No. 18 / June 2004 RWANDA: THE IMPACT OF CONFLICT ON GROWTH AND POVERTY The human, social and economic costs of Rwanda's Genocide have been staggering. Although the country has made remarkable progress over the last ten years, especially in terms of recovering some of the ground lost on education and health, GDP per capita remains much lower than what it would have been without the Genocide. Per capita GDP today would probably be between 25 and 30% higher if the conflict had not taken place. About one fourth of the population in poverty today can be said to be poor as a result of the Genocide. Conflict and Recovery. In 2000, one in five people in Substantial progress has been achieved in education. sub-Saharan Africa lived in a country affected by World Bank estimates show that only five years after conflict. Over the last four decades, more than a third the end of the conflict, the number of primary school of the region's countries have experienced civil strife. students has rebounded to its pre-Genocide long term The cost of these conflicts in terms of loss of life, trend-line. Rwanda's gross primary enrollment ratio, human and social capital has been enormous, and the at 107%, is higher today than that observed in other psychological impact of the violence will take a long Sub-Saharan countries of similar income levels, and time to heal. Yet in many areas, the populations of the the number of secondary school students has almost affected countries have demonstrated remarkable tripled since 1996. While there are issues of quality resilience in getting back on their feet after the end of (including high repetition rates), and some of the loss conflicts. in human capital (due to deaths, injuries, and permanent school dropouts) cannot be recouped, the Rwanda is a good example of such resilience. At least country has recovered rapidly. 800,000 people (10% of the population) died in the 1994 Genocide, and 3 million people sought refuge in Key health indicators have also improved substantially. neighboring countries. The war led to widespread World Bank estimates of infant mortality suggest that destruction of property, infrastructure and resources after increasing from 85 to 137 per thousand between such as livestock. Large-scale migration into 1988-92 and 1992-94, the infant mortality rate dropped bordering countries and less populated areas of the to 104 per thousand over 1998-00. Similarly, the child country weakened social networks. A high incidence mortality rate increased from 150 to 247 per thousand of rape contributed to the spread of HIV/AIDS, and the between 1988-92 and 1992-94, but it fell to 178 per victims of rape and other violent crimes suffered thousand in 1998-00. The rate of malnutrition among extensive trauma. Today, 85,000 households are still children under five was lower in 1998-00, at 24%, than headed by children, women-headed households are in 1988-90, at 27%. prevalent, and orphanhood is widespread. Overall, while additional efforts will be required to At the same time, considerable progress has been meet the education and health targets of the achieved over the last ten years in a range of areas. Millennium Development Goals, recovery has been Peace and stability have been maintained. Traditional impressive. Social indicators are clearly improving for Rwandan values, such as community participation, Rwanda's next generation. For children being born group solidarity, support to the poor, and Gacaca--the and entering the education system, the impact of the concept of conflict resolution through communal conflict is not necessarily high a few years later, at efforts--have been instrumental in advancing least as measured through the type of indicators reconciliation and accountability following the reviewed here (the psychosocial effects of the Genocide. Genocide will undoubtedly still be felt many years one if there is civil war and zero otherwise. Because from now). of the cross-country averaging involved in estimating the impact of conflicts with panel data, one year of The impact of the Genocide on the economy, however, civil war will typically have the same impact on GDP may prove to be longer lasting. Measuring the growth no matter which country is under consideration economic costs of conflicts is not easy. Violent and how severe its conflict was. conflicts may affect a country's short- and long-run growth prospects. In the short run, casualties and Economic time series are frequently affected by damage to physical capital (such as infrastructure) outliers and structural changes. These may be caused directly affect production. In the long run, a civil by events such as wars, natural disasters, or policy conflict can inhibit markets and foreign investors, and changes, which manifest themselves as aberrant thus constrain the country's access to external savings observations or level shifts that are inconsistent with and technology transfers that are crucial for growth. an econometric model thought to be appropriate for the Furthermore, it is possible that the economic overwhelming majority of the observations. The consequences of conflict in one country extend beyond statistical literature in outlier identification and its borders. For example, if regional trade or risk correction has been concerned with the econometric perceptions are effected then one could find negative implications of analyzing data affected by outliers. For spillovers resulting in lower regional growth (this is example, in time series analysis the existence of clearly the case in some regions in Africa). outliers is known to cause significant bias in the sample autocorrelation and partial autocorrelation Methodology. The objective of the paper on which functions reducing the efficacy of these statistics as this note is based is to provide a measure of the tools for model identification. As a result, and economic cost of the Rwanda Genocide using a depending on the size and position of the outliers technique for the identification and correction of selected, models can be under- or over-specified. As outliers in time series. Specifically, the detection of an for the properties of the time series