Report No. 27061-EC Ecuador Poverty Assessment April 2004 Poverty Reduction and Economic Management Sector Unit Latin America and the Caribbean Region Document of the World Bank TABLE OFCONTENTS RESUMENEJECUTIVO ............................................................................................................................................ EXECUTIVE SUMMARY ................................................................................................................................... ..i XXVII INTRODUCTION ....................................................................................................................................................... li 1 MACROECONOMICDEVELOPMENTSANDPOVERTY . ......................................................................... 1 THEIMPACTOFECONOMIC GROWTHANDECONOMICVOLATILITY ONPOVERTY .................................................... 2 Sources of economic growth................................................................................................................................ 3 Evolution of GDPper capita. 1970-2002............................................................................................................ 2 Determinantsof economic volatility andpoor growth ........................................................................................ 5 Growth, volatility and poverty ............................................................................................................................. 9 Theeffect of economic volatility on GDPper capita growth............................................................................... 8 12 THEIMPACTOFTHE 1999CRISISONPOVERTY ....................................................................................................... Promoting stable GDP growth throughfiscal discipline................................................................................... 13 Sources of the crisis........................................................................................................................................... 13 Social spending and the crisis............................................................................................................................ 15 Poverty and inequality during the crisis............................................................................................................ 13 THEIMPACTOFDOLLAREATION ONPOVERTY ....................................................................................................... 15 Formal and informal dollarization .................................................................................................................... 16 17 CONCLUSIONS......................................................................................................................................................... The impact of dollarization on consumptionandpoverty.................................................................................. 23 2 NATURE,DISTRIBUTION AND EVOLUTION OFPOVERTYIN1990-2001 . ........................................ 25 POVERTY UPDATEAND POVERTYPROFILE ............................................................................................................... 25 Poverty and inequality in Ecuador in 2001.. ................................................................................................ 26 Living conditions and the characteristics of 27 Correlatesofpover ty ......................................................................................................................................... poor........................................................................................ 30 CONSISTENTESTIMATES OFPOVERTY, 1990-2001 .................................................................................................. 31 Poverty Mapping in Ecuador: A Brief Introduction .......................................................................................... 31 TheEvolution and Distribution of Poverty in 1990-2001:Main trends and canton-level changes................... 34 LOCALDETERMINANTSOFGROWTHAND POVERTYREDUCTION 39 Initial conditionsand changes in poverty.......................................................................................................... .............................................................................. 40 CONCLUSIONS......................................................................................................................................................... Changes in employment and changes in poverty............................................................................................... 42 44 3. URBANPOVERTY,LABORMARKET DYNAMICS AND FORMAL EMPLOYMENT CREATION 45 MAINLABOR MARKET DEVELOPMENT INURBAN AREAS, 1997-2002 Laborforce participation, employment and unemployment.............................................................................. ................................................................... 46 46 Labor income and wages................................................................................................................................... 48 URBAN POVERTY AND LABOR MARKETS .................................................................................................................. 52 Household income and urbanpoverty trends.......... ..................................................................................... 53 Whoare the urbanpoor? TheRole ofhbor Market Outcomes........................................................................ 56 LABOR INCOME AND THE DEMANDFOR SKILLS ....................................................................................................... 65 CONSTRAINTSTOEMPLOYMENT CREATION:ANANALYSISOFTHEMANUFACTURING INURBAN SECTOR AREAS71 Employmenttumover and employment creation: Whatfirms do versus whatfirms want................................. 72 74 Actual employment creation and labor productivity.......................................................................................... Constraintsto employment creation and business expansion............................................................................ 77 79 CONCLUSIONS......................................................................................................................................................... Policies to increase labor productivity and employmentcreation..................................................................... 80 4. RURALPOVERTY.AGRICULTURALPRODUCTIVITY.AND THEDISTRIBUTIONOFLAND 81 ... WHOARETHERURAL POOR?................................................................................................................................... 81 CONSTRUCTING MEASURES OFAGRICULTURAL PRODUCTIVITY AND LAND DISTRIBUTIONFOR ECUADOR ...............83 83 Tumingyields into dollars................................................................................................................................. 87 Yieldsper hectareandper worker-hour: Afirst approximation........................................................................ Estimating agriculturalproductivity.................................................................................................................. 90 Small-, medium-,and large-scalefarms are all moreproductive in high-productivity areas........................... 92 Thedistribution of land in Ecuador is extremely unequal................................................................................. 93 AGRICULTLJRAL PRODUCTIVITY, HOUSEHOLD INCOMES, AND POVERTY: WHO STANDS TO BENEFIT FROM POLICIES TO INCREASE PRODUCTIVITY AND ACCESSTO LAND? Self-employedfarmers ....................................................................................................................................... 95 ........................................................................................ 95 Agricultural laborers......................................................................................................................................... 96 The relationship between agricultural productivity andpoverty across cantons.............................................. 98 POLICIES TO RAISEAGRICULTURAL PRODUCTIVITY ................................................................................................. 99 Closing the gap: How to get on the Production Possibility Frontier ................................................................ 99 Theimpact of policy reforms on output: Simulation results............................................................................ 101 POLICIES TO INCREASEACCESS TO LAND ............................................................................................................... 105 Promoting tenure security ............................................................................................................................... 105 Encouraging land transactions........................................................................................................................ 105 THE RURALOFF-FARM SECTOR INECUADOR ......................................................................................................... 106 CONCLUSIONS ....................................................................................................................................................... 110 5. SOCIALSERVICESAND THEPOOR ........................................................................................................ 112 SOCIAL OUTCOMESIN ECUADOR: COMPARATIVE A PERSPECTIVE ....................................................................... 113 Intemational comparisonsof health and education outcomes........................................................................ 113 Within country variation in health and education outcomes........................................................................... 116 LEVEL, COMPOSITION, CYCLICALITY AND INCIDENCE OF SOCIAL SPENDING 119 Trends in Social Spending: Level, Compositionand Cyclicality ..................................................................... ....................................................... 120 Incidence of Social Spending:Does Social Spending Benefit the Poor?......................................................... 122 RE-TARGETINGSOCIAL PROGRAMS: ANEVALUATIONOFTOOLS AND PROJECTS ................................................... 128 130 Using the SelBento re-target the Bono de Desarrollo Humano (BDHJ.......................................................... Using the SelBento re-target the GasSubsidy - A Simulation ....................................................................... 132 CONCLUSIONS....................................................................................................................................................... 133 BIBLIOGRAPHY ................................................................................................................................................... 135 ANNEXES ................................................................................................................................................................ 144 ANNEX1.ECUADOR THEANDEANCONTEXT .................................................................................................... 144 ANNEX2.ECONOMIC IN GROWTH ANDECONOMIC VOLATILITY - METHODOLOGICAL ISSUES147 ................................. ANNEX 3 COMPARABLE SMALL-AREA ESTIMATES OFPOVERTY: TECHNICAL NOTE 151 ANNEX 4 MEASURINGMONITORINGPOVERTY, SOCIAL OUTCOMESAND PROGRAMS .. .............................................. AND .................................... 160 ANNEX5.NOT WANTING TO REINVENTTHE WHEEL: THEECUADOR POVERTY ASSESSMENT OTHER RECENT AND WORK ON ECUADOR ........................................................................................................................................ 162 DATAAPPENDIX .................................................................................................................................................. 164 ACKNOWLEDGEMENTS The Ecuador Poverty Report was prepared by a team led by Carolina Sanchez-Paramo and including Maria Caridad Araujo, Daniel Dulitzky, Mauricio Leon, Norbert Schady and Raimundo Soto. Many thanks are due to: Tamar M. Antinc, JeskoHentschel and Ana Revengaas peer reviewers. Peter Lanjouw, Carlos Larrea, Pilar Larreamendy, Donald Larson, Rinku Murgai, and Thomas Pavefor valuable inputs and comments. Gal0 Arias, Estuardo Alban, Remigio Burbano and Eduardo Encalada from the Instituto de Estadisticas y Censos del Ecuador (INEC); and Patricio Davila and Victor Hugo Bucheli from the Proyecto SICA - Ministerio de Agricultura for granting the team access to the data used for the analysis. Juan Ponce from the Secretaria TCcnica del Frente Social; Roberto Salazar and Diego Martinez from the Ministerio de Economia y Finanzas; and members of the Ecuador Country Team and staff at the World Bank office inQuito for support inWashington and Ecuador. Anne Pillay for invaluable help inthe preparation of the final document. Finally, financial support from the Rural Development Strategy for Ecuador, managed by Peter Werbrouck and Jose Maria Caballero, the Norwegian Trust Fundfor Poverty Mapping, managed by Quentin Wodon, and the Human Development sector inLatin America i s gratefully acknowledged. ECUADOR:EVALUACION DELA POBREZA RESUMENEJECUTIVO Durante las dos dltimas dtcadas Ecuador experiment6 bajas tasas de crecimiento del PIB, a causa de las cuales no present6 aumentos en el PIB real per &pita. Especificamente, mientras el PIB real creci6 a una tasa anual del 2 por ciento entre 1980 y 2001, el PIB real per c8pita disminuy6 medio punto porcentual por aiio entre 1980 y 1990 y se mantuvo casi constante desputs de ese aiio. Entre 1980 y 2001, el crecimiento del PIB y del PIB per csipita fue lento PIBreal PIB real per chpita Tasasde crecimiento anualizadas 1971-1980 1981-1990 2.09 -0.47 1991-2001 2.09 0.01 Fuente:CAlculosde 10s autores apartir de datos del BancoMundial. Engran medida, el lento crecimiento del PIB durante este period0 se considera comdnmente una consecuencia de la alta volatilidad del PIB, generada principalmente por la vulnerabilidad externa y por inestables politicas internas. Una serie de perturbaciones externas -vinculadas a la volatilidad de 10s precios del petrdleo y a las variaciones en 10s flujos de capital- y de desastres naturales, combinada con una pobre gesti6n econbmica, generaron desequilibrios macroecon6micos que tenian una alta probabilidad de tener un impact0 negativo sobre el crecimiento. Sin embargo, como lo explicamos m8s adelante, el pobre desempeiio econ6mico de Ecuador no se debe tinica ni principalmente a la alta volatilidad, sino a una pobre gesti6n econdmica y, especialmente, a1dtbil aumento de la productividad. L a relaci6n entre productividad y crecimiento econ6mico se ha hecho adn m8s importante en 10s Cltimos aiios, desputs de que Ecuador decidiera adoptar el d6lar como la moneda nacional (en 2000), con l o que renunci6 a la opci6n de utilizar l a politica cambiaria para generar aumentos temporales de la competitividad y el crecimiento. Aunque no cabe duda de que la decisidn de dolarizar la econom'a mejor6 el clima de inversibn, tranquiliz6 a 10s posibles inversionistas y, por ende, potencialmente elev6 la capacidad de la econom'a para generar empleo y reducir la pobreza, en el futuro se necesitarh aumentos sostenidos de la productividad para mantener tasas positivas de crecimiento y tasas decrecientes de pobreza. Debido a eso, este informe se enfoca en el crecimiento de la productividad y en su efecto sobre el empleo, el ingreso y, sobre todo, la pobreza. El informe presta especial atencidn a la relacidn entre la pobreza y 10s sectores productivos, tanto desde un punto de vista macroecondmico como desde la perspectiva microecondmica, y tanto en las ireas urbanas como en las rurales. A1 adoptar esa perspectiva, no solamente complementa la anterior Evaluaci6n de la pobreza en Ecuador (Banco 1 Mundial, ~OOOC),que se concentrabaprincipalmente en la pobreza y 10s servicios sociales, sino que proporciona conclusiones importantes sobre la relacidn entre el crecimiento econ6mic0, la productividady la generaci6nde empleo, por una parte, y la reducci6n de la pobreza por otra. Ademis, a1reflexionar sobre la pobreza, el informe se concentra en 10s aspectos monetarios del bienestar, en lugar de 10s no monetarios, pues parecen estar m8s intimamente vinculados con la evoluci6n del PIB y el aumento de la productividad y, por ende, han presentado pocas mejorias en 10s tiltimos aiios: concretamente, aunque en Ecuador 10s indicadores sociales y el acceso a 10s servicios bisicos han mejorado de manera lenta per0 continua desde 1980, la tasa de pobreza basada en el consumo pas6 del 40 a1 45 por ciento entre 1990 y 2001, como l o explicamos m i s adelante, y present6 aumentos mayores en las keas urbanas. Por iiltimo, el informe apela a una gama de fuentes, tanto cuantitativas como cualitativas, y a estudios existentes para recomendar politicas que ayuden a Ecuador y a su gobierno a diseiiar una estrategia eficaz de reduccidn de la pobreza que se base en el crecimiento econdmico y el aumento de laproductividad. Acontecimientos macroecon6micos y pobreza iCdmoseexplicalabajay altamentevola'til tasadecrecimientodelPIB(ydelPIBper chpita) de Ecuador? Elbajocrecimiento delPIBper capita enEcuador nose puede atribuircompletamente a las perturbaciones relacionadas con 10s tCrminos de intercambio... Growth and ExternalShacks 5% 4 990-2001 5 Ea I 13 D -2$6 -3% a 4 B 12 16 20 24 StandardDeviationaf Terms of Trade Fuente: Ciilculosde 10s autores conbaseen datos del Banco Mundial. .. 11 Las perturbaciones externas, medidas como perturbaciones de 10s tCrminos de intercambio, que se han considerado tradicionalmente una de las causas miis importantes de la baja tasa de crecimiento del PIB y de su volatilidad, tuvieron un impacto negativo sobre el crecimiento per0 no puedenexplicar plenamente el pobre desempefio de la econom'a ecuatoriana. Si se le compara con otros paises de la regibn, la exposici6n de Ecuador a la volatilidad relacionada con 10s tkrminos de intercambio fue relativamente moderada, mientras sus tasas anuales de crecimiento del PIB per chpita estuvieron entre las miis bajas de AmCrica Latina. D e la misma manera, las perturbaciones relacionadascon politicas del gobierno -medidas por el de'ficit fiscal y las sorpresas monetarias- que se suelen considerar 10s otros posibles responsables del pobre crecimiento econ6mico del pais, tuvieron un efecto negativo per0 limitado. Los resultados de las simulaciones muestran que el crecimiento del PIB per cipita habria sido superior al nivel actual si 10s dCficits fiscales y las sorpresas inflacionarias se hubieran eliminado entre 1980 y 2001, per0 la tasa promedio de crecimiento anual durante ese periodo todavia habria sido baja, especialmentedurante lade'cadade 10s 90. ...ni alasfluctuaciones delas politicasinternas. Tasa anual real de crecimiento del Tasa anual pronosticada a la que habna PIB per capita crecido el PIB per capita si no hubiera habido dCficits fiscales ni sororesas monetarias I 1981-85 I -0.6 1.4 I 1986-90 -0.4 1.4 1991-95 I 1.2 0.3 Entonces jcuiil es el elemento que falta en este cuadro? La respuesta es: el aumento de la productividad, o mis bien la ausencia de aumento. Entre 1980 y 2002, el comportamiento del PIB ibade la mano con laProductividad Total de 10s Factores(PTF), una medida de laeficiencia econ6mica o productividad que refleja la calidad de 10s insumos y de las instituciones, asi como la calidad de varias politicas econ6micas. La PTF present6 tasas de crecimiento negativas durante ese periodo y con frecuencia compensd 10s aportes positivos de la mano de obra y la acumulaci6n de capital a1crecimiento. Cambios observados Contribucih de 10s Contribucidn de 10s Contribucih de 10s en el PIB per capita cambios en el empleo cambios en la relacidn cambios en la PTF efectivo capital-product0 Tasas anualizadas de crecimiento (%) 1981-1985 -0.5 0.0 1.3 -1.8 1986-1990 -0.2 1.9 0.1 -2.2 1991-1995 1.o 0.7 0.6 -0.3 1996-2000 -1.1 -0.4 1.8 -2.5 2001-2002 2.4 3.9 -0.6 -0.9 ... 111 En consecuencia, las politicas orientadas a preservar la estabilidad con disciplina fiscal y, especialmente, a elevar la productividad econ6mica y la competitividad, tienen el potencial de promover el crecimiento positivo y sostenido. El us0 de medidas fiscales para alcanzar esos objetivos se discute brevemente a continuaci6n1,mientras otros tipos de politicas se examinan en mayor detalle m5s adelante. Paraproteger a lapolitica fiscal de las perturbacionestemporales es necesario, entre otras cosas: 0 Que el ingreso fiscal sea menos dependiente de 10s ingresos petroleros, tanto en tkrminos de 10s niveles como de las fluctuaciones a lo largo del tiempo. Teniendo en mente ese objetivo se puedenconsiderar varias medidas. Enprimer lugar, la base de ingresos no relacionada con petr6leo debe ampliarse, mejorando el cumplimiento en el pago de impuestos y la efectividad de su recaudo. En segundo lugar se deberian modificar las reglas que rigen a1 Fondo de estabilizacih de 10s precios del petr6leo para mejorar su capacidad y hacerlo m5s efectivo. En especial, 10s precios de intervencidn del limite superior de la banda que provocan el desvio de 10s ingresos procedentes del petr6leo hacia el fondo deberian alinearse con las estadisticas histdricas sobre precios del petr6leo y se deberian suministrar pautas claras para el us0de 10s fondos disponibles. 0 Que se reduzcan la preasignacibn y la destinaci6n especifica de 10s gastos para aumentar la flexibilidad en el us0 de 10s recursos existentes y minimizar la necesidad de apelar al gasto discrecional. A1 hacer esto se debe garantizar la proteccih de programas fundamentales, como ciertos programas sociales y enbeneficio de 10s pobres. En tkrminos m5s generales, las politicas fiscales para mejorar la eficiencia en el us0 de 10s recursosdeberian incluir, entre otras: 0 La armonizacih y simplificacidn del sistema tributario. La actual proliferacih de impuestos, que en su mayoria tienen baja capacidad de generaci6n de ingresos, tiene un impact0 negativo sobre la eficiencia tributaria. S i se derogaran algunos de 10s impuestos menores y a1mismo tiempo se simplificaran y fortalecieran 1os.impuestosa las utilidades de las empresas, el impuesto a la renta y el impuesto a las ventas, se reducirian 10s esfuerzos administrativos, se elevaria la transparencia del sistema tributario y se reducirian las distorsiones. 0 La eliminaci6n de 10s subsidios a empresas ptiblicas de varios sectores. Las transferencias a empresas p6blicas no estiin vinculadas a sus indicadores de producci6n o de calidad de servicio, lo que implica una protecci6n artificial de esas empresas frente a las fuerzas de la competencia y desalienta la rendicidn de cuentas y la eficiencia. Esas transferenciasdeberian eliminarse o utilizarse para suministrar incentivos a la rentabilidad, que est& condicionados a1suministro de mejores servicios. La crisis de 1998-1999,la dolarizacidn de 2002 y sus efectos sobre lapobreza La crisis macroecon6mica de 1998-1999, que fue la peor en m6s de dos dkcadas, tuvo efectos devastadores y duraderos, sobre todo entre 10s residentes de las ireas rurales de la Costa, debido a El Nifio, y entre 10s de la clase media urbana, especialmente afectados por el colapso del sistemabancario y financier0 (Banco Mundial, 2000c; Vos, 2002; Halac y Schmukler, 2003). 'Remitimos a1lector a1Repaso del gasto pu'blicode Ecuador, que se est6 preparando actualmente, si busca una discusidn m6s detallada de la implementacidn de esas medidas. i v En el corto plazo, la adopci6n del d6lar como la moneda nacional en respuesta a la crisis contribuy6 a1control de la inflaci6n y provoc6 cambios significativos de 10s precios relativos, a1 igual que una reducci6n del costo de la canasta familiar promedio. El precio de 10s bienes comercializables se redujo respecto al de 10s bienes no comercializables, al igual que el precio de 10s bienes duraderoscomparado con el de 10s bienes perecederos, pues una alta proporcidn de 10s primeros se importa. La disminuci6n del precio de 10s bienes duraderos llev6 a una reduccidn del 16por ciento en el costo de la canastafamiliar del hogar ecuatoriano promedio y, por ende, a un aumento general del bienestar. Elcostorelativo delacanastafamiliar delhogar promedio se redujo despuCsde ladolarizacidn Figure9 RelativeCost of the Consumption Basket (December 1999= 100) .-. , 2000 2001 2002 Fuente: C6lculosde 10s autorescon baseen datos del IPC y de la ECV de 1999del INEC Sin embargo, en vista de que 10s patrones de consumo vm'an en grupos de diferente nivel de ingreso, no todos 10s hogares se beneficiaron por igual de la reducci6n en el costo de la canasta familiar. El costo de la canasta baj6 mucho m6s para 10s hogares que no se consideran pobres que para 10s hogares pobres (un 19 por ciento frente a un 2 por ciento). Esa diferencia se puede atribuir casi por completo a diferencias en la cantidad de recursos que 10s hogares que no se consideranpobres y 10s hogarespobres destinan a1consumo de bienes duraderos (46 por ciento y 20 por ciento del consumo total, respectivamente). V Los cambiosde preciosque generdladolarizacidnhanbeneficiadoen su mayoria a 10s hogaresque no se consideranpobres Todos 10s hogares Hogaresque no son pobres Hogarespobres Cambios porcentualesde precios G C G C G C Alimentos -4.44 -2.55 -3.87 -2.08 -5.46 -4.37 TransporteIComunicaciones 2.31 1.33 2.77 1.49 1.70 1.36 Productosno alimenticios -1.04 -0.60 -1.16 -0.62 -1.02 -0.81 Ropa -0.04 -0.02 -0.05 -0.03 -0.02 -0.02 Bienes duraderos (compras) -2.62 -1.51 -3.62 -1.94 -0.69 -0.55 Agua, gas, electricidad 4.97 2.86 4.13 2.22 7.84 6.28 Vivienda 10.94 6.29 11.15 5.99 7.82 6.26 Bienes duraderos(consumo) -22.31 -24.34 -10.50 Total I 10.08 -16.51 9.36 -19.32 10.18 -2.35 Fuente:Ctilculosde 10s autoresconbaseen datos de laECV de 1999ECV. G: Gasto.C: Consumo. Todavia son inciertos 10s efectos que la dolarizacidn tendri a mediano plazo sobre el crecimiento, el consumo y la pobreza. Ha contribuido a brindarle credibilidad a las politicas econdmicas (por ejemplo, la percepci6n de riesgo del pais se ha reducido), lo que genera condiciones mhs favorables para el crecimiento econ6mico sostenido y para unos mayores niveles de ingreso. Sin embargo, tambi6n ha afectado la capacidaddel gobierno ecuatoriano para implementar politicas econ6micas anticiclicas a la vez que lograba muy poco en el sentido de eliminar la volatilidad del crecimiento. En ese contexto, el crecimiento sostenible dependeri estrechamente de aumentos de la productividad que ayuden a mantener la tasa de cambio real dentro de limites competitivos. como lo explicamos a continuacibn, para que se den esos aumentos se necesitari invertir en capital fisico y humano, asi como tener un entorno econ6mico e institucional mis estable y transparente. La naturaleza, distribucidn y evolucidnde lapobrezaentre 1990y 2001 Caracteristicasde 10s pobres y tendenciasde la pobreza entre 1990y 2001 La tasa nacional de pobreza basada en el consumo era del 45 por ciento en 2001, mientras en 1990 era del 40 por ciento. Durante el mismo period0 el n6mero de personas que viven en la pobreza aument6 de 3.5 a 5.2 millones*. Los aumentos de la pobreza no estaban distribuidos de manera uniforme en todo el territorio nacional. Fueron mayores en las zonas urbanas de la Costa y de la Sierra, en las que la tasa de pobreza aument6 en mis de un 80 por ciento entre 1990 y 2001. En cambio, la pobreza se mantuvo constante en las keas rurales de la Costa y se elev6 en un 15 por ciento en las keas rurales de la Sierra. ~ ~~ ~~~ Las comparaciones entre 1990 y 2001 no incluyen a la regi6nde Oriente debido a la carencia de datos sobre esa &ea en 2001. vi Aumentos de la pobrezaen generaly, en particular, dela pobrezaurbana 1990 2001 Tasa de recuento Entodo elpais 0.410 sin la regi6n Oriente 0.403 0.452 Quito 0.222 0.243 Guayaquil 0.382 0.386 Costa urbana 0.258 0.464 Sierraurbana 0.213 0.467 Costarural 0.505 0.504 Sierrarural 0.528 0.617 Areas urbanas del Oriente 0.192 0.598 En consecuencia, el ndmero de personas que viven en condiciones de pobreza en las Areas urbanas se elev6 de 1.1 millones a 3.5 millones, de tal manera que el ndmero de pobres de las keas urbanas super6 al de las Areas rurales en 2001, lo que se traduce en una urbanizacih de hecho de la pobreza. AI mismo tiempo, las mis altas tasas de pobreza se continuaron viendo en keas rurales, donde viven 10s mis pobres de 10s pobres. Por ultimo, 10s pobres Vivian en hogares m6s grandes, tenian niveles inferiores de educacibn, padeciande niveles mis altos de desempleo y tenian menos acceso a 10s servicios bisicos que 10s que no eran pobres. En las keas urbanas, 10s pobres tendian a estar empleados en el sector informal, mientras en las rurales tendian a estar empleados en el sector agricola. Esos resultados son consistentes con 10s que se discuten en las Evaluaciones de la pobreza anteriores (Banco Mundial, 1997y 2000c) y en otros estudios (SIISE, 2002c y 2002d). L a urbanizacih de lapobreza: causas y consecuencias La urbanizaci6n de la pobreza fue el resultado de: (i) migratorios del campo a la ciudad, flujos (ii)naturalezaparticulardelacrisisde1999,queafect6especialmentea10shogaresdeclase la media urbana, y (iii) 10s cambios en el nivel y lacomposici6n del empleo en las diferentes keas. Entre 1990 y 2001, aproximadamente del 30 a140 por ciento de la poblaci6n ecuatoriana migr6 tanto en el interior del pais como hacia fuera. Los flujos migratorios respondian a la existencia de condiciones de vida relativamente mejores y a las mejores oportunidades econ6micas en las keas urbanas de Ecuador, asi como en otros paises como Espaiia e M i a , 10s destinos mis populares de 10s recientes emigrantes ecuatorianos. En la medida en que a1principio el migrante promedio no podia conseguir un salario igual a1 del residente urban0 promedio, esos desplazamientos contribuyeron a un aumento en la pobreza en las Areas que registraron tasas netas de inmigracih respecto a las Areas que presentaron tasas netas de emigraci6n. AdemBs, vii tstas filtimas tambitn se beneficiaron del voluminoso flujo de remesas internacionales, lo que ayud6 a mitigar 10s efectos de lacrisis y sus consecuencias. El colapso del sistema financier0 y bancario que ocasion6 la crisis de 1999 tuvo un impacto particularmente negativo en las Areas urbanas y, dentro de ellas, entre 10s hogaresde clase media. Mientras 10s grandes inversionistas no resultaron tan afectados por 10s efectos negativos de la crisis, 10s pequeiios y medianos inversionistas (por ejemplo, 10s de la clase media urbana) sufrieron un impacto significativo (Halac y Schumekler, 2003). En vista de que en 2001 apenas habian transcurridos dos aiios desde la crisis, es muy posible que las cifras que se presentanen el informe sobre ese aiio todavia est& de alguna manera contaminadas por sus repercusiones, especialmenteen el cas0 de las keas urbanas. Por iiltimo, 10s cambios en el nivel y la composici6n del empleo tambitn han tenido un impacto sobre la pobreza. Sobre todo, 10s cambios positivos en la participaci6n del empleo agricola guardaron conelacidn con 10s aumentos en la pobreza, mientras 10s cambios positivos en la participaci6n del empleo no agricola de baja y alta productividad estaban vinculados a disminuciones (0 aumentos menores) de lapobreza. Los cambios en la naturaleza y la distribuci6n de la pobreza y de 10s pobres tienen implicaciones importantes para el desarrollo, tanto urban0 como rural. En las ireas urbanas, el rhpido aumento de lapoblaci6n y el ripido aumento de lapobreza plantearh desafios importantes en ttrminos de la generaci6n de empleo y la generaci6n de ingresos, y en ttrminos del suministro de servicios bisicos. Este informe comenta politicas orientadas a promover la creaci6n de empleos (ver pr6ximas secciones) mientras otros estudios en progreso se dedicarh a 10s asuntos de servicios bisicos y vivienda3. Ademis, mientras 10s diferenciales de ingreso y pobreza entre las ireas urbanas y rurales sigan siendo tan altos como son en la actualidad, la gente seguiri estando atraida por las primeras, abandonandolas segundas y elevando la presi6n que existe sobre las ya agobiadaseconom'as de las Areas urbanas. Una alta proporcidn de 10s residentes pobres de las keas rurales depende todavia del sector agricola para sobrevivir y la mayoria de ellos no tiene acceso a la tierra o trabaja en tierras de baja productividad. Este informe examina 10s factores determinantes de la productividad agricola y comenta las politicas destinadas a incrementar el acceso a la tierra de aquellos que carecen de ella (ver pr6ximas secciones). Lapobrezaurbana,lasdinhmicasdelmercadolaboraly la generaci6nde empleo Elempleo constituye la principal, y con frecuencia la dnica fuente de ingreso para la mayoria de las familias que viven en keas urbanas y por eso la mayoria de las veces la carencia de empleo lleva a la pobreza. Los ingresos laborales representanmis del 90 (80) por ciento del gasto total y mis del 75 (80) por ciento del ingreso total de 10s hogares pobres (que no son pobres) de las Areas urbanas de Ecuador. En consecuencia, las politicas para mejorar la capacidad de la Los asuntos de servicios bhsicos y vivienda para 10s pobres se estudiarhn en el Estudio regional de la pobreza urbana, que est6 en preparacibn, en el cual se incluyen datos sobre Ecuador, y en el Proyectopara la reduccidn de la pobreza urbana en Ecuador. viii econom'a urbana para generar empleo e ingreso (salarios) se convierten en las herramientas mhs importantes para reducir la pobreza urbana. Tendenciasdel Mercado laboral y pobreza urbana entre I997 y 2002 Elcomportamiento de 10s mercados laborales urbanos y, por ende, de la pobreza urbana, result6 profundamente afectado por la crisis de 1998-1999 y por la dolarizaci6n de 2000 (Fretes et alia, 2003; Lebn, 2002). Los niveles de empleo y el ingreso laboral real cayeron en picada como resultado de la crisis y s610 regresaron a 10s niveles previos a la crisis en 2002. Esos acontecimientos provocaron un aumento de la pobreza urbana entre 1997 y 1999 y llevaron a 10s hogares pobres de las zonas urbanas a apelar a una serie de estrategias para enfrentar la situaci6n, tales como una mayor tasa de actividad y la migracicin. Las tasas de pobreza se redujeron despuCs de 2000, per0 lo hicieron lentamente, debido a1 hecho de que la generaci6n de empleo, y sobre todo la generacicin de empleo formal, era dCbil. La tasa de empleo permanecid casi constante entre 2000 y 2002, lo que indica que la generacidn de empleo apenas era suficiente para acomodar a1aumento de unpunto porcentual en la tasa de actividad que se observ6. De la misma forma, aunque la participacidn del empleo formal aumentd ligeramente durante ese periodo, en 2002 estabamuy por debajo de su nivel de 1997. ITasade actividad 1 56.8 58.5 60.2 57.5 63.6 58.5 ' Hombres 71.1 71.8 73.2 70.4 74.5 70.3 1 Mujeres 43.3 46.2 48.0 45.2 53.0 46.9 ITasade empleo 90.8 88.5 85.6 91.0 89.1 90.8 1 I Hombres 93.3 92.1 89.7 94.0 93.2 94.7 Mujeres 87.6 84.4 80.7 87.2 84.1 87.0 Tasade desempl D 9.2 11.5 14.4 9.0 10.9 9.2 Hombres 6.6 7.8 10.2 5.9 6.7 5.2 Mujeres 12.4 15.5 19.2 12.7 15.8 12.9 IIngresolaboralpor hora(2000, US$) 1.06 0.72 0.48 0.55 0.70 0.83 I I1 1.08 0.74 0.52 0.59 0.77 0.95 Hombres Mujeres 1.03 0.68 0.44 0.48 0.60 0.64 ix ...a1 igualque lapobrezaurbana 0.50 0.45 0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 1997 1998 1999 2000 2001 2002 Fuente: CAlculosde 10s autores a partir de laEEDS, 1997-2002 Las tasas de pobreza miis altas se presentaron entre 10s hogares cuyo jefe de familia estaba desempleado o empleado en el sector informal, lo que hace de la creaci6n de empleo (formal) un requisito indispensable para la reducci6n de la pobreza en las ireas urbanas. Productividad laboral, generacidn de empleoy pobreza urbana: el papel de la tecnologia, de las destrezasy de las instituciones L a generaci6n de empleo, y especialmente la generaci6n de empleo en el sector formal, est5 estrechamente vinculada con 10s aumentos en la productividad laboral. Esos aumentos estfin en funci6n de la cantidad y calidad de 10s insumos de producci6n, asi como del marco institucional en que operan las empresas. De esos factores depender8,el grado a1 cual la econom'a urbana podra promover la generaci6n de empleo, mientras el grado a1cual 10s pobres se beneficiarin de este proceso dependera de quC tan apropiadas Sean sus destrezas respecto a las que demanden las empresas. La exposicidn a la competencia intemacional y el acceso a mejores tecnologias guardan correlacih con 10s mayores niveles de productividad laboral y, en consecuencia, con 10s mis elevados niveles de empleo. Las empresasexportadoras y aquellas que tienen acceso a la tecnologia extranjera son un 30 por ciento mis productivas que sus contrapartes, mientras un aumento del 10 por ciento en la productividad laboral genera unaumento del 1por ciento en el empleo. De la misma manera, un aumento del 10por ciento en la participacidn de 10s trabajadores educados se traduce en un aumento del 5 por ciento en la productividad, lo que ha conducido a 10s aumentos sostenidos en la demandarelativa de mis trabajadores educados de 10s tiltimos aiios. X Lageneracihde empleo guardaunacorrelacihpositivacon laproductividadlaboral Variable dependiente Productividadlaboral Generaci6nde empleo (US$/trabajador) (Neta) Productividadlaboral 0.17 ** (0.08) Participacidnde la fuerzade trabajo con 0.16 ** 0.06 educaci6nsecundariao superior (0.07) (0.08) Acceso ala tecnologfaextranjera 9.19 * 12.59 ** (4.87) (6.34) Exportaciones 8.99 * 9.62 (4.74) (6.16) Acceso a1crCdito 0.95 (1.30) Variablesdic6tomas del tamafio de laempresa S i Si Nitmerode observaciones 250 245 ~ ~~~ Fuente: Cficulos de 10s autores con base en el Sondeo del clima de inversi6n - Ecuador, Banco Mundial(2003). ** (*I Significativamente diferente a cero a1nivel del 5 (10) por ciento. Entre las medidas para elevar la productividad laboral y, por ende, la generacidn de empleo, deberian incluirse: 0 La ratificaci6n de tratados de libre comercio y la racionalizacibn y reducci6n de las barreras arancelarias y no arancelarias. Estas medidas deberian contribuir a la eliminaci6n del sesgo existente en contra de las exportaciones, que tiene que ver con afios de aplicaci6n de politicas de sustituci6n de importaciones. 