ECONOMIC SHOCK AND CIVIL CONFLICTS: EVIDENCE FROM AFRICA

Abstract

This PhD Thesis deals with the problem of civil war in Africa. It clearly arises from the economic literature that the most robust predictor of a country's probability of civil war is its level of GDP. Low GDP empirically dominates all other correlates, including the level of democracy, the degree of ethnic diversity or the dependence on natural resource exports. The main intuition behind this nding is related to the opportunity cost of the rebels who decide whether to devote e ort to a productive endeavor or to violent predatory activities. When GDP growth is weak, there is a reduction in real wages that produces a gap in the returns from predatory and productive activities. In this situation of low opportunity costs, citizens are more likely to turn to predatory activities. After identifying main theoretical and applied works on determinants of civil war onset, we replicate one widely cited study of this eld: Edward Miguel, Shanker Satyanath and Ernest Sergenti (2004). We extend this important study covering the African countries during 1981-2009. The probability of civil war is estimated by using an instrumental variable approach with rainfall variation as instrument for economic growth. Our results are in line with the original study, and we nd a strong and negative relationship between economic growth and civil war. We test the robustness of the instrument with Stock and Yogo (2005) Test and we nd out that the instrument is weak, therefore the estimates are biased. We show that the weakness of the instrument is not related to the coding rule of the dependent variable, namely civil war, a dichotomous variable assuming 1 when more than 25 battle-deaths per year are reached. Further, we adopt the Zero-in ated poisson approach to deal with the abundance of zero (peace) in the dependent variable. We nd interesting and not biased results with this model. In the third chapter we deal with the critique of the coding rule of the civil war variable and we suggest a new indicator of civil war, based on violence escalation rather than battle-deaths threshold. Taking information from the Social Con ict in Analysis Database Version 3.1 (SCAD), from the Armed Con ict Location and Event Data Project 2015 (ACLED), and from the Uppsala Con ict Data Program (UCDP) we design a violence indicator ranging from 1 (peace) to 3 (civil war). We apply this new indicator by estimating Markov transition probabilities for countries, ethnic groups and non-ethnic groups violence status over time as function of observable characteristics. We propose 3 methods to estimate these Markov probabilities. The first method is a counting method, the second one predicts transition probabilities using ordered logit regression models and finally, we derive hazard rates from a non-parametric Kaplan-Meier estimator and a semi parametric proportional hazard (Cox) model. We also test whether the Markov assumption holds. Our results are in line with the economic literature, but, beyond the economic growth e ect, we nd a strong relationship between civil war and diamond exports, soil fertility and population growth.