Report No. 26275-ET Ethiopia Risk and Vulnerability Assessment August 2005 Human Development Group III Africa Region Document of the World Bank Ethiopia: Risk andVulnerability Assessment Table of Contents ACKNOWLEDGEMENTS ..................................................................................... i 1. EXECUTIVE SUMMARY .............................................................................. 1 Promising beginning ..................................................................................................................................... 1 Buthighvulnerability remains .................................................................................................................... 2 Reasons for high vulnerability ..................................................................................................................... 2 Effectiveness of risk management strategies ............................................................................................... 4 Growth and Welfare costs of ineffective riskmanagement ....................................................................... 5 Role for social risk management .................................................................................................................. 6 2. SOCIAL PROTECTION FRAMEWORK ....................................................... 9 3. COUNTRY CONTEXT ................................................................................ 13 Output Growth Trends ............................................................................................................................... 13 Poverty trends .............................................................................................................................................. 14 Risks. growth and poverty traps ................................................................................................................ 20 Identifyingthe vulnerable .......................................................................................................................... 21 The distribution of the vulnerable ............................................................................................................. 24 4 . DENTIFYING SOURCES OF RISKS IN ETHIOPIA .................................... 25 Covariate Risks InEthiopia ....................................................................................................................... 25 Low andunpredictable rainfall (or frequent droughts) ............................................................................. 25 Domestic and international price volatility............................................................................................... Crop and livestock diseases...................................................................................................................... 29 29 Unpredictable policy changes................................................................................................................... 33 Environmental degradation....................................................................................................................... 34 Idiosyncratic risks ....................................................................................................................................... 34 HIV/AIDS ................................................................................................................................................ 34 Malaria ...... .............................................................................................................................................. 34 Ill-health, disability and mortality ............................................................................................................ ~ 35 Structural and life-cycle factors ................................................................................................................. 36 5. ASSESSING EFFECTIVENESS OF RISK MANAGEMENT STRATEGIES 39 Riskmanagement Strategies by households ............................................................................................. Riskmitigationstrategies ......................................................................................................................... 39 39 Riskcopingstrategies............................................................................................................................... 40 Effectiveness of household strategies ......................................................................................................... 42 Markets for risk sharing ............................................................................................................................. 43 Public Risk management strategies ........................................................................................................... Risk-reducingprograms........................................................................................................................... 47 47 Public programsfor copingwith risk....................................................................................................... 51 Effectiveness of public risk management Programs ................................................................................ 58 Welfare cost of shocks ................................................................................................................................. 60 6. HELPING HOUSEHOLDS MANAGE RISKS BETTER: BUILDING BLOCKS OF A SOCIAL PROTECTION STRATEGY 66 SustainingGrowth.................................................................................................................................... ....................................... 66 Reversingor haltingenvironmentaldegradation...................................................................................... 68 69 Makingexistingsafetynets more effective andcomplementary.............................................................. Specialfocus oncovariatehealthrisks..................................................................................................... 69 7 . CONCLUSION AND NEXT STEPS ............................................................ 72 Lessons learned ............................................................................................................................................ 72 Next steps ..................................................................................................................................................... 73 A N N E X E S .................................................................................................... 75 List of Tables TABLE 3-11ETHIOPIANGROWTHBY SECTOR INTHE 1990s......................................................... 14 TABLE 3-2: TRENDS INNON-CONSUMPTIONPOVERTYINETHIOPIA, 1995-2000..................... 19 TABLE 3-3: VULNERABILITY TO CONSUMPTIONPOVERTYLINEDUETO RAINFALLAND INCOMESHOCKS, 1995/96-1999/2000.......................................................................................... 22 TABLE 3-4: TRANSITIONINAND OUT OF POVERTYBASEDONACTUAL AND PREDICTED EXPENDITURE,ETHIOPIA, 1995/96-1999/2000............................................................................ 23 TABLE 4-1: PERCENTAGEOF HOUSEHOLDSREPORTINGTHAT PARTICULARTYPE OF EVENTH A S CAUSED CONSIDERABLEHARDSHIP(LOSS OF INCOMEORWEALTH) DURINGTHE LAST20 YEARS ....................................................................................................... 28 TABLE 4-2: SOURCES OF RISKINETHIOPIA SUMMARIZED BY SUBJECTIVERISKINDEX OF TABLE 4-3: PRICEOF COFFEERECEIVEDBY THE FARMER(INBIRRPERKG)......................... INCIDENCEAND SEVERITY OF RISKS........................................................................................ 29 TABLE 5-1: TYPES OFACTIONS TO MANAGE RISKBY INDIVIDUALS AND HOUSEHOLDS...33 39 TABLE 5-2: OFF-FARMWORK PARTICIPATIONINTWO WOREDAS, TIGRAY REGION............40 TABLE 5-3: SOURCESTO GET 100BIRRFORUNFORESEENCIRCUMSTANCESINA WEEK...41 TABLE 5-4: PROPORTIONOF HOUSEHOLDSWHO CANGET 100BIRR INA WEEK FOR 43 TABLE 5-5: RISKMITIGATINGORRISK-PREVENTIONACTIVITIES............................................. UNFORESEENPROBLEMS............................................................................................................. TABLE 5-6: INDICATIVECATEGORYOF FOODINSECUREHOUSEHOLDSINETHIOPIA.........48 TABLE 5-7: PROGRAMS,ESTIMATEDCOSTSAND SIZE OF BENEFICIARIES,ETHIOPIA.........52 54 TABLE 5-8: ESTIMATEDEFFECTSOF RAINFALL AND TRANSITORY INCOMEON CONSUMPTION(1995196-1999/2000)............................................................................................. 63 Listof Figures FIGURE3.1: CONSUMPTIONGROWTHBYDECILE.ETHIOPIA. 1995-2000.................................... FIGURE3.2: CONSUMPTIONGROWTHINRURALAND URBANETHIOPIA. 1995-2000..............16 16 FIGURE4.1: AVERAGE ANNUAL RAINFALL. ETHIOPIA(1967-2000)............................................. 26 FIGURE4.2: AVERAGE STANDARDDEVIATIONOFANNUAL RAINFALL, ETHIOPIA(1967- 2000).................................................................................................................................................... 26 FIGURE4.3: AVERAGE ZONAL RAINFALLAND COEFFICIENT OF VARIATION, ETHIOPIA, FIGURE4.4: CEREALAND FOODPRICEINDEXES, ETHIOPIA,JULY 1997-APRIL2000..............27 1967-2000............................................................................................................................................. 30 FIGURE4.5: CEREALPRICESINADDIS ABABA WHOLESALE MARKET, ETHIOPIA, 1996-2002 ............................................................................................................................................................. 31 FIGURE4.6: PERCENTOFHEADSOF HOUSEHOLDSREPORTINGHEALTHPROBLEMSTWO MONTHS PRIORTO START OF SURVEY.................................................................................... 35 FIGURE4.7: PROBABILITYFUNCTIONSOF BECOMINGDESTITUTEBY SEX OFHOUSEHOLD FIGURE5.1: TOTAL SPENDINGONRISK-COPINGPROGRAMS(MILLION USD)......................... HEAD.................................................................................................................................................. 37 55 FIGURE 5.2: PERCAPITAFREEAID AND GIFTS, 1995196AND 1999100.......................................... 57 FIGURE 5.3: REALPERCAPITAAID FROMGOVERNMENTAND NON-GOVERNMENTAL ORGANIZATIONS............................................................................................................................. 57 FIGURE 5.4: THE EFFECTOF DROUGHT ONAGRICULTURALGROWTH, ETHIOPIA, 1981182- .................................... 61 FIGURE5.5: IMPACTOF 1984DROUGHT ONCEREALPRODUCTIONAND YIELD..................... 199912000.................................................................................................. 62 List of Boxes BOX 4.1: MULTIPLERISKS (BUNCHINGOF RISKS) WORSENVULNERABILITY........................ 32 BOX 4.2: DECLININGCOFFEEPRICESMAY DEEPENPOVERTY.................................................... 33 BOX 4.3: CHANGES INPOLICYHAVEA LONGLASTINGEFFECT................................................. BOX 4.4: EFFECTOF HOUSEHOLD SIZEONHOUSEHOLDVULNERABILITY I S AMBIGUOUS.33 BOX 5.1: HOW LONGCANTHE AVERAGE SUBSISTENCEHOUSEHOLDRELY ONOWN HARVEST?AT MOST7 MONTHS.................................................................................................. 44 List of Annexes ANNEX 1: METHODOLOGYAND DATA.............................................................................................. 76 ANNEX 2: TABLES.................................................................................................................................... 83 ANNEX 4: REFERENCES........................................................................................................................ ANNEX 3: FIGURESAND GRAPHS...................................................................................................... 100 120 TABLE A 1: TYPE AND TIME OF SURVEY........................................................................................... 77 TABLE A 2: SIZE OF AGE COHORTSAND AVERAGE SIZEOF CELLS. RURAL ETHIOPIA. 78 TABLE A 3: SUMMARYSTATISTICS FOR 1995/96AND 199912000.................................................. 1995/96-1999/2000.............................................................................................................................. 83 TABLE A 4: SUMMARYSTATISTICSOF RAINFALLVARIABLES(MM). ETHIOPIA. 1967-2000.85 TABLE A 5: ESTIMATES OF HEAD COUNT RATIO, POVERTY GAP AND SQUAREDPOVERTY 85 TABLE A 7: CONSUMPTIONGROWTHRATESBY DECILE, ETHIOPIA, 1995196-199912000........86 TABLE A 6: REGIONAL POVERTY COUNT, ETHIOPIA, 1995196-199912000,................................... GAP INDEXES, ETHIOPIA, 1995 AND 2000 (%). ......................................................................... 86 TABLE A 8: CONSUMPTIONINEQUALITY (GIN1INDEX), ETHIOPIA, 1995196AND `1999/2000.86 TABLE A 9: CHARACTERISTICSOF THE CHRONIC, TRANSIENT, AND NOT SO POOR(RURAL HOUSEHOLDS),ETHIOPIA, 1995196-199912000............................................................................ 87 TABLE A 10: DETERMINANTS OF CONSUMPTION, ETHIOPIA, 1995196-199912000. (DEPENDENT VARIABLE =LOGOF CONSUMPTION).................................... .......................... 90 TABLE A 11: DETERMINANTS OF INCOME, ETHIOPIA, 1995196-199912000.(DEPENDENT VARIABLE = LOGOF TOTAL INCOME). ........................................................ ..............................92 TABLE A 12: DETERMINANTS OF CONSUMPTIONINCLUDINGTRANSITORY INCOME, ETHIOPIA, 1995/96-199912000. (DEPENDENTVARIABLE =LOGOF CONSUMPTION).......94 TABLE A 13: PER CAPITA REAL AID AND GIFTSTO HOUSEHOLDS(INBIRR), ETHIOPIA, 1995/96................................................................................................................................................ 97 TABLE A 14: PER CAPITA REAL AID AND GIFTS TO HOUSEHOLDS(INBIRR), ETHIOPIA, 199912000............................................................................................................................................ 97 TABLE A 15: PER CAPITA REAL AID AND GIFTS TO HOUSEHOLDS(INBIRR), RURAL ETHIOPIA, 199912000............................................................................,.,,,,,,,.,,,..,............................98 TABLE A 16: PER CAPITA REAL AID AND GIFTS TO HOUSEHOLDS(INBIRR), URBAN ETHIOPIA, 199912000........................................................................................................................ 99 FIGUREA 1: CUMULATIVE DISTRIBUTIONOF CONSUMPTION, ETHIOPIA, 1995196AND 1999/2000.......................................................................................................................................... 100 FIGUREA 2: CUMULATIVE DISTRIBUTIONOF CONSUMPTION, ETHIOPIA REGIONS, 1995196 AND 199912000. ................................................................................................................. FIGURE A 3: SHIFT INCONSUMPTIONDISTRIBUTION, ETHIOPIA, 1995196- 1999120 FIGURE A 4: SIMULATED EFFECTS OF ASSETS, EDUCATION, DEMOGRAPHICS, SERVICES, AND SHOCKS (RAINFALL AND HEALTH) ONCHANGESINCONSUMPTION, ETHIOPIA, 1995196-1999/2000............................................................................................................................ 102 FIGURE A 5: VISUAL IMPACT OF VARIABLES ON CHANGESINCONSUMPTION,ETHIOPIA, 1995196-199912000............................................................................................................................ 103 FIGURE A 6: SHIFT INCONSUMPTIONDISTRIBUTION, RURAL ETHIOPIA, 1995196- 199912000. ................................................................................................................................................. 103 FIGUREA 7. SIMULATEDEFFECTS OF ASSETS, EDUCATION, DEMOGRAPHICS, SERVICES, AND SHOCKS (RAINFALL AND HEALTH) ONCHANGESINCONSUMPTION,RURAL ETHIOPIA, 1995/96-1999/2000....................................................................................................... 104 FIGUREA 8: VISUAL IMPACT OF VARIABLES ON CHANGESINCONSUMPTION, RURAL ETHIOPIA, 1995196-199912000:...................................................................................................... 105 FIGUREA 9: SHIFT INCONSUMPTION DISTRIBUTION, URBANETHIOPIA, 1995196- 199912000. ........................................................................................................................................................... 105 FIGUREA 10: SIMULATED EFFECTS OF ASSETS, EDUCATION, DEMOGRAPHICS, SERVICES, AND SHOCKS (RAINFALL AND HEALTH) ON CHANGESINCONSUMPTION, URBAN ETHIOPIA, 1995196-199912000....................................................................................................... 106 FIGURE A 11: VISUAL IMPACT OF VARIABLES ONCHANGES INCONSUMPTION, URBAN ETHIOPIA, 1995196-199912000. ........................1................,.,.,,...........,..,..............................,,,,.,.... 106 FIGURE A 12: SHIFT INCONSUMPTION DISTRIBUTION, TIGRAY REGION, 1995196- 199912000. ........................................................................................................................................................... 107 FIGURE A 13: SIMULATED EFFECTS OF ASSETS, EDUCATION, DEMOGRAPHICS, SERVICES, AND SHOCKS (RAINFALL AND HEALTH) ONCHANGES INCONSUMPTION, TIGRAY REGION, 1995/96-1999/2000........................................................................................................... 107 FIGUREA 14: VISUAL IMPACT OF VARIABLES ON CHANGESINCONSUMPTION, TIGRAY FIGUREA 15: SHIFT INCONSUMPTION DISTRIBUTION, AMHARA, 1995196- 199912000.........108 REGION, 1995196-1999/2000........................................................................................................... 108 FIGUREA 16: SIMULATEDEFFECTS OF ASSETS, EDUCATION, DEMOGRAPHICS,SERVICES, AND SHOCKS (RAINFALLAND HEALTH) ON CHANGESINCONSUMPTION,AMHARA, 1995196-199912000............................................................................................................................ 109 FIGUREA 17:VISUAL IMPACT OF VARIABLES ONCHANGESINCONSUMPTION,AMHARA, 1995196-199912000............................................................................................................................ 110 FIGUREA 18: SHIFT INCONSUMPTIONDISTRIBUTION, OROMIYA REGION, 1995196- 199912000.......................................................................................................................................... 110 FIGUREA 19: SIMULATED EFFECTS OF ASSETS, EDUCATION, DEMOGRAPHICS,SERVICES, AND SHOCKS (RAINFALL AND HEALTH) ONCHANGES INCONSUMPTION,OROMIYA REGION, 1995196-199912000........................................................................................................... 111 FIGUREA 20: VISUAL IMPACT OF VARIABLES ON CHANGESINCONSUMPTION,OROMIYA REGION, 1995196-199912000........................................................................................................... 111 FIGUREA 21: SHIFT INCONSUMPTIONDISTRIBUTION, SNNP REGION, 1995/96- 199912000. ......................................... ............................. ............................................................................ * FIGUREA 22: SIMULATED EFFECTS OF ASSETS, EDUCATION, DEMOGRAPHICS, SERVICES, AND SHOCKS (RAINFALL AND HEALTH) ON CHANGESINCONSUMPTION, SNNP REGION, 1995/96-1999/2000........................................................................................................... 112 FIGUREA 23: VISUAL IMPACT OF VARIABLES ON CHANGESINCONSUMPTION, SNNP REGION, 1995/96-1999/2000........................................................................................................... 113 FIGUREA 24: SHIFT INCONSUMPTIONDISTRIBUTION, ADDIS ABABA, 1995196- 199912000. ........................................................................................................................................................... 114 FIGURE A 25: SIMULATED EFFECTS OF ASSETS, EDUCATION, DEMOGRAPHICS, SERVICES, AND SHOCKS (RAINFALL AND HEALTH) ONCHANGESINCONSUMPTION,ADDIS ABABA, 1995/96-199912000............................................................................................................ 114 FIGURE A 26: VISUAL IMPACT OF VARIABLES ONCHANGESINCONSUMPTION,ADDIS ABABA, 1995196-1999/2000................................................................................ ;.......................... 115 FIGUREA 27. EMPIRICAL DISTRIBUTIONOF AVERAGE ANNUAL RAINFALL, ETHIOPIA, 1967- 2000. .................................................................................................................................................. 115 FIGUREA 28. EMPIRICAL DISTRIBUTIONOF STANDARD DEVIATION OF RAINFALL, ETHIOPIA, 1967-2000...................................................................................................................... 116 FIGURE A 29: AVERAGEANNUAL RAINFALL, NATIONALAND BY REGION, ETHIOPIA (1967- 2000).................................................................................................................................................. 116 FIGURE A 30: STANDARD DEVIATION OFANNUAL RAINFALL, NATIONALAND BY REGION, ETHIOPIA (1967-2000).................................................................................................................... 118 Acknowledgements This report hasbeenpreparedby ateam comprisedo fMmesMessrs.: NadinePoupart (AFTH3, Team Leader) and Andrew Dabalen(AFTH3, Economist, Co-Team Leader), Antoine Bommier (Universityo f Toulouse), Pierre Dubois (University o f Toulouse), Tassew Woldehanna (Addis Ababa University), Tewodros Negash Kahsay (AddisAbba University), Patrick McGraw Kline (Michigan University), Jo-Anne Bour (JPA, MNSHD),and Diane Coury (Consultant). The team would liketo express its special appreciation and gratitude to Ato Getachew Adem, Head, Economic Policy and Planning Department, Ministry o f Finance and Economic Development (MOFED) andhisteam; and Dr.Abdulahi Hassen, General Manager, Central Statistical Authority, MOFED andhis team for facilitating this work. The report was preparedunder the general guidance o fMs.LauraFrigenti (Sector Manager, AFTH3), Mr. Ishac Diwan (Country Director, AFC06), Mr.ArvilVanAdams (Lead Sector Specialist, AFTHD), andMr.Kalanidi Subbarao (former Lead Sector Specialist for Social Protection, AFTHD). Peer reviewers were Ms.Margaret Grosh(LeadEconomist, HDNSP), Ms.GiovannaPrennushi (Lead Economist, PRMPR), Ms.Gillette Hall, (Economist, LCSHS), and Mr.Lant Pritchett (Harvard University). Invaluable comments were received from Mr.Harold Alderman (Lead Economist, AFTHD) and Ms.Trina Haque (Senior Economist, AFTH3). -1- 1. Executive Summary 1.1 This study i s a review o f risks and how they are currentlymanaged, by individuals, households, communities andthe public inEthiopia.Itstarts with the hypothesis that risks are important determinants o fpoverty and understanding how they are managedpermits us to assess the prospects and strategies for poverty reduction and sustainable development inthe future. The review focuses on the most common risks that affect individuals or communities. 1.2 The approach proposed to undertake this assessment i s the social risk management (SRM) framework. S R M refers to public intewentions that help individuals, households, andcommunities to better manage the diverse risks they face. Inthis framework, risk- inducedpovertytraps leadto highvulnerability, itselfdefined as the inability to prevent falling into poverty or destitution inthe future. Therefore, falling into povertytraps, are investments-not a drain onpublic coffers. all direct andindirect public expenditures, which protect or prevent households from 1.3 A risk and vulnerability assessment complements general poverty assessmentsin two ways. First, it takes a fluid and dynamic view o fpoverty and thereby expands the definition o f the poor to include the current poor and those who are at risk o fbeingpoor inthe future. Second, it accounts more explicitly for theimpact ofshocks onhousehold welfare. This risk-focusedperspective allows policy makers to devise a diversity o f strategies andinstrumentsto combat two sources o fhouseholdpoverty andvulnerability: one that i s due to low income or lack o f assets and typically the preoccupationo f standard poverty analysis, and the other that i s due to highvariance o f income. The welfare consequences o fboth situations can bemade worse by inadequate andineffective risk management instruments. PROMISINGBEGINNING 1.4 The review finds that the developments inthe 1990s haveprovided a promising platform to reduce poverty and vulnerability inEthiopia. The relative peace which was interruptedby a briefbut tragic border war with Eritrea, significant public investment, and a period o fhistorically favorable rains have ledto highgrowth rates inGDP, improvements inbroad measures o fwelfare, and evenmore dramatic improvements innon-consumptionmeasures ofwelfare. Startingwithconsumptionmeasures, we find that average consumption per adult heldsteady between 1995/96 and 1999/2000. We also find that the consumption o fthe poorest households edged up, but only slightly, compared to consumption o f the richest households. As a result, poverty rates didnot decline significantly between 1995/96 and 1999/2000. Incontrast to slow movements inconsumption, there was significant progress innon-consumptionmeasuresofpoverty inareas such as malnutrition, literacy, andaccessibilityto healthcare, water and education. Moreover, all this was done without measurable increase ininequality. - 1 - BUTHIGHVULNERABILITY REMAINS. 1.5 These positive developmentson welfare notwithstanding,many Ethiopians remainpoor andhighly vulnerable. About 45% o f the population finds itself inconditions o f mass poverty. Inthe same period that poverty rates came down, vulnerabilityto poverty, whether measured as the probability to fall into poverty inthe future or movement inand out o fpoverty betweenperiods, remainedhigh. About 70% of cohorts, defined as groups o fhousehold heads ofthe same age group livinginthe same administrative zone, who were followed between 1995/96 and 1999/2000 were predicted to have more than a 50% chance to fall into poverty inthe future. About two-fifth o fthis estimated vulnerabilitycan be explainedbythe knowledge o frainfall shocks alone. When looking at the transitions inand out o fpoverty we find that about 10% of the rural population remainedpoor inboth periods, while another 35% were predicted to move inandout ofpovertybetweenperiods. These numbers andthe recent drought of 2002/03, portray the fragility o f recent gains inimproving welfare. They suggest that inadditionto the 45% ofthe populationwho remainpoor, another 25% who arenotpoor now have a highprobabilityo f falling into poverty inthe futurewith a single large scale shock. 1.6 Furthermore, despiteimpressive gains innon-consumptionmeasures o fpoverty, major risks remain inkey areas. As an example, 57% o f children less than 5 years continue to be malnourishedand 31% severely so. Ifconditions remain as they are, every Ethiopian childbomnow faces these chances o fbecoming malnourishedinthe future. Such exceptionally highlevels o f malnutrition, compared to countries at similar levels o f income, have persistent long term impacts. Inparticular, malnutritiondiminishes the futureproductivity of survivors, thus makingthemmore likely to fall into poverty traps andtherefore vulnerability. 