0 Simplificaci6n de 10s convenios de concesidn de licencias y promoci6n de la inversi6n extranjera directa. Ecuador podria beneficiarse de manera significativa de las tecnologias existentes si estableciera incentivos apropiados para las licencias extranjeras y la inversi6n extranjera directa, junto con medidas efectivas para proteger 10s derechos de propiedad intelectual y las patentes. 0 Inversiones en educaci6n y en capacitacidn. Los niveles de educaci6n y las tasas de escolarizaci6n de Ecuador son bajos para el nivel de desarrollo del pais. Ecuador carece de una base amplia de trabajadores con educaci6n secundaria, necesaria para adoptar y adaptar de manera eficiente las tecnologias existentes, y ese dCficit no disminuiri en 10s pr6ximos afios a menos que se destinen mayores recursos a las escuelas secundarias. Ademis, el SECAP, el instituto p6blico de capacitacibn, necesita una reforma radical. El cum'culo que ofrece en la actualidad es obsoleto y, en consecuencia, 10s recursos del instituto est6n siendo subutilizados. La mayor competencia en el suministro de capacitaci6n podrfa ayudar a generar 10s incentivos necesarios para el cambio a la vez que amplia las alternativas de capacitaci6n a las que pueden apelar las empresas. Los pobres tienen menos educaci6n que 10s que no son pobres y tienden a estar empleados en firmas informales pequefias con escaso acceso a la tecnologia. En consecuencia, para que las politicas que describimos antes tengan Cxito en la reducci6n de la pobreza, deben estar acompafiadas de medidas explicitas a favor de 10s pobres, tales como: 0 Promoci6n de las vinculaciones entre las empresaspequefias y las grandes. Es mis probable que las empresas grandes se beneficien inicialmente del mayor acceso a 10s mercados extranjeros y a la tecnologia que las empresas pequefias, per0 tambiCn es mis probable que tengan menos flexibilidad para responder ripidamente a 10s cambios en las condiciones del mercado. Entonces, xi la promocicin de vinculos productivos entre empresas grandes y pequeiias podria contribuir a distribuir 10s beneficios relacionados con esos acontecimientos y transferir tecnologia a las empresas pequeiias, y a la vez que les brindm'a a las empresas grandes un mayor grado de flexibilidad. Creaci6n de centros de servicio para pequeiiasempresas. La adopcicin y adaptaci6n de tecnologia suele ser un proceso costoso. Los centros de servicio, o incubadoras de pequeiias empresas, penniten que 10s pequeiios negocios compartan el costo de una determinada tecnologia o servicio que de otro modo les resultm'a inaccesible y por ende les permiten aumentar su productividad Incentivos para la capacitacicin de trabajadores del sector informal. Todas las empresas de Ecuador deben aportar el 0.5 por ciento de su ncimina a1 SECAP. Sin embargo, las pequeiias empresas del sector informal por lo general se abstienen de hacerlo y, en consecuencia, no tienen acceso a 10s servicios del instituto. Se deberian promover programas especiales de capacitacicin para esas empresas y explorar la posibilidad de que la red de CBmaras de Comercio u otras asociacionesde empleadores 10s patrocinen. Lamayoriade lasempresas quisieracontratar m k empleados permanentes ... 70 All Small Medium Large Fuente: Sondeo sobre el clima de inversi6n-Ecuador (2002). xii Todas Pequefias Medianas Grandes (0 a 10) (11 a99) (100 +I Motivo para no aumentar (% de las empresas en el gntpo) Costosde despido 38.7 47.1 39.5 25 Costosno salariales 17.8 17.6 13.5 43.7 Trfimites anteel Ministeriode Trabajo 0.8 0.0 1.o 1.o 0.0 Sindicatos 1.5 0.0 6.2 Expectativasde ventas 41.1 35.3 44.7 25 Los aumentos en la productividad laboral podn'an no traducirse en empleos e ingresos adicionales cuando se presenten limitantes institucionales o incertidumbre. De hecho, 10s costos de despido y 10s costos laborales no salariales parecen ser limitantes importantes para la generacidn de empleo (permanente). L a creacidnde empleo permanentereal entre las empresas cobijadas por un sondeo de 2002 erade 0.1por ciento, mientras las mismasempresas informabanque deseabanunaumentodel 8 por ciento. La diferenciaentreesas cifras obedece principalmente a 10s costos de despido y 10s costos laborales no salariales. Desde un punto de vista m6s general, el escaso y costoso crkdito, la pobre infraestructura y la incertidumbre sobre el ambiente econdmico e institucional parecen ser las principales limitantes para la expansidn empresarial. Enespecial, el 40 por ciento de las empresas encuestadas declar6 que tenia dificultades para encontrar mano de obra calificada -un hecho que enfatiza la importancia de la inversi6n en educaci6n- y mhs del 60 por ciento se vi0 obligado a reconsiderar sus planes de expansi6n debido a la carencia de crkdito, al pobre suministro de servicios p6blicos y a la incertidumbre econ6mica e institucional. Entre las medidasorientadas a mitigar algunas de esas limitantes deberfan estar: Una reforma laboral para reducir 10s costos relacionados con la contrataci6n permanente. Durante 10s filtimos aiios, 10s costos laborales relativamente altos relacionados con la contratach permanentehan llevado al us0 abrumador de 10s contratos temporales por parte de 10s empleadores ecuatorianos y, en consecuencia, a un grad0 creciente de segmentacidn del mercado laboral. La legislaci6n laboral actual debe ser modificada para que esas dos cifras contractuales se acerquen. Ademhs, en vista de que la altemativa de hacer m6s estricta la regulacidn de 10s contratos temporales podria afectar desproporcionadamente a ciertos grupos vulnerables y dificiles de emplear, se deberia considerar la creaci6n de modalidades especiales de contrataci6n, tales como la vinculaci6n de aprendices o 10s contratos de re- vinculaci6n. Mejorar el acceso a1 crkdito para las pequeiias y medianas empresas. L a disponibilidad de crkdito para empresas medianas, y sobre todo para las pequeiias, es baja en Ecuador, lo que refleja la actual debilidad del sistema bancario del pais y la escasa capacidad de ahorro. El mayor acceso al crkdito de podria lograr por medio de la creaci6n de uniones de crkdito patrocinadas por 10s gremios (asociaciones de industriales) o las Charas de Comercio y de la promoci6n de compaiiias de capital de riesgo y de 10s vinculos entre las empresas grandes y las pequeiias. ... Xlll Lasempresaspercibennumerosaslimitantesparasu operaci6ndiariay su expansi6nfutura (Porcentaje de empresasque consideraque cada factor es unalimitante) --I I Finanung avaiiab,im/ . Financing.cost Rnancing- Telecommunlcalonr ElsctWiY Transpitallon Accessto land /.NI ~JSWII#Medium#Large/ /.All OSmall BMedum #Large] Recursos financieros Infraestructura Ambiente econ6mico Fuente: Cficulos de 10sautores con baseen 10s datosdel Sondeo sobre el clima de inversi6n-Ecuador, BancoMundial(2003). Pobrezarural, productividadagricolay distribuci6nde la tierra El40 por ciento de la poblacidn de Ecuador vive en las keas rurales y el 60 por ciento de ellos es pobre. Los pobres de las keas rurales tienden a concentrarse en el sector agricola, tienen un acceso limitado o nulo a la tierra y trabajan tierras de baja productividad. En consecuencia, en Ecuador, como en otros paises, el ingreso de 10s pobres de las keas rurales suele estar vinculado a la producci6n agricola, por lo cud las politicas orientadas a elevar la productividad agricola y el acceso a la tierra tienen el potencial de ser herramientas efectivas para reducir la pobreza en las iireasrurales. xiv Productividad agricola y pobreza rural La mayor productividad agricola, medida como el valor en d6lares de la producci6n agricola por hecthrea, guarda correlaci6n con un mayor ingreso y un menor nivel de pobreza, especialmente para aquellos hogares que obtienen la mayoria de sus ingresos en el sector ap'cola. Concretamente, un aumento del 1por ciento en la producci6n agricola se traduce en un aumento de entre el 0.16 y el 0.30 por ciento en el consumo per cipita entre 10s hogares cuyo jefe de familia es empleado independiente del sector agricola, o un aumento de casi uno a uno para 10s hogares de 4 o 5 miembros, que es el tamaiio del hogar rural promedio en Ecuador. TambiCn se pueden observar aumentos positivos para 10s trabajadores del campo, para quienes un aumento del 1por ciento en la productividad agricola se traduce en un aumento de salario de entre el 0.10 y el 0.30 por ciento. Por lo general, 10s trabajadores agricolas y otros que est& empleados en el sector agricola se encuentranentre 10s habitantes m6s pobres de las heas rurales, y por ello las politicas orientadas a elevar la productividad agricola, especialmente en las pequeiias fincas en las que se encuentra el mis alto nivel de pobreza, son instrumentos esencialespara la reducir lapobreza rural. Los niveles m6s altos de productividad agricola se encuentran en las keas que rodean a Quito y en el sur de la regi6n de la Costa y 10s mis bajos se encuentran en la regi6n de Oriente. Sin embargo, existen grandes diferencias en la productividadde cant6n a cant6n y de finca a finca dentro de cada cantbn, lo que refleja las diferencias en la cantidad y calidad de 10s insumos de produccih, asi como las diferencias en la eficiencia tecnol6gica. Laproductividadde latierraes altaen10s cantonesque rodean aGuayaquiiy Quito Fuente: Cilculos de 10s autores con baseen datos del Tercer Censo Agropecuario de 2001. xv Elus0 de 10s insumos y la eficiencia vm'an dependiendodel tamaiio de lafinca, a1igual que 10s rendimientos de diferentes tipos de insumos. Las fincas mis pequeiias (de entre 0 y 1hectkea) son por lo general mis productivas que las mis grandes (de mis de 10hectkeas). Elrendimiento de la mano de obra es bajo en las fincas pequeiias (una elasticidad de 0.05) y alto en las fincas grandes (una elasticidad de 0.45), mientras el rendimiento del capital es bajo (una elasticidad de 0.07) y el de la irrigaci6n y, sobre todo, el del us0 de fertilizantes y pesticidas, son altos (una elasticidad de entre 0.40 y 0.70), independientementedel tamafio de la finca. Los rendimientosrelativosde latierra, el capitaly lamanode obra varian s e g h el tamaiiode lafinca. Fincas de pequeiia escala Fincas de escala mediana Firmasde gran escala Estimados de lafuncidn de produccidn Cobb-Douglas Mano de obra 0.05 0.17 0.45 Capital 0.08 0.07 0.08 Tierras no imgadas 0.14 0.04 0.08 Tierras irrigadas 0.14 0.00 0.08 Us0de 10s insumosentierras no irrigadas 0.72 0.77 0.37 Us0de 10s insumos entierras imgadas 0.40 0.70 0.39 Escala (tierras imgadas) 0.99 1.04 0.99 Escala (tierras no imgadas) 0.67 0.94 1.01 Fuente: Cdculos de 10s autores con baseen datos del Tercero Censo Agropecuario de 2001. Una serie de politicas orientadas a elevar la eficiencia agricola tambiCn se estudiaron y se usaron tkcnicas de simulaci6n para evaluar el impact0 potencial de cada una de ellas. Entre las politicas que se estudiaron estaban el acceso a1 crCdito, a la educacidn formal y a la educaci6n tCcnica agricola, a 10s mercados y a 10s intermediarios, entre otras. El acceso a1crkdito es la politicamis importante para elevar laproductividad de 10s agricultores pobres de pequefia escala, entre 10s que se encuentra la mayoria de 10s pobres de las zonas rurales. La educaci6n te'cnica agricola y el acceso a 10s fertilizantes y a 10s pesticidas tambikn son importantes, aunque estos podrian tener consecuencias ambientales negativas que se deben tener en cuenta. Lo sorprendentees que la distancia a 10s mercados nojuega unpapel importante una vez que se tienen en cuentaotros factores que describen el nivel de apego de un determinado agricultor (por ejemplo, la venta de su producci6n al mercado versus el autoconsumo o el us0 de intermediarios). xvi Elaccesoa1crbditoy alaeducacidnagricolatienensuma or impacto sobre laproductividadde lasfincas pequeiias H % de Idas la: Pesticidas Provincia fincas 'aloresiniciale; y fertilizantes Extensi6n Mercados Crkdito Educaci6n agr. Aumento relacionadocon cada reforma (en US$/hecthrea) izuay 43.0 314.4 16.2 13.7 21.5 44.8 28.2 3olivar 15.1 170.0 12.8 7.2 8.6 24.6 15.4 Sanar 37.3 292.5 21.2 13.2 16.4 41.6 26.3 Sarchi 15.4 469.2 10.7 18.0 11.9 49.3 37.3 3himborazo 37.0 423.4 24.6 13.6 9.8 47.1 28.3 Sotopaxi 37.2 278.0 7.9 12.9 11.5 42.6 26.3 31Or0 13.6 427.7 22.2 16.1 11.5 52.2 31.2 3smeraldas 2.3 140.9 11.6 6.7 6.3 22.1 13.3 3uayas 13.4 386.7 10.9 13.1 9.2 32.5 24.5 lmbabura 49.0 218.0 15.1 9.6 14.4 30.1 18.9 ,oja 15.3 264.2 25.6 10.6 10.6 34.2 21.4 h s Rios 10.6 231.1 13.4 8.6 3.7 25.4 16.3 Manabi 16.8 204.2 10.4 9.5 9.8 29.2 17.9 MoronaSantiago 4.4 297.7 28.9 11.8 13.9 35.4 21.8 Vapo-Orellana 4.0 735.0 6.0 19.9 45.1 94.5 49.7 ?astaza 9.0 149.8 13.9 7.9 11.8 23.6 14.2 ?ichincha 39.4 313.9 19.0 14.6 24.3 46.7 29.0 (TIT) Annualized growth rates (%) 1970-2002 1.7 0.5 0.2 1.o 1975-1980 3.3 -0.1 1.1 2.3 1981-1985 -0.5 0.0 1.3 -1.8 1986-1990 -0.2 1.9 0.1 -2.2 1991-1995 1.o 0.7 0.6 -0.3 1996-2000 -1.1 -0.4 1.8 -2.5 2001-2002 2.4 3.9 -0.6 -0.9 Source: Authors calculationsbasedon data from the World Bank. l3More formally, we decompose GDPper capita growth between period t and t+s as follows: S S '"I S S where GDP per capita (Y/N) growth is a function of (i) changes in TFP (A), (ii) changes in the capital-output ratio (KN),and (iii) changes in hours worked per working-age person (L/N). l4In this sense Ecuador is no different from other countries in the region, such as Chile and Mexico, where a similar relationship betweenGDP per capita and economic efficiency has beendocumentedelsewhere (Bergoeing at alia, 2002). 4 Determinantsof economic volatility and poor growth 1.16 Our measure of TFP captures changes in the quality of inputs, as well as changes in (the quality of) macroeconomic and microeconomic policies and other temporary shocks, such as external shocks. In this section we investigate the commonly held belief that it i s the last two (domestic and external shocks) that explain TFP and economic volatility in Ecuador (Vos et alia, 1999; Solimano, 2003). Box 1.1: DataSources for the MainMacroeconomicand Policy Variables Variable Definition and construction A Source RealGDP percapita Ratioof totalGDPto total population IMF Openness (% of GDP) Ratioof exportsand importsto GDP IMF Government consumption(% GDP) Ratioof government consumptionto GDP IMF Inflationrate (Log) changesinCPI IMF Terms-of-tradegrowth Logdifference of the termsof trade ECLAC Terms-of-tradevolatility Standarddeviationof the terms of trade ECLAC A GDP is measuredin2002 USdollar value. Terms of trade are definedas customary. 1.17 Countries that are highly dependent on primary-goods exports, such as Ecuador, are likely to be exposed to external shocks generated by international price volatility. In addition, lack o f consistency indomestic policies or highvolatility indomestic policy outcomes can hamper economic efficiency and, hence, induce changes in TFP. A simple plot of the cyclical component of TFE' against time series data on the terms of trade and on the two main indicators of domestic economic policy, the fiscal deficit and inflationary surprises, i s suggestive of some degree of contemporaneous correlation between TFE' and these variables (Figure 1.2). In particular, the correlation between the TFP and terms of trade is 0.16, while the correlation between TFP and inflationary surprises and fiscal deficits i s 0.37 and0.10 respectively. Figure 1.2: There is evidence of contemporaneous correlation between TFP and external, fiscal and monetary shocks 312 -2, 2 1 1 0 0 1 1 2 2 3 1 1980 1985 1990 1995 2wo -TFP 5 1980 1985 1990 1995 20W (rightscale) -Terms of Trade (leftscale) 1980 1985 1990 1995 2000 -inflationarySurprises(invertedleft 1 -Deb'endedTFP (leftscale)-FiscalDeficit(rightscaie) -DetrendedTFP(rightscale) scale) I Source: Authors' calculationsbasedon datafrom the World Bankdata. Cyclical componentof TFP calculatedas the logdistanceto the TFP long-runtrendobtainedusingthe Hodrick andPrescottfilter (1997). 5 1.18 Contemporaneous correlation, however, i s not equivalent to causation. The fact that fiscal deficits are high at a time when TF'P i s negative does not necessarily imply that the first i s causing the second (i.e. low productivity i s a consequence of high deficits). It could very well be the reverse, with the government deciding to increase expenditure, in an attempt to stimulated the economy, in response to low (or negative) growth rates associated with negative TFP. In order to distinguish between these two possible chains of events, we perform a more formal test that allows us to determine the direction of ~ausality'~.The results of such a test show that indeed shocks to terms of trade and fiscal policy cause TFP fluctuations, while the converse it not true. In contrast, there i s no causality from inflation surprises to TFP shocks, suggesting that monetary shocks play a minor role (perhaps, only as a source of financing of the fiscal deficits). Table 1.2. TFPvolatility is due to fiscaldeficits and, especially, terms-of-trade shocks Nullhypothesis I F-Statistic I Probability I External Shocks TOT shocks do not Granger cause TFP fluctuations 3.25" 0.08 TFP fluctuations do not Granger cause TOT shocks 0.47 0.50 Domestic Shocks Inflation shocks do not Granger cause TFP fluctuations 0.12 0.73 TFP fluctuations do not Granger causeInflation shocks 2.06 0.16 Fiscal deficits do not Granger cause TFP fluctuations 0.33 0.57 TFP fluctuations do not Granger causeFiscal deficits 4.19* 0.05 1.19 Do external and fiscal policy shocks then fully explain TFP (and GDP per capita) volatility? Although useful to determine the existence of a causal relationship, these tests do not provide information on the potential magnitude of this effect. Inother words, they do not provide information as to how much of the observed TFP and GDP variation can be explained by external and fiscal shocks. In order to explore this question, we focus first on the role of external shocks and compare the experience of Ecuador to that of other countries in the region during the 1990s, asking the question: where external shocks in Ecuador large enough, relative to those received by other countries, to fully justified Ecuador's poor growth performance inthe 1990s? 1.20 To answer we plot the standarddeviation of the terms of trade against GDPper capita growth for 16 Latin American countries in 1990-2001. We find that both variables are negatively correlated, that is, more volatile terms of trade are associated with lower growth. There is, however, significant variation across countries (Figure 1.3). Inparticular, Ecuador exhibits the worst growth performance inthe region despite terms-of-trade volatility that is below the Latin American median. To illustrate this point, let us compare Ecuador and Argentina - two countries with, at most, mixed macroeconomic records. Terms-of-trade volatility i s similar for both, but Argentina's annualized GDPper capita growth rate for the decade is 2.5 percent, compared to -0.5 inEcuador. This implies that, while terms-of-trade shocks negatively affect TFP and economic growth in Ecuador during l5We use Granger-causality tests. These tests are of the following generic form: variable x does not Granger-cause variable y if the 2 (L)coefficients of the regression y, = q(L)y,-, -I-8(L)Xt-, -I-E, are statistically insignificant. The tests are performed on annual data (the only available for long run analysis). Hence, higher frequency causality is inevitably omitted from the analysis. 6 1990-2001, their magnitude was not large enough to fully account for negative growth rates, and suggests the existence of mechanisms and channels that contribute to amplify these shocks' negative effects. We address this question next. Figure 1.3: Economic volatility and poor growth inEcuador cannot be fully explained by shocksto terms of trade Gruwth and External Shocks iI19u-2oa1 Source: Authors' calculationsbasedondata from the World Bank. 1.21 Given our discussion so far, it i s likely that unstable fiscal and monetary policies play an important role as amplifiers of the effects of external shocks. To explore this idea we again turn to the experience of other Latin American countries and this time ask the questions: What i s the size of the effect of external shocks on GDP per capita growth? Does the size of this effect change once we take into account fiscal and monetary policies? 1.22 We use data for the 16 countries represented on Figure 1.3 to test for the impact of external and domestic shocks on economic growth during 1970-2001. Following the endogenous growth 1iteraturel6,we model changes in GDPper capita as a function of (i) the (log) level of GDPper capita in 1970, to control for the fact that initially poorer economies are expected to grow relatively faster"; (ii)setofcontemporaneousexplanatoryvariables,includingtermsoftradeandpolicyindicators; a (iii)setoftime-invariant,country-specificeffectstocontrolfordifferencesacrosscountriesthatdo a not vary over time, such as distance to the U S or access to the sea; (iv) a time-specific effect, to 16See Barro and Sala-i-Martin (1995) for a summary of this extensive literature. l7The "conditional convergence" hypothesis poses that, ceteris paribus, poor countries should grow faster than rich ones becauseof decreasingretums to factors inproduction. 7 control for changes in aggregatemacroeconomic conditions, such as Asian financial crisis of the end of the 1990s; and (v) a random disturbance18. 1.23 We estimate two different models, excluding and including the policy indicators and report the coefficients inTable 1.3. Model 1shows that negative terms-of-trade shocks slow down GDP per capita growthlg. In particular, a one-standard-deviation increase in terms-of-trade volatility would decrease GDP per capita growth by 0.1 percentage points. Incontrast, more favorable terms-of-trade and increasesinthe investment rate are bothpositively correlated with GDPper capita growth. 1.24 However, the effect of terms-of-trade volatility disappears (Le. becomes smaller and insignificant) once domestic policy indicators are included in the regression, suggesting that its negative impact i s channeled through (and amplified by) the vulnerability of domestic policies. For instance, an increase in the relative price of imports (external shock), caused by the depreciation of the sucre, would have a negative impact on industries usingimports as inputs and, as a consequence, would lead to lower growth. In addition, this increase would also translate into higher (unexpected) inflation and, thus, into lower growth, so that the negative effect of inflation would then compound that of the external shock''. Fiscal Fiscal Initial Terms of Trade Invest. revenue surplus GDP Growth Volatility Openness (% GDP) (% GDP) (% GDP) Inflation RZ Model 1 -0.052 0.051 -0.001 0.169 (-4.64) (2.64) (-2.56) (4.46) 0.64 Model 2 -0.029 0.075 -0.0002 0.001 0.130 -0.001 0.172 -0.113 (-3.27) (2.51) (-0.59) (0.817) (2.63) (-3.54) (4.85) (-7.14) 0.79 Datain 5-yearaverages for 1970-2001. Datafor Argentina, Bolivia, Brazil, Colombia,Costa Rica, Chile, DominicanRepublic,Ecuador, El Salvador,Guatemala, Honduras, Mexico, Nicaragua, Peru, Uruguay,andVenezuela. All regressionsincludefixed countryandtime effects. Robuststandarderrors are obtainedusingWhite's matrix.T-statisticsin parenthesis. The effect of economic volatility on GDPper capita growth 1.25 Given that the effect of volatility in fiscal and monetary policy on GDP per capita growth appears to be negative, the question then arises as to what would GDPper capita growth have been in Ecuador during this period had fiscal and monetary policy been more stable over time. We try to answer based on Model 2 above. In particular, we use the coefficients generated by the model to More formally, we estimate the following model: where z are the contemporaneous explanatory variables, p are the time-invariant country effects, h are the time effects, and E is the error term. 19This i s consistent with previous studies such as Easterly et alia (1993). 2oAll other coefficients exhibit the expected signs and statistical significance, so that overall the results are broadly consistent with the existing empirical literature on endogenous growth. 8 predict GDP per capita growth under the assumption of stable fiscal and monetary, defined as zero fiscal deficit and constant inflation of 3 percent (i.e. no inflationary surprises), while keeping all other variables at their real values. We can then compared predicted and actual GDP per capita growth and attribute the difference between the two to fiscal and monetary policy volatility. 1.26 The comparison of both series shows that predicted GDP growth i s (i) overall higher (on average, predicted GDP per capita i s 7 percent - or US$90 - higher than actual GDPper capita), and (ii)significantly less volatile than actual GDPper capita growth (Table 1.4). Since the only difference between both series i s the behavior of fiscal and monetary policies, we can conclude that poor GDPgrowth duringthe periodresulted partly from domestic policy instability. Table 1.4: Fluctuations indomestic policy are the maincause of poor GDPper capita growth and GDPper capita volatility Actual GDP per capita growth PredictedGDPper capita growth 1975-80 3.4 1.5 1981-85 -0.6 1.4 1986-90 -0.4 1.4 I1991-95 1.2 0.3 1996-02 -0.1 0.6 1.27 Let us return at this point to the original question that motivates the analysis thus far: what explains poor GDP per capita growth and GDP per capita volatility in Ecuador during 1970-2002? The different pieces of evidence we have discussed and pulled together throughout the last pages provide us with the following answer: TFP and GDP short-term fluctuations are mainly driven by terms-of-trade and fiscal policy shocks, while below-average GDP growth i s driven by low TFP growth, and poor fiscal and monetary policy. Growth, volatility and poverty 1.28 Unfortunately, such a disappointing growth record i s unlikely to have contributed much to poverty reduction during the period. What has then been the effect of poor growth and economic instability on poverty? What would poverty levels have been had the economy behaved in a more stable manner? The most logical way to answer these questions would be to correlate changes in GDP per capita growth and poverty over time to measure the effect of one on the other. Unfortunately lack of data on poverty rates does not allow us to follow this direct approach. Instead, we analyze the effect of macroeconomic performance on labor markets, income and consumption, and government finances, and then speculate about the connection between these different outcomes and poverty. 1.29 The choice of these three variables is motivatedby basic economic intuition. First, we would expect sustained economic growth to increase the demand for capital and labor, thus increasing employment and the returns to both inputs (capital returns and real wages). As a consequence of these increases, particularly employment and wage increases, permanent income and consumption levels should also rise. In fact, the elasticity of employment with respect to GDP i s 0.3, and that o f consumption i s 0.9 - that is, a Ipercent increase in GDP leads to a 0.3 percent increase in employment and a 0.9 percent increase in consumption. Second, an expanding economy allows the government to collect more revenue, which could then be devoted to poverty alleviation programs. 9 Box 1.2: Economic Crisis, Dollarization and Competitiveness: The Evolution of the RealExchange Rate, 1990-2002 The 1999 crisis of the balance of payments was the end result of fiscal mismanagement and continuous appreciation of the real exchange rate (RER), which hampered the capacity o f Ecuadorian exporters to compete internationally. Between 1992 and 1998,the RER -measuredwith respect to a US-based consumption basket- appreciated by around 30 percent. The pace of the appreciation accelerated during 1998/99,when the RER increasedby an additional 35 percent, fuelingspeculation and eventually leading to a financial crisis. Control of the crisis was primarily achieved with the dollarization scheme implemented in early 2000, when the Central Bank chose a largely devaluated parity of $25.000 to the US$ for the currency conversion. This devaluated parity was thought to provide a cushion for exporters in the event of a slow convergence in the price of non-traded goods to international inflation levels. Despite these precautions, however, a slower-than-expected process of inflation convergence has caused the RER to appreciate again, and this trend shows no sign of reversal. First inflation rates for 2003 and 2004 are expected to be above the US inflation rate, at 7 and 5 percent respectively. Second plans to allow for increases inthe prices of most utilities, currently kept at artificially low levels, would undoubtedly impactexporters. Figure B1.1.2: Real Exchange Rate (up=appreciatton) I 19921 1996 1 2000 1 To the extent that the recent appreciation i s misaligned with economic fundamentals, it represents a worrisome trend that should be corrected. Ina dollarized context, however, a significant (although temporary) reduction in the relative price o f exports can no longer be achieved with a nominal devaluation as used to be customary, but rather it needs to be based on a combination of real wage declines -a disheartening possibility for an economy with highpoverty levels- and productivity increases. On the other hand, there are reasons to believe that this appreciation, although undesired, may not'be as harmful as feared. First, a large fraction of the recent appreciation corresponds to a return towards pre-crisis levels. Second, measuring RER appreciation relative to the US$ neglects the importance of the other hard currencies (such as the Euro or the yen) in which Ecuador trades with the rest of the world, and overestimates the degree of appreciation given the relatively weakness of the US$ dollar against these currencies. Finally, production costs and competitiveness depend on a number of factors that are poorly represented in the RER (i.e. cost and availability of credit, distorting taxes and bureaucratic barriers, etc.) ~~ ~ 1.30 In order to measure the effect of GDP growth and GDP volatility on employment and consumption, we follow the same approach we used in analyzing the effect of domestic policy volatility on growth. Inparticular we ask ourselves what the levels of employment and consumption 10 would have been had GDP per capita exhibited a stable behavior - Le. GDP per capita growth under stable domestic policy. To answer this question we use the full-stabilization GDP per capita series presented in Table 1.4 to predict what the behavior of employment, unemployment, and private consumption would have been under that scenarioz1.Although this i s a rather simplistic approach, it proves useful as a benchmark for the cost of volatility and low growth. 1.31 The results from this exercise are striking. Predicted unemployment levels are between 20 and 50 percent lower than the actual figures depending on the period. Similarly predicted private consumption levels are significantly higher than actual private consumption, especially towards the end of the period (Table 1.5). 1.32 Although we cannot determine the exact distribution of extra employment and consumption across individuals, we would expect some of the increase to benefit poor households and therefore decrease poverty. In particular, to the extent that the poor are unemployed more often than the non- poor, as shown in Chapter 3, it i s likely that they would benefit relatively more from these developments than other groups. Actual Predicted Full stabilization Growth in Growth in GDP per Unempl. Private GDP per Unempl. Private capita rate consumption capita rate consumption 1975-80 3.4% 6.0% 6,019 1.5% 10.1% 5,640 1981-85 -0.6% 8.0% 7,532 1.4% 7.4% 7,062 1986-90 -0.4% 7.8% 8,189 1.4% 3.4% 8,324 1991-95 1.2% 8.3% 8,367 0.3% 5.1% 9,548 1996-02 -0.1% 11.4% 10,343 0.6% 6.5% 11,187 1.33 Finally macroeconomic volatility often causes instability infiscal revenues and, as a result, in the provision and administration of public services and programsz2.Infact, fiscal revenues and social expenditures, measured as a fraction of GDP, fluctuate significantly over time, and so do their coefficients of variation - a measure of volatility (Table 1.6). Although exposure to risk could be reduced usingdifferent types of financial instruments (Le. stabilization funds, etc.), such instruments are not available, or are ineffective, inEcuador. 