estimators, outlier in the GDP per capita time series that can be structural changes, such as level shifts, will produce traced to the conflict allows us to estimate the GDP inconsistent estimates of the parameters unless the losses associated with the Genocide. Outlier model is correctly specified. In cross section analysis, identification and correction, or intervention analysis, outliers and structural changes affecting the dependent is a commonly used procedure when working with variable of one of the cross section units in the sample time series. The inclusion of a dummy variable into a will create problems similar to those of an omitted model to measure the effect of a particular event could variables regression, that is, inconsistency of estimates. be considered the simplest form of intervention analysis. But the inclusion of the dummy assumes first Outlier detection and correction may also be that an event has been identified and second that the interesting for reasons that go beyond statistical issues impact of that event is of a particular type. In the if interest centers on the difference between an absence of precise information on the likely effects of observed series and the estimate of its hypothetical a shock, analysts have recently developed and resorted value in the case where an outlier had not occurred. to more refined procedures for outlier identification For example, if one detects an outlier in a GDP series and correction. In the paper on which this note is then it would be possible to estimate the GDP losses based, we exploit work by Tsay (1986, 1988) and (or gains) associated with that extraordinary Gomez, Maravall and Peña (1997) who have occurrence. In addition, if the outlier can be identified developed unified procedures for detecting and with a particular event, say a war, then the estimate of handling outliers with different patterns in univariate the losses could be interpreted as the GDP losses time series, such as our trend in GDP per capita. caused by the war. It is important to note that these estimates would be based on what would have seemed Note that our method is different from that used by a reasonable evolution of the series after observing the Murdoch and Sandler (2001) in their work on conflicts past and future dynamics of the series rather than any and growth. These authors rely on cross-country panel particular measure of the event in question. However, data in order to assess the average effect of a war index to the extent that we can control for other events that on GDP growth. When focusing on a given country, may have affected economic performance in a such panel analysis is likely to be somewhat limited, particular year, then it is possible to have a reasonable especially when it is considered that war indices are degree of comfort in the reliability of those estimates. usually discrete variables with low cross country In principle this approach may seem less than optimal variability. For example, the index proposed by the because of the limited information it uses. Also, for a Center for Defense Intelligence (CDI) takes a value of country that has not suffered a civil war, this approach 2 would prevent the evaluation of potential spillovers Conditions de Vie des Ménages au Rwanda) conducted from other countries in the region. Unfortunately, the between October 1999 and July 2001 by the Statistics available indicators of the nature of armed conflict Department of the Ministry of Finance and Economic present important problems which may produce Planning. Data collection took place between October misleading results. As noted above the CDI index 1999 and December 2000 in urban areas, and July takes a value of zero or one to indicate a war no matter 2000 to July 2001 in rural areas. Since 90% of the the damage caused by the conflict, making it unlikely population lives in rural areas, we will consider that that a standard regression approach alone can capture the survey took place in 2000-2001, and we will the GDP losses inflicted by the conflict. Furthermore, compare poverty with and without the Genocide using outlier identification and correction procedures can be observed and simulated levels of GDP per capita in extended to include regression variables in the 2001. information set. Three components are needed for poverty Figure 1 provides our estimates of the impact of the measurement: an indicator of well-being, a poverty conflict on per capita GDP using three different line, and a poverty measure to aggregate information at models, all based on the outlier identification the household level into a national poverty measure. technique. It can be seen that without the Genocide, The indicator of well-being used in Rwanda is annual per capita GDP today would be up to 30 percent above household consumption per adult equivalent. its current level. Household consumption was obtained by using the Once we have estimated what level of GDP per capita information on a large number of consumption goods would have been achieved without the conflict in the in the survey, including some imputed values (e.g., for year corresponding to our household survey, it is owner-occupied housing). relatively straightforward to estimate what poverty would have been without the conflict, provided we are Household consumption was adjusted for price willing to make a number of assumptions. These differences across regions and time using a monthly assumptions are necessary to translate our estimate of cost of living index by locality. To account for the impact of the conflict on GDP per capita into an differences in needs, the number of adult equivalents estimate of how consumption per adult equivalent may was computed within each household. A food poverty have changed as measured in household surveys. In line was estimated in order to capture the resources order to explain these implicit assumptions, we first needed to meet basic nutritional needs by assessing the briefly review how poverty was estimated in Rwanda. cost of a food basket providing 2500 kcal per adult per day. The non-food component of the poverty line was Figure 1: Impact of the Genocide on Per Capita estimated by assessing how much was spent on non- GDP food items by households whose food consumption was close to the food poverty line (plus or minus 10%). 