1.7 Vulnerability also differs geographically, with the three regions of Tigray, Amhara and SNNP exhibitingthe highest estimated levels. Furthermorebeingelderlyor livinginfamilies with highdependency ratio and inisolation (that is, further from water andhealth services, etc.), havingno assets or engaging insubsistence agriculture, was associated with higher vulnerability, suggesting localized impoverishment or possibility o f a poverty trap, despite positive per capita growth. REASONSFOR HIGHVULNERABILITY 1.8 The ability o f ahouseholdto reduce or prevent vulnerability depends on three variables. The first i s the severity and frequency o frisks it faces. The second concerns the level o fthe household's own resources, which can be financial (e.g. savings), skills (training, education), or physical (landand livestock). The thirdregardsaccess to social networks (family, relatives, communal associations, markets, etc.) or public programs. This report finds that vulnerability remains highinEthiopiabecauseo f (a) the presence o fmultiple and overwhelming risks, and (b) inadequate andineffective management o f these risks by households, through markets and the public domain. - 2 - 1.9 Sources of risks andvulnerability: The most compelling risks faced by Ethiopian households tendto be covariate, inthat they affect a network o f individuals or communities all at once. The most well-known o f these risks i s drought, andbetween 1978 and 1994, 15 droughts which have ledto the displacement, injury or death o fmore than a million people have beenreported. Drought affects all households regardless o f livelihoods andi s most often experienced as harvest failure, loss of livestock, food and water insecurity. Another common and disruptive risk i sprice risk, particularly o f grains (the mainconstituent o fthe diet o fthe majority population) andcoffee, the main export crop. Price volatility inevitably triggers conflicts o f interest betweennet producers or traders on the one handandnet consumers on the other. Periods o f low prices favor consumers as they buy the grain cheaply andtheir welfare improves, but they are injurious for the net producers and sellers. Whenprices rise sharply the opposite happens. That said, high, frequent and unpredictable price changes introduce uncertainty into market conditions for all concerned. Both farmers and traders hesitate to participate inunstable marketwithout adequateprotectionagainst adverseprice movements. Many farming households withdraw to self-sufficiency, while traders may not enter the market at all. The outcome may be low production, thinmarkets, and overall poverty and vulnerability. While price risk affects everybody, the poor are especially vulnerable to the adverse effects o fthis risk,becauseunlike the rich, they no savings with which to protect themselves. 1.10 Some risks, though considered individual, are overwhelming andhave spillover effects that can easily deteriorate into community-wide risks. One o f these i s malaria, a disease remains prevalent inlarge areas o f the country. It i s estimated that 40% o f the population i s at risk o fmalaria, and about 24% live inareas inwhich malaria i s at levels epidemic. Pregnant women andvery young children are the population groups at highest risk, since they may not have acquired significant levels o f immunity against malaria. Inareaswith unstable malaria exposure, pregnant womenmayhave2 to 3 times higher risk o f contracting malaria than non-pregnant women living inthe same area. Moreover, malaria exposure may leadto additional adverse outcomes such as low birthweight, abortion, and neonatal death. Fear o f exposure to malaria also leads households to avoid malariaprone areas, thereby creating challenges for population and landuse planning. Inthe pasttwo decades, HIV/'IDS has emergedas anew majorrisk.About 2 million Ethiopians between ages 15 to 49, or 6.4% o fthe adult population, have already been exposed to this disease. The exposure i s higher for urban areas, where adult prevalence rates approached 15%, and for the 25-29 years olds where the rates were 17% bythe end o f 2001. Inthe same year, 160,000 died from HIV/AIDS. Cumulative deaths o fthe past have ledto 1million orphans. 1.11 Vulnerability to poverty inducedby these risks can be reduced or abated ifthe strategies for managing risks are sufficient. However, inthe context o f Ethiopia, household vulnerability to risk-induced poverty remains highbecause the household's own resources, opportunities available inthe market or current public policies and strategies for managing risks are often inadequate and/or ineffective. - 3 - EFFECTIVENESSRISKMANAGEMENTSTRATEGIES OF 1.12 Household strategies to manage risks are often inadequate and ineffective. Many o f the well-known householdstrategies to deal with risk either before it occurs (e.g. multiple crops on one field, mixedcrop livestock systems), or after (e.g. selling livestock, reducing health expenditures), do not provide all the required protection. But since these strategies involve foregoing activities with highproductivity, innovation, or future investment, they come at a highcost to the individual household and to society. Traditional sources o fmutual insurance also tend to be localized (confined to networks within narrow geographic reach) and exclusionary (closed to poorer households). 1.13 Markets for sharing risk are poorly developed. After manyyears o f control under the Dergregime, the market systemis taking root inEthiopia, but it is still very small. There are only about 25 to 30 grain traders, andthey operate inonly a few marketsandtrade primary within a 200 kmzone. Furthermore, most markets, especially factor markets (land, labor, credit) andfinancial markets (banking, micro-finance, insurance), tendto be incomplete and characterized by high transaction costs. For example, land sales are prohibited and leasing andrenting have manyrestrictions. Because o fhighcosts ofhandlingandtransportation, maize farmers get only 1/3 o fthe marketprice. For the same reasons, food cantake up to 30 days to reach the final consumer. 1.14 Moreover, existing public risk management programs have their own limitations. The first is that even though the risk mitigation strategies such as supplemental irrigation, water harvesting, agro-ecological packages, and resettlements, provide an important step to address the problem o f food insecurity, they introducetheir own risks (inhealth, environment, and conflict) inaddition to the challenges o f implementation. Unless these risks are addressed at the outset, risk-averse households may not be willing to adopt the strategies. Second, even ifimplementation i s relatively successful, these strategies would not be sufficient to solve all the problems o f chronic poverty and vulnerability, since the estimated size o f the beneficiaries that can be helped from such programs i s only a fraction o fthose estimated to be indeep poverty and vulnerable. Third, some key strategies are not as effective as they could be, either because o f design or unintendedconsequenceso fpolicy. For instance, water-harvesting i s designed to reduce intra-annual, not inter-annual, variability o frainfall. Ifrain fails in one o fthe two seasons inthe year, the households still may not be protected from hunger for part o fthe year. Similarly, the well-intentioned policy o f scalingup fertilizer use by linkingits distribution to extension services has drivenout private participants, increased farmer debt, and created uncertainty over future fertilizer credit and use, factors which underminethe goals o fADLIand riskreduction. 1.15 Fourth, existing risk-coping strategies have a narrow focus on drought risk and its associated food insecurity. While this i s understandable given the frequency and devastation o f droughts, it has mutedneeded attentionto prevent or reduce other risks such as malaria, HIV/AIDS, andmalnutrition, all o fwhich have as muchpotentialto lock households into poverty traps. - 4 - 1.16 Fifth, the existingnationalfood security program, due (according to Government) to a limitedcountry income level, has a narrow coverage. It targets food-insecure households in 156 woredas inonly 4 out o f 9 regions and 2 administrative councils, thereby passing over millions of households that are food insecure or vulnerable to severe andlong-termpoverty. While understandable, boththe size o fthe needywho are currently excluded andthe implications o fthis position for poverty reduction inthe long term are difficult to ignore. Consider first, the size o fthe problem. The food security programnow targets about 5 million people, or 1millionhouseholds assuming an averagehousehold size o f 5, which i s about 10%o fthe population. However, in 1995/96,33% o f 10millionhouseholds inthe country qualified to be food insecure, accordingto the definition o f a minimumo f 2200 Kcal/day, and 17% as extremely food insecure, a consumption level o f less than 1650Kcal. Evenifthe current beneficiaries o f the program are all extremely food insecure, an additional 7% o f all households who are inthe same position as thebeneficiaries wouldhavenopublic assistance,which mightbe the only way for some to avoid sinking deeper into destitution. Inaddition, when inter- generational transfer of disadvantage i s taken into account, the children involved are more likely to have less education and poorer health, not to mentionpoor nutrition, leading to persistent poverty. 1.17 Sixth, the risk-copingprograms are primarily designedandusefbl for relief. While implementing them, potentially creative ways to use food aid for a muchbroader objective o fbuildingand protectingpublic andhousehold level assets i s missed. At present, there is no active practiceto integrate food securityprograms with efforts to buildandprotecthumancapital ofthe most vulnerable groups such as orphans inpoor households, girls, or households inpastoralist areas. To give two examples, food aid transfers directed to households or communities inexchange for sendingor keeping children inschool i s small scale andreaches only about 260,000 eventhough such programs may be suited to improving the education o f orphans, girls or children from pastoralist areas. Similarly, only 20% o f food aid distribution i s usedfor public works programs that are aimed at improving community assets. GROWTH AND WELFARE COSTSOF INEFFECTIVE RISK MANAGEMENT 1.18 Ineffective risk management results ina massive loss o fwelfare. InEthiopia, the most visible result i s large-scale physical death from two scourges: lives threatened by and lost to drought, havebeentragically high, especially inthe past; and HIV/AIDSand malaria have killedhundredso f thousands with little abating to date. These scourges have other impacts. Income losses have beenstaggering. Inthe most extreme cases, droughts can shrink household farm production by up to 90% o f output ina normal year, whichmagnifies the problems ofmalnutrition andinfant mortality. Diseasepreventionis also costly. Households mustpay up to 30% o f their farm income to preventmalaria for one year. Reducedincome inturn leads to lower consumption. Specifically, a 10% reduction inincome reduces consumptionby up to 7%. A decrease inrainfall by 10% can leadup to 10% loss inconsumption. For a single household, even a single experience o f one o f these shocks would be unbearable. But whenthey are bunched together, as i s often the case, the scale o fthe problembecomes magnified. Such losses, occurring to a large traditional agricultural system, would deter even the bravest households from adopting high-productivity buthigh-riskeconomic activities, such as non-traditional crops, vegetables, h i t s or flowers, all o fwhich havebeen identified as crucial activities for export and agricultural diversification and sustained growth for Ethiopia. ROLEFORSOCIAL RISKMANAGEMENT 1.19 To reduce poverty traps andhighvulnerability, abettermanagement o f risks i s necessary. The social riskmanagement strategy recommended inthis review seeks to achieve a balance between risk-reduction or prevention and assetprotecting strategies. While Ethiopia's SDPRI?i s already implementingthis approach, there are a number o f areas where special attention is needed to improve the effectiveness o f existing strategies. These areas encompass strategies aimed at reducing risks to sustainingrapidgrowth, reversing or halting environmentaldegradation,reversingthe impact o f key community-wide healthrisks,and improving effectiveness and coverage o f asset- protectingprograms. The importance of developing asset-protecting strategies stems from the observation that even the best-intentioned o fpublic programs does not always ensure the full participation o fthe poor and the vulnerable (e.g. orphans, girls, isolated grOUPS). 1.20 To achieve effectiveness of existing risk management strategies. Some policy obstacles mustbe removed while other new policies mustbe initiated. Various actions enhancing rapid growth mustbe undertaken by improvingthe functioning o f land, fertilizer and grain markets, complemented bypublic investmentsinroads. Suggested policy actions needed inthis area include: a The clarification o f tenure policies through public statements precluding future landredistributions. a Restrictions concerningthe lengtho frental contract and size o f landthat can be rented lettinghouseholds on either side o fthe rental market (rentinginand rentingout) determine themselves how muchlandto rent andfor how long. The government must establish a legal system that will enforce such arrangements. a Reversingnon-competitive market developments inthe fertilizer market, by simplifyingthe complex and anti-competitive biddingandprocurement process, levelingthe field to access credit and foreign exchange and hastening the separation of fertilizer distribution and extension service. a Reducing hightransaction costs inthe grain markets bypromoting standardization, increasingroadnetwork, increasing credit for storage and encouraging formation o f trust- based institutions such as traders association. a Reversingenvironmental degradation, by adopting a stronger population control policy, starting with satisfying existing demand for family planning services. 1.21 A second group o fpriority policy actions concerns programs that already exist, butwhose reach and effectiveness canbe improvedby scaling up. Suggestions inthis category are; - 6 - a Reducing the incidence o fmalaria and HIV/AIDS, by scaling up existing programs andpromotingnew technologies, such as insecticide-treated bednets. a Expandingthe coverage ofthe food security programs to all the food insecure households inthe country through adequate andpredictable funding o fthe program. 1.22 A number o f existing strategies, especially to buildor protect valuable household assets and public infrastructure, hold great potential to reduce risk andor help households manage riskbetter inthe future. However, their implementationto date i s limited, and their effectiveness hasnotbeencarefullymonitored. Hopefully, lessons from these pilots can be usedto form the basis for future scaling up. The suggestionsinthis category of strategies include: a Introducing anationwide nutritionprogram, drawing on lessons fromthe child growth monitoring programs inthe food security project andbest practice ideas from international experience. a A nationwide food-for-school program aimed at improving the education of girls, vulnerable children such as orphans, and children inpastoralist areas drawing on lessons learned inexisting school-feeding programs. a Establishing a cash-based nationwide public works program, based on lessons learned from the on-going schemes (under implementationbyNGOs) and internationalbest practice. a Promoting water-harvesting technologies, after a careful cost-benefit analysis of existing designs. a Encouraging voluntary resettlements, after determiningthe available size o f underusedland, eligibility criteria for beneficiaries, minimuminfrastructure needs and funding, andtaking care to avoid risks such as localized conflict, environmental damage. 1.23 There are also a number o f risk management strategies that are widely accepted as needingreforms, but where fbrther evaluation and analyses is warranted. The list o f suggestedstudies include: a Impact o f food aid on markets. What are the relative contributions o f food aid inflows and other market imperfections (e.g. poor storage, lack o f credit, isolation, etc.) to domestic food price volatility and availability? a The role o f Ethiopian Strategic FoodReserve (ESFR). What market stabilizing role can ESFR play, ifany? a Monetization of food aid. What are the benefits o fmonetizing food aid? What are the risks? What are the components o f a transition process to full or partial monetization? a Scalingup public works program. What i s the design and fundingmechanisms for makingpublic works program act as a more effective riskmanagement tool? a Weather based insurance. How feasible is aweather-based insurance institution for Ethiopia? Who should it target (regional or woreda governments, producers or traders associations)? - 7 - 1.24 This review has taken the first step to identifythe sources ofrisks inEthiopia, and how they are managed now, and how they can be managed better inthe future, with the active participation o fthe state. It i s hopedthat this review would be one of a number o f steps to develop a viable social protection strategy that will serve all the poor inEthiopia. As an immediatenext step, it is hopedthat the current and future dialogue over the PRSC policy actions can benefit from the findings inthis review. Inparticular, together with other existing assessments, it can serve as an organizing framework for initiating the difficult and complex discussion over objectives o f a broad social protection strategy, trade-offs betweenprograms and selectivity. - 8 - 2. Social Protection Framework 2.1 Riskand poverty. Risks are uncertainevents whose chanceso foccurring are not known inadvance. Their presence, whether realized as shocks or anticipated, induces behaviors that lead to poverty. Risks can be idiosyncratic, that is, affect a single person, household, or covariate, affect a group o f individuals or households at the same time. Allhouseholds face risks, butdiffer intheir ability to manage them. There are two main avenues through which household management (or lack thereof) o frisks leads to deep poverty. First,Ex-ante (before risks are realized), whereby a household, which perceives its exposure to a risk to be high, will take action to reduce or mitigate the future impact o f the risk. For example, ahouseholdmightresist diversification o fcrops and livestock, or the reluctantto adopt anew high-returntechnology, or persevere increatingredundancies (farming many small disjointed plots, keepingmany heads o f cattle, or having many children). While these actions offer some insurance inthe event o f a shock, they come at a highprice, inthat they lock households into activities with low productivity. Second, Ex-post,wherebythe householdmay respond to the shock by taking actionthat reduces its ability to respond to future economic opportunities. These responses (e.g., selling assets such as oxen or seeds, andpulling children out o f school, etc.) may also deplete the household's ability to face the same or other risks inthe future. 2.2 Poverty and vulnerability. Poverty and risks leadto vulnerability. Vulnerability refers to the potential o fbeing locked into long-term poverty or destitution. A single severe shock can send a poor householdinto destitution, andrepeated shocks can send a previously safe householdinto poverty. Being vulnerable can therefore be understood as the "propensity o f a society (households) to experience substantial damage and disruption as the result o fhazards (e.g. drought, flood) and difficulty (lack o fresources) to cope with andrecover from them" (GoE, 1999b, p. 8)." Itis the presence ofrisk, the ideathat hturewell-beingi s uncertain, that distinguishes the concept o fpoverty from vulnerability. 2.3 A riskandvulnerability assessmentcomplements standardpoverty assessment in two ways. First, it takes a fluid and dynamic view o fpoverty and thereby expands the definition o fthe poor to include the current poor and those who are at risk o fbeing poor inthe future. Second, it calls for amore explicit assessmentofshocks onhousehold welfare. This allows policy makers to devise different strategies and instruments to effectively combat two sources o f vulnerability: the one resulting from a low level o f consumption due to lack o f assets, and the other resulting from a highlevel o f variance o f consumptionandpoor coping instruments inthe event o f a shock. 2.4 The ability of a householdto reduce or prevent vulnerability depends on four variables. The first i s the severity and frequency o f the risk it faces. A single catastrophic drought has the capacity to overwhelm even a householdthat can be considered relatively wealthy (i.e. one with large tracts o f land, several heads o f cattle, etc.) inmany developing countries. When such a drought or a moderate version i s repeated more frequently, it can erode the ability of any householdto protect itselfover time. The second concerns the household's own resources. These resources can be - 9 - financial (e.g. savings), skills (training, education), or physical (land and livestock). It i s generally believedthat the more o f these resources the householdpossesses, the better its chances o fweathering adverse circumstances inducedby risks. The thirdregards access to social networks (family, relatives, communal associations, etc.). For a very poor household, the mutual reciprocity ofrelatives or the generosity o f a neighbor, are sometimes the only forms o f security standingbetween it anddestitution. But ifa repeated or severe shock occurs, even these avenues o f assistance canbreak down, thereby deepeninghousehold's vulnerability. The fourth and the final channel i s the availability o fpublic assistance. Whenthe household's own resources are insufficient even with the best o f effort, perhaps becausethey are elderly or disabled or just too poor, or the social networks breakdown, say inthe face o f a large covariate risk, it often turns out that the only line o f defense against tragic outcomes i s the response o f the state. 2.5 Welfare costs of risk andvulnerability. Inefficient andinadequate management o frisks byhouseholds can have largeprivate and collective costs. As mentioned above, when ahousehold's riskmitigating actions lead it to forgo the value o fspecialization and higherquality, it becomes lockedinto low productivity activities. Furthermore, the household' well-intentioned responsesto protect itself from a current shock, can leadto perverse consequences and undermine its very survival inthe future. For example, selling the family landto meet the cost o f a funeral can render a household landless. Similarly, pulling children out o f school can reduce the fkture stock o f education for the child withdrawn from school, but also o fthe household. The result i s reduced asset stock (land and humancapital), which reduce future inputsto investment, and contribute to persistent poverty. 2.6 The welfare costs o finadequate riskmanagement are not bornjust by individual households. When a large fraction o f the population i s trapped inpoverty, there i s little prospect for sustainable high growth. Instead, aggregate poverty will prevail. Mass poverty diminishes avenues for greater wealth accumulation, revenue collection, public investment and broad development. Put simply, poor protection against risk and vulnerability perpetuates the vicious circle. 2.7 The case for social risk management. Socialriskmanagement (SRM) refers to public interventionsthat help individuals, households andcommunities to better manage the diverserisks to which they are exposed (World Bank, 2001; Holzmann and Jorgensen, 2000). The case for moving social riskmanagement (SRM) to the center of the economic development agendarests on four crucial observations. First, risks and inadequate risk management strategies act as major sources o f poverty traps. Second, the poor are oftenthose most exposed to diverse risks and at the same time the least able to protect themselves against the risks. Third, because o f the huge welfare costs o frisk- inducedpoverty traps, the S R M approach maintains that accelerated poverty reduction and sustained economic and humandevelopment cannot be accomplishedwithout adequate and efficient protection against risks. Indeed, as it has beenwidely recognized inpovertyreduction strategy papers (PRSPs), a successful strategy to fight poverty must include helpingthe current andwould-be poor. For the latter, forward-looking poverty interventions are necessary, which must, by definition, bebased on an assessment o fvulnerability. Fourth and finally, direct and indirect public expenditures on social - 10- protection are investments -notjust a drain on public coffers -inso far as theyprotect or enable household investmentsinhumancapital, environmentalquality, new products andtechnologies. Because ofthese assessments,the SRM approach advocates a multiplicity o f instruments(to reduce, mitigate, and cope) as well as institutions (including informal, market, or public) to deal with risks effectively. 2.8 Objectives of the Study. The aim o fthis study i s to better understand the connection between risks and vulnerability inEthiopia and identifykeypolicy measures to reduce household vulnerability. The study attempts to assess risks and vulnerability by: (a) undertakinga literature review that summarizes existing qualitative and quantitative research on vulnerability inEthiopia, focusing on a) main sources o f risks; b) risk management strategies; and c) the possible existence o fpoverty traps, and (b) complementingit with a quantitative analysis o f vulnerability to weather and health shocks. The quantitative analysis: (i) measuresthe impact o f weather andhealth shocks on households' consumption; (ii) establishes a typology o f vulnerable groups; and (iii)examines the relative importance o f shocks, humancapital, assets, and access to public investments in influencingthe future probability to falling into poverty. 2.9 Expected BeneJts. First,bymaking the connectionbetweenrisk and growth explicit, the study hopes to raise awareness and elevate the role o f social risk management to the forefront o fthe country's development strategy. 2.10 Second, this assessmentwill be the first nationally representative study o f vulnerability inEthiopia. Its results are expected to inform policy dialogue on on-going activities such as: (i) Ethiopia's PovertyReduction Strategy Credits (PRSC), which will focus on issues related to growth and vulnerability, food security, capacity building, and decentralized service delivery, especially ineducation andhealth; and (ii) Public the Expenditure Review (PER), which i s focusing on social sectors, including safety nets. 2.11 Third, beingamongthe first vulnerability assessmentsinAfi-ica, the studymay serve as a methodological guide for other countries planning to do the same. It shows how poverty dynamics can be linked to the measure o fvulnerability inthe absence o f panel data, by following cohorts, or groups o fhouseholds with fixed membership, constructed from repeated and independent cross-section surveys, which are available for a number o f countries. 2.12 Organization of the report. The rest o fthe report is organized as follows. Chapter 3 provides the context to the rest o fthe report bytracing, briefly, recent developments inEthiopia. The first part o f the chapter provides a quick overview o fthe evolution o f consumption and non-consumption measures o fwelfare. The second part extends these measures, especially the standard consumption poverty measure, to include thenotion o fvulnerability andprovide aheadcount o fthe vulnerable. Inthe process we show that the numbers look different ifa dynamic view o fpoverty i s taken into - 11- consideration. Chapter 4 takes up a discussion o f risks, one o fthe main causes o f vulnerability to poverty. For brevity, it looks at a few core risks that are perceived and experienced as compelling inthe country. Chapter 5 discusses how these risks are managed byhouseholds (informal insurance) via marketsandbythe public sector and assesses the effectiveness o f these strategies. It also provides an estimate o f welfare losses from inadequate management o frisks. Chapter 6 provides an outline o f a social protection strategyto help households better manage risks. Chapter 7 concludes. - 12- 3. CountryContext 3.1 The impact o frisks on individuals, households, andcommunities andhow bestto respond to them can only beunderstood within a social, economic andpolitical environment. Therefore, this chapter provides abriefbackgroundregarding developments inlivingstandards inEthiopia inthe 1990s. 3.2 The 1990s were a decade o f great promise for Ethiopia. First,the longandtragic civilwar came to an endin1991. Except for abloodybut short borderwar with Eritrea in 1998 andagain in2000, the country has enjoyedpeace for most o fthe decade. As a result, significant resources, bothmaterial and intellectual, could at last be usedto improve the social welfare of the population. Second, major economic reforms were introducedbythe govemment that took power in 1991. The reforms movedthe structure o f the country's economy from a centralized, command systemtowards a more liberal, market-based one. The currency was devalued, trade restrictions and price controls were gradually relaxed, and inflation was tamed, all o f which spurredlong-neglected private activities. Third, a major transformation inthe governance structure o fthe countrywas introduced, where a federal system o f government replaced the communist-inspiredcentralized and concentrated political power of the past. OUTPUTGROWTH TRENDS 3.3 In the lastdecade, outputgrowthhasbeenhighbut erratic. Since the endof the civilwar, growthhas averaged about 6 percent per year, thusplacing Ethiopia among the fastest growingcountries inAfrica. When adjusted for population growth, this translates to an annual average growth o f 2.8 percent duringthe period 1991-2001. Butthepositive average growthmasks largeyear-to-year fluctuations. Between 1991/92 and 1997/98, the year-to-year growth rate rangedbetween 10% down to 1% (Easterly, 2002). 