1.34 Insum, we have argued inthis section that Ecuador's economic performance duringthe last 3 decades and, in particular, during the 1990s was at best poor, due mainly to poor TFP growth, and external shocks and domestic policy instability. Low growth and excess economic volatility in turn had a negative impact on both employment and private consumption levels and, as a result on poverty. Similarly, volatility in fiscal revenues induced by macroeconomic instability led to pro- cyclical social expenditure, which compromised the effectiveness of such spending and its capacity to helpthose who neededit most. 21The behavior of employment and consumption is modeled using econometric techniques. Inparticular, we construct a long-run labor demand model to predict employment and unemployment, and an error-correction model to predict consumption A detailed description of the labor demand model usedfor estimation and its results is provided in Annex 1. The error-correction model is basedon the permanentincome hypothesis following Campbell and Mankiw (1989). Results are not reported. 22See chapter5 for a moredetailed discussionon social expenditures. 11 Table 1.6: Fiscalrevenuesandsocialexpenditurefluctuatesignificantlyover time.. . Average level (% of GDP) Coefficient of variation* Fiscalrevenue Social expenditure Fiscal revenue Social expenditure 1970-1979 16.0 n.a. 8.8 n.a. 1980-1989 11.8 n.a. 13.2 n.a. 1990-1994 13.3 4.0 6.3 9.3 1995-1998 15.6 3.6 4.5 5.8 1999-2002 19.5 3.9 18.3 11.3 1.35 As a result, policies aimed at preserving domestic policy stability with fiscal discipline and, especially, at increasing economic productivity and competitiveness hold promise in promoting positive, sustained growth accompanied by declining poverty rates. We briefly describe below the use of fiscal measures to achieve these goals, leaving the discussion of other types of policies for subsequentchapters. Promotingstable GDPgrowth through fiscal discipline 1.36 Shielding fiscal policy against temporary shocks will require, among others: 0 Fiscal income to be less dependent on oil revenues, both in terms of levels and fluctuations over time. Several measures can be considered with this objective in mind. First, the non-oil revenue base should be increased by improving tax compliance and tax collection effectiveness. Second, the rules governing the existing oil-price stabilization fund should be modified to make it operational and effective. Inparticular, the upper-bound price triggers for diversion of oil-related revenues into the fund should be brought in line with historic oil price data, and clear guidelines should be providedfor the use of available funds. 0 Spending pre-allocation and earmarking to be reduced in order to increaseflexibility in the use of existingresources and, as a result, minimize the need to resort to discretionary spending. Indoing this, protection of key programs, such as certain social and pro-poor programs, should be ensured. 1.37 More generally, fiscal policies aimed at improving efficiency in the use of resources should include, among others: 0 The harmonization and simplification of the tax system. The current proliferation of taxes, most of them with little revenue-generating capacity, has a negative impact on tax efficiency. Repealing some of the minor taxes, at the same time that the business, income and sales taxes are simplified and strengthened would reduce administrative efforts, and increase tax transparency and decreasedistortions. 0 The elimination of subsidies to public enterprises in various sectors. Transfers to public enterprises are not linked to production or service quality indicators, artificially protecting these 12 enterprises from competitive forces and discouraging accountability and efficiency. These transfers should be eliminated or, alternatively, used to provide incentives for profitability and conditioned on improved service provision. 1.38 Inthe next two sections, we abandonthe long-run approach we have followed up to thispoint and focus on the most recent macroeconomic events, the 1998/99economic crisis and the subsequent dollarization, and their impact on poverty. For this purpose we will draw on previous work done by the World Bank and others, as well as original work prepared for this report. THEIMPACTOFTHE 1999CRISISONPOVERTY 1.39 The 1999 crisis was the most severe economic downturn experienced by Ecuador in the last two decades. As a result its effects were dramatic and lasting. A significant amount of work has already been done on the identification of the sources and consequences of the crisis. We briefly summarize that work here, emphasizing those aspects of the crisis that will be most helpful for us in understanding the current nature and level of poverty further along in this report. Additional referencesare providedfor the interested reader inthe text and inAnnex 5: Sourcesof the crisis 1.40 It has become customary to say that the 1998/99 crisis began with the natural disaster E2 NiRo, and was aggravated by the decline in oil export prices and the global financial crisis (World Bank, 2000c; Beckennan and Solimano, 2002). Although these were the primary shocks that triggered the crisis, they were not the maindeterminants of its depth and length, nor can they explain the extraordinary measures that were required to restore stability and growth. As discussed in the previous section, macroeconomic policies throughout the 1980s and 1990s were erratic and misleading (see Box 1.3 for summary of political events and policies during the decade). In the face of the 1997/98 shocks, the authorities lacked the ability to propose sound policies and the credibility to implement them, so that the initial shocks eventually led to a large fiscal deficit, exchange rate and price instability, a severe banking problem, and a deep recession that sent poverty rates up by 30 percent (World Bank, 2000~).Amidst growing inflation and a deep political crisis, the government took the extreme decision to dollarize the economy. Poverty and inequality during the crisis 1.41 Macroeconomic crises in developing economies are often associated with increases in poverty and income inequality, and Ecuador was no exception in 1998/99 (World Bank, 2000~). Moreover, since PPP-adjusted per capita income (inequality) was already among the lowest four (highest three) of the main 16 Latin American economies, the effects of the crisis were truly devastating, pushingpoverty and inequality to record highs. Two areas were particularly affected. In the Costa, El NiRo produced flooding, landslides, and damage to infrastructure, and resulted in large economic losses. According to Vos et alia (1999), it cost the lives of at least 286 people, an estimated 30,000 people lost their homes, and about 25 percent of the population incurred a serious increase in health risks associated with the spread of infectious diseases, resulting in part from reduced access to drinking water and sewerage. The magnitude of the damage i s estimated at between 2.7 percent and 13.6 percent of GDP (Vos et alia, 1999 and ECLAC, 2002). While almost everyone in the Costa suffered losses from the crisis, the rural poor were the most affected. 13 Box 1.3: Macroeconomic developments of the 1990s As mentionedabove, the 1999 crisis did not occurred in a void, but rather was the end result of a fairly turbulent decade both in political and economic terms. We briefly review here some of the decade's developmentsto give the reader a flavor for the circumstances. Politicaland institutional instability 0 Five differentpresidentsbetween 1988 andthe endof 1999 Fiscal policy 0 Numerousattemptsto apply fiscal stabilization measures, followed by periods of relaxation, leading to "adjustment fatigue" andloss of credibility 0 Highvolatility of fiscal revenuesassociated with chronic instability of oilprices 0 Large increases inpublic expenditure inthe secondhalf of the 1990sdue to increasesinpublic sector wages anddebt-relatedpayments 0 Increasingvulnerability of fiscal policy as the value of the sucre declineddue to the highlevel of debt-dollarization Monetary policy 0 Low credibility of the Central Bank`s capacityto control inflation 0 Highdefacto dollarization (see below) 0 Liberalization of financial sector without proper supervisionandcontrol 1.42 In urban areas, poverty increased as a result of the collapse of the banking system and the deterioration in labor market conditions. Low and middle-class urban households saw their lifetime savings disappear as a consequence of the collapse of the banking system, and were often forced to cash-in a significant portion of their assets (Halac and Schmukler, 2003). More generally, plummeting employment rates and real wages cause household income to fall, with those employed inthe informal sector, and those with less education beingthe most affected initially. 1.43 As the crisis progressed, however, all segments of society felt the severity of the downturn. Poverty increased as the real earnings and assets of the near-poor declined, and by the end of 1999 consumption-based poverty levels had reached 60 percent of the population. Households already in poverty also reduced their consumption, especially of food, health care, and education. Finally many were forced to migrate (internally and internationally) in search of economic opportunities. 1.44 The effects of the crisis, however, extended beyond the realm of household income and assets, as documented in World Bank (2000~).Both the quality of basic services and the capacity of households to access such services declined (see Box 1.4 for a summary on the effects of the crisis as reportedby the poor). While the percentageof children enrolled in school remained constant, average attendance dropped (i.e. the number of days missed per month climbed from 5 to 10 between 1998 and 1999) and child labor increased among those who were not enrolled, particularly in rural areas. Similarly the percentage of households that decided to defer medical treatment due to its cost rose from 50 to 72 percent. Finally, qualitative evidence suggests that negative externalities arose including violence and consumption of non-merit goods and stealing public utilities (Le6n and Troya, 2000). 14 Box 1.4: The non-monetary effects of the crisis as reported by the poor We summarizeherethe results from a set of interviews with poor householdsandfocus-group discussions on the effects of the 1999 crisis conductedfor the 2000Ecuador Poverty Assessment (World Bank, 2000). Education 0 Deteriorationof physical infrastructureandlack of didactic materials 0 Irregular attendanceamongteachers 0 Poor student performancedue to unmotivated teachers andadecline in healthandnutrition conditions Health 0 Deferral of medical attention 0 Increaseincost anddecreasein availability of medicinesandtreatment 0 Shortageof medicines inhospitals so that householdsneedto buy their own supply 0 Vaccination campaign sloweddown Nutrition 0 Decrease inconsumptionof proteinsand minerals 0 Fewer mealsa day 0 Foodpriority within the householdwas given to the main income earner 0 Families headedby women or older individuals hadtrouble infulfillingnutritional standards Source: World Bank (2000~) Social spending and the crisis 1.45 Inresponse to the crisis the Government attempted to protect some social expenditures, but failed overall in dampening the effect of the economic downturn on the social sectors as salary arrears in the social sectors had serious consequences for the delivery of basic social services. Spending per capita fell by 40 percent between 1997 and 1999, affecting primarily the health and education sectors, which saw their budget allocations drop by more than 30 percent. Chapter 5 will analyze the effect of these and other trends insocial spending inmore detail. THEIMPACTOFDOLLARIZATION ONPOVERTY 1.46 We devote the last section of this chapter to the analysis of the effect that the adoption of the US dollar as the national currency in 2000 has had on poverty, paying special attention to the role of changesinprice levels and changesinrelative prices as transmission mechanisms. 1.47 The dollarization process helped control and then eliminate hyperinflation. During 2000 the monthly inflation rate fell from 14 percent in January to 5 percent in May to 1.3 percent in August, preventing further purchasing power erosion, especially among those living on fixed incomes. Furthermore, because prices stabilized faster in some sectors than others as a consequence of dollarization, relative prices adjusted at the same time that inflation declined. 1.48 Price stabilization, combined with changes in relative prices undoubtedly impacted consumption patterns and poverty in various ways. First, dollarization caused a realignment of prices between tradable and non-tradable goods. Prior to dollarization, "foreign inflation", or inflation due to chronic exchange rate devaluations, was one of the major factors underlying endemic price increases in Ecuador. In contrast, the main cause of inflation after dollarization has been the steady growth of the domestic price component, which has led to an appreciation of the real exchange rate 15 and to a decline in the relative cost of tradable to non-tradable goods. Second, total credit and consumption credit has grown steadily since 2000, both in absolute terms and as a proportion of income, facilitating purchases of durable goods. Finally, a reduction in the risk associated with currency and monetary instability has translated into a decline in interest rates, hence lowering investment and borrowing costs. 1.49 In the medium run, the effects of the dollarization on growth, consumption and poverty are more uncertain. Dollarization could help provide credibility to economic policies, creating more favorable conditions for sustained economic growth and increased income levels. However, it could also impair the Ecuadorian government's capacity to implement countercyclical economic policy, while doing very little to eliminate growth volatility - most of which i s associatedwith variation in the price of oil and other external shocks, and natural disasters. Similarly, the dollarization could have lasting effects on the level and structure of employment associated with losses in competitiveness due to sustained RER appreciation, and with changes in the relative prices of both inputsandfinal goods. 1.50 We explore here the link between the short-run effects of the dollarization and poverty, leaving the question of medium and long-run effects open for future research. The extent to which consumption and poverty are affected by changes in relative prices will then depend upon the magnitude of the adjustmentsbrought about by dollarization, the nature of consumption patterns, and the differences in such patterns between the poor and the non-poor. We address these three issues below. Formaland informal dollarization 1.51 The dollarization process produced dramatic changes in economic policy and fiscal management. Its impact on consumers, however, may have been less significant since the economy was de facto already highly dollarized by the end of 1999, as Ecuadorians sought a stable unit of account for wealth and contractual relationships inresponseto the purchasing power instability of the sucre (Beckerman and Solimano, 2002). 1.52 Dollarization progressed slowly at first, but became pervasive by the mid-1990s, especially after exchange-ratedepreciations mechanically increased the sucre equivalent of US dollar balances. The U S dollar-denominated proportion of the money supply (including quasi-money) rose from 7.4 percent in 1990 to 36.9 percent in 1997 and 47.4 percent in 1999. Similarly, 19 percent of all on- shore deposits and 28 percent of all loans were already denominated inU S dollars in 1995, and these figures hadincreasedto 54 and 66 percent respectively by 1999(Table 1.7). 1.53 Furthermore these measures understate the extent of dollarization since the US dollar was increasingly used for domestic transactions and in offshore deposits, not included in official money- supply figures (Beckerman, 2001). In addition, foreign currency was increasingly available as remittances climbed from around US$0.5 billion in 1990to US$1.5 billion in2001. 16 Money & Quasi Money Deposits LoanPortfolio % of total denominated in US$ 990 7.4 13.3 1.5 995 24.3 19.2 28.3 997 36.9 23.6 45.1 999 47.4 53.7 66.5 1.54 Given all this, it i s very possible that formal dollarization did not alter consumption patterns radically -an observation we make use of when examining the impact of dollarization on consumption belowz3. The impactof dollarization onconsumptionand poverty 1.55 We have argued above that the dollarization process generated drastic changes inprice levels and relative prices. In this section we document changes in relative prices and discuss the extent to which households reacted to and were affected by them. On the one hand, changes in relative prices can induce changes inconsumption patterns24by making certain consumption articles relatively more expensive than others. On the other hand, these changes can increase or decrease the cost of a given consumption bundle, even in the absence of changes in consumption patterns. Since consumption patterns are different across poor and non-poor households, both developments can have an effect on poverty. 1.56 Our discussion on the extent of informal dollarization prior to the adoption of the dollar led us to conclude that most likely consumption patterns did not change dramatically after this happened. We assume this was indeed the case25,and focus on the effect of relativeprice changes on the cost of different consumption baskets, instead. 1.57 Using 1999ECV, we calculate the share in total expenditure of different categories of goods for the average Ecuadorian family (Table 1.8)26.Food, housing, and utilities account for the largest fraction of expenditures for the average household, with smaller fractions devoted to durable consumption goods (such as refrigerators, TV sets, and stoves), clothing and non-food groceries 23 This does not imply, however, that dollarization did not affect the level of consumption. For instance, saving- consumptiondecisions may be a function, amongother things, of the degree of economic stability. 24By consumption patterns we refer to the share of expenditure devoted to each good or group of goods. 25A formal analysis of changes in consumption pattems would require pre and post-dollarization data on these pattems, which is currently not available. The INEC is at the moment collecting information on income and expenditurein order to updatethe structureof the CPI. Once this data becomes available, it will be possible to perform this analysis and check for the validity of our assumption. 26 The 1999 ECV covers 5,824 households during the period September, 1998- September, 1999. The survey contains information on expenditure in 151different products, including 83 food items, 43 non-food types of groceries, 23 durable goods and indirect informationto determine expenditureson housing. Inorder to construct average expenditure shares, we follow the methodology described in Annex 3 and correct extreme or missing data for prices, quantities, and frequencies using mean values at the cluster level. 17 (ranging from soap to candles). These shares are inline with those reported for comparable countries (Gallego and Soto, 2001)27. All households Non-poor households Poor households E C E C E C Food 35.1 20.2 30.6 16.4 43.1 34.5 TransporVCommunication 6.7 3.8 8.0 4.3 4.9 3.9 Non-foodgroceries 5.1 3.3 6.3 3.4 5.5 4.4 Clothing 5.1 3.0 6.7 3.6 2.7 2.2 Durable goods (purchases) 8.2 4.7 11.4 6.1 2.2 1.7 Water, gas, electricity 15.7 9.0 13.0 7.0 24.7 19.8 Housing 23.5 13.5 24.0 12.9 16.8 13.4 Durable goods (consumption) 42.5 46.3 20.0 Total 100.0 100.0 100.0 100.0 100.0 100.0 Source:Authors' calculationsbasedondatafrom the 1999ECV. 1.58 Certain goods, however, are not entirely consumed at the moment of purchase, but rather over a long period of time. This i s the case of consumption durable goods such as cars or appliances. When a family buys a refrigerator, it does so under the expectation that it will work for years to come, so that, even though the money spent on the refrigerator gets accounted as part of the family's expenditure this year, `consumption' of the services provided by this appliance will be spread over future years. 1.59 We examine consumption patterns and compare them with expenditure patterns to better capture the difference between both. In order to do this, we need to make assumptions about the fraction of durables goods that i s actually consumed in each period, as well as about their rate of depreciation due to normal wear and tear. We follow the methodology proposed by Gallego and Soto for this purpose (see Box lS),and find that consumption patterns differ from expenditure patterns in interesting ways. In particular, the share of durable goods in consumption i s much higher (42 percent) than their share inexpenditure (5 percent). The magnitude of this difference reflects the fact the size of the stock of durable goods i s large compared to the size of the flow of goods bought in each period. Put simply: the purchase of durable goods tends to occur sporadically whereas services from these goods are consumed regularly. 27The average householdinArgentina, Mexico, Colombia, and Chile spentabout 5 percent on durable consumption goods inthe 1980sand 1990s.The equivalent number is developedeconomiesis higher, around 12percent, sincehigher income levels allow for mass consumption and higher replacement rates of these goods. For developing economies such as Argentina, Mexico, Colombia, and Chile, the estimated shares for purchasesof durable consumption goods hover around 5% of total expendituresin the 1980s and 1990s 18 Box 1.5: Estimatingdurablegoodsconsumptionflows usingexpendituredata When buying durable consumption goods (such as cars, appliances, and TV sets), consumers consider (i) the flow of services derived from the good, (ii) the average life-length expected for each good (Le., the period of time the good will be functional), and (iii)the indirect cost derived from the alternative use o f the resources investedin the durable good. Followingthis principle, the consumption o f the durable good, Cd,can then be modeledas: Cd, = (r, 6) Kd, + where Kd i s the stock of the durable good, r i s the interest rate, and 6 is the depreciation of the stock due to use and normal wear and tear. Since data on the stock of the durable good i s not available in this case, the following simplification i s used: Cd, =NG, * RV,/AL, where NG is the number of goods of each type in each households, RV is the replacement value and A L is the average life of the good. This approximation eliminates the "financial" effect arising from changes in the interest rate and neglects quality differentials that may arise in time that are not reflected in the 1998-99 prices. Average lives for the different goods were obtained from Gallego and Soto (2001). Source: Gallegoand Soto (2001). 1.60 The behavior of the average household, however, masks important differences between poor and non-poor households28.Non-poor households spend relatively less on food and utilities, and relatively more on durable consumption goods and housing than their poor counterparts. Similarly, their consumption of durable goods i s twice that of poor households. 1.61 Given these differences in expenditure and consumption patterns between poor and non-poor households, it is likely that changes in relative prices associated with the dollarization have affected them differently. We turn to this question next by documenting changes in the cost of the average consumption basket between the end of 1999 and the present, as well as changes in the cost of particular groups of goods, and by analyzing the potential impact of such changes on the welfare of poor and non-poor households (see Box 1.6 for a comprehensive description of the methodology used)29. 1.62 The relative cost of the consumption basket of the average Ecuadorian household has declined significantly (16 percent) since 2000. Interestingly most of this decline occurred during the months immediately following dollarization (Figure 1.4). Households are classifiedas poor and non-poor according to the poverty line developed in Chapter 2, and used in the rest of this report. 29 Prices are for urban areas only. It is, however, reasonable to expect prices differences between urban and rural areas. We do not account for those differences in our analysis and can only speculate as to what effect the may have had on the results we report here. For instance, expenditure in housing and utilities is lower in rural areas so that negativeeffect of increasesin the price of both on expenditure andconsumption would be smaller in rural areas. 19 Box 1.6: Relative price changes and their impact on consumption We use data on mean prices from the Consumer Price Index (INEC) to trace down the evolution of (relative) prices. The data i s the weighted average of observed prices in the country's main urban areas, so that it is adequatefor most urban households, but may be not be so for rural ones. In addition, product grouping in the CPI differs from that in the 1999 ECV. The CPI provides monthly prices for 197 (groups of) products, while the 1999 ECV considers only 185, of which only 118 are defined in a comparable fashion in both surveys. Luckily these 118 products represent over 90 percent of total expenditures, so that we feel confident the results discussed here are robust to existing methodological differences between these two surveys. The impact of price changes on consumption, Ct, can be studiedusingthe following expression: where C, i s the sum o f the values of all goods consumed (purchased and imputed) during the period, xCiQ: , Sq9,, i s the share of good i in consumption in 1999, and 1,i s the relative value of the 1999 consumption basket evaluated at time t prices. Figure 1.4: The relative cost of the average household consumption basket declined after the dollarization Figure9 RelativeCost of the ConsumptionBasket (December 1999= 100) 20 1.63 Although the evolution of the cost of the consumption basket i s a comprehensive measure of welfare, it hides important changes in the relative price of the different components of consumption. To the extent that households consume different amounts of and derive different utility from such components, these relative changes may have a differential impact on welfare. For instance, the cost of food and groceries has declined by 15 and 20 percent, respectively, since 20003°, potentially benefiting poor families who allocate a larger fraction share of their income to food more (Figure 1.5). Figure 1.5: The (relative) price of different goodshasvaried significantly over time costdTransportation&Uilitii I "~ ISource: Authors' calculations basedon data from the CPI andthe 199ECV (INEC) 30 Figures are seasonally adjusted to avoid nuisance from seasonal effects. 21 1.64 In contrast, the cost of transportation and telecommunications, and utilities has grown by more than 30 percent during the same period, due to price increases in public transportation3', telephone services, water and electricity. 1.65 The cost of clothing has remained flat, while the cost of housing (purchase and rental prices) has surged, after a small initial decline, reflectingthe more cyclical nature of the housing market. 1.66 Finally, the cost of durable consumption goods has halved since 2000, leading to important welfare gains for consumers. This decline i s mainly the result of the RER appreciation, since a large fraction of these goods are imported. In fact, the evolution of the relative price of durable to non- durable goods mimics closely that of the relative price of tradable to non-tradable goods (Figure 1.6). 1.67 In order to evaluate the differential impact that the price changes may have had on poor and non-poor households, we compute changes in expenditures and in the cost of the consumption basket for both types of households assuming that the expenditure and consumption shares of different groups of goods remain constant over time. Total expenditures have increased across the board by around 10percent, due mainly to increasesinthe cost of housing and utilities. 1.68 Incontrast, the cost of the average consumption basket has declined by 16percent, although changes inthe cost of this basket differ by household income level (Table 1.9). Non-poor households have seen the cost of their consumption basket decline by almost 20 percent, compared to 2 percent for poor households. This decline can be attribute almost entirely to changes in the cost of durable goods, of which non-poor householdsconsume a larger amount. Figure 1.6: The decline inthe cost of durable goods is associatedwith the decline inthe cost of tradable goods a5 t a4-99 mol ma2 JLk-toO JLk-tol JLk-ta2 JLCFa3 I Source: Authors' calculationsbasedondata from the CPIandthe 1999ECV (INEC) 31The saw-like shape of the transportation curve reflects the low frequency of adjustments in public transportation tariffs and the subsequent erosionof their real value due to inflation. 22 1.69 There are three reasons that explain the differential impact that changes in the price of durable goods have on poor and non-poor households. First, non-poor households spend relatively more on purchases of durable goods, and hence benefit relatively more from price declines. Second, and far more important in quantitative terms, higher income families own larger stocks of durable consumption goods (mainly, cars and appliances). Hence they derive more services from these goods. Third, the cost of credit decreased and its availability increased as a result of the dollarization. Since consumption credit i s allocated mainly to non-poor households, these changes contributed to expand their capacity to purchase and consumedurable goods. 1.70 Finally, to the extent that the dollarization i s a recent phenomenon, we cannot judge its impact on consumption (as described above) as definite. A number of positive outcomes are yet to be delivered. For example, dollarization will most likely lead to the expansion of the range of services offered by the financial sector. Likewise, firms that survived the 1999 crisis and that have adjusted to the dollarization are probably now in much better shape to undertake efficient investment projects, raising productivity levels, employment levels, and real wages. All households Non-poor households Poor households % changes E C E C E C Food -4.44 -2.55 -3.87 -2.08 -5.46 -4.37 TranspodCommunications 2.3 1 1.33 2.77 1.49 1.70 1.36 Non-foodgroceries -1.04 -0.60 -1.16 -0.62 -1.02 -0.81 Clothing -0.04 -0.02 -0.05 -0.03 -0.02 -0.02 Durable goods (purchases) -2.62 -1.51 -3.62 -1.94 -0.69 -0.55 Water, gas, electricity 4.97 2.86 4.13 2.22 7.84 6.28 Housing 10.94 6.29 11.15 5.99 7.82 6.26 Durable goods (consumption) -22.31 -24.34 -10.50 Total 10.08 -16.51 9.36 -19.32 10.18 -2.35 Source: Authors' calculationsbasedondata from the 1999ECV. CONCLUSIONS 1.71 We have argued in this chapter that poor growth and economic instability have crippled Ecuador's capacity to reduce poverty. We have also argued that growth behavior i s closely linked to TFP behavior - that is, economic growth depends closely on productivity growth, which in tum dependson the quality o f inputs, institutions and policies. 1.72 This connection between productivity and economic growth has become even more relevant inrecent years, after Ecuador decided to adopt the US dollar as the national currency, henceforgoing the option of using exchange rate policy to generate temporary increases in competitiveness and growth. Although the decision to dollarize undoubtedly improved the investment climate, reassured potential investors, and hence potentially increased the capacity of the economy to create 23 employment and reduce poverty, sustained increases in productivity will be required to maintain positive growth rates. 1.73 The chapter has also emphasized the negative impact that domestic policy volatility has had on economic and productivity growth. 1.74 As a result, policies aimed at preserving stability with fiscal discipline and, especially, at increasing economic productivity and competitiveness hold promise inpromotingpositive, sustained growth. The use of fiscal measures to achieve these goals was discussed earlier on in this chapter, while other types of policies are examined inmore detail later on inthe report. 24 2. NATURE,DISTRIBUTIONAND EVOLUTIONOFPOVERTYIN1990-2001 2.1 The first step to an analysis of the relationship between poverty and public policies i s an understanding of how poverty is distributed, both geographically and across households of different characteristics, and how (if at all) this has changed over time. In this chapter we describe the determinants of poverty and examine changes in poverty over the period using household-level data and two poverty maps developed for this purpose. 2.2 The use of poverty maps allows us to put poverty numbers under the microscope, inthe sense that they provide a very detailed picture of the distribution of poverty in space and of the changes in this distribution over time. This picture reveals a significant amount of heterogeneity that we then exploit to identify local correlates of growth and poverty. 2.3 Analyzing changes in poverty allows us to identify `hot spots', or areas where poverty has increased significantly over the decade, as well as to discuss the main challenges that these changes pose for such areas. In addition, analyzing correlates of growth and poverty reduction allows us to identify policies that may help these areas reverse negative poverty trends. 2.4 The main findings of the chapter are: The poor tend to live in larger households, be less educated, suffer from higher unemployment, and have lower access to basic services than their non-poor counterparts. The national consumption-based poverty rate stood at 45 percent in 2001, compared with 40 percent in 1990, although whether we can consider this to be a statistically significant increase depends on the statistical test chosen. 0 Some areas, however, did unambiguously experience significant increases in poverty. In particular, poverty rates in urban areas in the Costa and the Sierra were more than 80 percent higher in2001 than in 1990. The urbanization of poverty was the result of: (i) rural-to-urban and urban-to-urban migratory flows attracted by better living conditions and economic opportunities, and (ii) the particular nature of the 1999crisis, which specially affected urban middle-class households. Shifts in employment from the agricultural to the non-agricultural sector appear to be correlated with declines inpoverty. 2.5 The rest of chapter is structured as follows. The first section presents poverty and inequality numbers, as well as the poverty profile for 2001. The second section presents consistent estimates of poverty for 1990 and 2001 and discusses changes in poverty and its distribution for the period, using the poverty maps. The third section examines local determinants of poverty and poverty changes. The fourth section concludes. POVERTYUPDATEANDPOVERTYPROFILE 2.6 What are the levels of poverty and inequality in Ecuador? How are poverty and inequality distributed across the country? What are the characteristics of the poor? And, do these differ across regions? In this section we provide an overview of the poverty and inequality conditions in Ecuador in 2001 - the year for which the most recent data are available at the national level. We present updated poverty and inequality numbers, and examine the main determinants of poverty at the household level, postponing the analysis of poverty changes over time until the next section. 