds to the value of rents controlled by the government, leading to a rise in the incidence of conflict. From a methodological point of view, it is not possible to disentangle these two opposing channels because of extensive heterogeneities across countries and export prices. Nonetheless, the significant negative coefficient of the terms of trade indicates that, on average, the first channel of transmission is dominant. In other words, an improvement in export prices raises the opportunity cost of violence through higher wages, and thereby reduces the risk of conflict occurrence over the subsequent years. It is worth noting that a key difference between this paper and other studies is that we estimate a forward-looking measure of conflict recurrence risk, instead of the current incident of conflict.

UN peacekeeping operations are positively correlated with the estimated forward-looking risk of conflict recurrence. Using a dummy variable for peacekeeping operations, we identify a positive and statistically significant effect on the perceived risk of conflict recurrence.²³ In other words, the presence of UN peacekeeping forces appears to increase the risk of conflict recurrence. Moreover, this finding remains robust even if alternative measures of UN interventions, such as the number of UN troops in the country, are included instead of a binary dummy variable. This result is counterintuitive; we believe it can be explained by the endogeneity of peacekeeping operations: the UN initiates an operation when risk is apparent in generating a chaos in an attempt to support the development of an environment of economic and political stability and social cohesion, confirming the perceived risk of conflict recurrence.²⁴ Accordingly, we exclude this variable from the benchmark model, because the aforementioned endogeneity problem can potentially lead to biased estimates for other factors.²⁵ We also include decade dummies to capture shifting political relationships between countries during and after the Cold War. Following the end of the Cold War and the disintegration of the Soviet Union, there have been significant changes in the global political landscape, with far-reaching economic implications. The results indicate that decade dummies, except for the 1960s, are statistically significant in most specifications of the model, capturing the evolution of conflict recurrence risk over time.

The risk of relapsing into conflict declines with a lasting peace, but it remains higher in the subsequent years than to the pre-conflict probability. According to our estimations, in the first year after the end of conflict the probability of recurrence exceeds 20 percent, and in the five-year period following the probability exceeds 40 percent. While the probability of relapsing into conflict declines with a lasting peace, it is remains higher in the subsequent years than to the pre-conflict probability. The analysis presented in this paper highlights the importance of establishing appropriate policies and reforms immediately after the end of a conflict, to address the underlying causes and to accelerate the pace of economic recovery. As shown in the right panel of Figure 3, the cumulative distribution of conflict recurrence by region indicates extensive heterogeneity among countries and highlights the importance of taking into account country-specific effects in the empirical analysis. In addition, there is a systematically higher risk of conflict recurrence in Africa, Asia, and the Middle East than in the Americas and Europe. In particular, African and Asian countries start peace with a significantly higher risk of relapsing into conflict, although the probability declines over time with a lasting peace. In contrast, however, countries in the Middle East appear to suffer from a high risk of conflict recurrence without much decline over subsequent years.

View Full Size

Probability of Conflict Recurrence, 1960 – 2010

Citation: IMF Working Papers 2013, 002; 10.5089/9781475531152.001.A001

Source: Authors’ estimations. .

• Download Figure
• Download figure as PowerPoint slide

The risk of civil conflict peaked in the 1990s across the world, but remains high for countries with a history of conflicts. In Figure 4, the average risk of conflict occurrence in countries with at least one conflict (left panel) and no history of civil conflict (right panel) across the world is depicted. The probability of civil conflict peaked for all countries in the 1990s, increasing from an average of 4 percent in the 1960s to 10 percent in the 1970s, 15 percent in the 1980s, and 16 percent in the 1990s.²⁶ However, the risk of civil conflict incidence is more than 100 times greater in countries with at least one past episode of civil conflict: it was 22 percent in the 1990s for countries with a history of civil conflict versus 0.06 percent for countries with no history of civil conflict. Even though the risk of conflict occurrence declined over the past decade for all countries across the world, the probability of relapsing into a new bout of conflict remains elevated for countries with a history of conflict.

View Full Size

Variation in the Risk of Conflict Recurrence, 1962–2009

Citation: IMF Working Papers 2013, 002; 10.5089/9781475531152.001.A001

Source: Authors’ estimations.