3.4 The mainsources of growthhavebeenservices andindustrialsectors. Table 3-1 shows that most o fthe growth can be explainedby growth innon-agricultural sources. Since 1992, growth inthe agricultural sector, roughly 50% o f economic activity, averaged about 2.1%per annum, while services, with 40% share o f economic activity, grew at 9% per annum. Industry,despite its small size, contributed as much as agriculture to growth between 1992 and2000 (Easterly, 2002). Despitethe government's commitment to agriculture-led development, the contribution o f agriculture to growth has beenmodest. -13- (1992/93-1999/2000) Sectoral Share Agriculture 48.8 % Industry 11.0 % Services 40.1 % Average log growth per annum AgricultureGrowth 2.1% IndustryGrowth 8.5% Services Growth 9.1% GDP growth 5.5% Decomposition Agriculture Component 1.O% IndustryComponent 0.9% Services Component 3.6% GDP Total 5.5% 3.5 Policy and shocks have played major roles in the growth experience. O f the 2.8% per capita growth rate, it i s estimated that the permanent component (that i s the growth trend, which excludes the component o f growth associated with temporary recovery from year-to-year fluctuation) averaged only 1.1percent per year during 1991-2001. Part o fthe growthi s attributedto the recovery from inefficient policies o f the previous regime and the civil war. However, about 50% (1.4 percentage points) o fthe 2.8 percent per capita annual growth can be explainedby policy changes on inflation and exchange rate management, fiscal discipline, and infrastructure investment, initiated by the current government (Easterly, 2002). Bycontrast, the low growth inagriculture hasmuchto do with risks, such as unpredictable international prices, especially o f coffee, drought, deforestation, and soil erosion and degradation, which are very common inEthiopian agriculture. POVERTYTRENDS 3.6 Indevelopment practice, itisnow widelybelievedthat growthisimportant for poverty reduction and, to the extent that poverty and vulnerability are closely associated, to reducing vulnerability as well. Inthe context o f Ethiopia, ifexit from poverty is highly responsive to growth, the per capita growth rates inthe 1990swould imply large poverty reduction. Moreover, the benefits would be considered broad-based ifthe beneficiaries are drawn mostly from the ranks o fthe poorest. O nboththese counts the news from Ethiopia i s mixed. - 14- 3.7 First,there was littlegrowth inprivateconsumptioninthe 1990s. The average householdconsumptionper adult equivalent remainedunchanged between 1995/96 and 1999/2000. Figure3.1 shows the ratio o f consumptioninyear 1999/2000 compared to consumptioninyear 1995/96 by expenditure decile. Ratios greater thanone imply consumptiongrowth andvice versa. Ingeneral, consumption in 1999/2000was at about the same level as consumptionin 1995/96. 3.8 Second,the littlegrowththat was realizedwas capturedbypoorer households. Boththe richest andthe poorest households experienced stagnation in consumption growth. However, while the two bottom deciles saw a 1to 2 percent growth inconsumptionover four years, the consumption o f the richest decile decreased by 1percent (Figure 3.1). The average consumptiongrowthinallthe other deciles stayed somewhere between these two. The bottom 7 deciles recorded either small positive growth or maintained their consumption,'whilethe top 3 deciles experienced slight decreasesinconsumption. Effectively, there was little difference inthe experience between the rich and the poor with regardto changes intheir consumption. 3.9 A look at urbanandruralhouseholds separately shows that the samepatternof slow consumption growthprevailed(Figure3.2). But all ruralhouseholds, except those inthe top decile, hadsmallpositive consumptiongrowth, while all urbanhouseholds except those inthe bottom and top deciles had negative consumptiongrowth. Among urbanhouseholds, the middle deciles (4 to 7) experienced the largestdecreasesin consumption-up to 10% duringthe period. This suggests that, despitethe overall stagnation, rural households maintained or grew their consumptiona little, while urban households experienced a decrease inconsumption. 3.10 As inthe separation ofhouseholdsinto rural and urbanareas, an examination of consumptiongrowthpatterns byregion shows that households inAmhara had improvements inconsumption, those residing inSNNP saw no change inconsumption, while inTigray, Oromiya and Addis Ababa householdconsumptiondecreased (Figure A2). 3.11 Third, becauseof these slow consumptiongrowthrates,poverty rates remainedstagnantbetween1995/96 and 1999/2000. Although no official poverty line inEthiopiahasbeendeclared, the absolute poverty line of 1075birr in1995pricesthat was set inthe first poverty study usingthe 1995/96 household, income, consumption and expenditure survey(HICES), has come to be accepted as the de facto consumption poverty line (GOE, 1999b). Usingthis line, the Government's analysis showedthat the nationalhead count poverty rate declined from 46% to 44% between 1995/96 and 1999/2000 (see Table A5). A recent report shows that national poverty rates remained about 38% inboth 1995/96 and 1999/2000, ifone uses standard method o f calculating poverty-that is, assuming a single poverty line for all households', These trends in poverty rates implylow poverty elasticity o f growth. Furthermore, the national average ' See Well-Being and Poverty inEthiopia: The Role o fAgriculture, Agency and Aid, Report No. 29468-ET. For a discussion on the slight differences between the calculation o f poverty lines and the Government's study, see Box 1.3 inthe same report. - 1 5 - hidesthe fact that ruralpovertydeclinedslightlywhile urbanpovertyrose during the same period(see TableA6, Lowerpovertylines). Figure 3.1: Consumptiongrowth by decile, Ethiopia, 1995-2000. - CmlUlTptIml" 1898rtl31ve101985. "&,mal Source: World Bank Staff estimates from HICESiWMS 1995/96, 1999/2000 Figure 3.2: Consumption growth in ruraland urban Ethiopia, 1995-2000 Rural and urban consumptlongrowth 1 2 3 4 5 8 7 8 9 10 oxpnditumdeclles -rural mnsvmpllon In 1999100relaliveto 1995196-- -urban mnsumplion in 1999100relative10 1995196 Source: World Bank Staff estimates from HICESNMS 1995/96, 199912000 - 16- 3.12 The limitations o fheadcount ratio as a measure of welfare are well-known. For one, it i s sensitive to the size o fthe poor households near the poverty line. A small change inaverage per capita income can leadto a large change inheadcount ratio ifthere was a big initial bunchingof households near the poverty line, since the ratio i s drivenby how manyhouseholds cross the poverty line. The head count ratio also ignores the extent to which different households fall short o fthe poverty line. Therefore, we also lookedat the poverty gap index, which measuresthe aggregate shortfall ofthe poor people's consumption from the poverty line. The poverty gap, the per capita shortfall o f the poor, also followed the same trendbetweenyears: they remained unchangedduringthese periods (Table A5). 3.13 Finally, these slow rates of poverty reduction happenedwithout an increase inaggregate inequality. The most commonly usedmeasure o finequality, the Gini, decreasedslightly at the national level, from 0.30 to 0.29 (see Table A8). The decline o f inequality was not general. The Gini coefficient for rural areas declined from 0.28 to 0.27 while inurbanareas it increased from 0.35 to about 0.37. 3.14 A look at determinantsof changes inconsumption across years suggestthat distribution of rainfall, public investments, education and assets have contributed to improved consumption while distributionof illness was associated with lower consumption. These conclusions are drawn from comparing counterfactual distributions o f consumption. The idea i s simple. Fromthe actual distributionof consumption and covariates in 1995/96 and 1999/2000, we create one counterfactual distributionby holding all the covariates o f 1999/2000 at their 1995/96 level (called D(a)). We create another counterfactual distribution by holding all the covariates o f 1999/2000, except some (say assets or education or rainfall) at their 1995/96 level (called D(b)). The difference between D(b) andD(a) illustrate the impact o f variables that changed between 1995/96 and 1999/2000. As the impact o f a single or a few variables i s often small, it i s usually convenient to graphthe difference betweenthe densities. Note that the difference between the densities on average has to be zero, but the range o fwelfare values (inthis case, log o f consumptionper adult equivalent) for which it i s positive or negative show the different impact o f the variables o f interest over the whole distribution. For instance, ifthe differenceisnegativebelow somethreshold (say thepovertyline) andpositive above it, thenit means some households have moved above the poverty line and therefore welfare has increased. The resulting counterfactual distributions andthe impact of the variables that were allowed to change between 1995/96 and 1999/2000 for the whole sample, ruralhrban and for five regions are displayed ina series o f graphs (Figure A3-Figure A26). The graphs plottingthe difference inthe counterfactual distributions show the visual impact o f these variables. They leadto conclusions explainedbelow. 3.15 Favorable rainfallin 1999had a generallypositive impact on consumption. The 1995 to 2000 rainfall season appears to have been good for Ethiopian households. Average rainfall in 1999 was higher than the average rainfall in 1995. Furthermore, the average deviation o f the 1999 rainfall from the 30-year average rainfall was minus 8mm, compared to the average deviation o f 1995/96 which was minus 45". Moreover, the average rainfall ineach o f the years preceding 1999 (that i s 1996-1998) was higher -17- thanthe average in1999. Across space, rainfallhas amorepositive effect on consumption inAmhara andSNNP, but less pronouncedeffect inTigray and Oromiya. 3.16 Asset accumulationhada large effect on changesin consumption. Perhaps because of better rains between 1995/96 and 1999/2000 and general economic recovery, households reported increased accumulationo f assets such as farm and transport animals. We find a positive association between asset changes and consumption. 3.17 Publicinvestmentsinservices andinfrastructurecontributedto growthin consumption. Access to services and infrastructureis measuredby distance to schools, health centers, water, food marketsandtransportation services. Between 1995/96 and 1999/2000, the data shows improved access to some publicly provided goods (Table Al). The average Ethiopiannow lives closer to aprimary and a secondary school, ahealth center and water point. We find that the observed improvements inthese variables have led to positive improvements inwelfare and a reduction o fthe fraction below the poverty line. The positive effect o f infrastructure on welfare was strongerinrural areas, Tigray and Amhara, but less obvious inurbanareas, Oromiya and SNNPR. 3.18 The impactofthe incidenceof illnessbetween1995/96 and 1999/2000 was negative.The increase of consumptionaround the poverty line followed bydecrease in the uppertail o fthe distribution suggeststhat the distribution ofconsumptionresulting from changingreported incidence o f illness from its 1995/96 distribution to its 1999/2000 distribution, led to a decrease inconsumption, that is, a leftward shift o f consumption. The results are particularly strong for rural areas, as well as Tigray andAmhara regions. 3.19 These slow rates o f consumption growth and stagnatingpoverty are not confirmed unanimouslyby all recent analyses ofpovertychanges inEthiopia. Another studyusing panel data from 15 rural villages inthe central and southern parts o f Ethiopia found that food consumptionbetween 1989 and 1997 grew by an annual rate o f 9%, which i s significantly higher than the national average (Dercon, 2002b). Not surprisingly, these growth rates ledto substantial poverty declines inthese villages, from a food poverty rate of 61% in 1989 to 49% in 1994 (Dercon andKrishnan, 1998). A more recent estimate o f absolute povertyrates inthe same villages show a decline o f 30% to 25% between 1994/95 and 1999 (World Bank, 2004). 3.20 One possible reason for the slow decline inpoverty observed with national data sets i s that using a single poverty line disguises differences across space. Ifconsumption baskets andprices households pay for them are allowed to vary geographically, we find an overall decrease inconsumptionbut even higher decrease (increase) inrural (urban) areas (Table A3). The poverty gap measure also shows a decrease. 3.21 Tracing trends inhouseholdconsumption andpovertyi s harder inpractice than i s suggestedby theory because o fproblemso f data comparability, unevenintervals between surveys which render comparisons across time more challenging, and the choice o f the deflators to use inorder to make expenditures betweenperiods (and areas) comparable. Since all these hurdles exist for Ethiopia data sets, the debate and disagreements on the extent o f the changes inconsumption measures o fwelfare inEthiopiawill - 1 8 - continue. That said, there i s less disagreement on the magnitude andthe direction o fnon- consumptionmeasuresofwelfare. 3.22 Specifically,non-consumptionmeasuresof povertyshow significantprogress between1995/96and 1999/2000(Table 3-2). Stunting, ameasure oflong-run malnutrition for children less than 5 years, stood at about 58% in2000 for male children and 55% for female children, bothof which are extremelyhighby international standard. However, this reflects a decrease from 67% for male and64% for female in 1995 . (Christiaensen andAlderman, 2003). Furthermore, extreme stuntingdeclined from44% in1995 to 31%in2000. The decrease, inthe level ofmalnutrition,was experienced inbothrural andurbanareas, althoughthe reductioninthe latter was morepronounced. Regionally, more progress inreducing malnutritionhas occurred inTigray, S"PR and the two city administrativecouncils o f Addis Ababa and Dire Dawa. 3.23 School attendance rates inrecent years have beenexplosive. The gross enrollment rate, a widely usedmeasure of access to education, has improved from 30% in the late 1980s to 62% in2003. This can be said to be one o f the major accomplishment o fthe reformist government of Ethiopia. It will take time before these rising enrollment rates are translated into rising literacy rates, but already, the literacy rates have improved, albeit slightly, from 27% in 1995 to about 30%, in2000. This gain appears to come from a modest improvement inrural literacy, especially rural male literacy, from 19% in 1995 to 22% in2000. 3.24 The proportion o f the populationwith access to primaryhealth care, sanitation and clean water has also improved. The distances covered byhalfthe population(50%) to reach a healthcenter andwater source have all declinedbetween 1995 and 2000. Table 3-2: Trends innon-consumptionpovertyinEthiopia, 1995-2000 Poverty variables (%) 1995196 1999/2000 National Urban Rural National Urban Rural Stunting Stunted 67 56 68 57 45 58 Severely stunted 44 32 46 31 21 32 Literacy rate 27 70 19 29 70 22 Distance (Km) covered by 50% of the populationto the nearest: Health center 8 1 10 7 1 8 Clean water 2.5 0.1 3 0.7 0.1 1 Primary school 2 0 3 2.5 1 3 Secondary school 18 1 20 15 _ _ 1 18 _. Jotes: Stunting i s an anthropometrics index that are used to measure long run malnutrition for children less than 5 years old. A child i s stunted if the standardized height to age ratio i s two standard deviations away, and severely stunted if the standardizedheight to age ratio is three standard deviations away. Source: Dercon (1999), GOE (2002). - 19- RISKS, GROWTHAND POVERTY TRAPS 3.25 This briefreviewo f growth andpoverty trends provides anumbero fuseful lessons for the connection betweenrisk, growth and vulnerability inEthiopia. First, there was progress against broad measuresof poverty, although there i s less consensus on the size o f the decline inthe consumption measure o fpoverty. 3.26 Second, these positive developments notwithstanding, many Ethiopiansremain poor. The Government's estimate o f44% national poverty rate in2000, would translate to about 28 million Ethiopiansbelow the poverty line. This does not include additional millions who may bejust above the poverty line and could be forced to cross the line downward by the slightest o fmisfortune. Since, as we have argued, there i s a close connection betweenhighpoverty rates and vulnerability, this implies that large numbers of Ethiopiansremainvulnerable to poverty. 3.27 Third, despiteimpressive gains innon-consumption measures o fpoverty, major concerns remain. As stated above, 57% o f children less than 5 years continue to be malnourished and 31% severely so. Ifconditions remain as they are, every Ethiopian child bornnow faces these chances o fbecomingmalnourished inthe future. Such exceptionally highlevels o fmalnutrition, compared to countries at similar levels o f income, have persistent long term impacts. Inparticular, malnutrition diminishesthe future productivity o f survivors, thus making them more likely to fall into poverty traps and therefore vulnerability. 3.28 Fourth, even with a runof aperiod o f good macro-economic outcomes, explained inpartbygood fortune (a periodofgoodrains) andreasonablepolicies, therecent drought (2002/03) inEthiopia-which threatened up to 10 millionpeople with starvation -- is a vivid reminder o fjust how exposed the population i s to potentially catastrophic risks. 3.29 Finally, overall reduction of poverty does not rule out localized impoverishment. Our study suggests that there were spatial differences inthe gains from growth. A separate look at regionalpoverty trends shows that poverty rates declined only in Amhara (Table A6 andFigureA2), but increased or remained the same inmost other regions, a result that i s largely confirmedby the poverty study o f the government. Moreover, as already reported, the growth of average consumption inurban areas lagged behindat the growth inrural areas. 3.30 Purposely sampled surveys invarious parts o f Ethiopia also provide additional evidence on the possibility that specific groups may have beenlocked out o f the benefits o f growth, and therefore, may be trapped inpoverty. The ongoing studyby IDS/SC-UK on destitution inthe Northeastern Highlandsinthe Amhara region (IDS/SC-UK, 2002), found that in9 village sites (2,160 households), 20% o f the sample o f households that were not destitute (extremely poor) in 1992/93became destitute by 2001/02. The study also found the widespread feeling among the participants that: (i) the poorest people now are ina more severe situation than a decade ago; (ii) the proportion o fpoor households had increased; and (iii) the proportiono fhouseholds inthe `better-off group had fallen. - 20 - Those better-offhouseholds previously provided access to productive resources to the poor (e.g. use o f their land or oxenbythe poor), as well as assistance intimes o fneed (cash or food loans). This community-wide `slide' towards destitution greatly exacerbates the vulnerability o fthose already at the bottom endo fthe scale. 3.31 Inthe next section, we look at the size andthe characteristics ofthe vulnerable population. IDENTIFYINGTHE VULNERABLE 3.32 One way to see, unambiguously, who i s vulnerable to poverty (or another measure o fwelfare, say malnutrition) andwho i s not, as a result o f a shock, i s to observe the same individuals' consumptionand timing o fthe shock over a long period o ftime. This calls for panel data. InEthiopia, the only panel data available i s for about 15 villages inthe rural areas. While rich indetail about these villages, the data cannot give us a nationally representative story. For this report, therefore, we created apanel o f cohorts -a group o fhouseholdheads inthe same age grouping and living inthe same administrative zone. We usedtwo nationallyrepresentative surveys, HICES/WMS 1995/96 and 1999/2000, to create two period observations for each cohort. We also obtained information on rainfall receivedinthese zones for a period o f 30 years. These two sets o f information, allowed us to predict how the rainfall variability affected the consumptionvariability, and then, further,how the variability predicts consumptionattributable to rainfallvariability to predict the probability o f falling below the poverty line (for a fuller discussion o f the methodology and data sets used, see Annex 1). 3.33 Since the effect o f rainfall shock was mostly felt by agricultural households, we usedonly rural households inthe estimation. Furthermore, we droppedall observations from Addis Ababa, DireDawa and Harari, even ifthey were coded as rural principally because these were mostly urbanareas with small pockets o fperi-urban households classified as rural. Inaddition, we dropped all observations from Afar and Somali because only sedentary populations were surveyed inthese regions, excluding the majority o fthe residents there who were more likely to feel the effect o f rainfall shocks. Finally, we dropped all observations o fhouseholds that didnot earn any income from agriculture, even ifthey residedinrural areas. 3.34 A cohort whose probability o f falling below the poverty line is greater than 50%, i s defined as vulnerable to poverty. Table 3-3 shows estimated levels o f vulnerability. Ifwe usedtotal consumptionvariance, as manyas 70% ofEthiopiancohorts are estimated to be vulnerable (row 1, Table 3-3). However, we find that rainfall shocks alone account for 41% (that i s the ratio o f 29% to 71% inTable 3-3) o f the vulnerability. These results implythat 29% o fthe approximately 41 millionmral Ethiopians (or about 12 millionpeople) living inthe 6 regions inthe sample are likely to fall below the poverty line due to exposure to rainfall shocks. The results inrow 4 refer to the estimated size o f the vulnerable usingthe empirical distribution o frainfall. This measure i s obtained by a) simulating the log consumption o fhouseholdfor observed levels o f rainfall shocks, given a simple o f consumption that links consumption and shocks, and b) thencountingthe proportionofhouseholds with halfthe simulated log consumption -21 - fallingbelow the poverty line in34 years. The results show that usingthe empirical distributiono frainfall would implythat 15% o fhouseholds are vulnerable to poverty. Finally, we find that changes intransitory income account for up to 4 percentage points o f the vulnerable. Table 3-3: Vulnerability to consumption poverty line due to rainfall and income shocks, 1995/96-1999/2000 Variable Mean (%) Std. Dev. Total estimated vulnerability 0.71 0.46 Estimated vulnerability attributable to rainfall shocks alone 0.29 0.46 Estimated vulnerability attributable to transitory income changes 0.04 0.20 3.35 Another way to count the vulnerable is to look at movements inand out o f poverty. Table 3-4 shows the proportion o f cohorts who have beencounted inone ofthe four mutually exclusive states ofwelfare, usingthe rural sample. About 10% o fthe cohorts were poor inbothperiods. These cohorts form the group that would typically be classified as chronically poor. About 55% were classified as non-poor inbothperiods. That leaves another 35% to have experiencedan episode ofpoverty betweenperiods, the transitory poor. 3.36 Recent estimates o fthe fraction o fthe Ethiopianpopulationbelow the poverty line, show that poverty inrural areas declined from 48% to about 45% between 1995/96 and 1999/2000 (GOE, 2002, Table 3.6). While slow decline ofpoverty is consistent with the results inTable 3-4 (which shows a slight increase), there are two ways inwhich the dynamic view o fpoverty adopted here enriches the static estimates. First, we provide a clear distinction betweenthe chronic andthe transient poor. Since the static poverty measure i s silent about this distinction it masks the size ofthe population that could be caught inapoverty trap, a knowledge that is important for policy choices. With this distinction, it i s worth notingthat our estimates indicate that significant chuming took place within this period. Only 10% of the rural population was chronically poor, but three and a half times as many moved in and out o fpoverty. Second, by expandingthe definition o fthe poor to include the current poor andthose who are at highrisk o f becomingpoor, we show that the static poverty estimates potentially under-estimate the true level o fpoverty and therefore vulnerability. That is, while the population estimated to live below the officially accepted poverty line remainedbelow 45% inboth 1995/96 and again in 1999/2000, we estimate that about 71% o f the population has a highchance o f falling below the poverty line. This suggests that a simple comparison o fhead count ratios intwo periods may give the impressionthat povertyrate remained the same, when infact it mayhave increased inthe interimyears. - 22 - Table 3-4: Transitioninand Out of Povertybasedon actualand predictedexpenditure, Ethiopia,1995/96-1999/2000 Variable Mean Standarddeviation Year Poor only in 1995 0.10 0.30 Poor only in 1999 0.25 0.43 Poor inboth 1995 and 1999 0.10 0.30 Not poor inbothperiods 0.55 0.50 Static poverty measures (headcount poverty) mean Standarddeviation Headcount index 0.40 0.49 poverty gap 0.10 0.16 1995 poverty gap squared 0.04 0.07 headcount 0.42 0.49 poverty gap 0.11 0.17 1999 lpoverty gap squared 0.04 0.08 Source: World Bank staff estimates based on HICESiWMS 1995196 and 1999/2000. 3.37 The 15 village ruralpanelalso suggestedhighlevels of vulnerability. About one- thirdto one-half o fthemeasured poor households were no longer poor inthe following period. Specifically, for the period covering 1989-1994/95, about 35% o f households remainedpoor, 26% movedout o fpoverty and 16% moved into poverty. Moreover, ina"bad" year (defined as ayear inwhichthehouseholdexperiencesproblemsinmany areas, including rainfall at 50% less than the long runmean) nearly 60% o fhouseholds would be poor ignoring seasonality effects, and it is this fraction that were classified explicitly as vulnerable (Dercon and Krishnan, 2000a). Inthese studies, movement out o f poverty was linked to better education, proximityto all-weather roads, andpossessing crucial assets such as oxen and extensive land. 3.38 These quantitative measures o fvulnerability accord well with self-reported prospects o f future well-being. The household income consumption and expenditure survey o f 1999/00 asked households to report (a) whether their expected income one year aheadwould be worse than their current income, and (b) whether or not they could cope with anegative shock to their income o f 100birr, within a week, andhow they would do so. Overall the results indicated that households considered 1999/00 a good year, except inTigray andSomali. Infact the rainfall data inTigray andSomali appeared to havebeen lower in 1999 and 2000 compared to recent years (see FigureA29). Furthermore, the respondents expected income a year aheadto decrease. 