25 Povertyand inequality inEcuador in2001 2.7 We construct poverty and inequality measures for 2001 using the 1999 Survey of Living Conditions and the 2001 Population Census following the methodology described inparagraphs 2.24 to 2.28. We measurepoverty and inequality on the basis of consumption, rather than income. We do this for a number of reasons, the most important of which being that consumption fluctuates much less than income during the course of a particular month or year, and that people tend to report expenditures more accurately than income. 2.8 Poverty measuresare a function, among other things, of the poverty line. We set the value for the 2001 poverty line so that it i s comparable to that of 1990, and allows us to produce comparable poverty figures for the decade32.Because this value of the poverty line may differ from that used in other reports, the figures presentedhere may not be strictly comparable to the ones presentedinthose reports (inparticular, they are not comparable to those inWorld Bank, 2002). 2.9 The national poverty headcount in 2001 was 45 percent33.The poor, however, were not uniformly distributed across areas and regions. Poverty was more prevalent in rural areas, particularly in the Sierra, than in urban areas, and rural poverty was deeper (poverty gap) and more severe than urban poverty34.Regional differences were large inrural areas, but insignificant in urban areas (excluding Quito and Guayaquil). Finally, poverty was lowest in the country's two largest cities, although poverty levels inGuayaquil were double those of Quito. Table 2.1: Almost halfthe populationinthe Sierraand the Costawere poor in 2001. I I Incidence Gap Severity I PO PI P2 National (w/o Oriente) 0.45 0.18 0.10 Quito 0.18 0.05 0.02 Guayaquil 0.34 0.11 0.05 Urban Costa 0.46 0.17 0.09 Urban Sierra 0.46 0.19 0.01 Rural Costa 0.58 0.21 0.10 Rural Sierra 0.66 0.33 0.20 Calculationsbasedon full consumptionaggregate, as described inparagraphs2.24 to 2.28. SierraandCostaurbanmeasuresexclude QuitoandGuayaquil. 32 The poverty line is set at a consumption levelof 15,807 sucres (or US$1.3) per capita per day. 33 This figure does not include the Oriente. Unfortunately we cannot calculate poverty numbers for this region because it was not coveredby the 1999ECV. 34 Urban areas are defined following Ecuador's administrative division (Division Politico-Administrativa). As a result, although urban cantons defined in this manner tend to have larger populations and higher population densities, we do not follow a specific population-size cut-off as is customary. 26 2.10 Consumption was more unequally distributed in the Sierra than in the Costa, and, within regions, urban areas were more unequal than rural ones (Table 2.2)35. These patterns are consistent with those reported inprevious Poverty Assessments (World Bank, 1996 and 2000~). Table 2.2: Consumptioninequalityis higherinthe Sierraandinurbanareas GEO GE1 Quito 0.35 0.39 Guayaquil 0.3 1 0.34 UrbanCosta 0.31 0.35 Urban Sierra 0.4 1 0.46 IRural Costa I 0.26 0.33 Livingconditions and the characteristics of the poor 2.11 Although the poor everywhere live in marginal circumstances with respect to housing conditions and access to employment and basic services, living conditions vary widely across the country. We briefly discuss some of the differences and commonalities that exist between the poor and the non-poor, and across regions below. Our focus i s on demographic characteristics and labor market outcomes since these fit best with the scope of the report, while a complete description of potential determinants of poverty i s presentedinthe Data Appendix (Table DA.1). 2.12 Expenditure composition: As we discussed in Chapter I,poor households tend to spent relatively more on food and utilities, and relatively less on durable goods than their non-poor counterparts. 2.13 Household characteristics: There exist clear differences between poor and non-poor households in terms of their size and composition. Poor households are significantly larger than non- poor ones and, as a consequence, tend to have higher dependency ratios (Le. number of dependents per income earner)36. Households with certain types o f people are also more likely to be poor, irrespective of household size. In the case of Ecuador, however, these are not households headed by women or older individuals, as one may have expected, but rather households were there are larger numbersof children (Table 2.3). 2.14 Ethnicity: Defining who is and who i s not indigenous in Ecuador i s a complicated undertaking, since there does not seem to be a unique criterion that distinguishes the indigenous and non-indigenous populations. For the purpose of this work, we will classify people as indigenous if 35A comparisonof inequality measures in 1990and 2001 suggest that inequality has increasedover the decade. However, we needto interpret these results with caution since measures for both years are based on slightly different consumption aggregates and may therefore not be strictly comparable. See Lanjouw and Lanjouw (1997) for a detailed discussion on the issueof comparability. 36 Notice, however, that the relationship between poverty and household size i s sensitive to assumptions about economies of scale inconsumption. 27 they speak one or more indigenous languages. However, because this definition may not be satisfactory to all, we also present a brief discussion on the issue of ethnicity identification in Box 2.1. 2.15 Households in which one or more members speak an indigenous language are more likely to be poor, and to have limited access to basic services. This i s especially true in the rural Sierra, where the majority of the indigenous population i s concentrated. Box 2.1: Countingthe indigenousandafro-Ecuadorianpopulation According to the 2001 PopulationCensus, 7 and 5 percent of the Ecuadorian population are indigenous and of Afro-descent respectively (self-reported ethnicity). These groups, however, are not uniformly distributed over the Ecuadorian territory. The indigenous population i s concentrated in rural areas and in the Sierra region. In contrast the Afro population i s predominantly concentrated in the urban areas of the Costa region, especially in the province of Esmeraldas (Lebn, 2003). It is hard, however, to assess to what extent the size and location of these groups has varied over time due to the lack of a coherent and comprehensive criterion to define ethnicity. Traditionally surveys have classified individuals into different ethnic groups exclusively on the basis of language, and it was not until 2001 that a question on self-reported ethnicity was included in the Population Census and other surveys. Results obtained under both `definitions' vary significantly (Table B.2.1. l), and fall extremely short of the figures provided by indigenous organizations. These groups argue that neither the language nor the self- reported ethnicity criteria identify the true indigenous population correctly, and claim that according to their estimates a third of the total population belongs to this ethnic group. Table B.2.1.1: The multidimensionalnatureof ethnicity Self-reported Speaks native language Both Indigenous 6.1 4.6 6.6 Afro-Ecuadorians 5.0 n.a n.a Black 2.3 n.a n.a Mixed 2.7 n.a n.a These discrepancies make it extremely difficult to produce statistical information that is accepted by all. This is particularly unfortunate given the social and political significance that the indigenous movement has acquired inrecent years. It is therefore critical to make a hobit of including ethnicity identification questions in the different surveys administered by the INEC and other institutions that are comparable and, to the extent possible, satisfactory to all parties involved. 2.16 Education: The education level of the household head is strongly related to the household poverty status. The average poor household head in both rural and urban areas has not completed primary studies (6 years), and while literacy rates are close to 90 percent for the country, about 20 percent of poor households i s headed by an illiterate individual. The situation i s particularly worrisome inrural areas, and among indigenous households. 28 Table 2.3: Demographicscharacteristicsof poor and non-poorhouseholdsvary across regions and across urbanandrural areas Urban Rural Non - Poor Poor Non Poor - Poor Householddemographics Iousehold size Costa 4.80 6.67 4.79 7.23 Sierra 4.49 5.85 4.44 6.30 'emale head Costa 0.20 0.20 0.17 0.09 Sierra 0.18 0.16 0.13 0.13 ige of head Costa 45.45 45.84 46.48 47.48 Sierra 44.99 43.04 44.67 46.82 Ethnicity k 6 and older who speakonly 2uichua Costa 0.00 0.00 0.00 0.00 Sierra 0.00 0.02 0.00 0.03 k 6 and older who speak Spanishand Quichua Costa 0.00 0.01 0.00 0.00 Sierra 0.01 0.12 0.05 0.23 Education lead is illiterate Costa 0.04 0.22 0.11 0.20 Sierra 0.03 0.13 0.04 0.28 Srades completedby head Costa 9.76 5.47 6.53 4.22 Sierra 10.05 5.68 7.65 3.48 Source: Authors' calculations basedondata from 1999ECV. 2.17 Employment: Employment is the main source of income for the large majority of households, and thus one of the main determinants of poverty. Although there are no significant differences in household head employment rates between poor and non-poor households, the percentageof household members that is employed i s higher among the latter. 2.18 The type and sector of employment also have an impact on poverty. The informal sector provides employment to a higher share of the poor than of the non-poor, especially in the Sierra. Employment in agriculture i s positively correlated with poverty, while employment in the non- agricultural sector (particularly the high-productivity non-agricultural sector) i s negatively correlated with poverty. 29 Table 2.4: Employmenteffectivelyreducespoverty. Urban Rural Non - Poor Poor Non Poor - Poor Employment(for membersages 10andolder) Householdheadis employed Costa 0.89 0.84 0.88 0.89 Sierra 0.89 0.88 0.95 0.95 % of memberswho are employed Costa 0.51 0.42 0.52 0.45 Sierra 0.53 0.47 0.60 0.58 % of memberswho are informal Costa 0.35 0.35 0.44 0.4 1 Sierra 0.34 0.41 0.45 0.54 % membersinagriculturalsector Costa 0.03 0.09 0.19 0.25 Sierra 0.03 0.14 0.25 0.40 % membersin high-productivity ion-agriculturalsector Costa 0.35 0.19 0.21 0.10 Sierra 0.35 0.17 0.24 0.08 % membersinlow-productivity ion-agriculturalsector Costa 0.12 0.13 0.10 0.09 Sierra 0.13 0.15 0.10 0.08 Source: Authors' calculations based o n data from the 1999ECV. Correlatesof poverty 2.19 In this section, we use conditional probability models, in which household poverty status i s modeled as a function of `exogenous' variables, to examine more formally the relationship between poverty and some of the variables discussed above. This approach helps us identify variables that are directly correlated with poverty, once the effect of other factors i s controlled for. We estimate these models separately by region and area to account for variation in living conditions of poor and non- poor households. 2.20 We briefly discuss here the results from these estimations, while a more detailed discussion of the correlates of urban and rural poverty i s presented in Chapter 3 and 4, respectively. In particular, we identify here common correlates across urban areas and rural areas indifferent regions. The results tend to confirm the ones presentedabove regarding the determinants of poverty. 2.21 Common urban factors: Urban poverty in both the Costa and the Sierra appears to be associated with (i) larger household sizes, (ii) levels of education of the household head, and (v) low unemployment of household head and/or other household members. Other relevant factors are poor access to basic services, and the lack of home ownership. 30 2.22 Common rural factors: Rural poverty in both the Costa and the Sierra appears to be associated with (i) large household sizes and high levels of crowdedness, (ii) identification of the household head as indigenous, (iii)low levels of education of the household head, (iv) unemployment of the household head, and (v) employment of household head as an agricultural laborer. Other relevant factors are poor access to basic services, landand markets. CONSISTENTESTIMATESOFPOVERTY, 1990-2001 2.23 How did poverty change over the decade, if at all? And, if it did, where the changes distributed uniformly across regions and areas? We answer these questions here making use of the 1990 and the 2001 poverty maps. We analyze regional and canton-level trends in poverty, as well as changes in urban and rural areas, and provide some tentative explanations for the patterns we observe. Poverty MappinginEcuador: A Brief Introduction 2.24 The 1990s were a period of intense data collection activity in Ecuador - an initiative we benefited from significantly in preparing this report. In particular, the availability of two Population Censusesfor 1990 and 2001, and a series of Encuestas de Condiciones de Vida (Ecuador's survey of living conditions, ECV hereafter) for 1994, 1995, 1998 and 1999 allowed us to construct comparable poverty maps for 1990 and 2001 and, based on these maps, to analyze changes in poverty over the decade. These two maps constitute the first such panel, so that the work discussed here should be regarded as new and innovative. 2.25 Working with poverty maps based on census data, as opposed to working with a limited sample of the population, improves our understanding of the evolution and distribution o f poverty in two important ways. First it allows us to study poverty at a very disaggregated level - in the case of Ecuador, the level of the canton and the parish. Second, it makes it possible to construct standard errors for our poverty figures so that we can evaluate the level of accuracy with which we are measuring poverty and changes inpoverty. 2.26 The 1990 and 2001 maps were constructed following the methodology proposed in Elbers, Lanjouw and Lanjouw (2003). A detailed description of this methodology, and how it was adapted to build the map panel is presented in Annex 3. In a nutshell, each map i s the combination of an ECV and a Population Census - in our case, the 1994 ECV and the 1990 Population Census for the 1990 map, and the 1999 ECV and the 2001 Population Census for the 2001 map. An econometric model for household consumption i s developed using the ECV data, and the coefficients of this model are then applied to the Population Census data to predict consumption for all households in the population and, using this prediction, to calculate poverty. Because poverty estimated in this fashion i s really the product of information collected in two different periods, it i s important to evaluate whether the variation observed in the figures i s driven by a particular data source. In our case, most of the heterogeneity underlying the poverty numbers comes from the 1990 and 2001 Population Censuses, rather than the 1994 and 1999ECVs, so that we feel confident in treating them as accurate measuresof poverty in 1990 and 2001. 31 I Box 2.2: The needfor a newECV inEcuador Consumption and poverty estimates for 2001 were calculated usingthe 1999 Encuesta de Condiciones de Vida and the 2001 Population Census. For these estimates to truly capture the level and distribution of poverty in 2001, two conditions are necessary. First, most of the heterogeneity in poverty has to be generated by the 2001 data. We argue inthe text that this i s indeed the case. Second, the structural model used to predict consumption must be stable - Le. the returns, in terms of consumption, to household and neighborhood characteristics and assets must remain constant between 1999 and 2001. Due to the special nature of the period separating both datasets, we cannot be absolutely sure that the second condition holds. For this reason it becomes imperative to validate our results using more up-to-date consumption data. The availability o f these data depends on the collection on a new ECV for Ecuador, or at least on the collection of income and expenditure data that isfully comparable to that of the 1999 ECV. This data shouldcover all three regions: Costa, Sierra and Oriente. Annex 4 discusses data needs for the measurement and monitoring of poverty and social outcomes in more detail. 2.27 Ensuringcomparability across both maps, however, was tricky and required a bit more work than the previous paragraph may have suggested. In particular, the expenditure modules used in the 1994 and 1999 ECVs were not identical, so that we needed to construct comparable consumption aggregates and comparable poverty lines in order to produce comparable poverty figures. The comparable consumption aggregate includes 60 percent of total consumption in 1990 and 55 percent of total consumption in 2001. Since these shares may seem small, we tested the robustness of our results and found that the poverty numbers were not sensitive to the exclusion of those consumption items not included inthe comparable aggregate (Table 2.6). 2.28 Finally, it i s important to mention that we were unable to produce poverty figures for the Oriente in 2001, since no households living in this region were interviewedinthe 1999ECV. This i s an important caveat in our analysis and it should teach us a lesson - no econometric method or approximation i s a good enough substitute for primary data. 32 Box 2.3: The Unmet BasicsNeed Map Ecuador counts with two Unmet Basic Needs maps (UBNmap hereafter) constructed usingdata from the 1990 and the 2001 Population Censuses. These maps are often used for social policy-making. This Box compares the Unmet Basic Needs index, on which these maps are based, with the (monetary) poverty measure developed inthis chapter and that serves as the basis for the poverty map. The UBN index is a weighted summary of five types of household-level variables and indicators: (i) characteristics of the dwelling, (ii) of access to basic services, (iii) householddependency rate, (iv) degree the the presence of school-age children currently not attending school, and (v) the number of persons per room (crowdedness). Due to the limited nature of the variables underlying the UBN index and to the methodology used for its construction, this index may not be fully informative about monetary poverty. First, as a measure o f structural depravation or poverty, the UBN index correlates only partially with consumption poverty, which generally exhibits a more variable cyclical behavior (Figure B.2.3.1). Second, under this methodology households are considered poor if they have one or more unmet basic needs, even though the impact of different needs on household welfare is not necessarily identical. Since both measurescapture different aspects o f poverty it i s then possible for the evolution of the UBNindex to differ from that o f monetary poverty in 1990-2001, as it is actually the case. Structural poverty, as measured by the UBNdecreasedduring this period, while monetary poverty increased, as discussedbelow. Table B.23.1: The UBNindex does not fully capture monetary poverty, althoughit is positivelycorrelated with it. 200 m 150 3 C 0 -0 a, $ 100 n .-x 0 C BC 50 0 0 50 100 150 200 Ranking based on Headcount Source: Sistema Integradode IndicadoresSocialesdel Ecuador (SIISE, 2002d) and2001Poverty Map elaboratedfor this report. 33 The Evolutionand Distributionof Poverty in1990-2001: Maintrends and canton-level changes3' 2.29 A raw comparison of national poverty figures for 1990 and 2001 points towards a deterioration in welfare over the decade, with the national poverty rate increasing from 40 to 45 percent and the number of poor increasing from 3.5 to 5.2 million during the period (Table 2.5). However, as alarming as this information may seem initially, it needs to be interpreted with caution. Due to the indirect methodology applied for the poverty calculations (i.e. imputing consumption, rather than observing it), these figures are imprecise. As a result, whether the observed increase in poverty i s actually statistically significant will depend on the precision with which the 1990 and the 2001 poverty rates were estimated - that is, it will depend on how large or small the standard errors associated with each of them are. 2.30 We use two different tests to disentangle whether the increasein poverty at the national level i s real - a two-tail test (more stringent and, hence, more likely to reject the possibility that the increase i s real), and a one-tail test (less stringent and, hence, more likely to accept that the increase i s real). Unfortunately, the results of both tests are contradictory: under the first test we reject the hypothesis that poverty did increase, while we accept it under the secondtest. 1990 2001 2001 Total expenditures Comparable expenditures Total expenditures H C Std. Err H C Std. Err H C Std.Err National 0.410 0.020 w/o Oriente 0.403 0.019 0.452 0.023 0.451 0.024 Quito 0.222 0.021 0.243 0.016 0.185 0.012 Guayaquil 0.382 0.018 0.386 0.028 0.337 0.024 Urban Costa 0.258 0.015 0.464 0.013 0.464 0.031 Urban Sierra 0.213 0.017 0.467 0.029 0.459 0.020 RuralCosta 0.505 0.025 0.504 0.017 0.587 0.026 RuralSierra 0.528 0.019 0.617 0.034 0.663 0.028 UrbanOriente 0.192 0.020 RuralOriente 0.598 0.026 37 Headcounts ratios are the only poverty statistic, out of the FGT family, that can be compared across time given the methodology usedto construct the 1990and 2001 poverty maps (Lanjouw, Olson andLanjouw, 1997). 34 2.32 We turn now to changes in poverty across different areas and region, since national trends usually hide a substantial amount of regional variation. Urban areas, both inthe Sierra and the Costa, experienced significant increases in the headcount ratio of 100 and 80 percent re~pectively~~. In contrast, poverty rates in rural areas, which were highest in both 1990 and 2001, appeared to be stable over the decade39. 2.33 The same way that the poverty rates vary across space and time, poor people appear to be concentrated in certain areas and regions and to move across the country over time (Table 2.6). In 1990 the rural Sierra was home to the largest share of poor individuals (37 percent), followed by the rural Costa (28 percent). In 2001, however, the poor were more concentrated in urban areas (20 and 26 percent in the Sierra and the Costa, respectively). As a consequence, the absolute number of poor people increased by 300 per cent in the urban Costa and 500 percent in the urban Sierra over the period, while it decreased by 36 percent in the rural Costa and 13 percent in the rural Sierra. Guayaquil and, particularly, Quito also witnessed increasesin the absolute number of poor, although these changes were of lesser magnitude than the ones observed inother urbanareas. Table 2.6: The numberof poorinurbanareasincreasedsubstantially ... 1990 2001 Change 1990-2001(%) Total poor % of Total4' Total poor % of Total Poor National 3,620,934 wlo Oriente 3,418,306 100.0 5,223,115 100.0 52.8 Quito 150,208 4.4 339,115 6.5 125.8 Guayaquil 496,337 14.5 764,177 14.6 54.0 Urban Costa 347,541 10.2 1,370,293 26.2 294.3 Urban Sierra 171,504 5.O 1,032,990 19.8 502.3 Rural Costa 963,164 28.2 610,664 11.7 -36.6 Rural Sierra 1,289,553 37.7 1,112,195 21.3 -13.8 IUrban Oriente I 4,900 of ruralCosta. 38 Increases in urban poverty are significant irrespective of the test used. This result contradicts the evidence presented in Leon (2002), where the author shows that urban poverty declined in 1990-2001using the Encuesta de Empleo, Desempleo, y Subempleo (INEC). This is a labor force survey and, as such, includes information exclusively on (labor) income. Since consumption-based poverty measures are generally regarded to be more reliable and stable over time than income-based ones, we trust that the measures presented here constitute a more accurate description of the evolution o f urban poverty over this period. 39As with national poverty numbers, we reject the hypothesis of poverty increases in the rural Sierra between 1990 and 2001 under the two-tail test and accept it under the one-tail test. 4oTotal excludes Oriente for the sake of the comparison with 2001. 35 Table2.7: ...growingfaster than the urbanpopulation. % of national population 1990 2001 1990-2001growth National 37.0 Quito 8.0 12.0 106.0 Guayaquil 15.0 16.0 52.0 Urban costa 15.0 24.0 119.0 Urban Sierra 9.0 18.0 174.0 Rural costa 22.0 10.0 -37.0 Rural Sierra 1 28.0 15.0 -26.0 Urbanoriente 0.0 2.0 985.0 Rural oriente 4.0 2.0 -24.0 2.34 In addition, the growth in the number of poor in urban areas far surpassed that of the total population, which explains the significant increases in poverty rates in these areas reported above (Table 2.7). Box 2.4: InternalmigrationinEcuador This box describes the magnitude andpatterns of internal migration inEcuador usingthe 2001 Population Census. The analysis i s based on individual answers to questions regarding (i) place of birth, (ii) current residence, and (iii) place of residence five years ago, so that it excludes seasonal migrants. Only those, who were born inEcuador are considered. The main findings are, briefly: Over 30 percent of the population lives in a place different from the one they were born in. 0 Most migratory movements (over 60 percent) are urban-to-urban, rather than rural-to-urban, although there are some differences across regions (i.e. rural-to-urban flows are more common in the Sierra and the Oriente than in the Costa). This should not be surprising given that 73 percent of the population resides inurbanareas. 0 Quito and Guayaquil alone are the destination of 29% of all internal migrants (13 and 16 percent, respectively). 0 About 80,70 and 67 percent of the population lives inthe province, canton andparish they were born in, respectively. These figures suggests that about one third of all internal migration occurs within provinces, while two thirds occurs across provinces. There i s very little inter-regional migration in the Sierra and the Costa, compared to the Oriente, which appears as a net recipient of out-migrants from the rest of the country. 2.35 The urbanization of poverty over the decade has been accompanied by a broader urbanization process. Population growth in urban areas is, in turn, the result of three different processes: the natural growth of the urban population, rural-urban migration, and the re-classification of rural 36 parishs as urban due to the creation of new cantons4'. Box 2.4 examines the first process using the 2001 Population Census. 2.36 We now descend from the regional to canton level. The canton-level picture provided by the 2001 poverty map confirms what we have discussed so far, at the same time that it allows for a more nuanced analysis of within-region heterogeneity (Figure 2. l)42. Poverty incidence i s highest (above 60 per cent) in the cantons of the Sierra, and also in the provinces of Esmeraldas, Manabi and, to a lesser extent, Guayas and Los Rios. Although these areas cover a large fraction of the country's territory, it i s important to notice that poverty incidence i s lowest in the most populated cantons (Quito, Guayaquil, Ibarra, Manta, Ambato, Baiios, Riobamba, Cuenca, Machala, and Loja), so that population-weighted poverty figures are lower than `area-weighted' poverty figures. Figure2.1: Poverty is highestinrural cantons inthe Sierra, andinthe north area of the Costa n 0 . 2001 Total ExpendturePoverty 001 .om s a - 0 4 0 = S040-0M sea-om s m - i w Source : Authors' calculationsbasedondata from the 1999ECV and the 2001 PopulationCensus. 41According to data provided by INEC, 49 cantons were created between 1990 and 2001 (of a total of 220 cantons in 2001). There are complex administrative as well as political factors behindthe creation of new cantons.While the creation of new cantons does respond to the population growth of certain rural parishes, which demands a change in the local administrative organization, there are also politics involved since the creation of new cantons has to be passed by the Congress. When a canton i s created, a -formerly- rural parish i s turned into a canton head and other of the neighboring rural parishes(from the same or other cantons) are annexedto the new canton. 42Province and canton-level poverty figures are provided inAnnex 5. 37 2.37 Ifwe then compare the 1990andthe 2001 maps, we find that poverty increasedsignificantly in 44 cantons, out of a total of 220, over the decade (Figure 2.2)43.The largest increases (15 to 25 percentagepoints) occurred in cantons located inthe provinces of Azuay, Bolivar, Cotopaxi, Guayas, Loja, Manabi, and Pi~hincha~~. 2.38 The question then arises as to what drives changes in poverty and, in particular, significant increasesinpoverty in some (urban) cantons. We explore this issue inthe next section. Figure 2.2: Significantincreasesinpovertyin44 out of 220 cantons b 0 . u Source: Authors' calculationsbasedondatafrom the 1999ECV andthe 2001 PopulationCensus. 43Increases inpoverty in these cantons are significant irrespective of whether we use a two-tail or a one-tail test. 44There are only four cantons, out of 220, for which we find statistically significant declines in poverty under the one-tail test. These cantons are: Giron in the province o f Azuay, where out-migration has been highest in the country; Samborondon in the province of Guayas, where significant investments in infrastructure have taken place in the last few years; Atacamanes in the provinces of Esmeraldas and San Cristobal in the province of Galapagos, where tourism is the most significant source of income; and Valencia in Los Rios and Las Lajas inEl Oro. 38 Box 2.5: Urbanheterogeneityand poverty. Large cities are often home to the poorest and the richest within a country. Hence, city-levelpoverty numbers can hide important welfare differences across population groups or neighborhoods. With support from the Ecuador Statistical Institute (INEC), we constructed neighborhood-level poverty estimates for Quito, Guayaquil, CuencaandLoja. Although some of these `neighborhoods' are still as large as 300,000 people, this disaggregationallowsus to capturea significantamountof heterogeneity. Table 3.2.5.1 Poverty ratesvary significantlywithin cities. Richestneighborhood Poorestneighborhood City Name Poverty Name Poverty Name Poverty Quito 0.19 Iiiaquito 0.05 Turubamba 0.29 Guayaquil 0.34 FebresCorder0 0.10 Ayacucho 0.48 Cuenca 0.28 Huaynacfipac 0.18 HermanoMiguel 0.38 Loja 0.37 ElSagrario 0.23 Sucre 0.35 Differences across neighborhoods within the same city are substantive. For instance, poverty is six times higher in Turubamba (29 percent) than in Iiiaquito(5 percent), the poorest and richest neighborhood in Quito respectively (Table B.2.5.1). In addition, dispersion in poverty rates appears to be larger in bigger cities (Figure B.2.5.1) Figure B.2.5.1: Heterogeneityacross neighborhood-levelpoverty infour Ecuadorian cities 0.50 0.45 0.40 5> % 0.35 0.30 Guayaquil r 0.25 Quito g u 0.20 Cuenca `EU - 0.15 0.10 0.05 0.00 Note:Neighborhood-levelpoverty rates for all four cities are providedinthe DataAppendix (TableDA.4) LOCAL DETERMINANTSOFGROWTHAND POVERTYREDUCTION 2.39 We argued above important differences exist across, regions, areas, and cantons. In this section we make use of these differences to identify those factors that appear to be correlated with changes in poverty. For this purpose we focus on cantons, the smallest geographic unit for which we 39 have comparable poverty numbers, in order to exploit heterogeneity associated with higher levels of disaggregation. 2.40 We tackle this question in two different ways. First, we analyze the determinants of changes in poverty by correlating changes in canton-level poverty rates with a series of initial canton-level characteristics (Le. measured in 1990). Second we examining the role of labor market dynamics by correlating changes in poverty with canton-level changes in employment and employment composition. Initialconditionsand changes inpoverty 2.41 While most cantons experienced positive increases in poverty incidence between 1990 and 2001, these increases were statistically significant in only 44 of them. We examine inwhich ways, if at all, these cantons are different from the rest and we find that most of them had lower-than-average poverty levels in 1990 - in other words, during 1990-2001 poverty incidence increased the most in those cantons with the lowest poverty rates in 1990 (Figure 2.3). Figure 2.3: Poverty increasedmost incantonswith the lowestpoverty ratesin 1990 ... 80 60 8 2 -.- a, 40 c 1 se 20 0) 8 0 -20 .2 I 1990Incidenceof Poverty .4 .6 .8 Runningline smoother Source: Authors' calculationsbasedon data from the 1999ECV and the 2001PopulationCensus. 2.42 These cantons also had better endowments than those of the average canton in 1990 (Table 2.8). In particular, they had (i) higher levels of education among the adult population, (ii) higher a fraction of the labor force employed in non-agricultural activities, and (iii) higher access rates to basic services than the rest of the country. 2.43 These results may seem counterintuitive at first, since one would expect "better" endowments to lead to a decline in poverty rather than an increase. This, however, ignores two important developments of the past 10years -namely, the particular nature of the 1999 crisis, and the existence of large migratory flows, both internal and external (see Box 2.4 on internal migration and Box 3.3 on external migration). We briefly explore these two issues. 40 Table2.8: ...andthe bestendowmentsintermsof education,employmentandbasicservices. Nochange Poverty increased Difference Obs 131 44 Urbanpopulation 0.506 0.541 Adults ages 31-40 0.099 0.102 With some primary schooling 0.526 0.479 *** With some secondary schooling 0.127 0.149 ** With some post-secondary schooling 0.025 0.037 *** Years of educationof householdhead 4.461 4.793 Years of educationof householdhead's spouse 3.894 4.207 Labor inagriculture 0.182 0.140 *** Labor inlow-productivity non-agricultural sector 0.049 0.060 * Labor inhigh-productivity non-agricultural sector 0.086 0.113 *** Householdswith water in dwelling 0.460 0.480 Householdswith water out of dwelling and lot 0.284 0.227 Householdswith connectionto sewerage system 0.190 0.242 ** Householdswith no sewerage disposal 0.476 0.378 *** Householdswith electricity 0.620 0.638 Householdswith phone 0.054 0.082 *** Householdswho dispose garbage by leaving it innearby lot or rive1 0.469 0.390 * Householdsthat cook with gas 0.491 0.525 Householdsthat cook with wood 0.466 0.432 Householdsthat share a toilet with other households 0.053 0.074 ** Householdsthat have a toilet 0.366 0.474 *** Householdsthat havea shower 0.288 0.349 ** Householdsthat share a shower with other households 0.047 0.066 *** Householdswho renttheir dwelling 0.115 0.145 ** IHouseholds who live in dwelling in exchange for services 0.045 0.036 Difference is significantly different from zero at:*** 99%, ** 95%, * 90%. 