• Download Figure
• Download figure as PowerPoint slide

The estimated risk of conflict recurrence has a statistically significant and economically large negative effect on post-conflict growth. As illustrated in Figure 5, the average risk of conflict recurrence in the 2000s is highly correlated with that in the 1960s, indicating a highly persistent risk of conflict recurrence in countries with a history of civil conflict. While it is not possible to formally test for the direction causality, we reason that persistency reflects a wide range of underlying factors including unresolved grievances and hostilities as well as dysfunctional institutions and economic weaknesses. In addition, the relationship between the average risk of conflict recurrence in the 1960s and average real GDP per capita growth over the entire period 1960–2010 appears to be robust. Again, consistent with the economic theory, a higher initial level of conflict recurrence risk tends to reduce the growth rate of real GDP per capita in the future.

View Full Size

Risk of Conflict Recurrence and Economic Growth, 1960–2010

Citation: IMF Working Papers 2013, 002; 10.5089/9781475531152.001.A001

Source: Authors’ estimations.

• Download Figure
• Download figure as PowerPoint slide

V. Empirical Results

The empirical results are consistent with the economic theory and provide robust evidence for conditional convergence. The post-conflict data on real GDP per capita show feeble recovery after the end of a civil conflict, with an average growth rate of 1.5 percent per year over the subsequent five years for all countries in our sample. This is a very weak performance relative to the rest of the world economy, especially considering the low level of initial income and the depth of economic contraction during a conflict. Furthermore, the distribution of post-conflict growth rates indicates a diverse pattern, ranging from a five-year average of -10 percent in the case of Liberia to 20 percent in the case of Bosnia and Herzegovina. We present the results of the cross-country panel growth regression in Table 2 for the baseline model and in Table 3 for the model with the alternative definition of conflict recurrence risk, which is discussed in detail in assessing the robustness of the results in Section VI. Our findings show that the extent of autocorrelation is well below 1 and not persistent, implying that growth is less autocorrelated in the aftermath of a civil conflict than in normal times. Nonetheless, the coefficient on the level of per capita income at the end of a conflict is statistically significant and consistent with conditional convergence. The magnitude of the coefficient implies that a country with lower per capita income would converge towards richer countries at a speed of 3 percent following the end of a conflict.

Table 2.

Determinants of Post-Conflict Real GDP Per Capita Growth: System-GMM Dynamic Panel

Notes: The dependent variable is the growth rate of real GDP per capita at time t, starting with the end of conflict. The model is estimated with the system-GMM dynamic panel approach for the sample period 1960-2010. The sample includes only post-conflict observations. z-statistics are reported in parenthesis, with ***, **, and * indicating statistical significance at the 1 percent, 5 percent, and 10 percent levels, respectively. Source: Authors’ estimations.

Table 2.

Determinants of Post-Conflict Real GDP Per Capita Growth: System-GMM Dynamic Panel