3.39 Regarding the ability to cope, about 34% o f rural and 38% o furban residents reportedinability to raise 100birr ina week (GOE, 2002d). Inthe draft Participatory - 23 - Poverty Assessment (PPA) urbanhouseholds inAddis Ababa reported decreasedliving standards due to highunemployment, rising food prices, and increases incrime and prostitution (World Bank, 1998). The Consultations with the Poor in Ethiopia (World Bank, 1999) study foundthat respondents inbothurbanand rural areas reported decreases inliving standards over the past ten years. This may be due to the recall period or to the fact that Ethiopia experienced extremely volatile rainfall pattern, especially in the early part, inthat decade. Bothurban andrural households listed seasonalvariance in rainfall as the major cause o f vulnerability (World Bank, 1998). THEDISTRIBUTIONOFTHE VULNERABLE 3.40 Turningto the characteristics ofthe vulnerable, we findthat they are different from the non-vulnerableinseveral ways (Table A9). The results indicate those who would be considered vulnerable are older, have larger family sizes andhave higher proportions of female-headed households comparedto the non-vulnerable. They are more likely to live inAmhara andSNPPR and more likely to report higher incidence o f ill-health and decrease inlivestock assets. The reportedreductioninlivestock assets couldbe a general indication o f either loss or sale o fthese assets due to shocks. Fewer of themhavehadsomeprimary education compared to the non-vulnerable, andmore of themlive further away fromwater points. Finally, proportionately fewer o fthem engage incashcrop farming. Inparticular, while 27% ofthemproducecoffee, about 35% ofthe non-vulnerable do. Similarly, while only 2% o f them engage inchat production, 15% of the non-vulnerabledo. With regard to income from cash crops, we find that those vulnerable to poverty due to exposure from rainfall shock earn about 44% less income from coffee than the non-vulnerable. Geographically, vulnerability is widespread. Although, vulnerability i s common inthe larger regions o f Amhara, Oromiya and S"?, it is measurably higher insmaller andisolatedregions such as Benshangul-Gumuz and Gambella, andTigray which is encompassedbythe highlydegraded Northeast highlands. 3.41 The purposively sampledsurveys place risks at the center o fthe vulnerability that households face. For instance, both the IDS/SC-UK andthe 15 rural village panel, place weather conditions as a critical determinant of changes inpoverty status. These studies suggest that repeated rainfall failures are a source ofpovertypersistence or poverty traps. Inparticular, they show that althoughsome loss inconsumptiondue to adecreasein rainfall i s recovered the following period, a full recovery occurs only over several years. Ingeneral, persistentlypoor households appearedto live inunfavorable environments, generally inremote areas characterized bypoor road connections, and to be poorly endowed interms o f land andtechnical inputs. Highincidences o f illness inthese households have also strongly limitedtheir ability to grow andreap the benefits o f the reforms. Conversely, moving out from poverty was mainly associated with better geographic (good rain; more andbetter land), infrastructure (good access to roads and towns) and technical (higher use o f fertilizers) endowments andwith higherproducer crop prices. 3.42 Because risks are crucial to the varying trajectories towards, and degrees of, vulnerabilitythat households experience inEthiopia, we now tumto a review o fthe most common risks inthe country. - 24 - 4. IdentifyingSources of Risks inEthiopia 4.1 As was mentionedat the start, risks canbe correlated or covariant inthat they affect a network o f individuals or communities, either ina localized area or spread across the entire geography o f a country. Or they canbe uncorrelated-idiosyncratic-if they affect individuals within communities. Some risks are repeatedwith relative frequency, while others are very rare. Risks also differ intheir effect. Some can have catastrophic consequenceswhile others are mere nuisances. 4.2 Inthis chapter we relyonanumberofqualitative andquantitative surveys and existingstudies to identi@the different risks that individuals and communities in Ethiopia face, based on their geographic location and source o f livelihood. The discussion examines only key community and individual risks, as well as structural and life-cycle factors, which when bunched or interacted with other risks, makes living conditions harder for individuals. COVARIATE RISKS INETHIOPIA Low and unpredictablerainfall(or frequentdroughts) 4.3 There i s no doubt that drought risk i s the most well known o f the risks facing Ethiopia, andwe beginour discussion there. Rainfall inEthiopia, on average, i s low. A plot ofthe realizedannual average rainfall over the last 30 years shows a concentration towards the lower values (see FigureA27). The average annual rainfall i s estimated at about 11OOmm, all o fwhich i s realized within a period o f 5 months, split into two rainy seasons. But the most striking observation about the rainfall pattern inEthiopia, which maybe maskedbythe low average, i s the volatility. Figure4.3 shows the average annual rainfall over the 1967 to 2000 period. The year-to-year averages exhibit large swings, a clear indication o frisk. - 25 - Figure4.1: Average annualrainfall, Ethiopia (1967-2000) 120 ---z.-2- E 110 - E m 100- 2% c P : 90 - 80 - I I I I I 4.4 Another way to measure volatility i s to look at the standard deviation directly (see Figure 4.2). The figure plots average deviation from the national meanrainfall for a given year. The longrun(that i s 30 years) meandeviation i s about 400mm, which i s almost 40% o f the longrunmean annual rainfall. Because the deviation for each year i s obtained from average rainfall reportedover various zones o fthe country ineach year, the standard deviation demonstrates that there are potentially large spatial variations in average rainfall inthe country. Figure 4.2: Average standard deviationof annual rainfall, Ethiopia(1967-2000) 4.5 Furtherevidence o fthe extreme variability, andinequality o frainfall across space, i s shown inFigure 4.3, which plots the longrunmeanand the coefficient o f variation, the standard deviation o frainfall relative to the mean for each zone. Inother words, for each zone, the mean rainfall for 30 years andthe ratio o f that mean to the standarddeviation is obtained andplotted. The results implythat bothexpectedrainfall and their variability are very unequal. Spatial variability (comparison o f coefficient o f - 26 - variation across zones) ranged from a low o f 15% to a higho f 81%. These levels of variation are similar to Morocco, and appear to be amongthe highest inthe world (Peter Hazell, IFPRI, personal communication). Figure 4.3: Average zonal rainfall and coefficient of variation, Ethiopia, 1967-2000 8.0 I , 90.0 Zones =mean rainfall (in logs) +Mean CV (%) 4.6 Put together, the distribution o fmeanrainfall andits standard deviation implythat Ethiopiaexperiencesmore frequent episodes o flow andhighlyvariable rainfall. The highvariability of rainfallbothacross space andtime, presents problems and opportunities. A major problem i s that it leads to frequent, and sometimes catastrophic droughts. As an example, between 1978 and 1994 alone, there were 15 droughts and famines that have led to the displacement, injury, or death o fmore than a millionpeople (WorldBank, 2000). The pervasive nature of drought riskinEthiopiais evident inthe multiplicity o f experiences, told through quantitative and qualitative surveys, from which it is reported. 4.7 Drought risk is often experienced as harvest failure. Households inthe 15 village ruralpanelwere askedto identify the shocks they experienced over the past 20 years (Table 4-1). About 78% ofhouseholds reportedharvest failure, often causedby rainfall failure, as the most common type of hardship. The survey, conducted in 1994-95, also reveals that shocks have a long lasting effect informing household expectations that might shapetheir subsequent behavioralresponses: when households were asked to report the most recent occurrence of a particular shock that has ledto loss of income or wealth, the majority recalled the 1984 drought year (Dercon and Krishnan, 2000a). - 27 - Table4-1: PercentageofHouseholdsReportingThat ParticularType of EventHas Caused ConsiderableHardship (Loss ofIncomeor Wealth) Duringthe Last20 Years Percentageof households Modeyear of the most recent reportedly affected by type of event serious event Harvestfailure 78 1984 Policyproblems 42 1985 Labor problems 40 1993 Oxenproblems 39 1993 Other livestock 35 1984 Landproblems 17 1989 Asset losses 16 1985 War 7 1989 Crimehanditry 3 1986 4.8 Another way inwhich drought risk is experienced is loss of livestock. Livestock losses (oxen and other livestock), which also cause considerable hardship o f rural households, i s also causedby -- but not exclusively -- rainfall failure. Over 35% o f rural households reported this form ofhardship (Table 4-2). 4.9 A studytracking the distribution o fherdmortalityrates for seventeenyears among Borena households insouthern Ethiopia, found that average annual mortality o f herdswas 16% o fbeginningperiod stocks andpeakedat 46% in 1991/92, a drought year (McPeak and Barrett, 2001). Other evidence, from villages located inthe drought-prone upland and semi-arid lowlands, shows that average mortality could rise to 70% o fthe herdina drought year, compared to 11% ina goodrainfall year (Webb andReardon, 1992)'. 4.10 Drought risk is also reported as food and water insecurity. When a sample o f 49 groups (mostly pastoralists) from the arid and semi-aridparts of the Oromiya region, were asked to identifykeyrisks intheir lives, food andwater insecurity came to the top Figures as highas SO-80% for cattle and 30% for sheep and goats are often cited (Fafchamps (1998), Lybett et al, 2001). Yet as notedby Fafchamps, one need to remain cautious since these figures may include physical loss as well as distress sales. - 28 - I o f the list (see Table 4-2). Next to these hardships,the respondents cited poor livestock health, poor accessibility to health clinics andto schools, and unavailability o f inputs as the most important sources of vulnerability (Smith, Barrett, andBox, 2000). Table 4-2: Sources of RiskinEthiopia summarized by subjective riskindexof incidence and severity of risks Source of risk kll Samplt Ethiopia Kenya Female Poor Lgriiulturai 'astoral Sample size (number of groups) 120 49 71 59 -61 60 21 68 Food Availability 0.56 0.69 0.47 0.64 0.48 0.62 0.37 0.6 Water availability 0.53 0.46 0.59 0.46 0.6 0.52 0.65 0.51 Animal disease 0.31 0.30 0.32 0.24 0.21 0.26 0.16 0.35 Access to healthclinics 0.23 0.23 0.24 0.21 0.26 0.33 0.38 0.17 Availability of farm inputs 0.07 0.16 0 0.06 0.07 0.09 0.15 0.02 Access to schools 0.12 0.17 0.09 0.12 0.12 0.16 0.21 0.09 Livestockprices 0.15 0.15 0.14 0.08 0.21 0.08 0.07 0.18 Humandisease 0.09 0.12 0.08 0.09 0.1 0.12 0.06 0.15 Wildlife crop destruction 0.07 0.01 0.11 0.08 0.06 0.04 0.21 0.02 Conflict/violence 0.14 0.06 0.19 0.15 0.13 0.2 0.09 0.17 School fees 0.06 0 0.09 0.07 0.04 0.08 0.06 0.06 Crop failure 0.05 0.11 0.08 0.03 0.07 0.04 0.1 0.04 Access to transport 0.05 0.01 0.08 0.03 0.07 0.02 0.08 0.05 Pasture availability 0.04 0.05 0.04 0.01 0.08 0.03 0.05 0.05 Consumer goods availability 0.04 0.02 0.05 0.05 0.03 0.04 0.02 0.05 Spatial mobility 0.01 0 0.02 - 0 0.01 n 0.01 Source:FromTable 1 of ;h,Barreti idBox(2 E0.01T Note: The communities surveyed were allowed to-decide on their own what the most important sources o f vulnerability were and to make subjective ranking of the most important sources o f their vulnerability. Based on their responses, a riskindex ranging from 0 (no incidence) to 1(universally most severe risk) was constructed by dividing the risk incidence by the reported severity. Crop and livestock diseases. 4.11 Harvest failure, is caused not only by drought but also bypests, storage losses, frost, flood etc. The risk of crop disease i s underlined inthe survey on destitutionin 9 villages ("gott") inthe Northeastern Highlandso fAmhara region, which found that the overwhelming determinants o f good or badyears were natural factors affecting crop production: rain, but also pests, and crop diseases (IDS/SC-UK, 2002). Similarly, livestock losses can be caused by, diseases and a variety o f other factors, including theft andnatural death. Domestic and international price volatility. 4.12 Highlyunstable prices are very disruptiveto economic activities andliving standards. Instabilityarises from a variety o freasons including poor roadnetworks that - 29 - perpetuate isolation, poor storage facilities, and inflation. While price risk affects everybody, the poor are especially vulnerable to the adverse effects o fthis riskbecause, unlikethe rich, theyoften do nothave access to savings instrumentsto protect themselves. 4.13 InEthiopia, grain andcoffee pricerisksposethe greatest sourcesofvulnerability to households. The first, because amajority o f the population rely almost exclusively on grains for their livelihood, andthe second, because it i s the main export crop for the country. 4.14 Grainprices: A discussion on grain price volatility inevitablybringsup the well- known conflict betweennet producers or grain traders on the one hand and net consumers on the other. The interests o fthese groups are not always the same. Inparticular, periods o fprice collapse are good for consumers as they can buy grain at cheap prices and their welfare improves, but they are injurious to the net producers andsellers. Price hikes have the opposite effect. So, even though the discussion that follows is presented more from the point of view o fnet producers and sellers, these conflicting interests should be kept in mind. 4.15 To begin, Figure 4.4 shows the movements o f general food and cerealprices in the period from July 1997to April 2000. For the entireperiod, overall foodprice increase was low, reflecting the success o fthe Ethiopian government to control overall inflation inthe period. However, the cereal prices, which constitute the main consumption bundleo f the poor majority, rose higher than overall food prices. As the figure illustratesthe cereal price indexwas everywhere higherthan the foodprice index. Figure 4.4: Cereal andFoodPriceIndexes,Ethiopia, July 1997-April 2000 140 I I :i 60 Month. Year - 30 - 4.16 Additional evidence from a recent grain marketreviewinEthiopiapoints to the existence o fhighspatial and temporal volatility ingrain prices (Harrison, 2002). The month-to-monthmovementsof wheat andmaize prices variedwithin a narrow band between 1996 and 1998, but thenrose sharply in 1999, before falling dramatically in 2000 and2001 (Figure4.5). The prices o fteff followed the same pattern, although the decline inprices between 1999 and2000 was slower. Together, these three cereals constitute 80% o f cereals produced and consumed inEthiopia, so that their price movementsmirror the nature andmagnitude o f grain price risk faced by Ethiopians. An additional price risk comes inthe form o f spatial isolation, especially for households and communities located far from the centers o f cereal production such as DireDawawhere opportunities for price gouging are ideal. InDireDawaprices, especially ofmaize, were among the highest (Harrison, 2002, pg.12). Figure4.5: Cereal prices inAddis Ababa wholesale market, Ethiopia, 1996-2002 Comparison Between Nominal and Deflated Cereal Prices Addis Wholesale - 350.0 300.0 250.0 2=200.0 4 T e f f Deflated +Wheat Nominal 6B +Wheat Deflated tMaizeNominal 150.0 100.0 50.0 - 3 1 - Box 4.1: Multiplerisks (bunchingof risks)worsenvulnerability. Price risk alone i s often bad enough for the poor, but when it is bunched with other risks, it canbe devastating. For instance, drought risk can and often do magnify price instability. Grainprices fluctuate significantly throughout the year, but they are usually at their highest during the dry season or immediately preceding a badharvest (World Bank, 1998, pp. 34). Yet, this i s the period when significant wealth loss amongst the agriculturalists and pastoralists occurs thus exposing households to an additional vulnerability. For example, during a drought, pastoralists would be compelled to sell livestock inorder to purchase cereals. But if distress sale o f livestock floods the market, boththe increased supply and poor quality livestock can lead to as much as 90% d r o in ~ livestock mices3(Sellen et al.. 2001). 4.17 InEthiopia, suchbunchingofrisks is believedto have ledto arecent collapse of grain prices, from January 2001 to January 2002. First, unexpectedlygood rains inthe precedingseason, (October 2000 to January 2001), ledto an increase incereal output o f about 17% above average. Second, because o f a poor roadnetwork, the price differences between surplus and deficit areas couldnot be smoothed (arbitraged away), so that a spatial price differential remained. Finally, and at the same time, first food imports of 1 million tons (2000) and another half a million tons (2001), were brought into the country insuccession. Theresult was acollapse inprices, which while helphlto some consumers, such as low-wage laborers, ledto hugelosses to producers o f cereals (Harrison, 2002). Inthe year 2002, the same arbitrage problems and a failure o fthe rains beganto pushprices upwards. 4.18 So, whether bunched or not, unstable grain prices bringabout specific types of losses to farmers, especially those on the margins o fprofitable farming, low income consumers, and grain traders, which increases the vulnerabilityofhouseholds. 4.19 Coffee prices4.Although this risk i s not captured bynumerous household surveys, the Government o f Ethiopia identifies the collapse o f the world coffee price as a major shock to small farmers who are responsible for 95% o f the total coffee production. According to the SNNP Bureauo fAgriculture, the 2001/02 prices o f coffee and coffee beans declined by 62% and72% respectively compared to 1997/98. Similarly the report ofthe Oromiya Region shows that the price o fredcherry coffee has decreasedby 70% andthe one o fdry coffee by40% between 1998/99 and 2001/02 (GOE, 2002c, p. 196; Table 4-3). "During times o f drought, the livestocklgrain terms o f trade tend to collapse withthe herd, especially whenthe proximate cause is drought that also reduces grain supply" (Lybbert et al, 2001). Coffee is the most important export crop inEthiopia. Ethiopian Fiscal Year (corresponding international calendar years) 1990 1991 1992 1993 1994 (1997/98) (1998/99) (1999/2000) (2000/2001) (2001/2002) SouthernNNPRegion Red Cherry Coffee 2.25 2.50 1.go 1.60 0.85 Coffee whole 5.10 4.10 3.55 3.50 1.45 Coffee beans 10.20 8.20 7.10 7.00 2.90 Oromiya Red Cherry Coffee 2.33 1.40 1.oo 0.70 Coffee whole 3.02 3.20 2.80 2.46 Coffee beans 6.05 6.40 5.61 4.93 Box 4.2: Declining coffee prices may deepen poverty. There is a realpossibility that the developments inthe coffee market are not temporary phenomena, but instead long term expectations o f the price regime, with very little prospect o frecovery. According to the GoE, the coffee sector employs over 25 percent o fthe country's active population(directly or indirectly). Inthat case, the persistence o fthis shock, willmake a large fraction o fEthiopianhouseholds lose the return on their long term investment (perennial coffee trees) and deepen already highpoverty levels. Unpredictablepolicychanges 4.20 Households surveyedinthe 15 village ruralpanel rankedpolicy shocks (such as villagization, resettlement, banon migration andwage labor inthe 1980s) o fthe Dergregime, second only to harvest failure as events that had caused considerable hardship (see Table 4-2). Poorhouseholds interviewed inthe Consultations with thepoor study also identifypolicy changes as having ledto considerable losses inwelfare. Finally, 9 villages sampledinthe NortheastemHighlandsinAmhara region identify changes inland tenure, from the Derg landreform to the most recent redistributions, as major events affecting their livelihoods (IDS/SC-UK, 2002). Box 4.3: Changes in policy have a long lasting effect. It is clear fromDercon and Krishnan(2000a) and IDSISC-UK (2002) that changes inpolicy have a long lasting effect. Policy changes o f the 1980s are remembered, particularly, as having created severe problems. However, some recent policy changes to correct past policies have hadunintended consequences o f worsening the situation o f several groups. For instance, those who were once subject to forced settlement didnot get their land back inthe land re-distribution under the reformist government o f EPRDF when they returned. As a result, most have had to migrate since they could not findjobs other than porters (World Bank, 1999). While some households lost land, others were given poor quality plots (small insize andior less fertile). Peopletoo young at the time o fthe redistribution were also left out, as well as large families, as the size o f the household was not used as a criterion in allocation. 4.21 Moreover, the reduction o fplot size following the redistributionis seen as having jeopardized traditional agricultural practices aimed at maintaining soil fertility, such as fallowing and crop rotation. This is believedto havehad anegative impact on farm productivity, leading to households' andcommunities' impoverishment. -33 - Environmental degradation 4.22 A combination ofintensive farming practices and demographic pressure, have led to substantial landdegradation and growing numbers o frural households with inadequate landholdings. About 50% o f the highlandsare believedto be eroded, and 25% ofthem severely so. This has constrained agricultural productivity and rural income growth. It i s increasingly seen as a major cause o f vulnerability, especially among the youth (WorldBank, 1999). 4.23 A constant message inthis analysis is that seeminglyrational actions from the point ofview o f individual households couldhave large negative social consequences. Inthis case, the time-testedhousehold strategies to cut andbumtrees either to meetits own energy needs or to earn some income, as a coping mechanism, can leadto ruinous . effects on the environment and the naturalresource base for current and future generations. Insurveys, this strategy i s not identifiedas a significant source to cope with a 100Birr negative shock on income (Table 5-3, page 41), but this may be partly because these activities are considered illegal and punishable. IDIOSYNCRATIC RISKS HIVIAIDS 4.24 Strictly speaking HIV/AIDS affects individuals, but the key lesson from the last two decades i s that, much like the plagues o f the past, it has a huge social effect. InEthiopia,2 millionpeoplebetweenages 15to 49, or 6.4% ofthe adult population, have already been exposed to this risk. The exposure i s higher for urbanareas, where adult prevalence rates approach 15%, and for the 25-29 years olds where the rates were 17% bythe endo f 2001 (UNAIDS, 2002). This risk does not appear to be diminishingin the foreseeable future ifthe expressed fears o fthe population interviewedinConsultation with thepoor, i s any indication. In2001 alone, 160,000 Ethiopianshave diedfrom HIV/AIDS,and cumulative deathso fthepast have ledto 1millionorphans. Malaria 4.25 LikeHIV/AIDSmalarianot only affects individuals, buthas socialcosts. Malaria outbreaks tend to be spatially concentrated andbunchedwith other risks associated with rain. As a result, its impacttends to be magnified. InEthiopia, malaria remains a persistent risk inlarge areas o f the country. It i s estimated that 40% o fthe population i s at risk ofmalaria and about 24% live inareas inwhich malaria risk often exceeds expected levels (epidemic). Pregnant women andvery young children are the population groups at highest risk, since they may not have acquired significant level o f immunity against malaria. Inareas with unstable malaria exposure, pregnant women may have 2 to 3 times higherrisk o f contractingmalaria than non-pregnant women living inthe same area. Moreover, malaria exposure may leadto additional adverse outcomes such as low birthweight, abortion, andneonatal death. Malariarelatedadmissions to hospitals have risen from 20% in 1999 to 31% in2001,while an additional 20% receive malaria related outpatient care. Meanwhile, only 26% o fpregnant women are reportedto have visited - 34 - antenatal care at least once in2000. This means the majority o fpregnant women in Ethiopia do not have access to treatment, whichwould help them reduce the impact of the risk. Also, only 1%ofhouseholds havemosquito or insecticide treatednets andonly 3% ofunder-5 childrenreceive any anti-malarial treatment (WHOLJNICEF, 2003). 4.26 It is important to remember that for bothHIV/AIDSandmalaria, when many families become affected, the disease overwhelms the mechanisms available to households, communities, and even countries to mitigate its impact. Ill-health,disabilityand mortality 4.27 Although idiosyncratic healthrisks affect individuals and may sometimes appear to bepurelydue to chance, their contributionto increasedvulnerability is not necessarily small. For manyhouseholds on the marginso f destitution, a single episode o f illness to the mainincome earner for an extended period o ftime can be the difference between life anddeath. An accident that disables anadult canbringhardshipto the whole household and deepenthe potential o fthe disabled to fall into apoverty trap. The death o f abread- winner can have equally long lasting effect. 4.28 Figure4.6 shows the proportionofheadsofhouseholds who reportedhaving health problems two months prior to the start o fthe survey. The incidence o f illness rangedfrom 25% to 44% o fhouseholds. Giventhat there are 10millionhouseholds in Ethiopia, this impliesthat inany givenyear 2.5 to 4.0 millionEthiopianhouseholdheads experience an episode of idiosyncratic shock. Although, the questionnaire didnot ask for identification of the cause o f illness, the most common causes o f inability to work are reportedto be illnesses such as typhoid, tuberculosis, and HIV/AIDS. Such diseases make all families, irrespective o f income level, exposed to substantial risk. Figure 4.6: Percentof headsof householdsreportinghealthproblemstwo months prior to start of survey. Source: World Bank Staff estimates from HICESrWMS(1995196, 1997, 1998, 199912000) - 35 - 4.29 Ethiopia's burdeno f disease i s dominated byperi-natal andmaternal conditions andacute respiratoryinfections. About 18% ofall life years lost are dueto peri-nataland maternal causes. Acute respiratoryinfections account for another 16% o f life years lost. The burdenof illness i s also often extremely intense among children. Diarrhea and nutritional deficiencies result in 17% o f children's deaths (GOE, 1998). Evenifa child survives nutritional deficiencies, they stand a very highchance o fbecoming stunted (57%), which i s knownto lower productivity inthe long term. STRUCTURALAND LIFE-CYCLE FACTORS 4.30 Strictly speakingstructural and life cycle factors are not treated as risks, since they are not always randomevents. That said, they play an important role in compounding the impact o f risks. Structural factors are usedbroadly to refer to all economic, social, cultural and legal factors that place constraints on individuals' and households' livelihoods. Inthe Ethiopiancontext some o f the key structural factors includelongterm unemployment, layoffs, and culturalpractices that prevent girls from attending school. By contrast, life-cycle factors include old age, widow-hoodor female- headed status. 4.3 1 Structural factors: Inqualitative surveys, housewives point out events like their husbands'loss o fa government job or the demobilizationo fthe Dergsoldiers as the main causes o f insecurity. The youth, on the other hand, feel the direct effects o f government legalpolicies -particularly the policy that disbanded their cooperatives -as the main cause o ftheir insecurity. Apart from the redistributionpolicy, land shortage and vulnerability may also arise from poor inheritance laws or misguided marriage endowments. 4.32 More recently, indebtedness has become an issue. Many farmers indicate having three or four years' worth o f debt for fertilizers owed to the government and fear they maybe arrested at any time (World Bank, 1999). Finally, those rural farmers who have to rent out their land as a strategy o f last resort -aftera labor shortage and/or drought power or seeds following a crop failure- may also face severe destitution from decreased supplyof grain andanimal feed. (IDS/SC-UK, 2002) 4.33 Life-cycle events that magnify the impact of risks. Some social events can pushhouseholds into poverty. Widowsare especially vulnerable to extreme poverty. The purposely sampledsurvey on destitution, found that female-headed households have a greater likelihood than male-headed households o f falling into extreme poverty over the 10 years precedingthe survey. More precisely, female -headed households that were not destitute in 1992/93 have a cumulative probability o f about 40% to become destitute by 2001/02, compared to 15% for male-headedhouseholds(ISD-SC-UK, 2002, Figure 4.7). - 36 - Figure4.7: Probability functionsofbecomingdestituteby sex of householdhead. ` 1 YEAR Source: IDSISC-UK, November 2002 Note: This figure plots the cumulative probability for the sample o fhouseholds that were not destitute ten years ago o fbecoming destitute over the decade leading up to the present time, againbased on the self-assessment data for four years: 1992193, 1999100,2000101,2001102. The vertical axis inthe Figure represents the probability o f `survival', that is, the probability o f not becoming destitute. For consistency, the time-line on the horizontalaxis should show the periodbetween 1992193 and 2001/02. 