2.44 As discussed in Chapter 1, the collapse of the financial and banking systems caused by the 1999 had a particularly negative impact in urban areas and, within these, among middle class households. Unfortunately, the lack of a post-crisis household survey prevents us from formally testing this hypothesis using individual-level data. This, however, does not imply that there exists no support for it. Halac and Schumekler (2003) study the effects of financial crises using financial- sector data for Argentina, Chile, Ecuador, Mexico and Uruguay, and find that these crises have important distributional impacts on both financial-sector participants and non-participants. In the case of Ecuador they argue that, while large investors and borrowers were buffered against the negative effects of the crisis, medium and small ones (Le. the urban middle class) suffered significantly. Given that only two years separate 2001 from the crisis, it i s very possible that our 41 figures for that year are still somewhat contaminated by its aftermath, especially inthe case of urban areas. Since urban areas where better off that rural areas in 1990, this may partially explain why poverty increased more inthose cantons where it was lowest initially. 2.45 Inorder to explore the secondexplanation volunteered above, we construct proxies for total and recent net migration flows at the canton-level usingthe 2001 Population Census4', and correlate them with canton-level poverty rates in 1990. In a world were individuals pursue better economic opportunities, we would expect to observe positive net migration flows in cantons with relatively better initial living conditions (Le. lower poverty rates). In fact, there i s a negative and significant correlation betweenchangesinboth migrationmeasuresand 1990poverty levels (-0.43 and -0.40 for net migration and recent net migration, respectively). This implies that, to the extent that the average in-migrant has `worse' characteristics or endowments than the average resident in the area of destination, poverty increases in cantons with low poverty rates in 1990 were inpart the result of in- migration of relatively poorer individuals. Infact, recent migrants46 inurban areas are more likely to be poor than the rest of the population (see Chapter 3). 2.46 Rapidpopulation and poverty growth could pose important challenges for these areas, both in terms of employment and income generation and in terms of basic services provision. The analysis presented in Chapter 3 sheds some light on the first issue, while the second will be addressed in the upcoming Regional Study on Urban Poverty, in which data for Ecuador i s included, and the Ecuador UrbanPoverty Reduction Project. Changes inemployment and changes inpoverty 2.47 We discussed above that employment of both the household head and other household members was one of the main determinants of the poverty status of the household. We return to this issue here using a different perspective. We investigate whether changes in employment levels and the composition of employment are correlated with changes in poverty at the canton-level. For this purpose, we exploit the variation in the size of these different changes, rather than differences in trends, which are common to most cantons. 2.48 We classify cantons in two groups, according to population size (above and below the median size) since the nature of employment i s generally related to population density and agglomeration, and divide employment in three broad categories (agricultural employment, low- productivity non-agricultural employment, and high-productivity agricultural employment) following Elbers and Lanjouw (1997). We then correlate changes in the level of each type of employment and changes inpoverty separately for each group of cantons. 45The 2001 Population Census contains information on (i) of birth, (ii) residency, and (iii) of residency place current place five years ago for all individuals aged 5 and above. Usingthese three variables, we construct two indicators: net migration, defined as the percentage of the population living in a canton that was not bornthere (Le. (Population living in a canton - Population born in a canton) I Population living in a canton * loo), and net recent migration, defined as the percentageof the population living in a canton that was not residing there five years ago (i.e. (Population living in a canton- Population living there 5 years ago) I Population living ina canton 100). * 46We consider all individuals who migrated into a particular area in the three years prior to the survey to be recent migrants. 42 Figure 2.4: Poverty declineswere associatedwith decreasesinagricultural employmentand increasesin non-agricultural employment CANTONSIZE: BELOW MEDIAN CANTONSIZE: ABOVE MEDIAN Change in %of poor Change in% of poor 05 4 .% j -50 .30 .10 10 40 -20 -10 0 Change in % in agnc employment Change in % in agric. employment Running line smoother Running line smoother Change in %of pool Change in % of poor 85 45 5 -35 4 -16 Change in %in low prod0 non-ag 8 8 16 -16 Q Change in hin low prod.non-ag. 8 0 Running line smoother Running line smoother Change in% of poor Changein %of poor 851 85 45- 45 51 i 5 -35- -35 0 10Change in % in20 prod. non-ag high 30 40 I 10 30 Changein %in high prod. non-ag 20 Running line smoother Running line smoother Source: Authors' calculationsbasedondata from the 1999ECV and the 2001PopulationCensus. 43 2.49 We find that positive changes in the share of agricultural employment are correlated with increasesinpoverty. Since agricultural employment declined in 170of the 175 cantons for which this data exists and poverty increased in most cantons, this implies that cantons with the largest declines experienced the smallest increases in poverty. In contrast, positive changes in the share of low and highproductivity non-agricultural employment are correlated with decreases (or smaller increases)in poverty, although in the case of the latter this relationship only holds for larger cantons with a higher degree of agglomeration (Figure 2.4). 2.50 Although these correlations are only a first attempt to relate structural changes in local labor markets and employment distributionto changes inpoverty, they are suggestive of the important role these play in determining poverty levels and dynamics. Chapters 3 and 4 will explore this issue in depth for urban and rural areas respectively. CONCLUSIONS 2.51 We have shown in this chapter that poverty increased in urban areas in the Costa and the Sierra during 1990-2001, while it remained fairly constant in rural areas. These increases in urban poverty were the result of the particular nature of the 1999 crisis, which deeply hurt the urban middle-class, and the extensive in-migration flows throughout the decade, attracted by the better economic and social conditions offered by the areas. Increases in the number of urban poor, combined with increases in the size of the urban population, led to the urbanization of poverty over the decade. We have also argued that changes in the level and composition of employment go a long way in explaining changes in poverty. In particular, increases in the share of non-agricultural employment are correlated with declines in poverty. Given these results, we ask ourselves about the challenges aheadfor different regions and areas inthe country? 2.52 What are the challenges for urban areas? These areas will have to provide income and services to increasing numbers of people, a large fraction of which are poor. This can only be done through employment creation and improvements inthe provision of basic services and infrastructure. Chapter 3 examines the relationship between urban poverty and labor markets, as well as the constraints to employment creation. 2.53 What are the challenges for rural areas? As long as the income and poverty differentials between urban and rural areas remain as high as they are today, people will continue to gravitate towards the former, abandoning the latter. This in, turn, will increase the pressure on the already strained economy of urban areas. A large number of the rural poor still depend on the agricultural sector for survival, and most of them do not have access to land, or works on low-productivity land. Chapter 4 examines the determinants of agricultural productivity, and discusses policies to increase access to land among those who are landless. 44 3. URBANPOVERTY,LABORMARKET DYNAMICS AND FORMAL EMPLOYMENT CREATION 3.1 The last few years have witnessed significant increases in the number of people living in urban areas, as well as in the number of urban poor. As a result, despite the depth and severity of poverty being higher in rural than in urban areas, there are more poor people in urban areas. Thus povertyinEcuador i s becoming increasingly urban and will continue to do so at a rapidrate. 3.2 Employment constitutes the main, and frequently the only source of income for most families living in urban areas, so more often than not the lack of it leads to poverty. Labor income accounts for more than 90 (80) percent of total expenditure and more than 75 (80) percent of total income among poor (non-poor) households in urban areas. As a result, the focus of this chapter will be on labor markets and the capacity of the urban economy to generate employment and income (wages) and, therefore, reduce poverty. 3.3 The chapter will explore the relationship between labor markets and poverty from two different, complementary angles. The first part will take stock by focusing on the relationship between labor market outcomes and poverty, and the extent to which the impact of recent macroeconomic developments on the former have affected the latter. The second part will instead be forward looking, identifying potential constraints to employment creation, and examining the capacity o f poor households to access (high-quality) employment. 3.4 The main finding of the chapter can be summarized as follows: Urban labor markets were deeply affected by the 1999 crisis and the 2000 dollarization. Employment levels and real labor income plummeted as a consequence of the crisis, and did not recover back to pre-crisis levels until 2002. Urban poverty moved hand-in-hand with real labor income and employment, rising between 1997 and 1999, and then declining between 2000 and 2002. Poor households were more likely to be headed by a jobless or informally employed individual, and employment rates were lower and informality rates higher among members of poor households than among their non-poor counterparts. The relative demand for more educatedworkers increased during 1997-2002, although there were differences across sectors. In particular, the relative demand for tertiary educated workers increased inthe formal sector, and the relative demand for secondary educated workers increased inthe informal sector reflectingcross-sectoraldifferences inlabor productivity levels. Firing costs and non-wage labor costs appear to be the main reasons for low (permanent) employment creation, while, more generally, poor infrastructure, costly and scarce credit, and uncertainty about the economic and institutional environment appear to be the main constraints to business expansion. Employment creation i s positively correlated with labor productivity, and labor productivity i s positively correlated with the level of skills of the workers, access to foreign technology and exposure to international competition. 3.5 The chapter i s structured in five sections. The first section provides a detailed description of urban labor market developments during 1997-2002, relating them to macroeconomic events. The second section studies the relationship between urban poverty and labor markets, as well as household responses to labor market dynamics. The third section describes the evolution of the demand for and the returns to education, and examines the impact that these may have had on 45 poverty. The fourth section investigates the issue of constraints to formal employment creation in urban areas from the view point of firms, with a special focus on the investment environment and the role of skills. Finally the fifth section concludes. 3.6 The work presented in the first three sections is based on the Encuesta de Empleo, Desempleo y Subempleo 1997-2002, administered by the Instituto de Estadistica y Censos in Ecuador (INEC), while the fourth section relies on the Ecuador component of the Investment Climate Survey, administered by the Research Department at the World Bank. Both datasets are described in more detail in Box 3.1. Box 3.1: DataSources The empirical evidence presented and discussed in this chapter is based on two different data sources: the Encuesta de Empleo, Desempleo y Subempleo (EEDS), an employment survey administered by the Instituto de Estadisticas y Censos (INEC), and the Ecuador component of the Investment Climate Survey (ICs), a firm- level survey collected by the ResearchDepartment at the World Bank. The EEDS is collected annually in urban areas and contains information on the labor market status, employment characteristics and income of all individuals aged 10and above47.The survey experienced several changes during 1997-2002 that need to be taken into account when analyzing the data, the most important ones being (i)the adoption of a new sampling framework in 2002, basedon the 2001 Population Census, and (ii) the inclusion of self-consumption and in-kindpayments as components of labor income from 2000 onwards. The first change makes it impossible to construct time series in absolute value since the 1990 and the 2001 census-based sampling frameworks are not compatible. In practical terms this implies that we can provide a time series, say, for the annual share of the manufacturing sector in total employment, but we cannot provide a time series for the level of employment in the manufacturing sector during 1997-2002. The implications of the second change, as well as other methodological issues having to do with the income and poverty measuresused in this chapter are discussed in more detail in Box 3.2 below. The ICS-Ecuador was collected during 2002-03 and contains information different aspects of the production process for 450 firms in the urban manufacturing sector. The survey covers firms o f all sizes, as well as firms operating in the formal and informal sectors located in urban areas in the provinceso f Azuay, Guayas, Manabi, Pichincha and Tungurahua. MAINLABOR MARKET DEVELOPMENTURBAN IN AREAS,1997-2002 3.7 The macroeconomic events of the past few years have had a strong impact on urban labor markets. Inparticular, employment levels and composition, as well as real labor income levels were deeply affected by the 1999 crisis and the structural changes brought about by the dollarization. In this section, we examine the main labor markets trends for 1997-2002, paying special attention to changes inearnings across sectors and types of employment. Labor force participation, employment and unemployment. 3.8 Participation, employment and unemployment rates followed a marked cyclical pattern. Participation rates increased slightly in 1998/99 in response to the economic downturn, and continued to fluctuate afterwards around this level. These changes were common to both men and 47hformationon rural areas was also collectedin2000 and2001. 46 women. Employment and unemployment rates also followed a cyclical pattem. The former declined and the latter increased as a consequence of the crisis, returning to their pre-crisis levels by 2002. Employment rates were higher and unemployment rates lower for men than for women throughout the period, but exhibited similar cyclical patterns for both groups (Table 3.1). Table 3.1: Labormarkettrendswere sensitive to the 1999crisis andthe 2000dollarization 1997 1998 1999 2000 2001 2002 56.8 58.5 60.2 57.5 63.6 58.5 71.1 71.8 73.2 70.4 74.5 70.3 43.3 46.2 48.0 45.2 53.0 46.9 90.8 88.5 85.6 91.0 89.1 90.8 93.3 92.1 89.7 94.0 93.2 94.7 87.6 84.4 80.7 87.2 84.1 87.0 9.2 11.5 14.4 9.0 10.9 9.2 6.6 7.8 10.2 5.9 6.7 5.2 12.4 15.5 19.2 12.7 15.8 12.9 23.1 22.8 23.5 24.0 24.9 22.8 26.9 27.5 28.0 28.0 28.8 26.7 16.6 15.5 17.6 19.2 16.5 17.0 76.9 77.1 76.4 75.9 75.0 77.2 73.1 72.4 72.0 72.0 71.2 73.3 83.3 84.5 83.6 82.4 80.8 83.4 53.2 56.7 56.4 59.4 57.6 56.2 49.7 52.6 52.4 57.3 54.7 52.5 58.8 63.2 62.8 62.8 62.1 62.1 9.6 11.5 15.7 14.2 16.4 14.5 7.2 9.1 12.2 10.9 12.4 11.3 13.5 15.4 21.3 19.5 22.3 19.6 1.06 0.72 0.48 0.55 0.70 0.83 1.08 0.74 0.52 0.59 0.77 0.95 1.03 0.68 0.44 0.48 0.60 0.64 0.50 0.46 0.4 1 0.40 0.44 0.45 I month. Includesdomestic workers andworkers in the agriculturalsector (this category is only availablefor 2001 and2002). 47 3.9 Although these trends may suggest that the effect of the crisis and the dollarization was only temporary, other changes took place during the period that question this optimistic view. First, a large number of working-age adults migrated out of Ecuador, easing the pressureon the labor market (see Box 3.3). The fact that unemployment is lower than the national urban average in Cuenca and other cities that have experienced large out-migration i s only areflectionof this phen~menon~~. 3.10 Second i n f ~ r m a l i t yand~underemployment5' grew significantly as a result of the crisis and ~ the dollarization, and this increase was not entirely reversed afterwards - the informality and underemployment rates were 5 and 50 percent higher respectively in 2002 than in 1997. While increases in informality were similar for men and women, the male underemployment rate increased substantially more than the female one (56 percent compared to 45 percent) over the period. 3.11 All in all these changes suggest that the recovery of the employment rate described above was precarious. In fact, although total employment grew by 17 percent between 1998 and 2001, formal employment grew by only 11percent while informal employment grew by 24 percent. Labor incomeand wages5* 3.12 Remuneration to labor, inthe form of both labor income and wages, was also deeply affected by macroeconomic events52.Labor income halved between 1997 and 1999 as a consequence of the crisis, and then recovered by about 40 percent during 2000-2002 due to price stabilization associated with the 2000 dollarization. This recovery, however, was not large enough to bringreal labor income back to its 1997 level, particularly among women (Table 3.1). 3.13 The legal minimum wage also declined between 1997 and 1999, although less so than real labor income (20 percent versus 50 percent), and then recovered partially, reaching 1998 levels in 2002. That is, the real minimum wage followed a similar pattern to that of real labor income, but fluctuated less over the business cycle, which may have protected (formal) workers at the bottom of the wage distributionfrom further labor income deterioration duringthe crisis53. 3.14 Average real labor income changes, however, hide important differences between sectors and types of jobs. Although those workers with the highest (lowest) level of labor income in 1997 continued to be so in 2002, developments over the period involved some relative changes that is 48 Approximately 200,000 people left Ecuador in 1998l2001, doubling the stock of Ecuadorians residing and working abroad. Calculations presentedby the SIISE (2002b) show that the unemployment rate could be between 0.5 and 2 percentagepoints higher than it i s today had these migrants remainedinthe country. 49 The informal sector includes salaried workers in firms with 10or fewer employers, and self-employed individuals (exceptthose inprofessional and technical occupations). 50 Underemployedindividuals are those who desire to work 40 hours or more a week, but cannotdo so due to lack of labor market opportunities. 51 The analysis in this chapter is based on labor income, rather than on wages (Le. labor income for those who are salaried workers). Although we understandthat information on labor income is usually less precisethan information on wages, our choice is based on the premise that salaried workers represent only 50 percent of all employed individuals, and only 40 percent of the employed poor. Hence by working with labor income, we ensure that our results are a more accurate representation of the experience of most Ecuadorians living in urban areas, at the same time that we retain the option to present information disaggregated by employment status, distinguishing between salariedand non-salariedindividuals, whenever necessary. 52 See Box 3.2 for a detailed description of the methodology used to construct the (labor) income variables used in this chapter. 53 Cunningham(2003) argues that minimum wages are binding in the formal sector in Ecuador. 48 important to consider. Informal workers saw their real labor income decline significantly more than that of formal workers as a consequence of the crisis, most likely due to weak enforcement of minimumwage and other labor regulation in the sector, but also experienced a faster recovery inthe years following. Anecdotal evidence suggests that this recovery was associated with upward rounding in prices, which appears to have been larger and more common in the informal sector than inthe formal sector immediately after dollarization(Table 3.2). 3.15 Average real labor income in the tradable sector was lower than that of the non-tradable sector, and this difference remained more or less constant over the period. This evidence contradicts the commonly-held belief that changes in the relative price of tradable and non-tradable goods are largely driven by increasing differences in the cost of labor between both sectors. Table 3.2: The evolutionof reallabor incomevaried across sectors... Ratio Formal Informal Informalto formal 1997 1.32 0.82 0.62 1998 0.99 0.50 0.5 1 1999 0.69 0.32 0.46 2000 0.77 0.39 0.50 2001 0.94 0.52 0.55 2002 1.04 0.55 0.53 199911997 0.52 0.39 200211999 1.51 1.75 200211997 0.78 0.68 Ratio Non-TradableSector Tradable Sector Tradableto non-tradable __.__-_________ 1997 1.11 0.96 0.86 1998 0.72 0.75 1.04 1999 0.5 1 0.45 0.88 2000 0.56 0.52 0.92 2001 0.73 0.66 0.90 2002 0.80 0.73 0.9 1 199911997 0.45 0.47 200211999 1.56 1.62 200211997 0.72 0.76 Source: Encuestade Empleo, 1 sempleo y Subempleo (INEC), 1997-2002. A Realhourly labor income expressedinUS$2000: Tradable sector includesagriculture,mining andmanufacturing.Non-tradablesector includesconstruction,commerce, transportationandcommunications,FIRE andthe public sector. 3.16 Similarly the evolution of real labor income varied significantly with employment status. Public sector workers, the self-employed and, particularly, domestic and family-business employees experienced real labor income losses of 50 percent or more between 1997 and 1999, while private sector salaried workers and employers suffered more moderate declines of 30 and 40 respectively (Table 3.3). The speed of recovery, measured by the ratio between real labor income in 2002 and 1999, was also different for different groups, with domestic and family-business workers and public sector employees fairing best, followed by employers, private sector workers, and the self-employed. 49 3.17 A comparison of average real labor income in 2002 and 1997 for these different groups of workers shows the highest increases among public workers and employers, and the smallest among salaried workers in the private sector. In addition, labor income volatility (defined as the ratio of the variance to the mean for the period) was highest for public sector workers, followed by employers and those in the informal sector, whereas it was lowest among salaried workers and the self- employed in the formal sector. Table 3.3: The evolution of real labor income also varies acrosstypes of employmentA All employed Public Private Self-employed Em iloyer Other 1997 1.31 0.99 0.96 .82 0.5 1 1998 0.98 0.67 0.61 .37 0.25 1999 0.65 0.44 0.41 .04 0.14 2000 0.7 1 0.5 1 0.51 .49 0.36 2001 0.8 1 0.67 0.69 .47 0.48 2002 0.98 0.72 0.68 .79 0.43 199911997 0.49 0.44 0.42 0.57 0.27 200211997 0.74 0.72 0.70 0.98 0.84 200211999 1S O 1.63 1.65 1.72 3.07 Formal Public Private Self-employed Employer Other 1997 1.31 1.13 1.78 1.82 1998 0.98 0.83 1.36 1.37 1999 0.65 0.58 1.03 1.04 2000 0.71 0.68 1.28 1.49 2001 0.81 0.84 1.25 1.47 2002 0.98 0.91 . 1.23 1.79 199911997 0.49 0.50 0.57 0.57 200211997 0.74 0.81 0.93 0.98 200211999 1.50 1.56 1.19 1.72 Informal Public Private Self-employed Employer Other 1997 0.78 0.90 0.51 1998 0.48 0.57 0.25 1999 0.28 0.38 0.14 2000 0.33 0.45 0.36 2001 0.48 0.65 0.48 2002 0.49 0.64 0.43 199911997 0.36 0.42 0.27 200211997 0.67 0.79 0.84 200211999 1.75 1.68 3.07 Source: Encuestadc mpleo, Desempleoy Subempleo (INEC), 1997-2002. Includesdomestic workers andworkers inthe agricultural sector (this categoryis only available for 2001 and 2002). 50 Table 3.4: Real labor income varies with sector and type of employment. Dependent variable: (log) hourly real income Model (1) Model (11) Model (111) 1997-2002 1997-2002 1997 and2002 Female -0.186" -0.187" -0.173" (0.007) (0.007) (0.014) Primary 0.184*' 0.184*' 0.178" (0.020) (0.020) (0.041) Secondary 0.489" 0.489** 0.459** (0.021) (0.021) (0.043) Tertiary 0.940** 0.939" 0.888** (0.022) (0.022) (0.045) Experience 0.027** 0.027" 0.028** (0.001) (0.001) (0.001) Experiencesquared -0.001** -0.001" -0.001*' (0.000) (0.000) (0.000) Private -0.114" -0.096" -0.054** (0.012) (0.012) (0.03 1) Self-employed 0.047** 0.011" 0.148** (0.014) (0.025) (0.040) Employer 0.268*' 0.280'* 0.356** (0.016) (0.016) (0.045) Other -0.208" -0.203" -0.587** (0.017) (0.027) (0.231) Private* informal -0.264** (0.052) Self-employed * informal -0.197** (0.056) Informal -0.301" -0.51 -0.297** (0.009) (0.051) (0.028) Private2002 -0.059*' (0.040) Self-employed2002 -0.155" (0.051) Employer2002 -0.123** (0.057) Other 2002 0.449' (0.234) Informal2002 0.073" (0.034) Industrydummies Yes Yes Yes Occupationdummies Yes Yes Yes Year dummies Yes Yes Yes Regiondummies Yes Yes Yes Numberof observations 71,881 71.881 17,435 A fourth modelincludingtriple interactionsbetweenthe private sector, self-employedandother sectordummies, the informalsector dummy and the year 2002dummy was estimated.We do not report the resultsbecausethey are very imprecise due to the small numberof observation in some of the subgroups. 51 3.18 These differences reflect the flexibility (or lack of) that characterizes each one of these sectors, as well as the various institutional arrangements that regulate them and the incidence of informal employment. They also reflect demographic differences across workers in each group, and the fact that such characteristics (e.g. education and experience) may command different returns in the labor market. We use regression analysis to disentangle the effect of some of these variables. In particular we regress the logarithm of real hourly labor income on a set of individual and job characteristics (Table 3.4). 3.19 More educated workers earn higher real labor income, as do more experienced workers54. Women and Afro-Ec~adorians~~,however, make about 20 percent less than their counterparts in the . form of real labor income, even after controlling for demographic differences and type of employment. Finally, although not reported, real labor income appears to be 7 percent higher in the Sierra than in other regions, even after controlling for the presenceof the government inQuito. 3.20 In addition, the models in the first two columns confirm the differences in labor income levels across employment status, as well as the fact that workers in the informal sectors have lower labor income irrespective of their employment status in 1997-2002. Informal workers earn 30 percent less than their formal counterparts after we control for demographic differences (column I), and 30 percent less after we control for the employment composition of the sector (column II). Similarly, real labor income i s highest among employers and public sector workers and lowest among the self- employed and the informal salaried (column II). 3.21 The third column examines differences between formal and informal workers over time by comparing labor income levels in 1997 and 2002 for different types of workers. The results show that informal workers were indeed able to recover faster than formal workers (making a relative gain of 7 percent) and, therefore, were relatively better off in 2002 than in 1997. Their income, however, continued to be lower than that inthe formal sector since the starting difference between both groups was well above 7 percent. URBANPOVERTYAND LABORMARKETS 3.22 How did employment and real labor income changes affect overall poverty? Are the poor more likely to be employed in certain sectors or types ofjobs than in others? And, if so, what are the consequences? In this section we provide answers to these questions combining a dynamic and a static approach. First, we examine the behavior of employment, real labor income, household income and poverty over time, and the extent to which they appear to be correlated. Second, we analyze the role that individual and household-level labor market outcomes play as determinants of poverty, paying special attention to the distribution of the poor across sectors and types of employment and the potential impact that changes inrelativereal labor income across these may have had on the poor. 54 The implied returns to education are 3, 8 and 20 percent per additional year of primary, secondary and tertiary education, respectively, compared to a worker with no studies. These figures are within the ballpark for Latin American countries. 55Results for Afro-Ecuadorians based on the same models estimated with data for 2001 and 2002 only, and not reportedhere. The coefficient on the equivalent dummy variable indigenousethnicity was positive but insignificant. 52 Household income and urbanpoverty trends 3.23 Changes in household-level labor income, due to changes in individual employment and real labor income, explain most of the variation in total household income per capita during the period, and hencemost of the variation inpoverty (Figure 3.1.) 3.24 Poverty trends in urban areas duringthe 1997-2002period mimicked national poverty trends. Poverty increased significantly in Ecuador as a consequence of the 1998199 crisis and urban areas were no exception (Figure 3.2). This increase, however, was overturned by economic recovery in 2000-2002. Measuresof the depth and severity ofpoverty followed a similar cyclicalpatted6. 3.25 In addition, poverty trends were similar across Ecuador's three regions so that differences in poverty levels neither increased nor decreased significantly during the period (Figure 3.1). Poverty continued to be highest in the Costa and lowest in the Sierra, mainly as a result of higher rates in Guayaquil than inQuito, as seen in Chapter 2. Figure 3.1: Changes inemployment andlabor income drive changes inhousehold income per capita 90 1.1 80 0.9 // 70 s 60 0.7 5E L a L r 5 50 n ifi 0.5 u) u) a 40 3 2 0 0 0 0 0 N N 0 30 0.3 20 0.1 10 0 -0.1 1997 1998 I999 2000 2001 2002 Source: Authors' calculationsusingEEDS 1997-2002(INEC). 56For a detailed analysis of urban poverty in 1990-98,see Le6n and Vos (2000). 53 Figure 3.2: Poverty followed a cyclical patternin 1997-2002 0.50 0.45 0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 1997 1998 1999 2000 2001 2002 0.60 0.50 -8 0.40 0- a p8 0.30 r0 f g 0.20 0.10 0.00 1997 1998 1999 2000 2001 2002 1 Amazonia v Costa A Sierra -All] Source: Authors' calculations using EEDS 1997-2002(INEC). 54 Box3.2 Constructionof Urban(Labor) IncomeandPovertyMeasures. This box briefly describes the methodology used inthis chapter to construct time-consistent real income measuresand poverty measures usingdata from the Encuestade Empleo, Desempleo y Subempleo. Real(Labor)IncomeinUrbanAreas, 1997-2002 The Encuesta de Empleo, Desempleo y Subempleo (EEDS) contains information on labor and non-labor income for all individuals aged 10 and above. However, the way in which this information i s collected has changed during 1997-2002- in particular, more detailed questions on non-wage components of labor income and on various components of non-labor income were included in the survey from 2000 onwards (Table B3.2.1). TableB.3.2.1 Incomeinformationinthe Encuestade Empleo,Desempleoy Subempleo 1997 1998 1999 2000 2001 2002 Labor income Primary wage incomeA, X X X X X X Consumptionof own production A X X X In-kindpayments X X X Secondary wage income A, X X X Secondary non-wage income X X X Tax deductions and other contributions X X X Non-labor income Capital income X X X X X X Pensions X X X X X X Remittances X X X Bono Solidario X X X Other non-labor income X X X A Employers and self-employed Salaried workers and home servants Because new, more detailed information becomes available over time there i s a trade-off between time consistency and completenesswhen constructing an income measure. Moreover, data accuracy appears tc be higher for labor income items than for non-labor income ones (based on comparisons with the 1999 Survey of Living Conditions). Taking these caveats into account, and after running a series of robustness and sensitivity tests, we decided to use labor income, both monetary and in-kind, as our income measure for the period. This measure captures approximately 85 percent of total reported income, so that results are not qualitativelj sensitive to this choice, and i s highly correlated with the income measure officially used by the National StatisticalInstitute in Ecuador (INEC) . (see next page) 55 Box3.2 Constructionof Urban(Labor)IncomeandPoverty Measures(Continued). PovertyinUrbanAreas, 1997-2002 In the absence of an official poverty rate for Ecuador, and in order to ensure internal consistency across different chapters of the report, the poverty line for urban areas was chosen so as to obtain poverty rates identical to those calculatedwith the 1999 Survey of Living Conditions (Chapter 2). Poverty rates for the rest of the period are then calculated by expressingincome in 2000 US$ terms, and comparing it to the poverty line. Who are the urbanpoor?The Roleof Labor Market Outcomes 3.26 We examine here the characteristics of the urban poor and the determinants of urban poverty, paying special attention to the degree of labor market attachment of the household as a whole and, in particular, of the household head (Table 3.5). 3.27 Characteristicsof the urban poor. Poor households are significantly larger than non-poor ones and, on average, income earners in the former have to provide for a larger number of household members than their counterparts in the latter. The number of dependents, defined as individuals under 10 or above 65 and out of the labor force, is 50 percent higher in poor households. These households also exhibit lower labor force participation rates, especially among women. Less than 50 percent of those potentially active are actually so inpoor households, compared to 63 percent innon- poor ones (equivalent numbers for poor and non-poor men and women are 67 and 73 percent, and 37 and 53 percent). Similarly employment rates are lower and unemployment rates higher among active members in poor households than among the rest of the active population - 80 and 20 percent respectively inpoor households, compared to 92 and 8 percent innon-poor ones. 3.28 In addition informality and underemployment rates are higher in poor households. Sixty percent of those who are poor and employed have jobs in the informal sector, while this number i s only 47 for the non-poor. Similarly underemployment rates, defined as the fraction of workers who would like to work 40 or more hours a week but cannot do so, are 17 and 10percent among poor and non-poor workers respectively. 