Violent Conflict and Inequality work in progress Cagatay Bircan University of Michigan Tilman Brück DIW Berlin, Humboldt University Berlin, IZA and Households in Conflict Network Marc Vothknecht DIW Berlin and Humboldt University Berlin IZA/World Bank Conference Bonn, May 4, 2009 1
Overview 1. Introduction 2. On the Determinants of Inequality 3. On the Impact of Violent Conflict on Inequality 4. Data 5. Estimation Strategy 6. Results descriptive statistics econometric results 7. Summary and Policy Implications 2
Introduction Aim of our paper to investigate how mass violent conflict affects income inequality using cross-country panel data Innovations of our approach fill in gaps in our understanding of conflict-inequality nexus study reverse causality from conflict to inequality (conflict cycle) refine previous studies by using different variables better datasets improved techniques 3
Global Trends of Inequality 60 Adj. Gini 50 40 30 20 Latin America Africa Asia Eastern Europe OECD 10 0 1960s 1970s 1980s 1990s Early 2000s Significant cross-country variation Little variation of inequality within countries over time important role of structural, i.e. slowly changing, factors 4
Determinants of Inequality - 1 Level of income / economic growth Kuznets Hypothesis (1955): inverted U-shaped curve support: Barro (2000), Higgins/Williamson (1999), Milanovic (1994) critique: Atkinson (1997), Deininger/Squire (1996), Li/Squire/Zou (1998) possibly due to unequal middle-income countries in Latin America Deininger/Squire (1998) Political economy argument rich minority implements inequality enhancing economic policy Bertola (1993) constrained by civil liberties and higher levels of education Li/Squire/Zou (1998) government transfers are a related determinant of inequality Bulir (2001), Milanovic (1994) 5
Determinants of Inequality - 2 Capital markets argument Investments constrained by access to credit, especially for the poor Tsiddon (1992) depends on the initial distribution of assets (especially land) Deininger/Olinto (1998), Deininger/Squire (1998) and on the level of financial development Beck/Demirguc-Kunt/Levine (2004), Li/Squire/Zou (1998) Openness to trade impact of trade on inequality depends on factor endowments but empirical evidence is mixed Anderson (2005), Milanovic/Squire (2005), Barro (2000) History and geography e.g. colonization impacts on inequality Acemoglu/Johnson/Robinson (2000), Barro (2000) hence use country fixed effects 6
Impact of War on Inequality Indirect Impact through changes in structural determinants Growth (destruction of human, physical and financial assets) Civil liberties and social spending Education Pure War Impact Heightened insecurity and weaker market structures Distortion of prices War entrepreneurs are likely to benefit In general Systematic regional and sectoral patterns are likely Duration of war may be relevant Either effect could persist into the post-war period 7
Inequality Data UNU-WIDER World Income Inequality Database (WIID) v. 2.0b 5,000 Gini coefficients from a total of 156 countries we use annual observations for 128 countries for 1960-2004 maximise countries and years annual observations adjusted for comparability based on a regression approach, with 60 % of our data of high quality with multiple income Ginis, we employ a selection rule manual check whether the Gini covers the conflict regions This is best possible data source for inequality data 8
Conflict Data UCDP/PRIO Armed Conflict Dataset Codebook, 4-2007 Gleditsch et al (2002) Annual observations 1960-2004 being at war = 1,000+ battle deaths in at least one year being indirectly affected by war = country at war, but conflict areas are not covered by the Gini coefficient post-conflict incidence = 1 if last war less than ten years ago post-conflict duration = time passed since the last war in years 9
Estimation Strategy 1) Pooled OLS this is in line with most of the empirical inequality literature Gini it = W it α + X it β + ε it assume independent and identically distributed errors ε likely to be biased due to omitted factors (country characteristics)! 2) Fixed effects model to control for time-constant unobservables country-fixed effects allows us to capture a substantial proportion of cross-country differences in inequality focus on within-country variations for the effect of violent conflict 10