4.34 The elderly form another group facing greater risks. Manyo fthem are unable to adequately exploit their landas their own physical capacity and the external (labor supply diminishesover time. Inaddition, the tradition o fproviding childrenwith endowments o f land at the time o fthe marriage tends to deplete the landholdings and other storable assets from the elderly (IDS/SC-UK, Nov 2002). Oldpeople are also increasingly taking responsibility for children infected byHIV/AIDS and grandchildren. The cost of caring for the sick and relatedunexpected medical andhouseholdexpenses can impose considerable pressure on them to sell assets and evento become full time beggars (GOE, 2000). 4.35 Orphan-hood and children in extreme circumstances are becomingincreasin ly visible. The latest estimate suggests that the number o f childrenwho became orphans B regardless ofthe cause was 3.8 million in2001 o fwhich 1million (26% o f all orphans) were AIDS orphans This figure is expectedto increase to 43% o f all orphans by2010 (UNAIDS,2002). Whether orphan hoodcomes from AIDS mortality or from other deaths, the fact remains that, orphans suffer loss o f educational opportunities, malnutrition, and inadequate access to healthcare, which are the very factors that are knownto increase povertyandperpetuate inter-generational disadvantages. Orphans are also subject to property grabbing, abuse and exploitation, stigmatization and 5Orphans, inthis case, are defined as children aged less than 15 years who have lost either one or both parents (UNAIDS, 2002). - 37 - psychological trauma. As aresult there i s a strong likelihood for these children to endup on the street (along with the current estimated 150,000 to 200,000street children in Ethiopia), become HIV infected, or be engagedincriminal activities. Box 4.4: Effectofhouseholdsize on householdvulnerability is ambiguous. Household size i s also believed to be an important determinant o f vulnerability. "Large family size canbe either an advantage or a disadvantage. Someone with a small household may have no one to support himif he's sick, and no one to keep animals or pests away from more distant fields. Onthe other hand, ifyou have a large family you may have trouble feeding them all, so you maybe forced to sell animals to feed them" (IDSISC-UK, NOV 2002). - 38 - 5. Assessing Effectiveness of RiskManagement Strategies RISKMANAGEMENTSTRATEGIESBYHOUSEHOLDS 5.1 Ethiopian households, like many households insimilar circumstances around the world, employ a variety o f strategies to manage the risks they face. The actions they take canbe classified as falling into two broad strategies. Some actions are aimed at reducing or mitigatingthe risk. All actions taken before the shocks occur with the intention o f lowering the expected negative impact o fthe shock are classified inthis category. Other actions are meant to copewith risk and include those actions that are undertaken after the shock has occurred inorder to reduce or negate the impact o f the shock. Table 5-1 highlightsexamples o fthe types o f actions that are commonly employedbyhouseholds under each o fthe broad strategies. Table 5-1: Types of actionsto manageriskby individualsand households. Risk management Goal or objective of the strategy Typesof actions undertakenby strategy individualsor households Riskreduction Actions taken before the shock Diversification o f economic or mitigation occurs inorder to protect against activity (multiple crops, multiple the anticipated negative impact species o f livestock, multiplejobs) o f a shock Insurance (formal or informal) Migration(seasonal) - saving Riskcoping Actions taken after the shock has Reducing or delaying expenditures occurred inorder to negate, or investments (e.g. inhealth) minimize or survive the adverse Reducing consumption (e.g. forgo effects o f the shock. current nutrition) Withdrawing children from school Borrowing (informal -relatives, neighbors- or formal-banks, credit societies, micro-institutions) Migration Selling assets (land, oxen, jewelry, etc.) or drawing down savings Riskmitigationstrategies 5.2 Inthe context o fEthiopia, themost common risk mitigatingactions include crop diversification, where households grow cereals andpulses, coffee and food crops, and maintainingmixedcrops and livestock systems at the same time. A parallel diversification effort among the pastoralists include, keeping multiple species o f livestock (goats, sheep, camels, cattle). Households also informally insure, byjoining rotating credit groups (iqub) and funeral societies (idir), the latter inorder to meet the expected costs, well-known to be huge, inthe event o fthe death o f a householdmember. - 39 - 5.3 Bothpoor and wealthy households believe that income diversification enhances their ability to withstand external shocks, and thereby smooth consumption. In 1996197, it was estimated that as manyas 62% o frural households worked less than one hectare of land,makingitnecessary for manyofthesehouseholds to supplement their farm income with off- farm earnings to meet their consumptionneed(Resal, 2000, p.18). Table 5-2 shows off-farm participationrates as well as shares of income obtained from non-farm activity by 402 heads o f farmhouseholds intwo woredas inTigray Regionin 1996/1997. At least 81% of the households participatedinsome form o foff-farm work, which i s described as wage employment, self-employment, food-for-work, and skilled andmanual non-farm wage work. As indicated inTable 5-2, 35% o f all household income inthese two woredas was obtained through off-farm work. The two mainsources o f off-farm work were wage employment and food-for-work activities (Woldenhanna and Oskam, 2001). Table 5-2: Off-farm Work Participationintwo Woredas, Tigray Region. I Share of income Activities I ParticipationRates Ifromthe activityin 1 Riskcopingstrategies 5.4 Riskcopingactions arejust as widespread, and encompassadiversityof strategies such as selling productive assets (livestock, oxen, and donkeys), foregoing current investment and consumption, child labor, borrowing or drawing down savings. The 1999 WMS asked households to identify how they would cover a negative shock -a decrease intheir income equivalent to a 100Birr (12 USD). Table 5-3 shows that urban andrural households would have useddifferent strategies to cope with such a shock. The top three actions, available to 54% o frural households, include the sale o f animals and animal products, the sale o f other agricultural products, and loans from relatives. By contrast, the two main coping instruments for urbanpeople were withdrawal o f own savings andloans from relatives. - 40 - Table 5-3: Sourcesto Get 100 Birr for UnforeseenCircumstancesin a Week. Gift fromnonrelatives 0.09 0.33 0.12 Sale of household assets 0.42 1.85 0.63 Others (Not stated or Missing) 34.57 41.29 35.54 Total 100.00 100.00 100.00 5.5 One coping mechanism that has emergedi s an informal system o f landfutures that has arisen to deal with price risk. Leasing landtwo years inadvance is, done by the poor families to cope with emergency expenses. Land lease i s common inNegda and Dugda (World Bank, 1998). 5.6 Childlabor is aparticularly damaging form ofcoping. Demand for labor has been identifiedas the most important reason for not sending children to school in Ethiopia (Yamano, 2000, pp. 1). Not surprisingly, according to ILO, Ethiopia has the highestincidence o fchildwork inthe world with a 42% participationinfill-time productivelabor'. Rural girls are particularly at risk: a survey of 1,477 ruralhouseholds foundthat 60% of all children, and 80% o f 11-15 year old girls, havework as their primary activity. Some o f the child labor i s a strategic move to mitigate the impact o f a future shock byusingchildren's contributionto buildassets, while some is a direct action to manage the ex-post effects o f a shock. Indeed, each working childhas been shown to contribute between4% to 7% ofhousehold's income inrural areas (Cockbum, 2002). 5.7 Another damagingform o f coping strategy i s the reductionof food consumption. Intime of food stress, households cut back the number o fdailymeals. Almost 70% and 58% o f the poorer andricher households respectivelyinthe 3 lowland villages surveyed by Webb andReardon(1992), resorted to this solution. Inthe upland villages surveyed, 62% and 39% ofpoor andrich households cut down current consumption. Consumption o f "famine" food was also observed. Iqub is a traditional rotating credit and savings association. Idir i s a burial society that covers funeral expenses o f its members. This definition excludes domestic work. - 4 1 - EFFECTIVENESSHOUSEHOLDSTRATEGIES OF 5.8 Although households resort to a diversityo frisk management strategies, often they are not very effective. Ingeneral, there are four drawbacks to household's ownrisk management strategies, as observed inEthiopia. 5.9 First,they achieve only partialinsurance at a highcost. Although many households diversify their economic activities to attain self-insurance, they are still unable to do so. Consider, for example, the insurance value o f a common strategy, which involves undertakinga mixed fanning systemo f crops andlivestock. Livestock sales as a means to smooth consumptionwork best for idiosyncratic shocks. Inthe event o f a drought that wipes out the crops, the household expects to sell livestock to maintain its level o f consumption. However, ina study conducted inBurkina Faso, which provides a similar setting as Ethiopia, itwas found that livestock sales, at best, do compensate for 30% o f income shortfalls after rainfall shocks - 15% beingmore realistic. Indeedcattle sales seemed less respondent to income shortfalls (except for the households experiencing large negative shocks), compared to small stocks (e.g., goats and sheep) (Fafchamps, Udry and Czukas, 1998). Another example o f a household's attempt to self- insure i s the widespread practice o freducing riskbyplantingmultiplecrops, some o f them drought andpest resistant. Insemi-arid parts o f India, farmers forego as much as 25% o f income to reduce exposure to risk (World Bank, 2001). This suggests that the cost of attempting complete self-insurance i s too highfor individual households, especially when it i s considered that these households also forego innovation, experimentation and adoption o fhigh-productivity technologies inthe effort to diversify. 5.10 Second, they are localized and of limited scope. Borrowing from relatives i s a common ex-post coping strategy inurban and, to a lesser extent, inrural areas (Table 10). However, as stressedby Webb and Reardon(1992), most transfers and borrowing among kin-relatives generally are o f small amount, more to maintain social bonds, than significantly improve the income or nutritional status. Table 10 shows that only 8% o f urbanhouseholds and 3% o frural households obtained loans from non-relatives. Moreover, traditional sources o f social insurance, commonly known as Idir9andIqub", were found to be limited as sources o f consumptioninsurance for bothrural andurban households (GOE, 2002d). 5.11 Table 11shows that nationally, while rural households were slightly better at findinginformal insurance than urbanhouseholds, up to one-third o fhouseholds would have been unable to find 100Birr (about 12USD)within a week for unforeseen problems. Within regions, ability to insure i s also very diverse. Householdsthat are less likely to findinsurance tend to be locatedingeographically less endowedregions (Tigray and Amhara, Afar), geographically isolatedregions (Gambella and Benshangul- Gumuz), and the urban administrative area o fDire Dawa (see Table 5-4). Idir i s a burial society that covers funeral expenses o f its members loIqub is a traditional rotating credit and savings association - 42 - Table 5-4: ProportionofHouseholdswho can Get 100 Birr inaWeek for UnforeseenProblems. Total 66.85 61.95 66.14 5.12 Third, informalinsuranceoftenmarginalizesthe mostvulnerable. The already limitedscope o fIdirs to insure consumption shortfall is made worse bythe fact that they often exclude the most needy households, especially when mutual reciprocity i s a condition for participation. While entry costs insome communities such as Mecheke, are low, they are prohibitive inothers, e.g. Korate (World Bank, 1998, pp. 36). Another form o frisk pooling, inthe form o f labor groups which exist to share labor at plantingandharvest times to help community membersincase o f family illness, may however restrict employment by excluding certain members(such as widows) o fthe community fromjoining their organization. 5.13 Finally,informalinsurancehas highhiddencosts. Put simply, some o fthe householdriskmanagement strategies such as delaying nutrition, health care, keeping children at home rather than school (or withdrawing them from school) for their labor, have the potential to permanentlyreduce the human capital and future prospects o f the householdmembers, individually and collectively. 5.14 Manyo fthe shortcomings ofthe informal insurance strategies canbe overcome if markets for risk sharing functioned well. For instance, functioning labor and land markets, can be a conduit for landless individuals or households, to smooth consumption. Similarly, ifinsurance marketsexisted and worked efficiently, farmers exposed to weather risk can still smooth consumptionby buyingcrop insurance albeit at a high premiumgiventhe variability ofrainfall. Itis therefore appropriate to ask and explore how well markets address the shortcomings o f the informal insurance arrangements discussed above. MARKETSRISKSHARING FOR 5.15 After decades o fsuppression under a centrally controlled economy, developing a market economy has been a major goal o fthe new government. From the view o frisk management, a set o f fully functioning and integratedmarkets are important sources o f -43- risk pooling. Therefore, the commitment to a market economy should be a welcome development. 5.16 That said, it i s widely acknowledged that at present markets inEthiopia do not provideadequate avenues for sharing risk. Most markets are considered to bethin, inthat theyhaveonly a fewbuyersandsellers at anytime andplace. Markets are also considered to be incompleteand characterized by hightransactioncosts. These characteristics o fmarkets introduce substantial risk, rather thanreducing them, which in turnlimitsmarket development. 5.17 The limitations o fexistingmarketsto provideinsurance for households are best illustrated by looking at some marketsthat are crucial for the poor households, starting with the grainmarkets. First, only 28% o ftotal production i s marketed. As aresult there are few intermediaries and little value added. The market i s dominatedbythe government-ownedEthiopian Grain Trading Enterprise (EGTE). Private traders number no more than 25 to 30. Most manage small enterprises, have limitedhuman and physical capital assets, andpoor access to finance, all o f which limit their scale o f operations. Specifically, these traders operate infew markets,an average o f 9 contacts (75% of which are from the same region), and trade little outside o f 190km. Furthermore over 60% have loanrepayment problems, which suggest a risky environment (Gabre-Madhin, 2002; GOE, 2002b). The market i s also characterized by hightransaction costs. For instance, only 40 birr of a final price of 120 birr (33%) goes to the producero fmaize, the rest going to transport, handling, storage and brokerage fees. Furthermore, it takes 20 to 30 days for food to get from the producer to the consumer (Gabre-Madhin, 2002). 5.18 The reviewo frisks above has shown that farmers face unusually highweather risks. From this short assessmento fthe output markets for grain, we see that there i s no reprieve, since additional risks are introduced into the chain. Which means that both farmers and traders face highrisks andcostly transactions. The result i s a high probability o fmarket failure, as may have happened in2002. Box 5.1: How longcanthe average subsistence householdrely onownharvest? At most 7 months. The household income consumption and expenditure survey o f 1999100 asked rural households to report how long their own harvestproduce would last. On average, subsistence farmers' harvested produce lasted only 7 months, and 6 months inTigray, Somali, SNNP, Harari and Dire Dawa, leavingthem vulnerable to hunger and poverty for the remaining half o f the year, especially since such households often rely exclusively on their own harvest to sustain their livelihood (GOE, 5.19 Ifgrainmarkets donotprovideprotectionfromrisks, neitherdo existing insurancemarkets. These markets are poorly developed andhave yet to recover from the nationalization decrees of 1975. Overall insurance penetration, measured as value o f insurance relative to GDP, i s very low -about 1%o f GDP. In 1994, private actors were allowed back into the market, and so far 8 such firms are known to operate, reinforcing the thinness o fthese markets as well. And except for the Ethiopian Insurance Corporation, the dominant player, they are small andprovide insurance to only limited - 44 - markets, and almost none for agriculture except for goods inwarehouses and transit (World Bank, 2000). 5.20 For the poor, factor markets inlabor, land, and credit can serve as alternative sources to mitigate or cope with risk. For instance, those who are landless can rent, and those with abundant landwho cannot till it all can lease for additional income. Formal or informal credit markets canbe a source o f capital for able individuals who lack money to buyinputs for their farm or start-up capital for-non-farm enterprise. So how well do land, labor andcredit markets work inEthiopia? 5.21 Prior to 1991, landsales, leasing, mortgaging andbequests were prohibited, and only periodic redistributionbypeasant associations guaranteed access to landfor additional memberso fthe associations. This precluded any mutually beneficial exchange betweenhouseholds -between those well-endowedandthose less endowed inland. It also diminishedthe role o f landmarkets to act as away to sharerisk. Since 1991, regulations governing landmarkets have changed again. Underthe new landpolicy, temporary lease o f land i s permitted, althoughprivate ownership, sales, andbequests are prohibited. Despitethe new relaxedregulations, the potential for landmarkets to provide insurance have not yet been realized. 5.22 An important problemis that farmers feel that they do not have security oftenure andso are reluctant to engage inmany landtransactions includingrenting (Nega, Adenew, Gebreselassie, 2002). At present, householdshave use, butno ownership, rights. The government's position i s that limiting landrights to use rights, preventsa "fire sale"- massive distress sale -from households too desperate to survive, inthe event o f a shock. While this is understandable, giventhe prevailing social conditions inEthiopia, maintaining such a policy shouldbeweightedagainst historical evidence and longrunwelfare costs. 5.23 With regardto historical experience, Mexico's abolition o frestrictions on the land market, especially inrental markets,inthe 1992 constitution, didnot lead to the original fear o f a fire sale, but to higherproductivity and equity (World Bank, 2002). And in China, provisiono fmore security inthe original landuse rights following the 1978 Household Responsibility System i s shown to have ledto more mutually beneficial inter- household landtransactions (Deininger andJin, 2002). Turningto welfare impact, one unintendedconsequence of such apolicy standis that it creates expectations about future land redistribution. Ina 1999 survey, 10% o frural farmers surveyed (20% inAmhara region) expected landredistributionwithin the following 5 years. Moreover, producers who expected to gain land through redistributionwere twice as many as those who expected to lose (Deininger, Jin, Adenew, Gebreselassie, andDemeke, 2002). These expectations have three negative effects on agricultural productivity and poverty reduction. First, the fear of landredistributioncan lead to slow exit from rural areas, slow growth o f off-farm economy, more people per landand lower agricultural productivity, which would completely undermine the efforts o fthe government's declared intentionunder ADLI. Analysis o frecent rural surveys show that land owners with off-farm employment have a 10%-15% perceivedlikelihood ("subjective probability") o f losing land through redistribution inthe future (Deininger, Jin, Adenew, - 45 - Gebreselassie, andDemeke, 2002). Second, long term improvements (investment) on landwill be lower, ifhouseholds have the fear that they will not obtainthe fullbenefits o f their investments. Finally, the full potential o f already restrictedsecondarymarkets, such as rental landand sharecropping, will be underminedeven further ifthere i s a threat o f losing land that i s rented out. The evidence from Ethiopia, shows that those who rent in landhave higher expectations o freceiving landinthe future. By contrast, it is the productive farmers with part timejobs inthe off-farm sector who perceive a greater threat from land redistribution. 5.24 An additional constraint that limitsthe potential role o flandmarkets to act as sources to share risk i s that landper household, especially inthe heavily settled highlands where its risk-pooling role can be great, i s very small, averaging less than 3 hectares, thereby limitingthe scope o fthe leasemarkets. However, empirical evidence suggests that it i s not the size but, many restrictions imposed on secondary landmarkets that reduce their role as risk-sharing institutions. InOromiya region, for instance, a farmer cannot rent out more than 50% o f landheld, and maximum contract lengthsare 3 years for traditional and 15 years for modern technologies (Deininger, Jin, Adenew, Gebreselassie, andNega, 2002). However, despitethese limitations recent analyses show significantly positive effects o f landrental markets inEthiopia. First, as many as 24% o f households report to usingmarket transactions to get access to some other household's lands. To underscorethe risky nature ofEthiopianagriculture, the majority ofthese arrangements,turn out to be sharecropping, where the householdwho rents inandthe one who rents out, share the risks o fproduction. Second, the findings show that these transactions tend to transfer land from households with more landbut less agricultural ability to those with less landbut higher agricultural productivity. Additionally, households with lower landper capita are found to use the market mechanisms to gain access to land, while those with larger per capita land use the same institutionto rent out land. Together, these two findings implythat the market transactions are both efficiency and equity-enhancing, andthere i s really no need to fear the emergence o f land concentration ifthe restrictions are relaxed (Deininger, Jin, Adenew, Gebreselassie, and Nega, 2002). Insome villages o f Oromiya, landgifts and donations, were more likelyto be givento poorer, younger, andcredit constrained households (Pender and Fafchamps, 2000) for longer durations. Indeedthese results suggest that constraints to secondary landmarkets, which restrict the insurance role that these marketscanplay for many households, have much larger negative welfare effects than currently known. 5.25 Finally, labor markets, are characterized by lack o f spatial mobility. Use o f migration for risk-copinghas been suppressedfor many decades inEthiopia, and althoughthis has improved inrecent years there are few opportunities for wage labor. The Tigray data (Table 5-2) is an exception rather than the practice. According to a surveyof 6,000 households representative ofthe households inthe four largest regions which represent 85% o fthe population (Amhara, Oromiya, SNNPR, andTigray) andthe sedentary populations inAfar region, only 15% ofrural householdsearned income throughruralwage employment, andonly 9% throughhandicrafts. This represents only 5% o fthe rural population inthese regions (GOE, 2000, p.19). This rate is much lower whencompared to other African countries where non-farm activities provide for about - 46 - 20% to 30% o f employment to the agricultural labor force, and non-farm income is 30% to 50% o frural household income (Reardon, 1997). PUBLIC RISKMANAGEMENTSTRATEGIES 5.26 Ethiopia's social riskmanagement programs include indirect efforts to build assets and incomes o fthe poor -keepinginflation under control, stimulating the private sector, relaxing undesirable regulatory controls, sector specific investments, and so on- and direct instruments, such as cash and in-kindsupport to specific groups. 5.27 The government's proposed approach to riskmanagement is to tailor programs to specific agro-ecological zones. Itmakes a distinction between fertile andhighpotential areas with regular and reliable rainfall, and unproductive, highly degraded and drought- prone areas. For the former, it proposes to implement agricultural development led industrialization (ADLI),whose mainpurpose i s to increase agricultural productivity and rural incomes, while for the latter, it proposes to provide food security using a mix o f instruments. The thrust ofthe proposed strategies for highpotential areas are risk- preventingor mitigating by purpose, while the strategies for the low productive areas focus on coping. The next section provides a short overview o fthe main elements o f risk-mitigation inthe country. Risk-reducingprograms. 5.28 Most o fthe risk-mitigation programs are recent, andmostly targeted at rural population. Because they are new, we cannot yet evaluate the operational outcomes, such as the target areas, number o fbeneficiaries, or the difficulties encountered by beneficiaries. That said, several activities to reduce or prevent some o fthe risks identifiedabove havebeenproposed for implementationunder the sustainable development and poverty reductionprogram (SDPRP, Ethiopia's PRSP). Table 5-5 lists a selection o fthese activities, andbelow we discuss ina bit more detail some o f the programs that have been identifiedas either key or for scaling up inthe country. - 47 - Table 5-5: Riskmitigatingor risk-prevention activities. Drought Terms of Trade Grainprice risk HIV/MDS Environmental risk (coffee price degradation risk) Promote Encourage Encourage Increase Establish and support diversification o f rural banks, clinics providing seed small scale export products savings and voluntary nurseries, irrigation. (to horticulture, credit HIVIAIDS especially o f textiles, cooperatives, counseling and indigenous Water garments, and and micro- testing services species. harvesting. oilseeds and finance pulses). institutions. Promote Prevent Voluntary HIVIAIDS deforestation resettlement. Niche Introduce education in (through tree marketing o f warehouse schools planting, Increase organic coffee. receipt schemes. imposing delivery o f Strengthen fines on farm inputs. Lower cost Introduce prevention and illegal o f production for commodity control o f cutting o f Produce exports through exchange. HIVIAIDS by trees). andpromote lowering tariffs in increasing and drought power and Remove buildingcapacity Promote resistant telecommunicatio hndrances to of regional and efficient and crops. nsectors. improved woreda level plannedland finctioning o f HIVIAIDS use Promote Enhance the markets for councils. strategies. off-farm application of agricultural income export guarantee inputs. Promote generating scheme. alternative activities. Organize sources o f and promote energy (e.g. Expand cooperatives for heat-efficient food security marketing stoves). programto services. regions not yet covered. 