3.29 The same observations can be made regarding poor and non-poor household heads, who are normally viewed as the main earner in the household. Heads of poor households are less likely to be active, or employed, and, conditional on having ajob, more likely to work inthe informal sector and to be underemployed. As a reflection of the latter, they work fewer hours on average than their counterparts in non-poor households and hold fewer jobs. 3.30 Moreover, individuals heading poor households are more likely to be self-employed and less likely to be public salaried workers or employers than heads of non-poor households. In particular, 40 percent of poor household heads are self-employed, compared to 30 percent of non-poor ones. Similarly only 4.5 percent of poor household heads hold salaried positions in the public sector and 5.5 percent are employers, while these numbers are 15 and 11 percent respectively for their counterparts innon-poor households. 56 attachment and labor market outcomes Poor Non-poor Poor Non-poor 1997 1999 2002 1997 1999 2002 Householdcharacteristics Household size 4.5 4.2 4.5 4.8 4.1 4.5 4.2 4.1 Dependencyratio 32.4 19.5 33.5 30.6 36.3 20.0 19.0 19.7 Household LFFrate 48.5 67.2 38.6 52.3 42.7 63.9 69.2 66.7 Household female LFPrate 37.3 53.3 29.1 39.2 34.1 47.9 56.1 52.8 Household employment rate 82.3 94.4 81.2 79.4 80.8 94.8 93.0 95.4 Householdunemployment rate 17.6 5.5 18.7 20.5 19.2 5.1 6.9 4.5 Household informality rate 72.0 52.1 70.0 75.5 71.2 50.8 50.2 53.7 Householdunderemployment rate 17.9 10.5 14.8 18.7 19.8 7.4 11.6 11.7 Costa 60.6 52.8 60.6 57.5 57.6 57.8 51.7 52.2 Sierra 37.8 45.3 37.7 40.6 40.1 40.7 46.5 45.5 Amazonia 1.5 1.7 1.5 1.7 2.1 1.4 1.7 2.1 Householdhead characteristics LFPrate 71.8 92.5 62.1 77.0 64.9 92.1 93.5 92.1 Employed 89.0 98.4 90.1 86.2 85.5 98.3 97.6 98.4 Unemployed 10.9 1.6 9.8 13.7 14.4 1.7 2.3 1.5 Public sector 4.5 14.9 3.3 4.0 5.4 17.0 16.3 13.4 Private sector 35.6 37.4 32.2 42.3 37.7 35.1 38.7 37.9 Employer 5.5 11.3 4.7 8.9 5.3 13.2 15.6 14.0 Self-employed 42.2 30.1 52.4 41.4 43.4 32.4 27.6 30.8 Informal 70.8 46.6 73.9 67.5 72.7 45.5 41.8 49.1 Underemployed 18.3 7.9 16.0 17.4 24.5 5.2 8.2 10.0 Multiple jobs 2.5 4.4 2.4 2.4 1.6 4.6 4.9 3.2 Weekly hours worked 41.8 46.2 41.7 42.8 38.3 46.7 46.8 44.8 No studies 7.8 3.4 7.7 7.1 8.1 3.4 3.3 3.3 Primary 50.6 35.9 54.8 51.7 48.6 37.2 34.1 35.5 Secondary 30.9 34.8 29.2 31.2 31.2 35.2 34.8 35.7 Tertiary 10.4 26.6 8.2 9.7 12.0 23.9 27.6 25.4 Years of schooling 7.4 9.7 7.0 7.4 7.3 9.6 10.0 9.5 Gender (Female= 1) 28.8 16.9 32.6 25.7 33.9 15.6 16.5 18.2 Indigenous 5.5 5.3 Afro 4.3 3.0 Migrant 46.6 49.7 54.9 52.8 40.6 51.9 52.7 45.7 Recent migrant into area (53 years) 3.9 4.1 5.1 4.7 3.5 3.6 5.1 4.5 Source: Authors' calculationsbasedon ata from EEDS, 1997-2002(INEC). Labor force participation,employmentandunemployment ratesare calculatedfor those above age 10following INEC. A All salariedindividuals working in firms with 10or fewer employeesand all those who are self-employedare considered informal. All differences inmeansbetweenpoor andnon-poorare significantat the 5 percentlevel, with the exception of `indigenous' 57 3.3 1 Average labor income differs across sectors of employment, and these differences, combined with lower employment rates among members of poor householdshelp explain to a large extent who i s poor and who i s not. For instance, in 2002 a public sector worker earned 40 percent more than a private sector salaried worker and 60 percent more than someone who was self-employed, while the average employer earned more than double the salary of a public sector worker. Similarly mean labor income for those inthe formal sector was double that inthe informal sector. 3.32 As already discussed in chapter 2, heads of poor households tend to be less educated, and slightly older than those of non-poor ones. They are also more likely to be women57, indigenous58 or Afro than those heading non-poor households. Maybe surprisingly, the fraction of poor households that i s headed by a migrant or a recent migrant59i s slightly smaller than the corresponding fraction for non-poor households. Box 3.3 Searchingfor New Economic Opportunities: Indigenous and Afro Population inUrban Areas Over the past two decades urban areas havereceiveda steady flow of indigenous and Afro rural migrants in search of the economic opportunities and social mobility offered by cities. Limitedaccess to land and low income levels appear to be the reasons behind the decision to migrate either seasonally or permanently (see Chapter 4). The largest flows of out-migration are registered in those provinces where the indigenous (and Afro) population i s largest, while the provincesof Guayas andPichincha are the most favored destinations leading to high concentration of indigenous and Afro populations in Guayaquil and Quito. "Ihavecometothecity duetothesituation back in thecountiy...Land ownersandintermediariesdonotallow usto prosper there. Most of us have migrated because working in the agricultural sector is a dead end" Cited in Le6n et alia (2003) Employment. Upon arrival to the city these groups do not seem to be sorting themselves out into particular sectors or industries.Infact the distribution of indigenousandAfro workers across industries in urban areas is not very different from that of the rest of the population, and neither i s their employment status (Tables B.3.3.1 and B.3.3.2). However, low levels of educationamong these groups severely limit their ability to access skilled employment, so that most men are employed in low-skill occupations and most women work as domestic help. (see next page) The fact that female-headed households are more likely to be poor in contrary to the findings in Chapter 2. Three factors can potentially explain this difference: (i)the use o f different datasets, (ii) the use of different poverty measures (income based here and consumption based in Chapter 2, (iii) the differences in the periods covered (1997- 2002 here and only 1999 in Chapter2). 58Information on ethnicity was only collected in 2001 and 2002. The indigenous and Afro indicators are constructed based the language(s) spoken by the individual. These percentages, however, may not reflect the true proportion o f the indigenous and Afro groups inEcuador. 59Migrated into the area less than three years ago. 58 Box3.3 Searchingfor New EconomicOpportunities: IndigenousandAfro PopulationinUrbanAreas (Continued) Moreover the incidence of informality and temporary employment i s higher among indigenous workers than among the rest of the population. This is particularly true inthe commerce sector where about a third of all indigenous workers are employed - anecdotally, in Quito and Guayaquil activities such as street or small market vending are often referred to as "trabajo de indios". Lower wages and precarious employment, however, do not seem to constitute large enough obstacles to deter migration into urban areas, Table B.3.3.1 Industryof Employmentdoesnotvary muchwithethnicity ... Construction 11.2 8.4 7.7 Commerce 28.5 22.2 29.4 Transport 5.8 4.7 6.7 FIRE 4.1 3.7 5.2 Other 26.0 36.4 27.2 Table B.3.3.2 EmploymentStatusandIncidence of Informalityby Ethnicity Indigenous Afro Other Salaried -public sector 8.8 10.3 10.6 Salaried -private sector 38.7 42.0 40.2 Self-employed 27.4 22.7 27.5 Employer 5.3 3.2 5.8 Other 19.7 21.7 15.7 Informal 62.1 57.3 56.0 Migration and copying strategies. Indigenous and Afro groups are often confronted with racial and social discrimination upon their arrival in the city, which makes it difficult initially to find and keeF accommodation andor jobs. For these reasons social and family networks and connections are extremely important during the early stages of the migration process. These networks provide recent migrants with food, housing, and, if at all possible, employment. "Iquittedmyjobasadomesticworker becausethepay wastoolow, andmy cousintoldmetherewasworkfor me where she was employed" Cited in Le6net alia (2003) As individuals and families stay in urban areas longer, they become to some extent `assimilated' to the prevalent culture, and this lessens the effect of discrimination. In fact second and third generation indigenous migrants tend to fair significantly better than first generation ones. a way thatpeople understand.They makefun of you if you are wearing the `chalina'...and, actually, it is too hotfor "At the beginning I would wear the traditional costumes, but the city demands some different... You have to speak in this type of clothing here... You change a bit, and then a bit more, until in the end nothing is the same" Cited in Le6nat alia (2003) 59 3.33 Determinants of poverty. The discussion so far has focused on a single characteristic of the household or the household head at a time, emphasizing the extent to which differences betweenpoor and non-poor households and individuals exist. A more rigorous way of performing a similar exercise, taking all these characteristics into account at the same time, is to study the determinants of the probability that a certain household i s poor using regression analysis. Table 3.6 presents the results from a series of probability models estimated in this spirit. Since information on ethnicity i s only available in 2001 and 2002, we present the general results in columns (I) and (11) and separate results for these two years in columns (111) and (IV). 3.34 The regressions consider both household and household head characteristics as potential predictors for poverty, and again focus on labor market related outcomes. We present two models: the first one, in columns (I) and (111), only considers whether the household head i s employed or not, whereas the second one, in columns (11) and (IV), explores the effect of different employment Characteristics on poverty. Both models confirm to a large extent what has been pointed out so far, namely that employment i s negatively correlated with poverty, and that, among those employed, family business and domestic workers are the most likely to be poor. Other things being equal, employment of the household head decreases the probability that the household i s poor by 13 percentage points. This effect, however, varies with the type of job the household head holds - i.e. while working in the public sector decreases the probability of being poor by 14 percentage points compared to family business and domestic workers, being self-employed does so by only 5 percentagepoints. 3.35 Higher informality and underemployment are positively correlated with poverty at the level of the household. The same i s not true, however, for household heads - once the household informality rate i s taken into account, informal employment of the household head i s in fact negatively correlated with poverty. This is due to the fact that, while most informal secondary earners are salaried workers in the private sector, the majority of household heads in the informal sector are self-employed and this group earns about 50 percent more than informal salaried workers on average. Although this result may seem somewhat counterintuitive at first, it i s the reflection of the very varied nature ofjobs in the informal sector and, thus, the level of labor income associated with. 3.36 Finally the effect of demographic characteristics on poverty i s also similar to that described above. Households headed by less educated individuals, recent migrants into the area or women are more likely to be poor. The effect of ethnicity i s less clear - households headed by indigenous or Afro-Ecuadorian heads are more likely to be poor, but the effect is not significant in the case of the former. Perhaps surprisingly, after controlling for demographic andjob characteristics, there are not differences across regions in terms of the probability of being poor, despite the fact that real labor income appear to be slightly higher inthe Sierra (Table 3.6). 60 Table 3.6: Labor M a ket and DemographicDeterminants of PovertyinUrbanAreas Dependentvariable Poor household(=1). (1) (11) (111) (IV) Probit Probit Probit Probit 2001/02 2001/02 Household size 0.012 0.018 0.005 0.005 (0.001) (0.001) (0.002) (0.002) Dependencyratio 0.374 0.374 0.291 0.293 (0.010) (0.010) (0.017) (0.017) Employment rate -0.659 -0.664 -0.590 -0.603 (0.010) (0.010) (0.017) (0.017) Underemployment rate 0.162 0.137 0.156 0.129 (0.007) (0.011) (0.012) (0.018) Informality rate 0.189 0.201 0.127 0.148 (0.005) (0.008) (0.010) (0.014) Employed -0.130 -0.123 (0.008) (0.014) Public -0.144 -0.105 (0.006) (0.011) Private -0.088 -0.099 (0.006) (0.009) Self-employed -0.063 -0.057 (0.006) (0.010) Employer -0.106 -0.082 (0.007) (0.013) Informal -0.051 -0.050 (0.007) (0.012) Underemployed 0.027 0.022 (0.010) (0.018) Primary -0.053 -0.048 -0.067 -0.060 (0.009) (0.008) (0.016) (0.016) Secondary -0.117 -0.107 -0.119 -0.106 (0.008) (0.008) (0.015) (0.016) Tertiary -0.162 -0.151 -0.138 -0.128 (0.007) (0.007) (0.014) (0.014) Female 0.061 0.056 0.054 0.051 (0,005) (0.005) (0.009) (0.009) Migrant -0.043 -0.043 -0.072 -0.074 (0.004) (0.004) (0.007) (0.007) Recent migrant 0.036 0.034 0.016 0.014 (0.012) (0.011) (0.020) (0.020) Indigenous 0.026 0.029 (0.016) (0.016) Afro 0.054 0.052 (0.022) (0.021) Year dummies Yes Yes Yes Yes Regiondummies Yes Yes Yes Yes Number of observations 49,524 Source: Authors' calculationsusing :EEDS, 1997-2002(INEC). **49,524Significantlv 14,562 14,562 (*) different from zero at the 5 (10) oercent level. The referencehouseholdhasthefollowing characteristics: householdGcatedin;he Orienteregion,with a mak hokehold headwith no studies, under 25 years of age, non-migrant,andnon-employed.Note: Reportedcoefficientsrepresentthe marginal effect of each variable on the probability of beingpoor. 61 Labor ForceParticipation RateRatios by GenderandAge Male Female Poor Non-Poor Poor Non-Poor 199911997 1.28 1.04 1.34 1.16 200211999 0.82 0.95 0.86 0.94 Labor ForceParticipation Rate Ratiosby Gender and Age Male Female Age Group Poor Non-Poor Poor Non-Poor I 199911997 <25 1.59 1.10 1.40 1.23 200211999 I 0.72 0.90 0.69 0.86 199911997 25-45 1.01 1.oo 1.21 1.11 200211999 0.96 0.99 1.01 0.96 199911997 >45 1S O 1.01 1.49 1.16 200211999 0.74 0.98 0.84 1.02 Because the nature of available jobs varies with the business cycles, and generally deteriorates durin; times of crisis, new labor market entrants in 1998/99 were likely to findjobs that were `worse' than thl average existing job, hence undermining the effectiveness of increased labor force participation as ; coping strategy. This could explain the observed increase in the informality rate between 1997 and 199! among poor households, whose labor force participation rates increased the most, and its subsequen decrease to almost 1997 levels by 2002. 62 I Box3.4: Copingstrategiesof the urbanpoor (Continued) Moreover, given the poor performance of urban labor markets during the crisis in terms of employment creation, the likelihood that new labor market entrants would indeed find ajob, even a low-quality, low- paying one, was fairly small. Falling labor income and scarce employment opportunities at home were then the main reasons behind the significant increase in out-migration that occurred during and after 1998/99, as attemptingto find ajob abroadbecamea common coping strategyfor both poor andnon-poor households(see Box 3.5). Unfortunately the available employment surveys do not collect data on out-migration, and information on whether the household receives (domestic and international) remittances is only recorded for 2001 and 2002. According to this limited information, in 2001 14 percent of poor householdreceivedremittances, comparedto 7 percent of non-poor households, and these figures were 26 and 8 percent respectively in 2002. Given that remittances already representedabout 10 percent of total income in poor householdsin 1999 , it becomes clear that these resources have played an increasingly important role in containing further raisesinpoverty. 3.37 When we combined the lessons from this and the previous section, we find both reasons for optimism and despair. The poor tend to be employed in the informal sector, and as a consequence they are more likely to have smaller levels of labor income, more precariousjobs and a lower degree of labor market attachment than their non-poor counterparts. On the other hand, the more dynamic and flexible nature of the informal sector led to this sector being the sole generator of employment in the economy during 1997-2002, as well as to a more rapid recovery in real labor income among informal workers after the crisis. Two facts that could have potentially contributed to both prevent poverty rates from climbing to even higher levels during the crisis, and bring poverty rates down faster afterwards. 3.38 However, the margin for sustained labor income increases in the informal sector, on which further poverty declines driven by informal sector growth would be dependent, is limited since the sector is traditionally characterized by low levels of labor productivity and capital investment. As a consequence, we need to turn our attention to productivity increases, employment creation (formal and informal), and the capacity of the poor to benefit from both as the main determinants of future decreases inpoverty. 3.39 We turn to these issues in the last two sections of this chapter. In the next section, we consider the fact that the poor are generally less educated than the non-poor, and examine whether changes in the demand for skills may impact their capacity to access better jobs. In the last section, we discuss constraints to employment creation inthe manufacturing sector usingfirm-level data, and study the extent to which actual employment creation i s correlated with labor productivity. In doing this we pay special attention to differences betweensmall (maybe informal) and large businesses. 63 Box 3.5: Internationalmigration Out-migration flows increased significantly during the 1999 crisis, as large numbers of people sought better economic opportunities abroad, and have remained high since (Figure B.3.4.1). According to data collected at customs approximately 200,000 people left the country between 1999 and 2001, compared to half that figure between 1992 and 1998, and, although some of them returned back, the difference between registered exits and entries increased significantly during 1999-2001. Out-migration was more prevalent in some areas and among certain population groups, and was mainly directed to Spain, Italy and the US. Almost half of those who left the country since 1990 resided in the province of Azuay, compared to 6 percent from El Cafiar and Pichincha, the next largest sources. Similarly most migrants were young adults, both male and female, with medium levels o f education and skills. Figure B.3.5.1: Out-migration flows and foreign remittances increased after 1999 600,000 1,600 -- 1,400 500,000 -- 1,200 -m 400,000 i -- 1,000t a 3 YJ0,ooo -- 800 2 $cv 9 a -- 600 E 200,000 -- 400 100,000 -- 200 0 I O 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 [-Exits Entries-Remittances ( US$l,OOO) 1 The amount of money sent back to Ecuador in the form of remittances increased with the flows o f out-migrants, anc foreign remittances constitute today the second most important revenue item in the balance of payments, after oi exports. These resources were invested mainly in housing and the purchase o f durable goods, such as cars anc appliances (Banco Central del Ecuador, 2002d). Not surprisingly, the distribution o f these resources across provinces closely resembles that of the out-migrants s( that their impact has not been uniform across the national territory. Although the question has so far not beer analyzed in depth (due to lack of appropriate micro data), the sheer magnitude of these transfers indicates they mus have had a positive effect on the income level o f recipient households and areas, and hence on poverty. For instance they flowed mainly towards rural areas, which may explain why poverty rates did not increased as much as urbar poverty rates during the 1999 crisis. Second, the exit of large numbers o f potential participants undoubtedly ease( the pressure on local labor markets at a time when job creation was low and unemployment high. In fact consumption and price levels are high and unemployment levels low in those areas that experienced the largest out migration flows, compared to the rest of the country. Finally, given that out-migration was concentrated among young adults, likely to have left small and teenage children behind, it is possible for this phenomenon to have lasting social effects in the future that are hard to measure today. For instance, the number of households where the intermediate generation (parents) is missing has increased over time so that growing numbers of children and teenagers are being bought up by their grandparents. I t i s yet an open question whether this will have an impact on schooling and other outcomes. In addition, there i s anecdotal evidence o f a rise in the number of teenage pregnancies and of increases ingang violence. 64 LABOR INCOME AND THE DEMANDFOR SKILLS6' 3.40 More educated workers are generally considered to be more productive than their less educated counterparts, and hence are paid more on average. However, wage differences between workers with different education levels vary across time and across sectors, and these changes have the potential to affect both the level and composition of poverty. This section examines changes in the demand for and supply of individuals with different levels of education in urban areas and their effect on relative wages by sector and education group, and then tries to relate the evolution of the latter to changesinpoverty. 3.41 Highly educated individuals saw their real labor income decline relatively less than that of others during the 1998/99 crisis, and by 2002 their earnings had almost recovered to 1997 levels (Table 3.7, panel A). Although this relative advantage of more educated workers during economic downturns has frequently been emphasized in the literature, it i s at odds with existing evidence for other countries inthe Latin American region during the 1990s (World Bank, 2003~). Table 3.7: More educatedworkers havehigher labor incomeacross sectors A.- Over time Primary Secondary Tertiary 1997 0.77 0.94 1.73 1998 0.45 0.69 1.23 1999 0.29 0.42 0.97 2000 0.35 0.47 1.05 2001 0.48 0.59 1.24 2002 0.49 0.68 1.34 199911997 0.38 0.44 0.56 200211999 1.69 1.62 1.38 200211997 0.64 0.72 0.77 I B.- Across sectors I Tradable Non-Tradable Primary 0.5 1 0.45 0.62 0.42 Secondary 0.65 0.5 1 0.80 0.54 Tertiary 1.51 0.87 1.32 0.95 3.42 Both real labor income for all education groups and the returns to education, measuredby the coefficients of education dummies in an earnings regression, are lower in the informal than in the formal sectors61. Incontrast, once differences in the incidence of informality are taken into account, ~~~ ~~ analysis in this section does not include results for those with no studies since their number is very small - about 3 ' 6o Although individuals are divided into four education categories (no studies, primary, secondary and tertiary), the ercent of all individuals above 10years o f age (5 percent among the poor). Inparticular, real labor income for secondary (tertiary)-educated workers in the informal sector is 5 (30) percent . lower than that of their formal sector counterparts. 65 equivalent patterns cannot be found in the tradable and non-tradable sectors (Table 3.7, panel B)62. Given this and the fact that poverty rates for the employed are much higher among informal workers than among formal ones, irrespective of whether they work in the tradable or the non-tradable sectors, the rest of this section focuses exclusively on the differences between the formal and the informal sector (Table 3.8). Table 3.8: PovertyRatesare higher inthe informalandthe tradable sectors Formal Informal Tradable 14.5 27.5 Non-Tradable 9.98 24.0 3.43 Since individuals with different education levels are not equally distributed across sectors (Table 3.19 panel B), this variation could be due to a series of cross-sectoral factors ranging from differences in institutional arrangements to differences in the (relative) productivity of more and less educated individuals, to differences in the availability of different types of workers. In other words, time and sectoral differences in returns to education are the result of the interaction between demand and supply factors - i.e. more educated workers may become more productive over time due to technological change, and this in turn may increase firms' demand for them; public investment in education may translate into significant increases in the number of workers with, say, secondary education, making them relatively more abundant. Table 3.9: The distributionof educationvariesintime andacross sectors Primary Secondary Tertiary 1997 33.7 40.0 23.3 1998 35.3 39.0 22.7 1999 35.0 39.0 22.5 2000 34.5 38.8 22.6 2001 32.1 40.8 24.3 2002 32.9 38.5 25.4 B.- Across sector (% of allemployedineachsector) Tradable Non-Tradable Formal Informal Formal Informal Primary 32.4 44.7 14.8 44.2 Secondary 41.9 42.1 34.7 40.9 Tertiary 23.4 8.4 49.6 9.9 3.44 The pool of employed individuals appearedto be slightly more educated in2002 than in 1997 (Table 3.9, panel A), reflecting both general increases inthe level of education of the population and changes in the likelihood of being employed conditional on education. In fact, individuals with secondary and tertiary studies were significantly more likely to participate in the labor force and, 62The coefficients on the interactions between secondary and tertiary education dummies and a tradable sector indicator in an earningsregressionsare insignificant after controlling for the incidence of informality. 66 conditional on participation, to be employed in 2002 than in 1997 compared to individuals with primary education63. 3.45 Although there i s no equivalent and readily available information on the evolution of the demand for more or less educated individuals, we can infer it from the data on labor income and number of workers in each education group following the methodology proposed by Katz and Murphy (1992). A brief description of their methodology i s presented in Box 3.6 for the interested reader. For the rest, it suffices to know that the exercise roughly consists in subtracting changes inthe relative supply of different types of workers from changes in their relative labor income, making some assumptions about the degree of substitutability across these groups. The residual from this operation i s then considered to equal changes inrelative demand. 63The coefficients on the interactions between secondary and tertiary education dummies and a year dummy for 2002 are positive and significant when estimating probability models for the likelihood of labor force participation and employment, conditional on participation, in 1997 and 2002. 67 Box 3.6: ConstructingRelative DemandShifts: What Needsto BeAssumed andWhy. Changes in the relative wages of workers with different education levels are a function of changes in the relative demand for and the relative supply of these workers, as well as of the elasticity of substitution between workers with different education levels. This elasticity is a measure of the ease with which one kind of worker can be replaced for another in production. For instance, if the elasticity of substitution between engineers and technicians is high, a small increase in the relative wage of engineers would lead to a large substitution out of engineers and into technicians, whereas if the elasticity is low changes in the wages of engineers will lead to little substitution between both types of workers. More formally, if B is equal to x a 1 percent increase in the relative wage of engineers is associated with an x percent decrease in the quantity demanded of engineers relative to technicians. These ideas can be modeled simply as follows. Under the assumption of a common elasticity of substitution (CES), relative wages have to satisfy the condition: (1) Log[w1(t)/w2(t)l = (W[D(t)- log [Xl(t)/XZ(t)l where wl(t)/w.~(t) is the ratio of relative wages, xl(t)/x2(t) is the ratio of relative supplies, B i s the elasticity of substitution between workers in the two education levels, and D(t) i s the time series o f relative demand shifts (Katz and Murphy, 1992). Since both D(t) and B are unknown parameters additional assumptions about either the behavior of D(t)or the value of B are necessary in order to construct a time series of relative demand shifts using (1). Two different approaches have been proposed in the literature for this purpose. First, a working value of B can be estimated if relative demand shifts are assumed to follow a linear trend by runninga regression of the (log) relative wages on the (log) relative supplies and such a trend as follows: (2) Log[wl(t)/w2(t)] = a + (110) time trend - (l/o) log [xl(t)/x2(t)] The value of B can then be recovered from the coefficient on relative supply and plugged into (1) after re-arranging terms to obtain: (3) D(t>=0 log[~i(tY~2(t)llog [xi(tYx2(t)l + where all of the parameters on the right-hand side are now "given". Unfortunately, because all of the series for the countries in the sample are rising (or falling) almost monotonically, estimates of Bconditional on a time trend are very imprecise (with t-statistics of one or lower), and often wildly improbable, rendering this first approach invalid. The second approach relies on assumptions about B rather than on assumptions about the behavior o f D(t), thus avoiding estimation. In particular D(t) i s then estimated directly from (3) under plausible values of B. Inthis chapter B isassumed to be equal to 2, which is the range o f international estimates summarized in Katz and Autor (1999). An additional concern with simple estimates of B and D(t) obtained from equations (2) and (3) is the changing composition of the workforce. The five countries in the sample are relatively advanced in their demographic transitions. The fraction of workers who are older is therefore higher in later than in the earlier years. Older workers generally earn more than younger workers, and a life-cycle model of earnings determination suggests thai the wage gap between more and less educated workers should also increase with age (Mincer 1974; Heckman Lochner, and Todd 2001). The observed increase in the relative wage of the skilled could, therefore, be a produci not of the changing supply or demand for educated workers, but rather of the changing age profile. Many countries in Latin America are also witnessing important changes in the participation of women in the labor force. If the difference in wages by education is larger (or smaller) for women than for men, this too could distort uncorrectec estimates of D(t). Again following Katz and Murphy, compositional changes in the labor force are controlled for b) holding age and gender distributions constant over time in each country. Specifically, for any education group average share of total employment were calculated for 14 age-gender cells over the entire period and these weights were then used to construct mean wages for an education group in any given year. Source: Katz and Murphy(1992) 3.46 The relative demand for tertiary workers increased during 1997-2002, with the exception of a one-time decline in 2000. The relative demand for secondary workers followed a similar pattern, 68 although in this case around a negative trend (Figure 3.3)@. The period covered by the data i s too short to make significant inferences about long-term processes. However, it i s important to notice that the observed patterns are in line with those found in other Latin American countries and in the United States during the 1980s and 1990s (Sanchez and Schady, 2003; Katz and Murphy, 1992), especially regarding tertiary workers. These countries experienced sharp increases in the relative demand for tertiary workers, which were positively correlated with trade openness and, inthe case of Latin America, with access to foreign technologies. Some of them also experienced increases in the relative demand for secondary workers, although the evidence is more mixed (e.g. large increases in Mexico, but little evidence o f positives changes inBrazil). Also important i s the fact that the changes observed in the Ecuadorian case are, in general, fairly small compared to those observed in other countries65. 3.47 Average changes, however, may hide important differences across sectors, especially since there exists significant sectoral variation in education levels and returns. If we then repeat the exercise for the formal and informal sectors, we find that the formal sector looks very much like the urban economy as a whole, while the opposite patterns appear to hold in the informal sector - namely, in this sector the relative demand for tertiary workers decreasedand the relative demand for secondary workers increased between 1997 and 2002. Figure 3.3: Increases inrelative demandfor tertiary workers informal sector andsecondary workers ininformalsectors All employed 0.25 0.2 0.15 0.1 0.05 0 -0.05 -0.1 -0.15 /-Tert/Sec I- Sec/Prim/ Calculations are based on an elasticity of substitution of 2 -a value with the range provided by the empirical literature. Volatility inthe demand series is due to the small sample size. Itis possiblethat larger changes occurredduring 1990-1997at the time of trade opening. 69 Formal 0.5 0 3 - 02 - - 0 1 -' 0 - 1997 1998 ".1999 2000 * mb'k- 2002 -0 1 - z1 uflP. 2, *\*, -02 - % -0 3 - `B - 0 4 l I /-Tert/Sec -' Sec/Prim 1 Source: Authors' calculations usingEEDS 1997-2002 (INEC). Relativedemand curves calculated assuming and elasticity of substitution between education groups equal to 2. 3.48 There i s a multiplicity of factors that could explain why the demand for skills i s lower in the informal sector. First, labor-capital ratios are generally lower in the informal sector, and so i s the use of technology. Second, and related to the previous point, labor productivity and labor productivity growth tend to be lower inthe informal than inthe formal sector. 3.49 Changes in the relative demand for individuals with different levels of education, together with changes in the relative numbers of such individuals, determined to a large extent the evolution of earnings by education group across sectors. In particular, given that education upgrading took place across the board, the evolution of (relative) wages was closely linked to the evolution of 70 relative demand in each sector. Hence, while in the formal sector individuals with tertiary education faired significantly better than others, interms of real labor income recovery, inthe informal sector it was those with secondary education who didrelatively best (Table 3.10). Formal Informal 1997 0.9 1 1.07 1.80 0.7 1 0.81 1.41 1998 0.56 0.85 1.31 0.42 0.56 0.86 1999 0.38 0.5 1 1.05 0.25 0.34 0.66 2000 0.48 0.57 1.09 0.31 0.40 0.86 2001 0.50 0.69 1.35 0.47 0.52 0.89 2002 0.61 0.80 1.75 0.44 0.59 0.96 1999/1997 0.42 0.47 0.58 0.35 0.42 0.47 2002/1999 1.60 1.57 1.66 1.76 1.73 1.45 2002/1997 0.67 0.75 0.97 0.63 0.73 0.68 3.50 The question then arises as to what effect these changes have had on poverty, if any. As we stated above, the poor are less educated than the non-poor, even after we control for employment. Hence, overall increasesin the demand for highly educated would only make it more difficult for the less educated to access jobs without getting more education or accepting relatively lower wages, which in principle would have a negative effect on poverty. These changes are not limited to the formal sector, where increases in the relative demand for more educated workers may make it more difficult for the uneducated poor to access formal sector jobs in the future, but also affect the informal sector, where most of the least educatepoor are employed. 