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18
Real GDP per capita growth t-1	0.289***	0.291***	0.241***	0.168***	0.286***	0.150***	0.296***	0.256***	0.289***	0.206***	0.226***	0.277***	0.225***	0.289***	0.240***	0.230***	0.289***	0.208***
	(14.438)	(14.572)	(11.263)	(7.821)	(13.506)	(6.266)	(14.654)	(12.312)	(14.435)	(8.470)	(8.506)	(14.020)	(11.158)	(14.403)	(11.465)	(10.935)	(14.427)	(6.472)
Log real GDP per capita t	-0.030***	-0.033***	-0.028***	-0.021***	-0.022***	-0.025***	-0.037***	-0.035***	-0.030***	-0.058***	-0.071***	-0.031***	-0.036***	-0.032***	-0.033***	-0.038***	-0.029***	-0.088***
	(-6.824)	(-7.136)	(-5.874)	(-5.381)	(-5.685)	(-5.894)	(-7.665)	(-7.119)	(-6.741)	(-8.313)	(-8.640)	(-7.146)	(-7.882)	(-6.874)	(-7.144)	(-8.050)	(-6.607)	(-8.202)
Risk of conflict recurrence t	-0.089**	-0.087**	-0.093**	-0.121***	-0.103***	-0.079**	-0.083**	-0.112***	-0.096**	-0.115**	-0.103*	-0.098***	-0.146***	-0.091**	-0.134***	-0.168***	-0.087**	-0.241***
	(-2.327)	(-2.272)	(-2.465)	(-3.186)	(-2.592)	(-1.974)	(-2.140)	(-2.733)	(-2.481)	(-2.294)	(-1.699)	(-2.590)	(-3.890)	(-2.379)	(-3.432)	(-4.238)	(-2.259)	(-3.244)
Duration of conflict t	-0.001		-0.001**	-0.000	0.001	0.000	-0.000	-0.000	-0.001	-0.002***	-0.002***	-0.001	-0.001**	-0.001	-0.001**	-0.000		0.001
	(-1.416)		(-1.995)	(-0.108)	(1.248)	(0.017)	(-0.909)	(-0.714)	(-1.444)	(-3.253)	(-2.993)	(-1.438)	(-1.966)	(-1.072)	(-2.107)	(-0.617)		(0.617)
Time since the end of conflict t		0.004***																0.013***
		(3.016)																(2.616)
Depth of contraction during conflict t			-0.056															-0.495***
			(-1.034)															(-3.639)
Terms of trade t				0.043
				(0.920)
Temperature variability t					0.001													-0.002
					(0.848)													(-0.641)
FDIt						0.007***
						(5.305)
Landlocked							-0.093***											-0.059
							(-4.172)											(-1.597)
Global real GDP growth t								0.481***										0.014
								(5.826)										(0.044)
UN peacekeeping operations t									0.014									0.063*
									(1.043)									(1.757)
Trade openness t										-0.120*	-0.119*							-0.028
										(-1.743)	(-1.655)							(-0.374)
Trade Openness*GDP t										0.020**	0.018*							0.008
										(2.098)	(1.891)							(0.763)
Trade Openness *Global growth t											0.622**							0.011
											(2.274)							(0.017)
Foreign aid per capita t												-0.001						-0.013***
												(-0.391)						(-3.284)
Secondary school enrollment rate t													0.239***
													(2.900)
Ethnic fractionalization t														-0.036				-0.098**
														(-1.330)				(-2.285)
Cabinet Changes t															-0.009***			-0.012***
															(-3.989)			(-3.708)
Natural resource rent t																-0.048**		0.010
																(-2.404)		(0.338)
IMF Program t																	-0.005	-0.006
																	(-1.269)	(-1.206)
Constant	0.253***	0.253***	0.239***	0.182***	0.178***	0.216***	0.311***	0.276***	0.252***	0.470***	0.562***	0.262***	0.293***	0.278***	0.290***	0.317***	0.231***	0.701***
	(6.914)	(7.299)	(6.112)	(5.395)	(5.335)	(6.249)	(7.790)	(6.878)	(6.867)	(8.266)	(8.317)	(7.281)	(7.911)	(6.669)	(7.558)	(8.233)	(6.851)	(7.691)
Number of observations	1,916	1,916	1,563	1,618	1,617	1,166	1,856	1,748	1,916	1,261	1,103	1,908	1,761	1,908	1,691	1,767	1,916	773
Numberofcountries	89	89	83	83	78	81	84	89	89	73	73	89	89	88	89	89	89	57
Sargan Test	1943	1943	1605	1696	1667	1247	1865	1744	1942	1279	1125	1930	1738	1940	1669	1757	1942	797.8
x2	333.0	343.3	206.5	119.9	272.7	108.3	370.4	331.6	334.2	205.1	234.6	328.3	265.9	334.2	274.2	300.0	332.4	215.0

Notes: The dependent variable is the growth rate of real GDP per capita at time t, starting with the end of conflict. The model is estimated with the system-GMM dynamic panel approach for the sample period 1960-2010. The sample includes only post-conflict observations. z-statistics are reported in parenthesis, with ***, **, and * indicating statistical significance at the 1 percent, 5 percent, and 10 percent levels, respectively. Source: Authors’ estimations.

View Table

Table 2.

Determinants of Post-Conflict Real GDP Per Capita Growth: System-GMM Dynamic Panel