5.29 Waterharvesting. The government rightlyrecognizes that some parts of the country, especially drought-prone Northeast highlands and lowland pastoral areas require a diversification o f livelihoods inorder to reduce the drought risk. The primary instrument advocated i s rainwater harvesting, as a way to respond ex-ante to the risk o f crop failure that would result from a drought. 5.30 At present the mainwater harvestingtechnology is a cistern, butriver water diversion and small or micro-dams are also encouraged. The goal o fthe program i s to helphouseholds bridgewithin-season variability inrainfall, since the water from a cistern (typically about 100cubic meters) provides at most a three-week supply o fwater, ifuse i s restricted to drinking andplot irrigation. It i s also hopedthat the availability of a regular supplyofwater within a growing seasonwould encourage households to shift - 48 - their productionto high-value crops, such as vegetables and livestock, which would guarantee a minimumincome. 5.31 The government meets nearly all or most o fthe capital cost o f the water harvesting technology, which at USD200 per cistern, i s substantial. To ease the financial burdenonthe government, arotatingfind, which involves credit from the Food Security Program to individuals, families or a community, has been started. As it i s currently designed, water-harvesting i s expected to bridgethe intra-annual variability inrainfall only. So, a major concern i s the risk-reducing value o f a program that does not bridge inter-annual variability o f rainfall. Another i s that, becausethe program i s less thantwo years old, there i s little documentation on the profile o fbeneficiaries andbenefits. It i s unclear how many o fthe food insecure households are receivingthe water-harvesting technologies. 5.32 Supplementalirrigation. At present Ethiopia's irrigationpotential is considerably under-exploited and one o f the strategic goals of the government i s to use it as an instrument for reducingrisks andby consequencepoverty and vulnerability. In fact, among the manyobjectives that a well-developed irrigation system i s expected to accomplish are: attain food security and reduce dependence on food aid, reduce vulnerability to drought, erihance intensification andpromote high-value export crops. 5.33 Inthe mediumterm, the focus ofthe irrigation activities is expectedto beon supplemental irrigation: irrigation water kicks inonly when the rainfall i s interrupted mid-season, or fails. While the overall goals o f the program are well-articulated, it i s yet unclear whether the beneficiaries will be food insecure households, small-scale farmers, large capital-rich farmers, or evenprivate developers. Furthermore, there are still unresolved issues regarding how fast irrigationprograms canbe brought into operation and on a scale that will make a bigdifference among the millions o f food insecure households. 5.34 Resettlementprogram: Although not traditionally recognized as a safety net, the government o f Ethiopiahas made resettling populations as an important instrument to deal with poverty in general and food insecurity inparticular. This shift stems from the recognition that specific strategies to address the needs o f the chronically food insecure, primarily food aid, have proven inadequate. Resettlement i s also seen as part o f the government's development strategy -Agricultural Development Led Industrialization (ADLI)-whose major objective is to improve agriculturalproductivity. 5.35 The goal o fresettlements i s to move populations from areas that have low agriculturalpotential-highlydegraded, drought-prone, and small farm sizes -to those withbetter andmore predictablerains, more fertile soils, and low populationdensity. By design, these programs are supposed to be(a) voluntary, (b) intra-regional, and (c) to target areas that have been clearly identifiedto possessthe potential to receive migrants. The design o fthe program has beenon-going for a couple of years, but it is only in2004 that significant populations havebeen moved. - 49 - 5.36 A feasibility study exploring the potential for voluntary resettlement inOromiya region found that only 27,000 people showed willingness to resettle, and among these registered volunteers, only 1%was female heads o f households. This i s partly because of the receivedmessage regardingwho the target groups are and not necessarily because o f their unwillingnessto move. The study also showed that not all areas targeted as sources for settlers were truly "drought-prone." Moreover, ina numbero fworedas (districts) near to many areas targetedto receive settlers, problems similar to those experienced in settler "source areas", such as populationpressure on land and degradation o fhillsides andforest lands, were common. The concern is that failure to offer the option of resettlement to equally vulnerable households near to "destination areas" could leadto resentment o f settlers and conflict (Environment andDevelopment Group, 2002). 5.37 Agro-ecological packages. The government's strategy i s to tailor agricultural development packages to specific agro-ecological zones, inorder to maximize the absolute comparative advantage o f each ecological zone. The ideai s to develop specialization within each ecological zone, while at the same time striving for diversification across the country. 5.38 Farmers are supposed to choose among a menu o fproducts deemedsuitable for their ecological area. The drought prone areas are advisedto getpackages that will enhance their capacity to be food secure, including improvingproductivity, off-farm activities andprice policies (GOE, 2002~).For the areas deemed to have adequate rainfall the focus will be intensive agriculture, that i s geared towards crop and fodder, while the package for pastoral areas will combine water and livestock development. In fact the key promise o fADLI i s that the packaging will leadto higher agricultural input use, especially highrisk, but productivity enhancing inputs, which will inturn leadto higher output andfood security. 5.39 A key component ofthe inputpackage is fertilizer. Inthe early reformyears, the prospects for fertilizer market development and its widespread adoption looked good. The issuance o f a national fertilizer policy and deregulation o f fertilizer prices, including removal o f subsidies, encouraged competition and entry o fprivate sector participation, which quickly moved to supply up to 68% o f fertilizer by 1996. Just when the new participants were beginningto develop the infrastructureneeded to add value to the entire supply chain-build storage, bulk transfer, retail outlets, as well specific reputations and competencies -the government introduced a new way to deliver fertilizer input that began to erode these positive developments and longrunimprovement infertilizer use. 5.40 The new way involvedmobilization o f the government's agricultural extension services to deliver farm inputs -fertilizer, seeds, pesticides and other farm inputs-as a package. Also linked at the same time were credit andretail andwholesale o f farminput distribution. Inthis program, the regional governments provide credit guarantees to participating commercial banks. The banks, inturn, assess the credit-riskiness o f farmers, lendthem hnds, and collect the payments. But one o f the consequences of linkingcredit and inputdistribution has beenthat the extension agents (and not the banks) became, effectively, the managers o frisk -assessment o f credit worthiness o f farmers and subsequent delivery o fthe credit. - 50 - 5.41 Although the original intent o fthese policies were to promote and scale up the use o f farm inputs, the unintendedconsequences, may leadto opposite results, especially in the longrun. Take the linkingo f fertilizer input andthe extension service. Although different regions use different institutions -for instance, party-affiliatedtrading houses in Amhara, cooperative unions inOromiya -the net outcome o f linking extension services andinputmarketinghas beenthe introduction of significant market risks for private participants and farmers. Not surprisingly,there has been an almost complete exit o fthe private sector from the market, andboththe quality of information, interms o frelevance andtimeliness, and availability ofinputsfor farmers are believedto have gone down. Inaddition, bureaucratic objectives ofthe extension agentsthat arenottied to market incentives, have ledto excessive supply and undoubtedly inappropriate use o f fertilizers. Inone survey, while73% ofhouseholdsreportedusingfertilizers, only 19% and31% used improved seeds and chemicals (Deininger, Jin, Adenew, Gebreselassie, and Demeke, 2002). This cannot possibly be the intendedobjectives o fpackaging. Withregardto inter-linkedcredit andinputdistribution, one ofthe unintended consequence has been"over-supply" o f credit, through extension o f credit to non-creditworthy farmers, which inturnhas ledto increase inthe size o fnon-performing debt owed by farmers. 5.42 EthiopiaSocial RehabilitationandDevelopmentFund(ESRDF): The Ethiopia Social Rehabilitation and Development Fund(ESRDF), established nation- wide in 1996, plays akeyrole inthe riskmanagement strategy ofthe government. This $270 millionprogram(1996-2004) invests on average at least 50 km away from an all weather road, targeting isolated communities, who by definition face higher vulnerability. ESRDFhas already benefitedabout 21 million people or roughly one-third o f the total population. As a provider o fbasic services requestedby the poor themselves, ESRDF has financed about 4,000 projects that help households mitigate or prevent some o f the most important risks, such as drought and environmental degradation (through small-scale irrigation, soil conservation projects -- implemented through cash-for-work), human and animal diseases (through better access to health services, rural water supply andsanitation, andveterinary clinics) or strengthen households' ability to manage risks (though better access to education, and support to saving and credit associations). Unfortunately, despite ESRDF's unique expertise inhelping isolatedcommunities mitigate risks, the future o fthis fund i s uncertain. Publicprogramsfor copingwith risk 5.43 The mainfocus o fEthiopia's socialprotectionprograms for coping with the after effects o f a shock i s on food security. To be sure, there are pensionprograms for the public sector workers and electricity subsidies for homes and businesses, but they are small in scale. The food security program has been designed to respond to two types o f food insecurity: chronic and transient food insecurity. To address large-scale transitory shocks, such as nationwide rainfall failure ina given year, the government has relied on international community to protect the populations affected. To fight chronic food insecurity, the government targets 156 woredas infour regions -Tigray, Amhara, - 51 - Oromiya and SNNPR" -which are classified as food insecure. Within these woredas, the food security activities make further distinction between, (a) those that are unable to work -children, women with many children, the elderly and the disabled, and (b) those that are able to work but sometimes fall on hardtimes becauseo f a shock. Table 5-6 shows the distinguishingcharacteristics o fthese groups. Table 5-6: Indicative categoryoffoodinsecurehouseholdsinEthiopia. Chronic Urban Others Resource poor households Low income Households D Refugees land-less or land-scarce :mployed inthe informal D Displaced 0 oxen-less sector people 0 poor pastoralists 0 Those outside the 0 female-headed labor market, such as households butnot exclusive to: 0 elderly --- Elderly disabled & sick Disabled & sick poornon- Some female- agricultural headed households - households Street children 0 newly established settlers Transitor Less resource poor D Urbanpoor vulnerable Groups Y households but vulnerable to to economic shocks, affected by shocks, especially but not especially those causing temporary only drought food price rises civil unrest 0 farmers and others in drought-pone areas pastoralists 0 others vulnerable to economic shocks. e.g. in low Dotential areas Source: GOE 102a). 5.44 First, the program covers only rural households. Second, to a large extent, separate social protectionprograms are designed to address the needs o f these two categories ofrural households or communities. Those who cannot work typically receive puretransfers, while those who canwork, receive aid inexchange for labor. The majority o ftransfers reach those inneed through two major programs -fiee food distribution and employment generation scheme. Inaddition, there has been a long-standing food for work program, andinrecent years, a school feedingprogram, and a variety o f small-scale cash andwork-based safety net pilot programs runbyNGOs have emerged. "IntherecentSafetynetdesignadditionalworedasintheseregionsandthepastoralistsregionsofAfar and Somali and two woredas inHarari and Dire Dawa has increased the number o f woredas covered to 262. This is a significant progress and such enlargement is a recommendationo fthis study as well. - 52 - 5.45 Freefood distribution (or GratuitousRelief (GR))-The target group for this program i s all vulnerable individuals who are unable to work. Intheory, approximately 20 percent o f total donated food aid (which has averaged 700,000 tons a year for the last 15 years) is to be used for GR, but inreality as much as 80% i s used for that purpose. The distribution to direct beneficiaries i s the responsibility o f a district committee drawn from woreda officials and memberso fthe community. The interesting features o f this program are that (a) unlike other programs, it i s available throughout the year, both in good and bad years, althoughthe duration for which an individualmightreceive food rations can vary between 3 to 12months, (b) households are selected by communities (Peasant Associations) through `means-testing' ,usingadministrative guidelines on the specific groups at risk based on access to land andother productive assets, and (c) while the benefit as percent o f extremepoverty line i s not insignificant, the cost o f administering the program - comprising mostly the cost o f transport to the distribution center - i s also not insignificant, but as much as 55% o fthe total cost (T. Woldehanna, 2002). 5.46 Employment Generation Program (EGS). The intended goal o fthe EGS i s to protect individuals from the adverse effects o f a temporary shock, such as drought, while at the same time creating socially valuable assets. The ideai s to use timely transfer o f food inexchange for labor to prevent individuals or households from depleting their critical financial andphysical assets andbecome trapped inpoverty forever. Additionally, the labor can be usedto buildpublic assets such as rural access roads, soil and watershed conservation mechanisms, reforestation, activities, andbridges--in short, assets that have the potential to improve incomes inthe longterm. Therefore, EGS attempts to exchange relief for development or private insurance for public investment. 5.47 It is stipulated that 80 percent o f all food aid is to be allocated to EGS activity, and the balance o f 20 percent to GR.The Guidelines also stipulate the wage rate (15 kgper personper month) and other norms for participation for pregnant andnursing women. Inpractice, a number of evaluations have pointed out that inmost locations EGS has degenerated into GR, and food meant for EGS was distributed without a work requirement, largely due to lack o f any non-food budgetary support at the woreda to implement a workfare program. Wage rates offered were generally lower than the stipulated 15 kg/per personper month. Ingeneral, most participants never received more than 12.5 kgper personper month. The program operates almost entirely with donated foods; the cost to the government i s nilor negligible (Subbarao and Smith, 2003). 5.48 Foodfor WorksProgram (FFW): Unlikethe case o f EGS, there i s much greater participation o fthe government. The program combines geographic andhousehold targeting. It selects communities where the soil i s degraded and deforested andwhere there i s a shortage o fwater. Within these areas, poor people self-select themselves into the program at times when the programwage i s lower than the market wage. Moreover, since it i s long-tenn and covers many woredas, it i s able to reachmore o f the poor. At times when the program wage exceeds the market wage, andwhen the activities are undertaken during the slack agricultural season, some non-poor farmers may also participate. Because the program by its very nature benefits able-bodied poor, it benefits - 53 - labor-rich households. Some very poor households with no adult labor may not benefit from the program. 5.49 School Feeding Program (SFP). The objective o fthe SFP i s to increase student retention insome food insecure districts by offering nutritional supplements through the school system. At present the program's coverage i s small, reaching only 258,400 students in600 schools located inTigray, Afar, Amhara, Oromiya and Somali regions. Although it was first piloted in 1994, not muchi s known about its effectiveness, and potential for scalability. 5.50 For a variety o freasons, key among which i s inconsistent data and lack o f systematic reporting and the fact that most transfer programs are donor funded and therefore off-budget, it i s difficult to say with clarity how muchi s spent on eachprogram, or the characteristics o fbeneficiaries and the size o fthe benefits they receive. Within these constraints, Table 5-7 presents a rough guide to estimated size o fbeneficiaries and estimated program costs (see World Bank, 2003-Ethiopia PER2003). Table 5-7: Programs, estimated costs and size of beneficiaries,Ethiopia. Program Estimated size of Benefits to Beneficiaries Total costs beneficiaries (million USD) FreeFood 2-5 million during normal - 0 8.8 USDper person 70 to 500' (Gratuitous year, but up to 10millionin Relief)' bad years. Employment An average o f200,000 per 0 11.2 USD per person 2.3 to 6.9b Generation year, but 600,000 (2002). Scheme(EGS)2 5.1 millionperson days o f employment (2002) FoodFor Work 357,000 per year 7 USD per person 20 (FFWQ SchoolFeeding 600 schools or 258,000 0 15 USD per person 4.1 Program(SFP)4 students Note: 1) For FFW--expenditure is based on World FoodProgram(WFP) country program o f about 60 million USD over 3 years. Benefits per beneficiaries are based on distribution o f 134, 824 metric tons o f food over 3 years. Foodtransfers are valued at 2 birrkilogram, and 1USD=8.6birr. 2) EGS expenditures are based on WFP transfers o f 47.7 kg o f food per beneficiary, which are valued at 2 birr per kg. 3) GR benefits are evaluated usingWFP showing average receipt o f 37.5 kg/person, valued at 2 birr per kg. Also cost o f delivering food is estimated at USD 35O/metric ton. 4) School Feeding Program, based on WFP cost o fUSD 12.4 million over 3 years. ') The low estimates i s the annual estimate based o n 2 million receiving 12.5 kg/month each when food i s valued at 2 birrikg. The high cost is based on 5 million chronically food insecure receiving 37.5 kglmonth each and food i s valued at 2birrlkg. Bothvalues do not include administrative and transportation costs. b,The high cost is based o n the average o f 200,000 people receiving 37Skglmonth each for 4 months ina year when food i s valued at 2 birrikg, while the low cost i s based o n the same number o f people receiving 12.5kglmonth each for 4 months ina year, when food i s valued at 2 birrkg. - 54 - 5.5 1 There are three points that need to bekept inmindabout these numbers. First, these are very conservative estimates. Only direct benefits are estimated here, and even these are indicative as the size o f the beneficiaries and the benefits per person are known to vary widely. Second, indirect costs, o f administration, transportation, and implementation are not included. The average costs o fbuyingand shipping (internationally), and locally delivering food inEthiopia i s estimated at aroundUSD265 per metric ton (Smithand Subbarao, 2003). In2001, which was not abadyear, about 500,000 metric tons o ffood was distributed, suggestingthat food aid cost at least USD132 million. Once administrativecosts are added (typically 50% o fprogram costs), the costs may approach about USD200 million a year. This is about 10% o ftotal public spending,where the latter includes budgeted expenditures andfood aidinflows. More specifically, Figure 5.1 shows some recent estimates (again rough) o fpublic spendingon risk-coping programs. It confirms that the value o f these programs i s substantial, and subject to volatility. Finally, the majority o fprogram costs are fundedby donors. The government's contribution i s estimated to be at about 7% o ftotal program costs in the period 1996-2000 (Smithand Subbarao, 2003). Figure 5.1: Total Spending on risk-copingprograms(million USD). 6w - 55 - 5.52 Inadditionto totalprogramcost, Table 5-7 also gives aroughestimate ofthe benefit per beneficiary for eachprogram. However, because o f the data problems mentioned above, it i s often difficult to know who receives these transfers. To get some idea, we turn to household data. Inboth 1995/96 and 1999/2000 surveys, households were asked to provide informationon sources o f financing expenditure on commodities and services they purchased. One such source included "gifts andremittances received from governmental organizations, non-governmental organizations, other households, and from abroad." Inthe 1995/96 survey, the sources were not explicitly identified. Instead, we observe total "gifts and /or obtained for free." First, ineach household, we computed total transfers from all four sources (government, non-government, other households and remittances from abroad) per adult, after adjusting for spatial and temporal differences inprices. Second, for the 1999/2000, where attribution to each o f the sources i s identifiable, we calculate the per capitareal transfers from what we refer to as public (government andnon-government) sources. Finally, after obtaining total and public transfers, we calculate average per capita benefitby consumption groups (deciles), for all, urban and rural households. 5.53 Figure5.2 shows total transfers per capita, while Figure5.3 shows public transfers per capita, to households by income group. It i s important to bear inmindthat these numbersdo not all refer to food aid, although it i s reasonable to expect inthe context o f Ethiopia that a significant share would be, especially when we refer to governmental and non-governmentaltransfers. With this caveat, there are several observations worth noting from the data. First, all householdincome groups receive some form o f aid, including what we have called public aid. This last observation i s particularly interesting since, inprincipal only really poor households are supposed to receive public assistance. Second, not only i s there a lot o f "leakage" (that is, more than the target group receives assistance), the transfers arenot particularly pro-poor since each income group receives roughly the same level o ftransfers per adult. Third, separating rural from urbanhouseholds indicates that rural households receive higher public transfers per capita than urbanhouseholds, especially in 1995/96, andthis i s true for each comparable income group (Figure 5.2). Fourth, ingeneral public transfers are too small as a share o fhousehold expenditure per capita. As indicated inFigure 5.3, the average transfer per capita for all households i s at most 13 birr (but only 7 birr on average), which i s only 1% o f average per capita expenditureo f the poorest decile. An additional thing to note i s that, the largest component o ftransfers i s between households. Unfortunately, poorer households -that i s lower deciles -who need the assistance the most, also happento receive the lowest levels o ftransfers (see Tables A11-A14). This makes the case for strengthening the targeting effectiveness o fpublic transfers all the more important. - 56 - Figure 5.2: Per capita free aid and gifts, 1995/96 and 1999/00. 160 0 I 1400 - I: 1200. wI I! ' 1000- Figure 5.3: Real per capita aid from government and non-governmental organizations. '1 LI 3 5 0 - L 3 0 0 - I! ' 2 5 0 - 'i 2 0 0 - : L 1 5 0 - : 1 1 0 0 - : -... _------- I 5 0 - ---.-......- ......._.-- . I . , - - . .--..------. 9 10 5.54 These observations on public transfers are consistent with local assessmentso f risk-coping transfers inEthiopia. A number of studies have pointed out that many more people than originally intendedreceive aid. These studies report that local administrators are often unable to exclude any family inthe community fkom aid distribution, perhaps because the local norms do not permit such an action or the needy are more than estimated. As a result, benefits per recipient were often too little to have a significant impact on the incomes o f the recipients (Save the Children, 2001; Jenden, 2002). This thenbringsusto amore general assessmento fthe effectiveness o fthe public risk management strategies, to which we now turn. - 57 - EFFECTIVENESSPUBLIC RISKMANAGEMENT PROGRAMS OF 5.55 Before taking up the assessment o fpublic riskmanagement strategies, it i s important to remember two points. First, Ethiopia's most visible and relatively well- knownrisk management programs have beenand continue to be dominatedbyrisk- coping strategies, primarily financed by intemational donations. The risk-mitigation strategies discussed above are not completely new, but have not beenimplemented inthe past on a scale comparable to the risk-coping strategies. At present significantly more i s envisioned for these risk-mitigating andpreventionprograms, butknowledge o ftheir effectiveness i s many years away. Second, the combination o fvarious risk-coping programs, supplemented on occasion by rare national reserves, have evolved over time to save many lives inthe wake o fperiodic rainfall failures. The fact that there has been no major famine o fthe type that took place in 1984 goes to the credit o f the government's approach and an impressive aid delivery mechanism. Additional features, such as adoption o f community (peasant association) led identification o fparticipants alongside administrative andself-targeting approaches, undoubtedlyhelped food reachthe most vulnerable groups. And the use o f localNGOsto deliver programs with donated food (which reduced the burdenon the government), speak to the programs' flexibility and reach. However, boththe risk-mitigation and risk-copingprograms as they are currently conceived and implemented have several drawbacks. 5.56 We start with the riskmitigation andprevention programs. The first i s that even though the risk mitigation strategies such as irrigation, water harvesting, agro-ecological packages, and resettlements provide an important step to address the problem o f food insecurity, they introduce their own risks (inhealth, environment, and conflict) in additionto the challenge of implementation. Second, even ifthe implementationissues are executed relatively successfully, unless these activities are carried out on a large scale, they may not be sufficient to solve all the problems o f chronic poverty and vulnerability, since the estimated size o fthe beneficiaries that can be helped from such programs i s only a fraction o f those estimated to be indeep poverty, and therefore vulnerable. 5.57 More specifically, the relatively highunit costs o f establishing irrigation schemes eveno f a modest size -- up to USD 10,000 per hectare for largeand mediumschemes and USD5,000 perhectare for small schemes-- introduce ahighrisk offailure. For instance, at the same time that small scale irrigation expanded the area coveredby 21,000 hectares, some 41,000 hectares inmediumand large scale irrigated areas were lost between 1991-1998. 