3.51 What then underlies these changes in the demand for skills? We argued above that in other countries such changes appear to be correlated with increased access to better, foreign technologies that require a more skilled labor force to operate them and make more educated workers more productive. In the next section, we examine constraints to employment creation, as well as determinants of actual employment creation and labor productivity, and show that workers' education levels play an important role inEcuador. CONSTRAINTSTO EMPLOYMENT CREATION: AN ANALYSIS OF THE MANUFACTURINGIN SECTOR URBAN AREAS 3.52 Why has employment creation been so low in recent years? What are the constraints that firms face when thinking about growing and hiring new workers? Are there differences between small and large firms? Inthis section we try to provide answers to these questions using information on manufacturing employment. Although manufacturing represents only 20 percent of total urban employment, we feel there are important insights to be gained regarding the overall functioning of the urban economy. Neighboring manufacturing and service sector firms operate in very similar economic environments, are subject to the same macroeconomic shocks and face the same set of labor and business rules and regulations. Therefore, to the extent that the decision to grow or to be formal or informal depends on the economic environment, rather than on the sector the firm operates 71 in, observations on the behavior of manufacturing firms will be informative about the potential behavior of service firms. 3.53 The analysis is based on firm-level data collected in Ecuador by the Investment Climate Project (DECRG, The World Bank 2003). The sample contains information on 450 manufacturing firms located in urban areas in the provinces of Azuay, Guayas, Manabi, Pichincha and Tungurahua. For the purpose of the analysis, firms were dividedinthree groups according to their size: small, with 1 to 10 employees and (normally) not subject to labor inspection; medium, with 11to 99 employees; and large, with more than 100 employees. Large firms are more likely to have more than one establishment, to count with public or foreign participation and to engage in export activities than mediumand small firms (Table 3.11). Table 3.11: Firmcharacteristicsdiffer byfirm size All Small Medium Large (0 to 10) (11 to 99) (100 +) Number of firms 450 74 288 84 Average size 83.8 7.5 1 37.4 310.1 Number of establishments 1.3 1.o 1.2 1.6 Public participation 1.o 0.0 0.0 4.7 Foreignparticipation 12.3 4.0 10.0 27.3 Export 29.7 8.1 23.6 67.8 Share of production in exports 10.1 0.8 8.4 22.6 Total income 2002 (US$l,OOO) 7,27 1.2 482.2 3,246.1 26,601.9 Oil products 11.3 1.3 13.8 10.7 Food and beverages 25.1 28.4 21.8 30.9 Apparel and textiles 21.1 20.2 21.2 22.6 Wood 7.3 8.1 7.3 7.1 Chemicals 17.6 25.7 17.0 13.1 Employment turnover and employment creation: What firms do versus what firms want. 3.54 All firms were asked about the number of workers that had been hired and fired during the year prior to the survey, as well as about the number of workers that had voluntarily quit their jobs in that same period. They were also asked whether, if faced with no constraints, they would increase, decrease or maintain the actual number of permanent workers employed at the firm, and, when choosing to increase or decrease, by how much. Using this information we construct estimates of gross employment creation and destruction, employment turnover and net employment creation, and compare them with the firms' unconstrainedpreferences. 3.55 Both the rate of employment creation and the rate of employment destruction were above 25 percent, pushing the employment turnover rate to 50 percent - i.e. more than half the workers employed at some point during the survey year either joined or left the firm in that period (Table 3.12). However, becausethe number of new unions was very close to the number of separations, high worker turnover was not accompanied by net employment creation. This i s close to zero for the sample, reflecting the poor performance of small and, inparticular, medium-size firms in this regard with net employment creation rates of 0.8 and-1.3 percent respectively. 72 All Small Medium Large (0 to 10) (11to99) (100+) (% of workforce) Employment creation 26.8 35.8 26.3 20.9 Employment destruction 26.3 35.0 27.6 13.2 I Employment turnover 53.1 70.8 53.9 34.1 Net employment creation 0.5 0.8 -1.3 7.7 3.56 Firmsare also askedto report the number of permanent workers employed in2000, 2001and 2002. Net permanent employment creation calculated using these data produces figures that are very close to the ones reported above. This implies that, while firms are making high use of temporary contracts (hence the large worker flows), they are barely transforming any of these temporary unions into permanent ones. 60 -- 5 50 -- el 2 .-c .-E40 -- LL c 0 & 30 -- e 0 $i p 20 -- I O -- 0 - All Small Medium Large Source: InvestmentClimate Survey -Ecuador(2002). 73 3.57 Incontrast, when questioned about their unconstrained preferences, more than 30 percent of all firms declared they would like to increase the number of permanent workers they employed, compared with less than 10percent that would decrease it, and 60 percent that would maintain it at its current level. Since firms are also asked to report the magnitude of the desired change, we can calculated a hypothetical `unconstrained' net employment creation rate, which stands at 8 percent (Figure 3.4). 3.58 To the extent that differences between actual and desired net employment creation rates are a reflection of existing constraints, small and medium-size firms appear to face higher obstacles to employment expansion than do larger firms. They more (less) frequently report they would like to increase (decrease) the number of permanent workers they employ, and, conditional on a desire to grow, they also report higher `ideal' increases (42 and 30 percent for small and medium firms, compared with 25 percent for large ones). Constraints to employment creationand businessexpansion 3.59 The reasons given for the difference between actual and desired hiring and firing behavior vary somewhat with firm size, and do not appear to be symmetric with respect to desired course of action. This i s not surprisinggiven that small firms are less likely to be unionized andto comply with certain labor regulations than larger firms, as long as they remain small (Table 3.13). For instance, firing costs appear to be a greater constraint for small firms when thinking of expanding than when thinking of contracting, while the opposite is true for medium and large firms. Although this may initially seem to be a contradiction, it merely reflects that fact that as a small firm grows firing costs will be more binding in the event of a future contraction. Similarly, non-wage costs, such as Social Security payments, pose a strong constraint to the expansion of large firms, which are more likely to be compliant at presentthan small ones. Table3.13: Firingcostsand non-wagecosts are the mainreasonsfor nothiringandfiring permanent workers as desired All Small Medium Large (0 to 10) (11to 99) (100 +) (% of firms in group) Not increasing Firingcosts 38.7 47.1 39.5 25 Non-wage cost 17.8 17.6 13.5 43.7 M O L procedures 0.8 0.0 1.o 1.o 0.0 Unions 1.5 0.0 6.2 Sale expectations 41.1 35.3 44.7 25 Not decreasing Firingcosts 56.7 NA 70.0 42.8 Non-wage cost 16.2 NA 10.0 14.3 M O L procedures 0.0 NA 0.0 0.0 Unions 0.0 NA 0.0 0.0 MOL procedures:Ministry of Labor procedures. NA: Not availabledue to the smallnumber of firms inthis categoryreportingdesireddecreases(3 observations). 74 Figure 3.5: Firms perceive numerous constraints to daily business operation and future expansion ... (Percentage o f firms that perceives each factor to be a constraint) .- . .. . Labor mgulatiano WoMorce skill level jmAU FdSmall.Medium .Large] Labor FinancialResources Taxes- Mlw Taxes-pccsave mum bcaxsmdpsm*s 1.N Osmall .Medm .Large1 Taxes, customs and permits Source: Authors' calculations based on data from the Investment Climate Survey - Ecuador, The World Bank (2003). 75 3.60 Uncertainty regarding future sales also plays an important role in shaping hiring and firing decisions, comparable to that of firing and non-wage labor costs. Finally unions and bureaucratic procedures with the Ministry of Labor do not seem to impose to heavy a burden on either hiring or firing. 3.61 More broadly, when asked about constraints to business expansion, 30 percent of all firms, and 35 percent of small firms still identifies labor legislation as a significant obstacle to growth. This factor, however, does not appear to be the main barrier to expansion, nor the main labor-related constraint since a large fraction of firms find low skill levels to be a more acute problem. Almost 50 percent of all firms declare to have problems finding workers that are adequately qualified. This i s important since, as we will discuss below, skills and labor productivity appear to be positively correlated and, as we will discuss in the next section, the demand for skills has been rising in Ecuador, particularly inthe formal sector (Figure 3.5). 3.62 Scarce and costly financial resources, tax payments and processing, poor infrastructure (telecommunications and electricity, in particular), and uncertainty regarding the economic environment business operate are generally viewed as constraints for both current operation and future expansion. For instance, 65,55 and 60 percent of all firms declare the cost of credit, the value of taxes and access and provisionof electricity to be aproblem, respectively (Figure 1.5) Table 3.14: Firmsof different sizes face different constraints to growth andexpansion Small Difference Medium Difference Large (S) S andM (M) M a n d L (L) Labor Labor regulation 0.36 * 0.28 0.28 Workforce skill level 0.46 0.50 ** 0.35 Infrastructure Telecommunications 0.46 0.48 * 0.39 Electricity 0.63 0.59 0.61 Transportation 0.28 0.31 0.33 Access to land 0.25 ** 0.15 0.12 Financialresources Availability 0.70 * 0.63 * 0.53 cost 0.69 0.68 0.64 Access 0.63 * 0.54 0.55 Economic environment Political andregulatory uncertainty 0.86 * 0.80 0.76 Macroeconomicinstability 0.81 0.75 0.71 Judiciary uncertainty 0.69 0.63 0.63 Corruption 0.46 0.44 0.47 Taxes, customsandpermits Tax value 0.58 0.56 0.50 Tax procedure 0.47 0.48 0.50 Customs 0.24 ** 0.41 ** 0.53 Licensesandpermits 0.34 ** 0.23 0.25 Source: Authors' calculationsbasedon InvestmentClimateSurvey -Ecuador, The World Bank(2003). 76 3.63 Interestingly some of these constraints appear to be more binding for certain types of firms than for others. Small firms declare to be more constrained by labor regulation, access to land, availability of and access to credit, and political and regulatory uncertainty than medium and large firms. Similarly, medium firms are more constrained by the lack of skilled workers, poor telecommunication infrastructure, and credit availability than large firms. Finally, large firms are the most constrained by custom problems and inefficiencies, which should not be surprising given that it i s precisely these firms that are most likely to export all or part of their production (Table 3.14). Actual employment creationand labor productivity 3.64 We turn next to the analysis of the determinants of net employment creation. We argued in Chapter 1 that GDP and employment growth appeared to be driven at the macroeconomic level by changes in productivity, measured by TFP. In other words, they appeared to be driven by improvements in the quality of inputs and the institutional environment. Our examination of the constraints to employment creation above suggests that these same factors are also important at the firm level. We complement those results here with a more formal analysis of the relationship between labor productivity and actual employment creation, as well as the relationship between labor productivity, skills and technology (Le. quality of inputs). Figure 3.6: Labor productivity increases with firmsize andover time All Small Medium Large Source: Authors' calculationsbasedon data from the InvestmentClimateSurvey-Ecuador, The World Bank (2003). 3.65 Labor productivity, measured as the dollar value of production per worker, i s positively correlated with firm size - workers in large firms are twice as productive as their counterparts in small firms according to the survey. These differences are the result of a variety of factors. Large 77 firms are more likely to use capital and to have access to better technologies than small firms - for instance, 35 percent of large firms use foreign technology compared to only 12 percent of small forms. They are also more likely to devote part of their production to exports and hence to be subject to international competitive pressures(Figure 3.6). 3.66 Moreover, labor productivity has increased over time, irrespective of firm size. Most of the increaseoccurred between 2000 and 2001, particularly for small and large firms that saw increasesin labor productivity of US$2 and US$6 (or 16percent) per worker, respectively. 3.67 Lower levels of labor productivity among small firms could potentially be responsible for their poor performance in terms of employment creation during the period, especially taking into account that it i s precisely these firms that appear to be the most constrained by labor costs and regulations. We explore the determinants of labor productivity and the relationship between this variable and employment creation more formally below. 3.68 Access to better technology, measured by access to foreign technology, and exposure to international competition, measured using a exporter dummy, are both positively correlated with labor productivity. So i s the level of skills of the workforce employed by the firm, where workers with completed secondary studies or higher levels of education are considered skilled (Table 3.15). In particular, exporting firms and firms with access to foreign technology are 30 percent more productive than their counterparts. Similarly, an increase of 10 percentage points in the share of educated workers leads to a 5 percent increase inlabor productivity. Table 3.15 Employmentcreation is positively correlated with labor productivity DependentVariable Labor productivity (Net) Employment Creation (US$/worker) (1) (11) Labor productivity 0.17 ** (0.08) Share of workforce with secondary or 0.16 ** 0.06 higher education (0.07) (0.08) Access to foreign technology 9.19 * 12.59 ** (4.87) (6.34) Exports 8.99 * 9.62 (4.74) (6.16) Access to credit 0.95 (1.30) Firm size dummies Yes Yes Number of observations 250 245 Source: Authors' calculationsbasedon InvestmentClimateSurvey-Ecuador,The WorldBank (2003). ** ('ISignificantlydifferent from zero at the 5 (10) percentlevel. 3.69 Inturn, higher labor productivity is associatedwith positive net employment creation, and so i s access to foreign technology. In particular, a 10 percent gain in labor productivity (10 percent of US$30/worker in 2001) would cause a 1percent increase in employment creation. The direct effect of skills and international competition on employment creation, however, i s weak, once labor productivity i s taken into account (i.e. - they mainly affect employment creation through their effect on labor productivity). 78 3.70 Maybe surprisingly, access to credit i s not significantly correlated with net employment creation despite having beenidentified by firms as an important constraint to businessexpansion (see above). Policies to increase labor productivity and employment creation 3.71 The discussion above helped us identify factors that directly or indirectly affect employment creation. Higher levels of skills and access to better technologies are correlated with higher levels of labor productivity and, as a result, higher levels of employment. Firms, however, face a number of constraints, mainly institutional constraints, that reduces their desire and capacity to increase labor productivity and employment. Measures aiming to increase labor productivity and, as a result, employment generation should include, among others: Ratification of Free Trade Agreements and rationalization and reduction of tariffs and non-tariff barriers. These measures should contribute to the elimination of the existing anti-exports bias, associated with years of import substitution policies. Simplification of licensing agreements and promotion of FDI.Ecuador could benefit significantly from existingtechnologies by puttinginplace the appropriate incentives for foreign licensing and FDI,combined with effective property rights andpatent protection. Investments in education and training. Education levels and enrolment rates in Ecuador are low for the country's level of development. Ecuador lacks a broad based of secondary-educated workers, necessary to effectively adopt and adapt existing technologies, and this deficit will not diminish in future years unless more resources are devoted to secondary schools. In addition, Ecuador's public training institute, SECAP, i s in need of radical reform. The cumculum it currently offers i s obsolete and, as a consequence, the resources of the institute are under-utilized. Increased competition in the provision of training could help generate the necessary incentives for change, while expanding firms' choices regarding training options. 3.72 In addition, the poor are less educated than the non-poor and they tend to be employed in small, informal firms with low access to technology. As a result, if the policies described above are to be successful in reducing poverty, they need to be accompanied by explicitly pro-poor measures such as: Promotion of linkages between large and small firms. Large firms are more likely to initially benefit from increased access to foreign markets and technology than smaller firms, but they are also more likely to have less flexibility to quickly respond to changes in market conditions. Promotingproductive linkages between large and small firms could then help distribute the gains associated with these developments and transfer technology to small firms, while providing large firms with additional degrees of flexibility. Creation of service centers for small firms. The adoption and adaptation of technology i s often an expensive process. Service centers, or small-firm incubators, allow small businesses to share the cost of aparticular technology or service that would be otherwise inaccessible and hence increase their productivity. Incentives for informal workers training. All firms in Ecuador are expected to contribute 0.5 percent of their wage bill to fund the SECAP. Small, informal firms, however, generally do not do so and, as a result, do not have access to the institute's services. Special training programs for these firms should be promoted, and the possibility of sponsorship from the network of Chambersof Commerce or other employer associations should be explored. 79 3.73 Finally, firms face numerous institutional constraints and economic uncertainty. Measures aimed at mitigating some of these constraints should include: 0 Labor reform to reduce costs associated with permanent hiring. Relatively high labor costs associated with permanent hiring have led over the past few years to an overwhelming use of temporary contracts by Ecuadorian employers and, as a result, to an increasing degree of segmentation in the labor market. Existing labor legislation needs to be modified to bring these two contractual figures closer together. In addition, because making the regulation on temporary contracts stricter may disproportionately hurt certain vulnerable, hard-to-employ groups, the creation special contractual forms, such as apprenticeship or re-entry contracts could be considered. 0 Increase access to credit for small and medium firms. Credit availability for medium and, particularly, small firms i s low in Ecuador, reflecting the country's banking system current weakness as well as its poor savings capacity. Increased access to credit could be achieved through the creation of credit unions sponsored by the "gremios" (industrial associations) or the Chambers of Commerce, and the promotion of venture capital enterprises and linkages between large and small firms. 3.74 Some of these reforms and policies are already underway, in some cases with financial and technical support from the World Bank (e.g. labor, education, competitiveness and business environment). Some others, however, will have to be actively pursued by the Government of Ecuador in the near future if the country is not to miss the window of opportunity that the more favorable investment climate brought about by dollarization has created. CONCLUSIONS. 3.75 We have argued in this chapter that poverty i s increasingly becoming an urban phenomenon and that the fate of the urban poor is intimately linked to the behavior of urban labor markets and, particularly, to their capacity to generate productive employment. We have also examined the determinants of and constraints to employment creation among small and large firms, and found the quality of inputs and institutions, or lack thereof, to be the main forces driving firms' willingness and capacity to increasepermanent employment. 3.76 The poor tend to be less educated than the non-poor and they tend to work smaller, informal firms, which have low levels of labor productivity and appear to face the most constraints to employment creation and expansion. Training policies and policies aimed at the promotion of micro and small enterprisesmay then go a long way inhelpingthe urban poor. 80 4. RURALPOVERTY,AGRICULTURALPRODUCTIVITY,AND THE DISTRIBUTION OF LAND 4. I Forty percent of the population of Ecuador lives inrural areas. Two thirds of them are poor, and a large fraction of these are extremely poor. This chapter i s devoted to the analysis of rural poverty inEcuador. The focus of the chapter i s on the income-earning opportunities of households in rural areas. In Ecuador, as elsewhere, the low incomes of poor households in rural areas are frequently tied to the very low output of farmers, and to the fact that the poorest households do not have access to land. Often, the root causes of low incomes in rural areas are low productivity and poor distribution of land. 4.2 The mainpoints madeinthe chapter are five: 0 There are large differences in agricultural productivity across cantons and across farms within cantons inEcuador. Self-employed farmers with higher productivity are significantly less likely to be poor. 0 Access to credit i s the single most important policy intervention to raise productivity among small-scale, poor farmers. Agricultural education and access to inputs such as fertilizers and pesticides are also important, although these could have negative environmental consequences that needto be taken into account. 0 The distribution of landinEcuador i s very unequal, and has essentially remained unchanged in the last twenty-five years. Agricultural laborers (peones) are among the poorest households in rural Ecuador. They standto benefit both from policies to increase productivity, some of which are passedon as wage increases, and from policies that facilitate access to land. 0 Policies that improve tenure security, and those which facilitate landtransactions such as rental or sale holdpromise as a means of reducing poverty among agricultural laborers. 0 Poverty inthe rural off-farm sector i s significantly lower than inthe on-farm sector inEcuador. There i s evidence that aproductive on-farm sector comes hand-in-hand with a vibrant off-farm sector, inparticular inactivities such as agricultural services and food processing. 4.3 The rest of the chapter proceeds as follows: Inthe first section we expand briefly on the rural poverty profile presented inChapter 2, paying special attention to the effect of sector of employment (on-farm or off-farm) and type of agricultural employment (self-employed or laborer) on the probability of being poor. The second section presents estimates of the distribution of agricultural productivity and measuresof the concentration of land across cantons. Inthe third section, we show that low productivity in the on-farm sector i s closely related to a high probability of being poor for households who work their own land, as well as for agricultural laborers. The fourth part of the chapter discusses specific policy interventions to raise agricultural productivity, while the fifth section discusses policies to increase access to land by the landless. The sixth section considers the importance of the off-farm sector inrural areas inEcuador, and section seven concludes. WHOARE THERURALPOOR? 4.4 We begin the chapter with a very brief description of the distribution of poverty inrural areas in Ecuador, with emphasis on the employment patterns of poor and non-poor households. (Other characteristics of the poor, including the rural poor, are presented in Chapter 2.) Table 4.1 breaks down the sample into the rural areas of the Costa and the Sierra, and presents estimates of the probability of being poor calculated on the basis o f the 1999 ECV. The table shows that the incidence of rural poverty i s higher in the Sierra than on the Costa. Inboth regions rural households 81 working in agriculture are significantly more likely to be poor than those in other sectors. Among those in the agricultural sector, poverty rates are higher among agricultural laborers than they are among self-employed farmers. When the sample i s limited to wage-earning households, excluding agricultural laborers, the incidence of poverty is significantly higher among households employed in the agricultural off-farm sector (for example, agricultural services) than those employed in the non- agricultural off-farm sector (for example, small-scale manufacturing). In the Sierra, too, poverty rates are significantly higher among the indigenous than the non-indigenous. (We cannot make similar comparisons for the Costa because there are not enough indigenous households in the ECV sample.) Table 4.1, finally, shows that there are no clear differences in the probability of being poor between households headed by men and women. This could be the result of an absence of discrimination against women in the labor and other markets, or a reflection that only the most able women can start independent households. In any event, it limits the effectiveness of poverty targeting basedon the gender of the household head. Table 4.1: The ruralpoverty profile inEcuador I Rural Costa I I Headcountindex Difference Headcountindex Head in agriculture 0.61 *** 0.44 Headnot in agriculture Head self-employed farmer 0.62 0.67 Head agricultural laborer (peon) HHowns morethan 1ha. of HHowns less than 1ha. of land land (1) 0.55 0.57 (1) Headin agricultural off-farm Head in non-agricultural off- sector 0.62 0.42 farmsector Headindigenous 0.52 Headnot indigenous Head male 0.53 * 0.42 Head female RuralSierra Headcountindex Difference Headcountindex Head in agriculture 0.79 *** 0.57 Head not in agriculture Head self-employed farmer 0.80 0.76 Head agricultural laborer (peon) HHowns morethan 1ha. of HHowns less than 1ha. of land land (1) 0.68 0.76 (1) Head inagricultural off-farm Head innon-agricultural off- sector 0.87 *** 0.54 farm sector Headindigenous 0.87 *** 0.64 Headnot indigenous Head male 0.69 0.73 Headfemale Source: Author' calculations basedondata from the 1999ECV. ***Difference significantat 1%level; ** at 5% level; * at 10%level. (1) Sample limited to those working in agriculture. A Only for those workingin on-farmsector. Only for those working innon-farmsector. 82 4.5 Highpoverty rates are a direct result of low income and associatedlow consumption. Table 4.1 makes clear that a poverty-reduction strategy for rural Ecuador must address the limited income- earning opportunities of both self-employed farmers and agricultural laborers-increasing agricultural productivity and facilitating access to land. In addition, since poverty rates in the off- farm sector are lower than those in the on-farm sector, stimulating the off-farm sector may allow some diversification of income-earning opportunities out of the on-farm sector. CONSTRUCTINGMEASURESOFAGRICULTURAL PRODUCTIVITYAND LANDDISTRIBUTIONFOR ECUADOR Yields per hectare and per worker-hour: A first approximation. 4.6 We start our analysis of productivity and the distribution of land across cantons with the 2000 ThirdAgricultural Census. This Census, not a "census" inthe true senseof the word inthat it did not visit all producers, nonethelesscollected information at about 150,000 farms across the country. The sample design and size mean that meaningful averages can be constructed at the level of the canton. The Census includes detailed information about the amount of crops planted and harvested, as well as the surface area and the number of workers used on the farm. It i s therefore relatively straightforward to calculate measuresof yields per hectare, and yields per worker. 4.7 Figures 4.1 through 4.5 present maps of the average yield per hectare for five crops-rice, potatoes, bananas, cocoa, and coffee. Bananas, cocoa and coffee are the mainexport crops, together with fresh flowers (see Box 4.1 for a discussion on the flower industry), and account for approximately 40 percent of the total value of the country's agricultural production (see next point). Rice and potatoes are the main Costa and Sierra crops for domestic consumption and account for 9 and 2 percent of the total value of agricultural production, respectively. 4.8 We use potatoes as an example to illustrate how to interpret the information in the maps66. Potatoes are produced mostly in the Sierra. Potato yields per hectare, however, vary significantly across cantons, ranging from low (inlight gray) to high (in black). For instance, Nabon, a canton in the province of Azuay, and Ambato, a canton in the province o f Tungurahua, are both important producers of potatoes, but the yield per hectare in Ambato i s four times the comparable yield per hectareinNabon (1.7 tons per hectare, compared to 7.0 tons per hectare). 66InFigures4.1 to 4.5, cantons are sorted into quintiles, with each quintile accounting for 20 percentof the total yields. 83 Figure4.1: Riceis cultivatedmainlyinthe southernCosta 84 Figure4.3: Bananasare cultivatedmainly inthe southern Costa 0 Figure 4.4: Cocoa is cultivated inthe Costa Source: Authors' calculationbasedon data from the 2001ThirdAgricultural Census. 85 Figure 4.5: Coffee is cultivatedmainly inthe northern Oriente Gata@aps $23 0 0 - (0.026 0.083) (0.085 0.136) (0.140 0.189) (0.193 0.237) 5 (0.253-----0.364) Source: Authors' calculationbasedon datafrom the 2001ThirdAgriculturalCensus. 86 Box4.1: Women andthe flower industry inEcuador Perhaps the most dramatic changes inthe agricultural sector in Ecuadorhavetaken placeinthe context of the growth of the flower industry. Between 1985 and 1997the real value of flower exports fromEcuador grew from $ 0.5 million to !$ 120 million per year, and Ecuador became the third biggest exporter of flowers in the world (after Holland andColombia). The flower industry also generateda large number of jobs-36,000 by 1998. Almost two-thirds of thesejobs were held by women. Inclosecollaboration with the NationalCouncil of Women of Ecuador(ConsejoNacionaldeMujeres- CONAMU), the World Bank launched a study of the social and economic impacts of the growth of the flower sector in Ecuador-a study which was particularly innovative in its combination of quantitative andqualitative methods. Some of the main findings include: The flower sector has provided alarge boost to employment and to wages-especially for women. In the two study areas which were unaffected by the flower industry-cotocachi and Pesillo-the hourly wage of women was barely a third of that of men. By contrast, in the two study areas where flowers were grown-cayambe andTabacundo-there was a similar disparity in wages betweenmen and women not employed in the flower industry, but no difference in the wages of men and women working inthe flower sector. Employment in the flower sector allowed women to view themselves, and their relationship with men, in a different light. Holdinga payingjob gradually came to be viewed as "n~~~nal":"Now it is normal. Before, when it was beginning, it was not seen well-how are you going to go work on a plantation-now work in the greenhouse only with men, Iam their supervisor... like it or not, they have to it is normal", says a woman from Cayambe; and another woman in Cayambe reported: "I listen to me ... Irespect them and they respect me." By contrast, refemng to her male relatives, a woman from Cotacachisays: "He doesn't want me to go out.. .He argued about it.. . Yes, that was the problem, (my husband's) family didnot want to talk to me-why do you haveto go out to work when you can work at home?' Women working in the flower industry also had more control over householdexpenditures, and saveda large fraction of their earnings. Repeated exposure to chemicals in pesticides was seen as the most obvious negative aspect of employment in the flower industry-reported in interviews by men as well as women. There are no scientific studies of the health damages associated with chemical exposure in Ecuador, but careful analysis of the health costs to workers, as well as any larger environmental costs, should be a high priority for the future. Source: Newman, Larreamendy and Maldonado (2002). Turningyieldsinto dollars 4.9 Calculations of yields per hectare are a useful first step, but they have important limitations. For instance, one cannot make comparisons across crops, or across cantons producing different crops. That is, the maps in Figures 4.1 through 4.5 are not informative about whether the value of rice or that of bananas i s higher in a given canton in the Costa, or whether the overall value of farm output is higher or lower in Nabon or in Ambato. To make these kinds of comparisons, we have to transform output in tons into a measure that i s comparable across crops-a measure such as the total dollar value of production per hectareor per worker-hour. 87 Source:Authors' calculationbasedondatafromthe 2001ThirdAgriculturalCensus. Figure 4.7: ...while labor productivity is highinthe interior of the Costa region *ovincial Source: Authors' calculationbasedondata from the 2001ThirdAgriculturalCensus. 