140 On average, these households allocated 29.4% of their Counterfactual per capita GDP 120 total consumption to non-food expenditures. Taking 100 into account such a provision for non-food index 80 consumption, the poverty line was set at FRw 64,000 GDP 60 per adult equivalent per year. Actually observed per capita GDP capita 40 Per 20 Assumptions. In Rwanda, as in many other countries, poverty measures have been estimated by comparing 0 adult equivalent household consumption to a poverty 1960 19621964 19661968 19701972 19741976 19781980 19821984 19861988 19901992 19941996 19982000 Year line representing the cost of food and non-food basic Source: Authors' estimates (3 methods for needs. In order to use the results from our GDP per counterfactual). capita simulations for poverty measurement, a first necessary assumption is therefore that GDP per capita growth as measured in the National Accounts is Poverty Estimates. A detailed explanation of the essentially correlated perfectly with average growth in method used to obtain the estimates of poverty consumption per adult equivalent at the household presented in Rwanda's Poverty Reduction Strategy is level. That is, we use our estimates of the impact of given in Ministry of Finance (2002). The estimates rely the conflict on per capita GDP as our best bet for the on data from the Integrated Household Living impact of the conflict on mean per-adult equivalent Conditions Survey (Enquête Intégrale sur les household consumption. 3 A second assumption is that we can rely on the poverty indicate that without the Genocide of 1994, Rwanda's lines used for measuring poverty in the 1999-01 per capita GDP could have been between 25 and 30% household survey in order to assess the impact of the higher in 2001. In turn, under a number of simplifying Genocide. The fact that we do not change the poverty assumptions, this suggests that one fourth of the lines for our counterfactual poverty measures without population in poverty in 2001 could have been non- the conflict means that we assume that the conflict did poor if the Genocide had not taken place, and more not affect relative prices and consumption patterns in than 40% of the extreme poor could have avoided such a way that other poverty lines would have had to extreme poverty. be used in the absence of conflict. A third assumption is that inequality in per-adult These results suggest that the economic losses due to equivalent consumption has not been affected by the armed conflict are long-lasting even though some other Genocide, so that we only need to incorporate the social indicators, such as the rates of enrollment in impact of the conflict on mean consumption for our primary school or the rate of child mortality can poverty simulations. With only one survey in Rwanda rebound fairly rapidly after conflict. While our results without pre-Genocide comparable household level demonstrate the highly negative economic impact of data, we cannot assess the impact of the conflict on conflicts on standards of living, they may be inequality, so it is best to assume that inequality has underestimated given that our econometric remained unchanged. methodology did not account for the longer-term economic costs associated, among others, with the If we accept these assumptions, the procedure for losses in human and social capital in the country. assessing the impact of the conflict on poverty is straightforward. We first compute poverty in the Figure 2: Impact of the Genocide on Poverty traditional way using the 1999-01 household survey data. Then we compute our counterfactual poverty 70% Actual measures after scaling up the adult equivalent 60% Counterfactual consumption aggregate for all households in the survey 50% by a factor equal to the ratio of the estimated per capita 40% Actual GDP without the conflict to the observed per capita Counterfactual 30% Actual GDP at the time of the survey. The poverty measures Counterfactual 20% used here are the headcount, the poverty gap, and the 10% squared poverty gap. 0% Headcount Poverty Gap SquaredPoverty Poverty Measures Gap Results. According to the methodology used in Rwanda's Poverty Reduction Strategy, 60.3% of the Source: Authors' estimates (3 methods for population was poor in 2001, i.e., with a level of counterfactual). consumption per equivalent adult below the poverty References line capturing the cost of basic food and non-food needs. The share of the population in extreme poverty, Gomez, V., A. Marvall and D. Peña. 1997. "Missing i.e., not able to meet basic food needs, is 41.6%. Observations in ARIMA Models: Skipping Strategy versus Additive Outlier Approach," Bank If the Genocide had not taken place, so that per capita of Spain, Working Paper No. WP9701. GDP would have been higher by close to 30 Ministry of Finance. 2002. A Profile of Poverty in percentage points, poverty would be much lower, Rwanda, Government of Rwanda, Kigali. around 45-47%, instead of 60.3%, while the share of Murdoch, J. and T. Sandler. 2001. "Economic Growth, the population in extreme poverty could be at 26-28%, Civil Wars, and Spatial Spillovers," unprocessed instead of 41.6% (see Figure 2). The impact in draft, University of Texas. proportional terms is even larger for the poverty gap Tsay, R.S. 1986. "Time Series Model Specification in and the squared poverty gap than for the headcount. the Presence of Outliers," Journal of the American For extreme poverty for example, the squared poverty Statistical Association, 81,132-141 gap could have been about only half of what it is today This note was prepared by Humberto Lopez if the conflict had not taken place. (PRMPR), Quentin Wodon (AFTPM) and Ian Bannon (CPR). The Note was also published as Conclusion. We estimated the economic cost of armed Social Development Note No. 94. For additional conflict in Rwanda and the impact of the conflict on copies, e-mail your requests to: poverty using a methodology for the identification and socialdev@worldbank.org. correction of outliers in time series. Our estimates 4