5.58 Equally grave risks have emerged from the unintendedconsequences o f scaling up input use by linkinginput delivery and extension services. The low-priced, public- sector input delivery system had three consequences. One, it crowded out the private sector participants and inthe process underminedmarket development and deepening in an area that i s crucial for the success o fthe ADLI, the government's linchpin for the take- off o f rural development. Complex procurement rules, sometimes requiring up to 220 days to clear, and cumbersome foreign currency requirement have added to the difficulties o f the private sector entrants. Two, linking o f fertilizer sales and credit has led - 58 - to inappropriate and, insome cases, excessive application o f the fertilizer, especially on marginal farms. This inturnhas ledto excessive farm debt and potential disruption o f future production. Three, the federal government has begundeductingthe amount o fthe un-recovered credit guarantees from regional block grants, who have inturn cut back on their commitment to guaranteecredit. This, plusthe significant size ofnon-performing farm debt heldby commercialbanks, have created uncertainty regardingthe future availability o f credit to obtain these inputs. Already, local availability o f inputshangs in the balance. At present, there are hardly any private fertilizer distributors left, and the party-affiliated companies have reduced their activities because o f the credit-tightening. As aresult, fertilizer use, which rose from 150,000 tons in 1990to 290,000 in2000, is expected to decline to 260,000 in2003. Because the retail intermediaries supplieduseful technical andmarket information, their disappearance has resultedinloss o ftransfer o f valuable production knowledge to farmers. Also, at the local level, the bulkprocurement o f inputs from wholesalers bythe extension services, often failed to take account o f supply and demandconditions andhas ledto major input spoilage and losses. 5.59 When it comes to resettlement, Ethiopia's past experience alone has revealed a number o frisks associated with resettlement. One i s death. It i s estimated that up to 20,000 people died en route to their new location andmanymore from disease after arrival. Inaddition, property and social capital was lost as community life became disrupted. Furthermore, agricultural and other production goals were not achieved because complementary infrastructure andservices failed to materialize. Finally a bad situation became worse with social conflict between immigrants and local residents. Therefore, unless these risks (inresettlement, irrigation, andinput use) are addressedat the outset, risk-averse households maynot be too willing to adopt the strategies. 5.60 Third, and turningto the risk-coping programs, the overwhelming reliance on foreign sources o f finance underminesthese programs' sustainability and their predictability-- two essential characteristics o f a safety net inthe long run. Ifa catastrophic risk occurs and internationalresponse or logistical delivery o f aid fails, many households may simply drop through the net. Fourth, the targeting approach o f the current programs misses millions o fhouseholds who are food insecure. At present the program targets food insecure households in 156woredas inTigray, Amhara, Oromiya and SNNP regions. This list has recently been expanded to 262 woredas, under the new Safety Net Program, to include Afar, Somali and one woreda each from Dire Dawa and Harari. The government's position i s that the country's income level does not support a national food securityprogram that reaches all the food insecure households, and therefore it i s forced to makethese painful choices. While understandable, boththe size o fthe needy who are currently excluded and the implications o fthis position for poverty reduction inthe longterm are difficult to ignore. Consider first, the size o fthe problem. The food security program now targets about 5 million people, or 1million households assuming an average household size o f 5. However, in 1995, 33% o f 10million households inthe country would have qualified to be food insecure, since they could not obtain 2200Kca1, defined as the minimumlevel o f calories essential to function, which also defines the food security line. Moreover, 17% would have been classified as extremely food insecure, since they consumed less than 1650Kcal. Evenifthe current beneficiaries o fthe program are all extremely food insecure, an additional 7% o f all - 59 - households who are inthe sameposition as the beneficiaries would have no public assistance, which mightbe the only way for some to avoid sinkingdeeper into destitution. Inaddition, when inter-generational transfer o f disadvantage i s taken into account, their children are more likelyto have lower education and health, not to mention nutrition, leading to persistent poverty. 5.61 Fifth,the current safety net programs have anarrow focus ondrought shocks and its associated impact on food insecurity. To be clear, drought i s one of the biggest risks faced by Ethiopia and given its devastation inthe past and its continuing threat, it i s understandable that overwhelming national effort goes into dealing with it. There should be no mistake that such effort must continue. But there are other risks, some old and some new, that claimjust as much lives and that require concerted public action. Among these risks are malaria, HIV/AIDS, andmalnutrition, all o f which have the potential to lock households into poverty traps. 5.62 Sixth, there i s little to no synergy between existing programs. For example, there i s no attempt to integrate food security programs with efforts to build and protect human capital o f the most vulnerable groups such as orphans inpoor households, girls, or households inpastoralist areas. Similarly, workfare programs are not integrated with activities at various levels o f government. Foodmeant for public works program i s often distributed free, and its disbursement among more than intendedrecipients reduces its impact on the incomes o f the poor. Furthermore, implementation o f these projects i s left to communities or NGOs, at times. With no resources of their own, andwith no provision o f government h d s for implementation, the design and supervision o fprojects suffered. As a result, use o f food aid as a "dual purpose instrument o frelief and development" did not materialize and neither did the creation o f assets consistent with regional (community) needs and priorities (Subbarao and Smith, 2003). WELFARE COST OF SHOCKS 5.63 The review o fhouseholdrisk management strategies has shownthat although households actively call upon a diversity o f strategies to insure against risk, many o f these strategies tendto be weak and ineffective since they do not provide full insurance. The review also revealed that markets -most o fwhich are thin, incomplete or missing, and characterized by hightransaction costs -fail to provide insurance to the large majority o f Ethiopians, and even introduce additional risks that leadto complete market failures. The report then examinedifthe existing public risk management strategies meet the needs o fthe poor that could not bemet through informal and market mechanisms and found them to fall short bothintheir targeting (of risks and beneficiaries) and intheir choice o f instruments to protect the poor. 5.64 Whenrisks are not adequately protected against, they leadto hugewelfare losses. At worst they cause large-scale physical death. Inmany instances they induce asset depletion (land and livestock sales, land degradation, etc.) or delay investments (in education, nutrition, etc.), which propel households into a spiral o f irreversible losses and eventually poverty traps andhighvulnerability. H o w large are these risk-induced losses inEthiopia? Andhow canthese lossesbereduced? Inthenext section, we look at the - 60 - first question, and inthe following chapter we discuss the outline o fbetter risk management through a reformed social protection strategy. 5.65 Welfare costs of droughts: Inthe last three decades droughthas occurred with alarmingregularityinthe country. Ineveryinstance, ithas threatened the physical survival of one to seven millionpeople. The mostwell knowni s the 1984 drought, which i s believedto have led to the death o f 1millionpeople. Eveninyears with favorable rains, Ethiopiaremains food insecure. In2001 about 6.2 millionpeople needed emergency food aid, and in2000 about 10 millionwere inneed. In2001/02, a year o f exceptionally good harvest, the country still needed food aid for six months for 5.2 millionpeople, or 9% o fthe rural population, most o fwhom are concentrated inthe pastoral areas of southern Ethiopia andthe belg(March -May agricultural season) dependent areas o f eastern Tigray, Amhara and Oromiya (FEWS, 2002). 5.66 At themacro-economic level, droughtsleadto large income losses. Figure5.4 shows that droughts in 1983/84, 1984/85, 1987/88, 1991/92 and 1997/98 have all ledto bigcontractions inagricultural output (excluding forestry and fishing), someby as much as 25% (Easterly, 2002). Figure 5.4: The effect of drought on agricultural growth, Ethiopia, 1981/82-1999/2000. Agricultural GDP Growth and Drought Source: William Easterly (2002) 5.67 At the householdlevel, a careful studyby IFPRIbased on ahouseholdsurvey in seven drought prone areas encompassing different farming systems andkey ethnic and religious groups, found that income losses were devastating (Figure5.5). On average, cereal yield, measured as output (inkilograms) perhectare inthe 1984 drought year, was only 35% of the yield ina normal year. Relatively better-off households (that is, less poor) inthe highlandareas did a bit better since their cereal yield fell to 65% of a normal year, although their production, or output per person, fell to 54% o f normal. By contrast, - 61 - productiono fpoorer households inthe lowlands collapsed completely, while cereal yield droppedto only 10% o f the normal (Webb andvon Braun, 1994). 5.68 Similarly, duringdrought, livestock losses may bemassive and the consequences for livestock output, such as milk production, may be dramatic, especially for poorer households, who experienced severe drops inmilkproduction from a 4.8 liters per day to a one liter per day1*,compared to richer households,who were able to sustain a production o f 5 liters o f milkper day, becauseo f their ability to purchase enoughfeed for their livestock(Webb andReardon, 1992). Figure5.5: Impactof 1984 Droughton CerealProductionandYield. 1000 Source: Webb and vonBraun, 1994, Table 4.5 5.69 Our study, which constructs apanel of cohorts-that i s groups o fhouseholds with fixed membership, usingthe HICEWWMS 1995/96 and 1999/2000 -also show high sensitivity o f consumptionto rainfall variation. The study finds that consumption elasticity o f rainfall i s around one. Specifically, it showsthat a 10% decrease inrainfall can leadto between 7% and 10% decrease inconsumption (Table 5-8). Another study, usingonly rural householdconsumptionchanges, showsthat a 10% decrease inrainfall leads to about 8% decrease inconsumption (Dercon, 2002~). 5.70 Losses inconsumption can also be estimated by assessingthe impact o f transitory income -that is, changes inincome due to variation inrainfall and temporary changes in health-on changes inconsumption. UsingHICEWWMS, we find that a 10% decrease inincome attributable to changesinrainfall andill-healthwill leadto asimilar decrease inconsumption (7% ) as decreaseinrainfall(Table 5-8). l2Because poor households do rely relatively more on their livestock for milk(andor blood) [because o f their smaller number], their livestock are likely to be weaker and exposed to greater risksduring drought (Lybbert et al, 2001). - 62 - Table 5-8: Estimatedeffectsof rainfallandtransitory incomeonconsumption (1995/96-1999/2000). Between Estimates (standard errors) Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Annual rainfall 1.o 0.9 (0.025) (0.029) Deviation from long run 0.8 0.4 mean (0.39) (0.18) Income shocks 0.70 1.23 (0.56) (0.34) Other variables included Yes Yes Fixed Effects Estimates (standard errors) Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Annual rainfall 1.o 0.90 (0.005) (0.007) Deviation from long run 0.80 mean (0.06) 0.2 (0.04) Income shocks 0.70 0.4 (0.19) (0.15) Other variables included Yes Yes Number of groups 237 231 231 231 231 231 Minimum 5 3 3 3 3 3 Average 50 42 42 42 42 38 Maximum 125 105 105 105 105 102 Source: World Bank staff estimates from survey data. 5.71 Such huge consumption losses attributable to droughts addto the already high direct welfare costs o f droughts. A study inthe Gode District o f Ethiopiabetween December 1999 and July 2000 found that duringthe famine, the crude mortality rate was 3.1 per 10,000 people per day compared to the normal 0.5 per 10,000 per day, a six-fold increase. This i s a staggering number because it implies that duringthose seven months, about 6,000 excess deaths occurred (Salama et al., 2000). 5.72 Welfare costs of HIV/AIDS. Most common estimates ofthe economic costs o f HIV/AIDScompare the annual GDP growth with andwithout HIV/AIDSandattribute the gap to HIV/AIDS. The estimated macroeconomic costs for Africa range from 0.3% to 1.5% (Amdt and Lewis, 2000; Bonnel, 2000; Sackey and Raparla, 2000). Although these losses add up over time to significant income losses, they are still considered modest, inpart because the underlyingassumption i s that the increased mortality from AIDS is to increase productivity o fremaining people-fewer people per unit of land or physical capital. 5.73 However, AIDS does more than destroy existing humancapital. Bykillingyoung adults and creating many orphans, it disrupts social channels through which leaming of traditional fanning systems and new technology can bepassedon. At sufficiently high levels o fprevalence and mortality, it could lead to a complete economic collapse. - 63 - A recent studyofthe economic impact ofHIV/AIDSinSouth Africa, that takes into consideration the importance o fthese socially disruptiveeffects o f AIDS, shows that if there i s no public spending to containthe disease and treat the infected, and ifno effort i s made to aid orphansto acquire education, then a complete economic collapse will result inonly four generations (Bell,Devarajan, andGerbach, 2003). While tragic outcomes are possible ifno action i s taken, the example o fUganda offers a hopeful sign for all poor countries. It shows that, aggressive policy to contain the spread o fthe disease and innovative ways to increase the schooling levels o f orphans (butnot exclusive to AIDS), can prevent collapse. 5.74 Welfare costs of Malaria. The welfare cost o f malariahas tendedto be equated with direct economic losses, or the sum o flost productivity andmedical costs, associated with the disease. However, such costs neglect the physicalpainand suffering, and the cost o f the disease on all those who are not engaged inproductive activities such as children, elderly, andpregnantwomen. Inparticular, it ignores that malaria is one ofthe maincausesof low birthweight which leadsto highermortality andlower adult productivity o f the survivors. 5.75 The only known source to infer the economic costs o fmalariainEthiopiai s a carehl study that sought to measure what households inTigray would be willing to pay to preventmalaria altogether. The study askedhouseholds to revealwhat they would be willing to payfor a hypotheticalvaccine and contrasted that with the direct costs of preventingthe incidence ofmalariausinginsecticide treatedbednets(ITNs) (see Cropper, Haile, Lampietti, Poulos, and Whittington, 2000). As the authors state, usingwillingness to pay for a malariavaccine, inprinciple captures all the benefits o f malaria control: the vaccine eliminates the risk o f death, time lost to engage inproductive activities and leisure due to illness from the disease, medical and other expenditures related to the disease, and discomfort from pain. 5.76 Among the 18 villages covered inthe survey, 7% o f the households had lost someone to, and 78% o f respondents had an episode of, malaria intheir lifetime. Furthermore, the malaria incidence was evenly distributed across households members split into adults, teenagers and children. The out-of-pocket costs o ftravel to see ahealth practitioner, consult, andbuy medicine, was estimated at USD 3.5 to 5 per household, assumingthat an adult, a teenager and a childhad an episode ofmalariainthe same year, which amounts to 4% to 5% ofhouseholdfarm income (or 2% to 3% o ftotal median household income). Inaddition, for every episode o fmalaria, 21, 26 and 12 working days are lost for an adult, a teenager, and a child, respectively. When these lost days are converted into monetary equivalent, the lost productivity ranges from USD 9 to USD 31, or 9% to 31% o fhousehold farm income. The households were also asked what they would be willing to pay for a vaccine that would preventthe incidence o f malaria for a year. On average, households were willing to pay USD 36 (the median price i s USD 25), which again i s about 36% o fhousehold farm income (or 20% of total median income). These households are also willing to pay USD 18 for ITNs (Cropper, Haile, Lampietti, Poulos, and Whittington, 2000, pg. 21). - 64 - 5.77 Incomeloss fromidiosyncratichealthshocks. Thereis no doubt that HIV/AIDSandmalaria constitute two ofthebiggest sources ofburdeno fdiseasein Ethiopia, andtheir cost on welfare are staggering. But they are not the only sources o f illness inthe country. Other infectious diseases are common and a few days work lost to one o fthese maybe all it takes to reverse the fortunes o fhouseholds. From the household surveys data, it i s possible to estimate the income loss fi-om these idiosyncratic health risks. We findthat a simple separation o fhouseholds into those reporting an incidence o f poor health from those who are not, reveals as much as 10% less income for the former, which cannot be considered small, especially for poor households who inadditionmay have to incur extra expenditures to improve their health. 5.78 To summarize, this chapter presented a briefreview o f existing household and market strategies for risk sharing and foundbothto beweak 'andlargely ineffective. It then discussed, the welfare cost o finadequate protection focusing on keyrisks such as drought and health (HIV/AIDSand malaria). It showed that the losses inlife and incomes are intolerably high. Highlevels o frisk ina poorly developed market for risk sharingcreate enormous strains on the households, sometimes leadingto catastrophic social outcomes. There i s therefore a needfor public action to help reduce or mitigate the risks faced by households and to strengthen their coping efforts. The next chapter provides suggestions for a social protectionstrategy. - 65 - 6. Helping Households Manage RisksBetter: BuildingBlocks of a Social Protection Strategy 6.1 To overcome the shortcomings o frisk management strategies and reduce massive welfare losses, Ethiopia needs to develop an alternative social riskmanagement strategy that achieves a balance between copingwith risks and reducingor mitigatingrisks. In the Ethiopian context, we suggest that this strategy should include: SustainingGrowth. 6.2 Ethiopia's poverty reduction strategy document (SDPRP), recognizes that the long runprospects to end hunger and vulnerability inthe country rests on industrializationunderpinned by increasing rural productivity, hence ADLI.Early empirical evidence indicates that the rural reforms andinvestmentso f the 1990s have already ledto observable poverty reduction (Dercon andKrishnan, 1998; Bigsten, Kebede, Shimeles and Tadesse, 2002), but that muchvulnerability remains and much needs to be done inorder to sustain the momentum and achieve the true promise o f ADLI. To do this,the developinganddeepening ofkeymarketsis essential, andbelow we propose three such markets where large gains can bemade. 0 improvingthe functioningof landmarkets. Thereport's analysis reveals that the two mainproblems impedingthe fullpoverty-reducing potential o f landmarkets i s uncertainty over tenure and restrictions on secondary markets, such as leasing o f land. The available empirical evidence shows that despite the structural reality o f small plots and administrative restrictions on landmarkets, households continue to use the marketmechanisms to undertake mutuallybeneficiallanduse arrangements. More specifically, productive land owners with little land have found access to land from larger but less productive land owners. And this is as it should be. These observations suggest that having already achieved a fairly egalitarian land "ownership" structure, further interventions to maintain the status quo may neither benecessary nor beneficial for the country. 6.3 The needto make landmarkets work is all the more important inview o fthe problems besettingother markets such as output, labor, oxen and credit. At this point several policy actions are available to the government. It can officially announce the ceasing o f further land redistribution. Inaddition to eliminating the uncertainty surrounding the status o f existinguse rights, such a move would encourage expansion o f the secondary markets and off-farm sector. The beneficial impact o f such a policy stand could be increase by other policy changes, such as turning existing use rights into ownership rights and thenletting the market process work. Another alternative would be to maintain the existinguse rightsbut remove the current restrictions on secondary land markets. Together or incombination, these landpolicies could go a long way to increasingrural production, improving risk-sharing, andreducing vulnerability. - 66 - a Developing fertilizer markets. The fertilizer market startedon a promising note, buthas since become problematic. The entry of sizeable private participants at all stages o fthe market (bulk procurement, wholesale, retail, etc.) and the creation o f competencies along the areas o f operations hadthe imprint o f a process where risk-sharing was beginningto take shape ina natural way. But the policy decision to use extension services as the conduit for fertilizer sales, permitted party-affiliated andpolitically connected companies to enter the market and impose large foreign exchange requirementsto procure fertilizer. This has ledto increased risk to farmers. As things stand, interruptionsto fertilizer supply from the few firms left inthe market and uncertainty regardingwhat the government will do with risingfarmer debt, have increasedthe risk inproduction. A potentially undesirable response to this risk i s that farmers will avoid using fertilizer to reduce the risk.Already, it i s estimated that fertilizer use will drop from 290,000 tons in2002 to 260,000 tons in2003. This will be detrimentalto the efforts to increase productivity, improve incomes, andreduce poverty traps and vulnerability. 6.4 There are three policy moves that can help reduce the risk inthe fertilizer markets and eventually leadto higherand more economically rational use o f fertilizer. The first i s to promote competition inthe market. The government has introduced a uniform bidding document inthe high-fertilizer consuming regions, which i s expected to remove bureaucratic hurdlesfor all firms. But for this to work, transparent rules for selection o f bidwinners and equal access to credit will also be necessary. The secondpolicy action would beto separate fertilizer distribution and extension services. This will reduce the excess supply and the growth o fbad debts. The government i s already making efforts in this areaby encouraging cooperative unions and societies to bethe channels for distributingfertilizer. Care mustbetaken, however, to ensurethat the cooperatives themselves do not turn into monopolies. Finally, there is a need to reform the rules o f accessto credit and foreign exchange, so that private entrants can have a level playing field. a Deepeningthe operations of grain market institutions. Many aspects o f the grain markets, especially for the main staple, teff, appear to work relatively well, especially following the reforms implementedby the EPRDF government. This has beenobserved for short term (spot) transactions. Unfortunately, grain markets continue to be volatile, as the most recent price collapse (in2002) testify, markets are still thin, andtransaction costs are still highdue to lack o f standardization, large distances between production and consumption areas and poor road network, inadequate storage facilities necessary to arbitrage, and lack o f localized market information (regional) bytraders. 6.5 Policy changes should target the obstacles inall or most o f these areas. Many traders do not have the training or the finance to participate inmarkets that are large and diverse (that is, markets outside o ftheir regions). Therefore, availability o ftraining to improve the business and human capital skills o ftraders, inaddition to credit availability can improve competition and efficiency inthe market. Regarding credit, a major - 67 - constraint is that manytraders do not have enough physical assets that to serve as acceptable collateral for commercial banks. One suggestion mightbe to use grain stocks as collateral: the trader could deposit grain ina warehouse, which would issue a receipt indicating the value o fthe grain. The trader could thenuse the receipt to obtain credit from a commercial bank. Such a system could serve as an important initialphase o f developing the market untilmore predictable assets could be accumulated by traders. However, to ensure bank participation, reliable market price forecasts from apublic information system managed byprofessional bodies such as trader associations wouldbe necessary. Alternately, professional associations could also act as guarantors o f credit for memberslacking acceptable collateral by commercial banks. Poor roadnetworks, especially inrural areas, may be the single most important factor for hightransaction costs. A key policy action inthis area, o f course i s to continue to widen the reach o f these roads. Inadditionto reducingthe transaction costs o f obtaining market information, or makinga trade, etc., roads reduce the isolation o fmany households, promote labor mobility, and integrate markets. Diversification of the economy. a To sustain growth, the government i s already pursuingdiversification o f the economy. Plans to diversify the export base to include horticulture, textiles, and tourism could bring significant pay-offs. The success o f some o fthese activities, such as horticulture and textile manufacture, would hinge on how successful the government i s inharnessing water resources. Reversing or halting environmental degradation. 6.6 The long term impact o funchecked environmental damage can be observed in many areas o f the country, but most vividly inthe areas declared to be chronically food insecure. It i s commonly maintainedthat these areas have beendepleted through years o f population growth with no land-quality-enhancingtechnological innovations, and it is feared that ifthis pace o f erosion continues with no environmental measures, many more areas will be depleted, thereby increasing the size o f the population caught ina poverty trap. Therefore, special efforts are needed to prevent firther damage to the quality o fthe naturalresources. Unlike many other shocks, the effects o f environmental degradation are often harder to reverse. 6.7 A keypolicy action is to have a morepro-active population control policy. For this nothingradical is suggested. Indeed, simplymeetingthe existing excess demand for family planning services will go a long way inachieving some o f the objectives expected for this policy. However, general population control policy i s but one o f several activities to achieve environmentalprotection. Others include, carefilly plannedresettlements, affordable and sustainable supplemental irrigation, and water harvesting. But all three activities carry risks, which have been highlighted inprevious pages o f this report. For them to be successfbl, suchrisk must be addressed. Water harvesting, for example, could significantly improve the welfare o f Ethiopianhouseholds ifall or most o f its promises are realized: inaddition to enhancing resource conservation, it will improve agricultural - 68 - output, reduce the burdeno f domestic work (water fetching) on women andchildren, and improve nutritional status through diversification o f diet consumed. However, at present there i s little knowledge on the risk-reducing value o fthis program and the characteristics o f the beneficiaries interms o f food security. Therefore, a suggestion i s to conduct a cost benefit analysis o fwater-harvesting interventions and also to estimate the size o f food insecure households benefitingfrom them. Special focus on covariate health risks. 