Potential Econometric Problems 1) Endogeneity: reverse causality from inequality to conflict? distinction between vertical and horizontal inequality cross-country studies consistently found that vertical income inequality does not increase the risk of violent conflict (Collier and Hoeffler 2004; Fearon and Laitin 2003; Sambanis 2005) 2) Selectivity bias sample selection bias if the observed Ginis are chosen non-randomly there are, in fact, relatively fewer Ginis in war-affected countries hence apply Heckman two-stage procedure Probit model on the observability of Gini coefficients inclusion of the IMR from this estimation in the primary model 11
Results I. Descriptive Statistics Inequality in Different Phases of War Gini (adjusted) Mean (S.E.) Obs. No War (War is >5 years away) 39.97 (0.29) 1603 Pre-War (5 years) 41.78 (1.27) 71 At War 46.89 (0.73) 116 Early Post-War (5 years) 47.00 (1.04) 64 Late Post-War (5-10 years) 43.50 (1.28) 51 12
Results I. Descriptive Statistics (II) Within-Country Variations A) Beginning of War Deviation from Within-Country Mean -2 0 2 4 6 8 Pre-War War Within-Country Mean No pre-war increase in inequality No endogeneity Inequality increases in the course of war -5-3 -1 1 3 5 7 9 Phases of War Onset (years) Median of Deviation Median of Deviation (Mov. Average) 13
Results I. Descriptive Statistics (II) Within-Country Variations B) Transmission from War to Peace Deviation from Within-Country Mean -4-2 0 2 4 End of War Within-Country Mean -5-3 -1 1 3 5 7 9 11 Post-Conflict Phases (years) Significant average increase in inequality in the early post-war period Decreasing trend in the late recovery phase Median of Deviation Median of Deviation (Mov. Average) 14
Results II. FE Regression Results Dep. Var.: Gini Coefficient (Adj.) Base Regressions + War Duration War Incidence 1.63* (0.98) War Incidence Indirect -1.06 (1.50) -0.71 (1.49) Post-Conflict Incidence (10 yrs) 2.13** (0.93) 2.44*** (0.94) War Incidence: Short War 1.09 (1.11) War Incidence: Long War 2.33** (1.14) GDP p.c. -0.29*** (0.05) -0.33*** (0.05) Gov. Share of Real GDI -0.29*** (0.05) -0.26*** (0.06) Trade/GDP -0.03** (0.01) -0.03** (0.01) IA Trade & Developing Country 0.07*** (0.02) 0.07*** (0.02) IMR -0.06 (0.59) -0.08 (0.59) Constant 49.10 (1.70) 49.30 (1.73) Decade Dummies Included Yes Yes Observations 1504 1504 R-squared 0.12 0.12 No. of Groups (Av. Obs. per Group) 128 (11.76) 128 (11.76) Inequality rises during violent conflict and especially in the postwar Inequality periodrises significantly during longer wars Control variables show the expected signs 15
Results II. FE Regression Results - 2 Dep. Var.: Gini Coefficient (Adj.) Post-Conflict Period War Incidence 1.74* (0.99) 1.56* (0.94) War Incidence Indirect -0.94 (1.50) -1.17 (1.48) Early Post-Conflict Phase (< 5 yrs) Late Post-Conflict Phase (5-10 yrs) 2.77** (1.08) 1.53 (0.95) Post-Conflict Duration (in years) 1.21*** (0.44) Post-Conflict Duration Squared -0.12*** (0.05) GDP p.c. -0.29*** (0.05) -0.29*** (0.05) Gov. Share of Real GDI -0.29*** (0.06) -0.29*** (0.06) Trade/GDP -0.03** (0.01) -0.03** (0.01) IA Trade & Developing Country 0.07*** (0.02) 0.07*** (0.02) IMR -0.05 (0.59) -0.01 (0.59) Constant 50.26*** (1.62) 50.31*** (1.61) Decade Dummies Included Yes Yes Observations 1504 1504 R-squared 0.12 0.12 No. of Groups (Av. Obs. per Group) 128 (11.76) 128 (11.76) Highest increase in inequality in the period of early recovery Inverted-U Relationship 16
Results III. Robustness Checks War Variables War Incidence (Gini covers conflict regions) Post-Conflict: Early Recovery Post-Conflict: Late Recovery Post-Conflict Duration Post-Conflict Duration Squared Base Specification OLS Developing Countries M2/GDP & Inflation Included 25+ Battle Deaths Threshold 1.63* 3.29*** 2.59* 1.75 0.79 (0.98) (0.86) (1.32) (1.06) (0.75) 2.77** 3.76*** 2.37 2.05* 1.59** (1.08) (1.13) (1.58) (1.12) (0.68) 1.53 0.47 1.34 1.10 0.83 (0.95) (1.05) (1.35) (1.04) (0.63) 1.21*** 1.23** 1.18** 0.96** 0.83*** (0.44) (0.53) (0.59) (0.46) (0.28) -0.12*** -0.14** -0.12** -0.10** -0.09*** (0.05) (0.06) (0.06) (0.05) (0.03) Observations 1504 1160 514 1163 1504 Intertemporal Kuznets curve in post-war period confirmed OLS in line with FE results 17
Summary and Policy Implications Violent conflict has an inequality-increasing effect rising inequality especially during long wars and in the early postconflict period highest five years after the end of a conflict (3.0 Gini points) Impact not permanent, but temporary decreasing inequality in the course of post-war reconstruction effectively, the post-war period lasts 10 years Focus on effects of war on structural factors rebuilding of security and infrastructure revival of business activities and markets 18
Thank you for your attention! 19