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18
Real GDP per capita growth t-1	0.289***	0.291***	0.241***	0.168***	0.286***	0.150***	0.296***	0.256***	0.289***	0.206***	0.226***	0.277***	0.225***	0.289***	0.240***	0.230***	0.289***	0.208***
	(14.438)	(14.572)	(11.263)	(7.821)	(13.506)	(6.266)	(14.654)	(12.312)	(14.435)	(8.470)	(8.506)	(14.020)	(11.158)	(14.403)	(11.465)	(10.935)	(14.427)	(6.472)
Log real GDP per capita t	-0.030***	-0.033***	-0.028***	-0.021***	-0.022***	-0.025***	-0.037***	-0.035***	-0.030***	-0.058***	-0.071***	-0.031***	-0.036***	-0.032***	-0.033***	-0.038***	-0.029***	-0.088***
	(-6.824)	(-7.136)	(-5.874)	(-5.381)	(-5.685)	(-5.894)	(-7.665)	(-7.119)	(-6.741)	(-8.313)	(-8.640)	(-7.146)	(-7.882)	(-6.874)	(-7.144)	(-8.050)	(-6.607)	(-8.202)
Risk of conflict recurrence t	-0.089**	-0.087**	-0.093**	-0.121***	-0.103***	-0.079**	-0.083**	-0.112***	-0.096**	-0.115**	-0.103*	-0.098***	-0.146***	-0.091**	-0.134***	-0.168***	-0.087**	-0.241***
	(-2.327)	(-2.272)	(-2.465)	(-3.186)	(-2.592)	(-1.974)	(-2.140)	(-2.733)	(-2.481)	(-2.294)	(-1.699)	(-2.590)	(-3.890)	(-2.379)	(-3.432)	(-4.238)	(-2.259)	(-3.244)
Duration of conflict t	-0.001		-0.001**	-0.000	0.001	0.000	-0.000	-0.000	-0.001	-0.002***	-0.002***	-0.001	-0.001**	-0.001	-0.001**	-0.000		0.001
	(-1.416)		(-1.995)	(-0.108)	(1.248)	(0.017)	(-0.909)	(-0.714)	(-1.444)	(-3.253)	(-2.993)	(-1.438)	(-1.966)	(-1.072)	(-2.107)	(-0.617)		(0.617)
Time since the end of conflict t		0.004***																0.013***
		(3.016)																(2.616)
Depth of contraction during conflict t			-0.056															-0.495***
			(-1.034)															(-3.639)
Terms of trade t				0.043
				(0.920)
Temperature variability t					0.001													-0.002
					(0.848)													(-0.641)
FDIt						0.007***
						(5.305)
Landlocked							-0.093***											-0.059
							(-4.172)											(-1.597)
Global real GDP growth t								0.481***										0.014
								(5.826)										(0.044)
UN peacekeeping operations t									0.014									0.063*
									(1.043)									(1.757)
Trade openness t										-0.120*	-0.119*							-0.028
										(-1.743)	(-1.655)							(-0.374)
Trade Openness*GDP t										0.020**	0.018*							0.008
										(2.098)	(1.891)							(0.763)
Trade Openness *Global growth t											0.622**							0.011
											(2.274)							(0.017)
Foreign aid per capita t												-0.001						-0.013***
												(-0.391)						(-3.284)
Secondary school enrollment rate t													0.239***
													(2.900)
Ethnic fractionalization t														-0.036				-0.098**
														(-1.330)				(-2.285)
Cabinet Changes t															-0.009***			-0.012***
															(-3.989)			(-3.708)
Natural resource rent t																-0.048**		0.010
																(-2.404)		(0.338)
IMF Program t																	-0.005	-0.006
																	(-1.269)	(-1.206)
Constant	0.253***	0.253***	0.239***	0.182***	0.178***	0.216***	0.311***	0.276***	0.252***	0.470***	0.562***	0.262***	0.293***	0.278***	0.290***	0.317***	0.231***	0.701***
	(6.914)	(7.299)	(6.112)	(5.395)	(5.335)	(6.249)	(7.790)	(6.878)	(6.867)	(8.266)	(8.317)	(7.281)	(7.911)	(6.669)	(7.558)	(8.233)	(6.851)	(7.691)
Number of observations	1,916	1,916	1,563	1,618	1,617	1,166	1,856	1,748	1,916	1,261	1,103	1,908	1,761	1,908	1,691	1,767	1,916	773
Numberofcountries	89	89	83	83	78	81	84	89	89	73	73	89	89	88	89	89	89	57
Sargan Test	1943	1943	1605	1696	1667	1247	1865	1744	1942	1279	1125	1930	1738	1940	1669	1757	1942	797.8
x2	333.0	343.3	206.5	119.9	272.7	108.3	370.4	331.6	334.2	205.1	234.6	328.3	265.9	334.2	274.2	300.0	332.4	215.0

Notes: The dependent variable is the growth rate of real GDP per capita at time t, starting with the end of conflict. The model is estimated with the system-GMM dynamic panel approach for the sample period 1960-2010. The sample includes only post-conflict observations. z-statistics are reported in parenthesis, with ***, **, and * indicating statistical significance at the 1 percent, 5 percent, and 10 percent levels, respectively. Source: Authors’ estimations.