6.8 Among the sector developments, reducing the incidence o fHIV/AIDS and Malaria should be considered a priority item. The welfare costs o f HIVIAIDSinterms of lives lost has already taken its toll. The results from the study on SouthAfrica show that inaction can leadto a complete economic and social collapse. Regardingmalaria, the Ethiopia study impliesthat to the extent that the vaccine i s bought inanticipation o f preventingthe future cost o f illness, the willingness to pay price mustbe viewed as an indicator o fhousehold evaluation o fthe real cost o fmalaria, measured as income loss. Based on these results we can infer that an episode o fmalariaincidence has the potential to reduce householdincome by 30% every year. This estimate i s about 3 times the expected losses inhousehold income from the disease, that would be obtained by looking at only direct medical costs andproductive days lost. This should place malaria as one o f the keyrisks that reduces household prospects to escapepoverty. Infact, a study o f neighboring Kenya showed that reducing incidence o fmalaria ineach community to less than 10%will reduce vulnerability (the probability o f consumption shortfall inthe future) by20% (Christiaensen and Subbarao, 2001). 6.9 Itis important to remember that althoughthere isno vaccine for malaria inthe market at the moment, these households are nonetheless expressingtheir value for a vaccine or a program with comparable results. At the moment, bednets provide such an alternative, but since their price i s not very sensitive to demand, it would be necessary for the government to subsidize these product for it to achieve significant market penetration. Making existing safety nets more effective and complementary. 6.10 Existing social safety net programs only reach a small fraction o fthe very needy (except when there i s a widespread emergency), and they are totally dependent on donor funding. As aresult they are not covered bynormal government budget, and thus fail to take advantage o f obvious complementarities across programs. As a result, the programs neither protect livelihoods nor create productive assets, such as roads, dams, water catchment's areas, or humancapital. 6.11 To rectify these weaknesses, we offer several recommendations. First, expand the reach o f the safety net programs to benefit all those who are food insecure, rather than only 156 woredas. Since ESRDFi s bothnationwide andhas built capacity to deliver community initiated public programs, one possibility i s to use it as the instrument to scale upthe food security program andreach households beyondthe 156woredas. Second, to improve sustainability andreliability o f these programs, it i s crucial to provide adequate funding-preferably a lineitembudgetprovision inthe federal budget. Third, to make - 69 - these programs serve the dual purpose o f helping cope with shocks and generate valuable assets (that is make them enhance productivity o fpublic investments), reform the design and increase the coverage o fthe existing public works program. The public works program should be timedto kick inwhen protection for the poor i s greatest. But to be effective, they have to pay a wage (or equivalent compensation) that will attract only (or mostly) the poor, and they have to have adequateprovision o f complementary inputs (say tools). And although they can be selected by the community, they must be integrated with woreda (district) level development programs. Finally, a long term strategy o freforming food aid (or cash equivalent ifmonetizable) to serve the purposes o f protecting assets should be givenpriority. 6.12 As to which assets shouldbeprotected, we notethat eventhough Ethiopia continues to undertakepublic investmentsfor reducingbroad groups o frisks, the most vulnerable groups often do not get the benefit o fthe protection. The investments mustbe carefully targeted to prevent potentially irreversible asset losses. We suggest targeted transfers aimed at protecting education, health, and nutrition. InEthiopia, this can be done by reshaping existing food aid transfers as an instrumentto maintain or improve these assets. A few examples here include: 0 Education: The transfer of food aid (or cash equivalent) to poor or isolated households whenthey send or keep children inschool, which, inturn, reduces child labor, endures schooling o f orphans, improves girls' enrollment, expands education o f isolated andpastoralist people, and so on. Orphansneed special attentionbecause even whenbroad free education programs are available, they still receive less education. A study of 10African countries, including neighboring East African countries o fKenya, Uganda, and Tanzania showed that orphaned children had lower school enrollment. This did not appear to be due to theirpoverty level(Case, Paxson andAbleidinger, 2002), butrather to discrimination. This provides the scope for public action. There are good reasons to target reduction o f childlabor inEthiopia. First,the country continues to have one o f the highest child labor incidence inthe world, despite positive government action to promote flexibility in school calendars inthe country. Second, food aid has been shown to be effective inreducing child labor. Inone study, the probability o f child farm labor supply, the mainreason cited for child labor inthe country, decreased from 58% to 37% when the per capita value o f food aid received by food aid recipients increased by USD 4.2 (Yamano, 2001). 0 nutritionprogram: Although malnutrition levels have fallen inrecent years, they remainvery highrelativeto comparable countries, perhaps because chronic malnutrition has not been addressed. One way to do this i s to start providing better nutrition to children early intheir lives by targeting children less than 30 months. Indeed, inEthiopia, food aid has been shown to have a large positive effect on growth o f children inthat age category. Studies findthat food aid can mitigate the negative effects on child growth that would result from plot damage (Yamano, Alderman, and Christiaensen, 2003). Inaddition to early childhood development and growth monitoring, an effective nutrition program should include community management, expanded immunization, and increased education o fmothers. - 70 - 6.13 It must be emphasized that these targeted transfers have to betreated as complementary to the broad risk-reducingprograms ineducation and health that are already on-going. When donejointly, these types o f interventions offer greater potential for effective targeting and better cost-effectiveness o fthe programs. By integratingthe instruments from each o f the broad strategies that complement one another, the country can increase the overall returns to public investmentsand reduce the populations vulnerability to poverty. -71 - 7. Conclusion and Next Steps LESSONS LEARNED 7.1 A numberofuseful lessonshave emerged from this review ofrisks and vulnerability inEthiopia. First, the patterno fwelfare dynamics indicates that several measures o f well-being have improved inthe 1990s. However, the extent o fthe decline inthe consumptionmeasureis still unclear. While our analysis ofthedata show stagnation and support the recent official estimates which report very slow rates o f poverty reduction (GOE, 2 0 0 2 ~other studies show substantial declines. The factors ) ~ determiningpoverty reductioninclude several years o ffavorable rains (what we have called "good luck"), peace, and public investments. 7.2 Second, the improvements inwelfare were not observed inevery village or every region. Tigray, Oromiya, and SNNP andpossibly smaller and isolated regions such as Afar, Somali, Benshangul-Gumuz and Gambella, witnessedslower reductioninpoverty between 1995/96 and 1999/2000. Our data suggests (though insufficient to claim positively) that Tigray's ascent from poverty was delayed duringthis periodbecause o f the degradation o f its highlands and poor rainfall, while the SNNP region-where the rainfall was good -sufferedfrom collapsed world market prices for its main cash crop o f coffee. 7.3 Third, despite the general improvement inwelfare measures, large fractions o f Ethiopians remainvulnerable to poverty from a multiplicity o f risks. The most pervasive of these are community wide risks such as weather, malaria andthe newly emerging HIV/AIDS, and idiosyncratic health shocks. Moreover, we find that the welfare costs o f these risks, whether measured as physical death, income, or consumption losses are very large. As a result, risk-induced vulnerability to poverty i s very high. Specifically, we findthat about 10% ofEthiopian cohorts are chronically poor, while an additional 35% move inand out o fpoverty inany given year. Additionally, we find that the most vulnerable tend to be older, live inhouseholds with more dependents, and live inremote and isolatedpopulations places, far from public-provided services. 7.4 The value added o fthis risk and vulnerability assessmentis to point out that the definition o fthe poor shouldbe expanded to include notjust the persistentlypoor but also those currently non-poor who stand a highchance o fbecoming poor inthe future. An additional contribution o f this risk-focusedperspective is to show more explicitly the impact o f risks on the welfare o fhouseholds. - 72 - 7.5 Interms ofpolicy, thisperspective arguesthat itis not enoughto only movethe poor out o fpoverty, but that it i sjust as important to prevent them from falling back into poverty. This means that a sustainable poverty reducing strategy should have a balance between risk-reducing or -prevention activities and risk-coping activities. Secondly, the risk and vulnerability analyses call for a multi-sectoral approach to preventing, reducing or coping with risks. For example, dealing with drought riskmay involve closer integration and coordination o frisk-reducing activities such as water management, resettlement, income diversification, rural roads, and risk-coping strategies such as food or cashbasedpublic works, school feedingprograms, nutritional supplementation and so on. Therefore, from the view o friskmanagement, all public expenditures --risk-reducing and-preventing and risk-coping- are investmentswhen their objective i s to prevent households from falling into poverty traps. 7.6 Interms ofstrategy, the focus onreducingrisk-inducedpoverty traps impliesthe needto make risk andvulnerability assessments as an integral part o fthe regular monitoring and evaluation o f development policy. They can be done as stand-alone analytical document or as part o fthe standardpoverty assessment. However they are done, they shouldbeginwith an identification o fthe mainrisks and those vulnerable to those risks. They shouldthen evaluate existingrisk management strategies by households, communities, and the public sector. Such evaluations should examine the existingbalance betweenex-ante riskmanagement strategies and ex-post (risk-coping) strategies. This knowledge should provide the efficiency o f a country investment strategy indealing with all aspects o friskmanagement. After evaluating existingrisk management strategies, the assessment should develop an implementationplano f sustainable public interventions that are likely to bepursued. NEXT STEPS 7.7 This review is intendedto be a first step inwhat is hopedto be a continuous process o f dialogue about how to builda robust and cost-effective social protection strategy for Ethiopia. As an immediate next step, it i s hoped that the current and future dialogue over the PRSC policy actions can benefit from the findings inthis review. Inparticular, together withother existingassessments, it canserve as anorganizing framework for initiating the difficult and complex discussions over objectives o f abroad social protection strategy, trade-offs betweenprograms, and selectivity. 7.8 To informthe dialogue on makingpublic risk-management strategies more effective, there are several outstanding issues requiringmore knowledge inkey areas where there is a felt needfor reform. These include: 0 Impact o f food aid on markets. What are the relative contributions o f food aid inflows and other market imperfections (e.g. poor storage, lack o f credit, isolation, etc.) to domestic food price volatility andavailability? 0 The role o fEthiopian Strategic FoodReserve (ESFR). What market stabilizing role can ESFR play, ifany? - 73 - Monetization o f food aid. What are the benefits o f monetizing food aid? What are the risks? What are the components o f a transition process to full or partial monetization? Scaling up public works program. What are the design and fundingmechanisms for makingpublic works program act as a more effective risk management tool? Weather-based insurance. How feasible i s a weather-based insurance institution for Ethiopia? Who should it target (regional or woreda governments, producers or traders associations)? Finally, although environmental degradation i s a major risk, there i s no focused effort to combat it. What are some internationalbest practices inenvironmental rehabilitation that Ethiopia can draw upon? - 74 - A N N E X E S - 75 - Annex 1: Methodologyand Data 1. Inthis annex we discuss, inabit more detail, the data andthemethodology underlyingthe analysis inthe mainreport. Inthe first part, we describe the dataused, including their comparability over time. Inthe second part, we start with the methodology and demonstrate the role that shocks play inthe variability inconsumption andtherefore uncertainty of futurewelfare, andhow to use the knowledge o fthe distribution o f shocks to estimate vulnerability to poverty. DATAUSED 2. Ethiopia has several large and nationally representative household surveys. The householdincome, consumption, and expendituresurveys (HICES) as the name suggests, collect information on expenditureandincomes. Since 1995, two such surveys havebeen conducted. The welfare monitoring surveys (WMS), collect information on a broad set o fhousehold characteristics, including education, health, housing, and distances to markets andpublic services (clinics, schools, etc.). A recent demographic and health survey (DHS) gives information on maternal and childhealthand family planning services; and a labor force survey (LFS)presents information on the working world. 3. To understand the changes inconsumption or non-consumption (health, education, malnutrition, etc.) measures o f welfare, it i s important to use data sets that have rich information on these variables and are comparable ifavailable, over time,. 4. Datacomparability:Inadditionto householdcharacteristics, the WMS includes a short module on consumption, and, because there are four years o f such surveys, it would seem reasonable to use them, together with HICES, to do a richer analysis o f welfare changes inEthiopia. However, the information on consumption inWMS and HICES are not comparable. The data collected under WMS was muchmore limitedthan inHICES, andthe recallperiod, (voicedbyquestions such as "how muchhave you spent on clothes inthe last one week, one month, and so forth?") was not the same. Finally, the 1998 W M S was done a bit later inthe year than the other three years. By contrast, the combined HICES/WMS o f 1995/96 and 1999/2000 are comparable. As indicated in Table Al, the months o f the year the survey was inthe field were the same so that issues o f seasonality were handled the same way inbothperiods. Also, the recall period for the information, especially on important variables like consumption, was the same. - 76 - Table A 1: Type and Time of Survey. I Survey I Year I Sample size I Month I I I I WMS 1996 12,260 July 1995,January 1996 15,034 February1997 1997 1998 45,123 MarcWApril, 2000 2000 25,917 JanuaryIFebruary2000 HICES 199912000 17,283 JulylAugust, 1999 199511996 12,260 JanuarylFebruary, 2000 July 1995 January, 1996 DHS 2000 February-May LFS 1999 March1999 (?) 5. Price date fromthe various surveys was also not comparable. Because the surveys were separated intime, and households live indifferent geographical areas that have differentprices, real expenditures could not be computed. We usedreporting level specific poverty lines to account for geographic differences inconsumptionbaskets and prices. For temporal changes inprices we used inflation numbers computed by the Government o f Ethiopia Central Statistical Authority (CSA). Eachreporting level poverty lines and therefore deflators were derived usingthe expenditure data collectedin the respective years (GOE 199b, 2002d). For these reasons, we use the combined HICES/WMS 1995/96 and 1999/2000, a summary o f which i s provided inTable A1. 6. To get an idea o f the dynamic nature ofpoverty andto implement the concept o f vulnerability as discussed inthis report, one requires panel data. Although, HICES/WMS 1995/1996 and 1999/2000 surveys are notpanel data, large cross-sections are repeated over time and can beused to follow groups o fhouseholds. With such surveys, the same households cannot be followed over time, as fresh households enter the surveys each time. Instead, cohorts or groups o fhouseholds with a fixed membership canbefollowed over time. Once cohorts are identified, panel data methods can be applied and vulnerability o f cohorts estimated. 7. Creation of cohorts: First, we group individuals into their administrative zone o f residence. This i s a level o f geography lower than the region, but it provides uswith significantly more observations. This alone provides us with about 48 cohorts from the 6 regions that are included inthe sample. Ineach zone, we group the data into 10-year age bands for individuals over 20 years old. We are able to identify 5 such age groupings. This gives us a total o f240 cohorts. Thenwe have 2 years o f data, which provide us with 480 cohort-year observations. The cohorts are groups o fhouseholdheads belonging to the same age group living inthe same administrativezone. We droppedsome cohorts that hadmissingrainfall andother variables. We also dropped all cohorts with less than 5 members.Altogether 474 cohort-year observations were used inthe estimation. The average cohort had37 members in 1995/96 and36 in 1999/2000. Table A2 shows the sizes o four cohorts. - 77 - Table A 2: Size of age cohorts and average size of cells, rural Ethiopia, 1995196-199912000. Source: World Bank staff estimate from survey data. 8. Data on shocks. The HICES/WMS data has very little information on shocks. The only valuable shock data collected i s health shocks. Inthe healthmodule, each individual was asked whether s/he experienced an episode o f illness two monthsprior to the start o f the survey. We use this data on incidence o f illhealthto estimate the impact o fhealthshocks on consumption. However, we do not estimate vulnerabilityto health shocks since there are not sufficient observations to calculate the variance o f consumption due to health shocks. Table A3 shows that between 1995/96 and 1999/2000,more people reportedto have suffered an episode of illhealth. The proportion ofpeoplewho reportedhavingbeen sick two months prior to the start o f the surveyrose from 25% to 35% inthat period. 9. Rainfall data: Although not available inthe HICES/WMS, we were able to obtain rainfall data recorded over 30 years for each o f the 55 zones inthe country. Bymerging the rainfall data to the HICES/WMS data, we are able to estimate the sensitivity of consumptionto rainfall and the vulnerabilityo f cohorts to this shock. Table A 4 shows summary statistics o frainfall, while FigureA28 plots distribution byregion. METHODOLOGY 10. To motivate the discussion, consider a simplemodel, where consumption (inlogs) o f an individual i inperiod t ,which we label as Ci, i s determined by a set of observed endowments inperiod t , Xi,,and a random event, Si,,that is; logc, = a +ax, +ps, 11. To make the exposition simple and relevant, we shall refer to the random variable, Si,,as a shock, and the constants a,6 and p as returns to the fixed characteristics, observed endowments and shocks respectively. - 78 - 12. Inaworldwithout uncertainty(whererandomevents or shocks do not occur), the log consumption o fthis individual i s perfectly predictable by the level ofhidher endowments and the returns to those endowments ineachperiod. Usingequation (1.l), and setting the last term on the right handside equal to zero, we get, 13. Where we have used"bars" to distinguished the true and often unobserved relationship o f equation (1.1) from the predictable log consumptioninequation (1.2). Intypical settings, the fixed characteristics would includefactors like gender, while endowments may include education levels, experience at work, or occupation. Ifwe now introduce shocks to this world, log consumption inthe hture (t=l, 2,3, ...,T years ahead) will be the sumo fthe predictable log consumptionin(1.2) plus the size o f the realizedshock inthe relevant year. Since the shock will vary randomly over time, each household faces a log consumption level eachperiod that has the potential to deviate quite substantially from the predictable level. We refer to the variation o f log consumption around the predictable level, Var(ZogCiJ,as consumptionrisk, and the chance o f falling below a given poverty line as vulnerability. 14. For empirical purposes, a long series o f observed log consumptionper individual coupledwith detailed knowledge o fthe timingo f shocks would constitute an ideal data set inorder to obtain the impact o f shocks on the variation inlog consumption. This matters because ifall the data came from the same distribution, it i s possible to predict the average log consumption and its variance over time anduse this informationto predict the probability that the shock will hand this household a log consumption level below the poverty line, which i s exactly what we would want. 15. But since such data sets are almost never collectedanywhere, andsince calculation o fvulnerability requires a reasonable knowledge o fbothpredictable and variance o f log consumption, we have to designa less perfect, but nonetheless credible procedure to predict the probability o f falling below the poverty line. Two questions come to mind. First,how does one estimate log consumption variance? Second, how does one resolve the absence o f a panel? Inthe data section, we discussed the issue o f how to obtain a panel, inthe absence o f one. Inthe next section we provide an answer to the first question. 16. Measuringvariance of log consumption: Recallthat our interest is to find some way to associate movements inlog consumptionto incidents o f shocks. There are two practical difficulties to obtaining total variance o f log consumption. One i s the well- knownproblem o f measurement error and the other i s the fact that we do not have a long panel (that i s many years o f observation) o f individuals. To see the first problem, suppose that - logc, =log ci,+Ei, (1.3) - 79 - 17. Where observed log consumption (first term on the left hand side) i s a fbnction of true log consumption (first term on the right hand side) and measurement error (last term). Ifwe wanted to obtain the variance o f log consumptioninthis environment, we would get, - R, (.) = var(logCit ) = var(1og C,) -+v a r ( ~ ~cov(1ogC,, .zit) + ~ ) (1.4) 18. Since the true log consumptioni s not observed directly, even though the last term i s zero which will be the case ifmeasurement error i s independent o f the log consumption level, the variance o ftrue log consumption i s over-estimated bythe presence o f the variance o fmeasurement error (the middle term on the right). Since measurement error can account for up to 50% o fthe variance inlog consumption (Dercon, 2002c), this i s a major problem. Furthermore, the common way to deal with this problem, which involves finding a variable or set ofvariables that canpredict true log consumptionvariance but are not correlated with measurement error, face practical difficulties o f their own. 19. The second challenge we face whenwe try to estimate total variance o f consumption i s that while we do have many years o f data on rainfall (the shock o f interest), we do not have a long series o f observations on log consumptionper individual. Therefore, rather than estimate directly the total variance inlog consumption, we take the modest route o fmeasuring the portion o fvariance inlog consumptionthat i s explained by the variance inrainfall. To see this, consider the variance of log consumptionimplied inequation (1.l), inorder to focus onthe role ofshocks, ignoredthe terms for and observable characteristics. Then, R,(S) = var(logCit) = p' var(Sit) (1.5) 20. Our parameter o f interest i s p which measures the effect o f changes inthe shock on the changes inlog consumption. Unlikethe difficulties o f obtaining total log consumption variance mentioned above, all the data neededto calculate R,(S) i s available. We can obtain p byregressing log consumption on rainfa1ll3. We use the repeated cross-section method o fDeaton (1985). Put simply, we obtain two periods of observations on log consumption at the cohort level. We also obtain corresponding rainfall records for the two periods. We treat these observations as a two-period panel. Inthecontextofthisannex, theideaistouseequation(1.l), obtainthe dataatthe but cohort (discussed above), rather than individual level. We then apply a difference method, that i s regress changes inlog consumption on changes inrainfallbetween the two periods. We obtain two results -with andwithout controlling changes inother observable variable. We can also obtain the variance o f the shock (Yar(S,,) ) usingthe 30 years o f observed rainfall data. The two (equation 1.5) gives usthe likely size o f log consumption variance implied byrainfall shock. As given, R,(S) can be interpreted l3 Since, ideally, this parameter should be obtained by a long series of consumption to match the rainfall years, we have to live with the assumption that the period relationship (what we have) i s stable and reflects the long runimpact of rainfall on consumption. - 80 - as the portion o f consumption risk explained solely by variation inthe shock. Thus, from an analysis o fvariance perspective, the ratio Rc(s$c (.) is equivalent to RZof consumption regressed on an observable shock after having partialled out the effect o f other variables. 21. To focus onthe impact o f shocks, we hadput aside the connection between shocks and observable characteristics. It is important to recognize that inreality, the observed or reported log consumptiondoes not always go up or down bythe exact size o f the shock because, inthe event o f a negative shock to log consumption, households may rely on coping mechanisms to make up all or some o fthe impliedshortfall. Similarly, in the event o f apositive shock, they may lend or save some o fthe increase. Becausewe rarely observe these responses, for practical purposes, we try to estimate the total shortfall inlog consumptionthat cannot bebridgedwhen a shock occurs. 22. Measuringconsumptionriskfromincomeshocks. So far, we have argued usingrainfall shock as a direct measure o flogconsumption shock. However, it is more likely that rainfall shocks enter into log consumption function only indirectly. In particular, it i s fairly well-known that there i s a stable relationship betweenpermanent income and consumption, andshort-run fluctuation inconsumptioncome fiom short run changes inincome -that is, transitory income. Therefore, inthis paper we estimate permanent and transitory income and use the latter as the measure o f consumption risk. We do this intwo stages. Firstwe estimate the size o fthe transitory income by running a regression o ftotal income on quantities that determine permanent and transitory income. To estimate transitory income, we use the deviation o fcurrent rainfall from the longrun regional meanrainfall, the standard deviation o f current rainfall, and an indicator o f a negativehealth shock duringthe survey year. Inthe second stage, we use the predicted values o ftransitory income from stage one to estimate the sensitivity o f consumption to the predicted transitory income. Thenwe use the coefficient o f and the value o f the predicted transitory income inthe consumption model to predict consumption risk. 23. EstimatingVulnerability:Vulnerability is definedas the probability o f falling into an undesirable condition (poverty inthis case) due to a shock. A shock i s defined as the realization o f a random event, such as a drought, ill-health, suddendeath, etc. Then, formally, V can be stated as, v PL =Pr(1ogC,