88 4.10 We do this by using data collected by INEC for the Producer Price Index (PPI) (see Box 4.2). productivity') and 4.7 (mean dollars per worker - `labor prod~ctivity')~~.Although these two The results from these calculations are summarized in Figures 4.6 (mean dollars per hectare- `land measures capture different notions of agricultural productivity (e.g. land productivity i s higher in Ambato than in Nabon, while the opposite i s true about labor productivity), they are positively correlated68and this correlation generates some common patterns across both figures. In particular, there seem to be two high "productivity belts" in the country. The first of these i s concentrated in cantons on the southern Costa near Guayaquil, the second in cantons around Quito. By contrast, the dollar value of productioni s low in the southern Sierra, inthe northern Costa regions (inparticular in the provinces of Manabi andEsmeraldas), and inmost of the Oriente. Box 4.2: Turning tons into dollarsinthe Agricultural Census The 2000 Third Agricultural Census collected detailed information on the quantities of crops produced but, with the exception of flower production, no information on their prices. How, then, do we turn the data on yields into data on the dollar value of production? For this purpose, we used the information collected by INEC for the Producer Price Index (PPI). The PPI data i s useful, but it too has two important shortcomings for this conversion of tons into dollars. First, The PPI collects information on the price of only 43 crops, while the Agricultural Census collects data on the quantity produced of more than 180crops. Fortunately, the 43 crops on which the INEC collects data account for more than 90 percent of the total cultivated area for medium- and large-scale farms, and more than 80 percent of the corresponding area for small-scale farms. The second pro,blem with the PPI data i s that it i s collected in only a handful of cantons which account for a large fraction o f the production of a given crop. However, there appear to be large regional differences in prices of a given crop across cantons, so it is not clear what price to apply for those cantons for which there are no price data for a given crop. To deal with the first of these two problems, we calculate a baseline scenario in which crops for which price data are missing are given the mean per-hectare canton value of those crops for which data are available. To deal with the second problem, when price data on a given crop are not collected in a particular canton, our baseline scenario assigns the price of that crop in the nearestneighboringcanton. We then contrast this with three other scenarios: (i) the missing crops are assigned a value equivalent to that crop in a canton which has the highest dollar value per hectare; (ii) the missing crops are assigned a value equivalent to that crop with the lowest dollar value per hectare in a canton; (iii) the missing crops are assigned the mean per-hectare canton value of those crops for which data are available, but all cantons are assigned the nationalaverage of the price per ton of a given crop for which data are available. We then calculate Spearman correlation coefficient to test the degree to which the rank-order of cantons i s the same across different scenarios. These correlation coefficients between the baseline scenario and the alternatives are 0.79 for the alternative scenario in (i), 0.99 for the alternative scenario in (ii), 0.99 for the alternative scenario in (iii). conclude from these and We calculations that our results of the dollar value of production across cantons are robust to alternative ways of dealing with the missingcrop and spatial data. Source: Le6nand Schady (2003). 6' InFigures 4.6, 4.7 and 4.8 cantons are sorted in quintiles, with each quintile accounting for 20 percent of total output. The Spearman correlation coefficient, a measure of the extent to which the rank-order of cantons is the same across both maps, is 0.48, significant at the 1percentlevel. 89 Estimatingagricultural productivity 4.11 Conceivably, the dollar value of production per hectare in Ambato could be higher than in Nabon because farmers in Ambato use more labor and more capital per hectare. Similarly, the dollar value of production per worker in Ambato could be higher than inNabonbecause the meanfarmer in Ambato has more land and more capital at his disposal. We explore the extent to which this i s the case next. 4.12 We first compare productivity levels across farms that use similar amounts of inputs and produce the same products and find that important differences exist even within these narrowly defined groups. For example, when we limit the analysis to farmers with plots smaller than 1hectare who do not use a tractor and do not hire any non-family labor, the potato yield per hectare in Nabon i s 3.0, while that in Ambato i s 7.8-suggesting that potato yields inAmbato are higher than in Nabon even among this more narrowly defined group of farmers. 4.13 An alternative to this kindof within-group comparison is to formally model the way inwhich different farms use inputs, and then use the results from this model to factor out productivity differences associated with variation in input quantity and use. Put simply: What fraction of the differences in output across farms cannot be explained by differences in the amount of labor, capital or land? The residual differences inoutput from such an exercise would then be a `purer' or `cleaner' measure of productivity, closer to the standard meaning of this term. Moreover, if we assume that farms with the largest positive residuals - that is, the largest productivity once inputs are taken into account - are farms that operate inthe most efficient way possible, we can interpret these differences as a measure of the distance a given farm is from the Production Possibility Frontier - an indication of the "technical efficiency" or "productivity" of a given farm. Box4.3: Estimating a Cobb-Douglasproductionfunction on the basisof the Agricultural Census The Cobb-Douglasproduction function is the most commonly estimatedproductionfunctioninthe economics literature. It hasthe followingalgebraic form: Q = A * L"*K~*HX Where Q is output, A, a, and are constants, L is labor, K is capital, andHis land. The function is said to p, behomogeneousof degree a + px+x, since multiplicationof L,KandHby some constant will raiseoutput by a proportion ka + + x. If the three exponents sum to unity, the Cobb-Douglas function is said to be homogeneousof degree one, exhibitingconstant returns to scale. We use data from the Agricultural Census to estimate such a Cobb-Douglas production function. The Agricultural Census is not ideally suited for this estimation, because it did not collect information on the quantity of many of the inputs used in production, or on their prices. We measure labor as the number of workers usedon the farm, bothskilledandunskilled, andincludingfamily labor; capitalby the numberof farm machines, mostly tractors, available in the farm; land, finally, is broken down into four categories, corresponding to irrigated and non-irrigated land, with and without the use of inputs such as fertilizers and pesticides. 4.14 We estimate a Cobb-Douglas production function in which differences in total farm output are explained by differences in the amount of labor, capital, and land that a farm uses (see Box 4.3). Because production processes may vary by farm-size, we present separate results for small farms (less than 1 hectare), medium-size farms (between 1 and 5 hectares), and large farms (more than 5 90 hectares)69. Results from these estimations, presented in terms of elasticities, are summarized in Table 4.2. As the reported values are elasticities, they correspond to the percentage change in output predicted for a 1percent change in the input in question: For example, for small-scale farms, a 100 percent increaseinlabor i s associated with (only) a 5.4 percent increase inoutput. 4.15 The results of Table 4.2 can best be summarized as follows: 0 There are very small returns to increasesin labor for small-scale producers (an elasticity of 0.05), butmuchlarger returns for large-scale producers (anelasticity of 0.45). 0 Returns to increases in capital use for farmers of all sizes are also relatively small (elasticities between 0.07 and 0.08). This may be explained in part by the inadequate measure of the capital stock that can be constructed from the Agricultural Census. 0 The bulk of differences in output among small-scale producers can be explained in terms of the amount of land they have at their disposal, and the fraction of that land which uses inputs such as fertilizers and pesticides. A one percent increase in the amount of non-irrigated land leads to a 0.14 percent increase in output if no inputs are used in this land, and to a whopping 0.86 percent increase in output if this land i s farmed with inputs". The corresponding elasticities for irrigated land are 0.14 percent when inputs are not used, and 0.54 percent when inputs are used in production. Table 4.2: Cobb-Douglasproductionfunctionestimatesof the relative returnsto land, capitaland labor I Small-scalefarms Medium-scalefarms Large-scalefarms Labor 0.05 0.17 0.45 Capital 0.08 0.07 0.08 Non-irrigated land 0.14 0.04 0.08 Irrigated land 0.14 0.00 0.08 Input useon non-irrigatedland 0.72 0.77 0.37 Input use onirrigated land 0.40 0.70 0.39 Scale (Irrigated land) 0.99 1.04 0.99 Scale (Non-irrigated land) I 0.67 0.94 1.01 I Source: Authors' calculations basedon data from the 2001 Third Agricultural Census. 4.16 We next aggregate our measure of technical efficiency (or residual productivity) for individual farms into canton-level averages. These averages are weighted by the total area cultivated by a farm-so that larger farms, in effect, receive more weight. This, our most complete estimate of the distribution of productivity across cantons in Ecuador, i s presented in Figure 4.8. The figure shows that many of the same cantons with the highest output per hectare, and (especially) highest output per worker, are also those in which farmers produce most efficiently. For example, output per worker and technical efficiency are 100 and 40 percent higher in Ambato than in Nabon, respectively. 69 The size of the average plot varies significantly across different areas and regions, so that this particular breakdown is not motivated by an attempt to reflect accurately the national distribution of plot sizes. Rather it responds to our desire to pay especial attention to farmers and workers in small plots, who are much more likely to be poor than their counterpartsinlarger plots. 70This correspondsto the sum of the elasticity of output with respectto non-irrigated land (0.14) plus the elasticity of output with respect to use of inputs on non-irrigated land (0.72). 91 4.17 More formally, the Spearman correlation coefficient between the measure of output per hectare and the measure of technical efficiency across cantons i s 0.46, while the corresponding coefficient between output per worker and technical efficiency i s 0.77-both of which are significant at the 1percent level. Figure4.8: Technical efficiency is highinthe interior of the Costa region ncial ource: Authors' calculationbasedon data from the 2001Third Agricultural Census. Small-, medium-, and large-scale farms are all more productiveinhigh-productivityareas 4.18 Do differences in productivity across cantons hold for farms of different sizes? This is an important policy question. If, by and large, some cantons are more productive than others across all farm sizes then there i s greater hope that interventions which boost productivity in the lagging cantons-precisely those policies which we consider in the fourth part of the chapter-have the potential to benefit all farmers. To answer these questions, we calculate canton-level means of technical efficiency for small-, medium- and large-scale farms. Spearman correlation coefficients between these three measures are all positive and significant at the 1 percent level. (These correlations are 0.63 when we compare small-and medium-scale producers, 0.70 when we compare medium-and large-scale producers, and 0.38 when we compare small- and large-scale producers.) Clearly, farmers of all sizes are more efficient in some cantons than in others. O f course, some of the differences in efficiency we observe are likely to be a function of differences across cantons in their "natural endowments"-things like the quality of the soil, or rainfall patterns. As we show later in the chapter, however, differences in efficiency persist even after we take into account these differences in natural endowments. 92 Table4.3: Landdistributionis very unequalinEcuador Latin America Land Gini Argentina 85.6 Bolivia 76.8 Brazil 84.1 Colombia 82.9 CostaRica 80.6 Ecuador 80.9 Guatemala 85.3 Honduras 76.5 Jamaica 80.3 Mexico 60.7 Panama 80.4 Peru 92.3 Paraguay 85.7 Uruguay 81.3 Venezuela 91.7 Other developing countries Egypt 54.9 Indonesia 55.5 India 61.4 Jordan 67.7 Kenya 75 South Korea 33.9 Malaysia 64 Pakistan 55.6 Philippines 56 Senegal 49.3 Thailand 42.6 Tunisia 64.6 Uganda 54.9 Developedcountries Australia 85.3 Canada 55.2 Spain 84.5 Finland 49.4 France 54.4 Japan 43.2 Norway 39.1 United States 73.1 Source: DeiningerandOlinto 2000 The distributionof land inEcuador is extremely unequal 4.19 The mean dollar value of output i s an approximation to the size of the on-farm pie in a given canton. Who benefits from differences in productivity, or from future productivity improvements i s determined by who owns the land, the capital, and the labor which are the main inputs into 93 production. Land is distributedvery unequally in Ecuador: The Gini coefficient of land ownership was 0.81 in 2000, barely unchanged from its value of 0.85 in 197471. This i s very high by international standards-but, as Table 4.3 shows, more or less the average for Latin America, a region notorious for the skewed distribution of land (with the exception of Mexico, the only country inthe region where landreform significantly changedthe underlying distributionof land). 4.20 Figure 4.9 shows that in Ecuador, the land Gini tends to be highest in many of the high- productivity area~~~-namely,the Costa areas around Guayaquil and the northern Sierra areas around Quito, as well as in southern parts of the Sierra. The correlation between the measure of technical efficiency and the Gini i s small, but positive (0.14), and borderline significant. This need not mean, however, that productivity in these cantons i s higher because of a larger concentration of land. The positive association between an uneven distribution of land and high productivity in Ecuador likely has more to say about the ability of wealthy land-owners to acquire the highest-quality land, and to ensure that government provides services which further increase the productivity of that land, than about any underlying causal relationship between inequality and productivity. Figure4.9: Landdistributionis moreunequalinthe SierraandaroundGuayaquil G a l a b p 0 , Source:Authors' calculationbasedon datafromthe 2001Third Agricultural Census. 4.21 In sum, there are large differences across cantons in the total dollar value of production and in productivity-with some cantons where small-, medium-, and large-scale farmers all have higher yields. There are also large differences inthe distribution of land across cantons. We now turn to an analysis of the relationship between productivity and land ownership on the one hand, and measures of consumption, earnings, and poverty on the other. 71The Gini coefficient is calculated using data on plot size. Hence a large value o f the Gini indicates that a large fraction of the land is concentrated in a small number of hands. Land quality and value, however, are not necessarily correlated with size so that this coefficient i s not informative about the concentration of wealth embedded in land. 72InFigure 4.9 cantonsare sorted inquintiles, with each quintile accountingfor 20 percent of total cultivated area. 94 Box 4.4: Access to landamongthe indigenous andAfro population. Approximately 50 percent of the rural indigenous population did not have access to agricultural land in 1996, and those that did often cultivated on poor quality land-for instance, only 13 percent of all irrigated land was inthe handsof indigenousfarmers (COMUNIDEC, 1996). Although equivalentfigures are not available for the rural Afro population in the province of Esmeraldas, high tenure insecurity and poor titlingpolicies most likely led to similar results for this group. Moreover, a large fraction of the land used by the indigenous population is in communal holdings, and hence potentially exposedto overuseandunder-investment. Inan attempt to recognizethe critical natureof this problem, the Constitution of Ecuadorwas amended in 1998 to acknowledge the `ancestral' right of the indigenous and Afro population to the landthey inhabit. It is difficult to assess, however, whether this measure contributed in any way to improve access to agricultural landamongthese groups. Other institutions, such as NGOs and the World Bank, have also tried to increase access to among the indigenous andAfro populationsusingtitling schemes andother strategies. We argue in this chapter than landless agricultural workers are at the bottom of the income distribution in rural areas. Low access to land among the indigenous and Afro population may then go a long way to explain high poverty rates among these groups, while policies aimed at increasing access may hold promiseas effective tools to reduceit (see the discussionbelow). AGRICULTURAL PRODUCTIVITY, HOUSEHOLDINCOMES, ANDPOVERTY: WHO STANDSTO BENEFIT FROM POLICIES TO INCREASEPRODUCTIVITY AND ACCESS TO LAND? 4.22 A critical distinction when analyzing the extent to which possible increases in productivity materialize as increases in household incomes and reductions in poverty i s between self-employed farmers who fa& their own land and agricultural laborers. In the case of agricultural laborers, it i s also important to distinguishbetween "permanent" employees-workers with stable contracts, whose wages may be set in part through collective bargaining (at least in larger farms)-and "temporary" employees with no such stability. Self-employedfarmers 4.23 Approximately 47 percent of the employed poor in rural areas are self-employed in the on- farm sector. For self-employed farmers, increases in agricultural productivity should translate directly into increases in household income. This is so almost by definition-unless there are large offsetting behavioral changes (such as a reduction in labor supply) or general equilibrium effects (such as a decrease in the price which farm output can command). Calculations on the basis of the 1999 ECV suggest that a one percent increase in the value of total on-farm output is, on average, associatedwith between a 0.16 and 0.32 percent increase inper capita consumption for households in the on-farm sect0r.7~ 73Ina weighted regressionof the log of per capita expendituresonthe log of total farm output, with the weights given by the expansionfactors in the survey, the coefficient i s 0.164, with a standard error of 0.024. On-farm output may be badly mis-measured and the coefficient on a regression of expenditure on output would then be biased towards zero. When output is instrumented with variables for whether the farmer received technical assistance, and uses improved seeds, fertilizer or pesticides,the coefficient on the regressionroughly doublesto 0.315, with a standarderror of 0.037. 95 4.24 To put these values in context, it i s important to remember two things. First, since we are using consumption per capita, rather than household consumption, as our measureof welfare, we will only observe a one-for-one increase (Le. a coefficient equal to 1) for households with a single member. The average household size in rural Ecuador i s about 5-6 people, so that a dollar-for-dollar increase would give us a coefficient of 0.16-0.20 under the assumption of no returns to scale in consumption. Second, it i s likely that increasesinoutput are the result of increasesininputsthat need to be paid for, so that net profits are only a fraction of the total increase in output. Taking all of this into account, the increases reported above seem fairly large. In addition, households running farms which are more efficient also have significantly higher per capita con~umption?~ Agricultural laborers 4.25 We next consider the link between productivity and household income for agricultural laborers, who constitute 20 of the employed poor in rural areas. Recall that the results in Table 4.2 suggest that the marginal product of labor i s low in small-scale farms, but high in large-scale farms. Ifmarkets are competitive andworkers arepaidtheir marginal product, we would expect agricultural wages to be higher in large- than in medium- or small-scale farms (hypothesis 1). Moreover, to the extent that average labor productivity (measure as output per worker) i s correlated with the marginal product of labor, we should also observe that farms where output per worker i s higher pay higher wages (hypothesis 2). Table4.4: Largefarmspayhigherwagesthansmallfarms Small-scalefarms Medium-scale farms Large-scalefarms Permanentunskilled employees Number of farms 413 1,704 15,885 Mean weekly wage ($) 14.5 19.5 21.5 Temporary unskilledemployees Number of farms 1,522 5,511 21,216 Mean weekly wage ($) 7.9 12.6 23.6 Permanentskilled employees Number of farms 70 441 8,478 Mean weekly wage ($) 31.2 42.0 41.6 Temporary skilled employees Numberof farms 60 165 803 Mean weekly wage ($) 24.2 41.5 64.9 Source: Authors' calculationsbasedon datafrom the 2001ThirdAgricultural Census. Note: Weightedmeans, with the weights givenby the number of workers ineach farm. Monthly wages for permanent workers havebeen transformedinto weekly wages dividing themby 4.3. 4.26 1n.order to test for the first hypothesis, we summarize mean weekly wages of permanent and temporary workers, skilled and unskilled, by farm size, in Table 4.4. The results show evidence in support of the hypothesis that large farms pay higher wages. Temporary employees are paid almost twice as much in medium- as in small-scale farms, and twice as much again in large- as in medium- 74We use the 1999 ECV to estimate Cobb-Douglas production functions, and calculate the technical efficiency parameter as before. The correlation between this measure of technical efficiency and log per capita expenditure i s positive (0.158), and significant at the 1percentlevel. 96 scale farms. These differences are also present for permanent employees, although they are not as large, especially between medium- and large-scale farms. It i s possible, however, that these farms are able to offer additional non-wage compensation, such as subsidized health care or schooling, so that overall their pay package i s higher than that of smaller ones. 4.27 To test for the second hypothesis, we turn once again to the data from the Agricultural Cen~us.7~The census collected information on the number of permanent and temporary employees, skilled and unskilled, hired by a farm, as well as on the average wage paid to these workers, by month for those who are permanent, and by week for those who are temporary. Inprinciple, we can use these data to relate the average wage to averageoutput per person at the level of the farm76. 4.28 Table 4.5 presents the results of regressions of mean farm wages on mean output per worker, and other controls. The table shows that a one percent increase in output i s associated with an increase in wages of between 0.10 and 0.32 percent (see the table notes for a detailed explanation on the different model specifications). Thus, like self-employed farmers, agricultural laborers stand to benefit from increasesinproductivity inthe farms that employ them. 1 2 3 4 5 6 7 Permanent unskilled workers Coefficient--0LS 0.10*** 0.10*** 0.04 0.14*** 0.06 0.04* -0.01 Coefficient--1V 0.32*** 0.2a*** -0.14 0.05 0.15 0.06 0.04 Number of farms 12,587 2,910 142 556 473 309 881 Temporary unskilled workers Coefficient-OLS 0.08*** -0.00 0.06* 0.09** 0.03 0.05* 0.0 1 I Coefficient-IV 0.24*** 0.24*** 0.04 0.22** 0.03 -0.01 0.24*** Number of farms 24,294 13,670 606 2,499 1,353 1,448 2,416 I 4.29 The relationship between farm-level output and wages among laborers is, however, weaker than that between farm-level output and consumption per capita among the self-employed. There are various possible explanations for this finding. It i s possible that wages of agricultural laborers are set more or less uniformly at the village, parish, or canton level, rather than at the farm level-either because there i s a norm-based system whereby laborers are paid what i s considered a "fair" or "acceptable" wage in a community, no matter what their productivity, or because output per worker The ECV surveys cannot be used for this purpose, as they collect information from households, but not from the farms inwhich theywork (unlessthey happento beself-employed). l6The data do not allow us to estimate the output (or the wage) of any single worker on a farm, and we cannot properly net out the contributions o f family labor to output. 97 i s hard to observe by the employer. If this were the case, we would expect to see that wages are similar within cantons, parishesor villages, and possibly quite different across them. 4.30 A decomposition of the variance of wages for unskilled permanent employees into within- and between-canton components (the lowest level of disaggregation possible with these data) shows that more than four-fifths (83.5 percent) of the variance in wages can be explained by differences within cantons. This suggests that wages are not set at the canton level, although we cannot rule out that the possibility that they are set at a lower level. A better understanding of wage setting mechanisms for agricultural laborers would require an in-depth study of agricultural labor markets beyond the scope of this chapter. The issue, however, i s of great importance and should be addressed inthe context ofthe upcomingEcuador RuralDevelopmentStrategy. 4.31 Perhaps a more convincing explanation for the patterns we observe involves the role of measurement error in the output measure, which would tend to bias the regression coefficients to zero. This seems likely because the coefficients of those regressions in which output i s instrumented with dummy variables for whether the farm uses imgation, improves seeds, fertilizers or pesticides, are generally significantly larger than the comparable OLS coefficient^^^. The relationshipbetweenagricultural productivity and poverty across cantons 4.32 We end this section of the chapter by considering the relationship between agricultural productivity and poverty at the canton level. Ifhigher productivity leads to higherhousehold income and consumption, and therefore to lower poverty, we would expect to find a positive correlation between the canton-level measures of output, productivity, and mean consumption. Moreover, we mightexpect that the relationship between higher agricultural productivity and a lower probability of beinginpoverty would be strongestfor households directly involved infarming. Table 4.6: Higher agricultural productivity is associatedwith lower ruralpoverty at the canton level Correlations with headcountindex Total Rural Self-employed Agricultural Agricultural population population farmers laborers semces Log -0.18 -0.25* -0.54* -0.63* -0.30* (dollar value of output per worker) Mean technicalefficiency -0.24* -0.40* -0.61* -0.63* -0.33* 4.33 We present suggestive results in Table 4.6. Specifically, we report the correlations across cantons between output per worker and productivity, on the one hand, and the headcount rates for the whole canton, for households in rural areas of the canton only, and (separately) for households in which the household head i s a self-employed farmer, an agricultural laborer, or employed in agricultural services. Table 4.6 shows that, as we would expect, the correlation coefficient between output or productivity and the probability of being poor i s larger for the rural population than for the population of a canton as a whole, and larger for households directly involved with farming than others. One should be careful not to attribute causality to these associations too quickly, as there are 77 These seem like plausible instruments: they are all highly correlated with output per worker, but arguably uncorrelated with wages other than through their effect on output. 98 likely to be a whole host of variables which affect both poverty rates and productivity in a canton, and which are not captured in these simple correlations. Nonetheless, the correlations do provide some evidence that cantons with higher productivity also have lower poverty, especially in rural areas, and especially for those households whose livelihoods dependdirectly on agriculture. 4.34 In sum, we find that there is a clear relationship between agricultural productivity and measures of welfare such as income and consumption for the self-employed as well as for agricultural laborers. In the next sections we discuss, separately, those policies which are likely to have the largest impact on productivity, and those which should improve access to land and the functioning of rural landmarkets inEcuador. POLICIESTO RAISEAGRICULTURAL PRODUCTIVITY Closing the gap: How to get on the Production Possibility Frontier 4.35 We argued earlier on in this chapter that differences in agricultural productivity (or output) could not be fully accounted for by differences in input use, and constructed a measure of technical efficiency based on total productivity. In this section, we try to explain differences in technical efficiency across farms interms of "policy variables" and "natural endowments". Table 4.7: The impact of different "policy variables" on technical efficiency varies with farm size I I Coefficients (distance from frontier) 0.307 -0.064 Assistance received -0.312 -0.137 -0.253 eceives credit -0.938 -0.559 -0.254 I ote from markets (90 minutes or more) -0.201 1s to intermediatebuyers -0.05 1 0.331 -0.65 1 -0.860 of formal education -0.024 -0.050 fagricultural education -0.069 -0.05 1 indigenouslanguage 0.088 -0.082 infall coefficient of variation 0.764 0.387 0.202 -0.098 0.202 -0.067 -0.154 -0.147 e of Canton Area subject to soil erosion 0.508 0.354 re of farmland devoted to usesother than field crops 2.249 1.966 re of crop area lost to weather, disease or pests 1.835 1.816 Source: Authors' calculations based on data from the 2001 ThirdAgricultural Census Note: All coefficients are significant at the 5 percent level or better except those shaded in the table. 99 4.36 As policy variables, we consider whether a farm is owner-operated, the fraction of land which has a formal title, whether a farm receives technical assistance or uses credit, measures of the degree to which a farm i s integrated into the formal economy (as given by three variables, whether a farmer sells output, sells to intermediate buyers, and whether the farm i s more than 90 minutes from an output market), and measures of the human capital of the farm operator (as given by his years of formal education, years of agricultural education, and whether he speaks an indigenous language). Similarly, we consider the rainfall coefficient of variation in the canton, whether the climate i s arid, humid, or dry, and the share of the canton which is subject to soil erosion as a measure of "natural endowments". The estimations also include variables for the share of farmland which i s devoted to uses other than field crops, and the share of crops which was lost to weather, disease, or pests. Results are presented in Table 4.7 and the mean values of each variable for different types of farms are presentedin Table 4.8. B y convention, these results are presented in terms of distance from the frontier, so that variables with negative coefficients will therefore have a positive relationship with output. (This can be confusing, as the interpretation we would generally give to coefficients in a standard regression would be that a negative coefficient indicates a negative relationship with output.) 4.37 The results of Table 4.7 can best be summarized as follows: Owner-operators are not more efficient than others Titling land is associatedwith higher productivity for medium- and large-scale farms, but not for small-scale farms. As we discuss below, however, land titling may be important for this group because it increases the security of tenure, and thus promotes investments. It can also increase access to landby the landless. Technical assistance i s associatedwith increasedproductivity for farms of all sizes. Availability of credit is associatedwith higher productivity for farms of all sizes, but particularly for smaller farms78. Unfortunately, no informacion i s available of the use made of the additional resources that credit provides. Controlling for other variables, being "remote" (more than 90 minutes away from a market) does not have an effect on productivity for small- and medium-scale farms, but increases productivity among large-scalefarms. Farms that sell part of their output are more productive; among small-scale farms, selling to intermediate buyers i s associated with higher productivity, while the effect i s reversed for large- scale producers. Years of agricultural-related education always boosts productivity, and the effect i s largest for small-scale farms. Formal education of the farm-operator matters for medium- and large-scale farms, but not for small-scale producers Speaking an indigenous language has a positive impact on productivity among small-scale farmers. One potential explanation for this result i s that, after controlling for other differences between indigenous and non-indigenous farmers, speaking an indigenous language i s an indication of the social capital available to small-scale producers. The result i s reversed for middle-scale producers, but i s quantitatively very small. 78A word of caution is necessary here since the variable used for the analysis is "uses credit", rather than "has potential access to credit". Current use of credit may be correlated with unobserved farm characteristics that result in higher productivity and are not controlled for in the analysis. Unfortunately, the available data does not allow to correct the endogeneity problem this poses, so that the estimated effect o f credit on productivity may be biased upwards. 100 0 Small-scale farms seem less able to cope with variable rainfall, even after adjusting for irrigation; they are also most sensitive to climate endowments, and to potential erosion problems. This could suggest the needfor catastrophic risk insurance. Lossesdue to nature hurt farms of all sizes, but hurt small farmers the most. Table 4.8: The meanvalue of "policy variables" varies with farm size Small Medium Large All 0.91 0.92 0.69 0.7 1 ~~ Owner operated Portion of land titled 0.66 0.68 0.82 0.81 Receives technical assistance 0.07 0.08 0.34 0.32 Receives credit 0.09 0.15 0.23 0.22 Time to market (in minutes) 20.79 36.44 79.83 76.49 Sells output directly to markets 0.12 0.07 0.02 0.03 Sells output to intermediaries 0.61 0.80 0.56 0.58 Years of formal education 4.25 4.36 9.62 9.22 Years of agricultural education 0.03 0.04 0.91 0.84 Speaks indigenous language 0.15 0.10 0.04 0.05 JUUILC. fiULIIUI5LaLuIdlIuII5 uaacu UII Udld llUlll LIIC LWl 11111ufiglILulluldl Lcllau5. The impact of policy reforms on output: Simulation results 4.38 The results in Table 4.7 provide clear indication of those policies which are, and are not, significantly correlated with higher farm productivity. However, because the units in which the policy variables are measured are not constant, it i s hard to compare the magnitude of effects. To make possible these kinds of comparisons, we conduct a series of simulations. Specifically, for each province, we first estimate mean output per hectare for all farms of a given size when all of the inputs (land, capital, labor), the policy variables, and the natural endowments are set at the mean value for that province. This gives the baseline scenario for the simulations. We then simulate the effects of various possible policy interventions, turning on policy interventions one at a time. The policy interventions or scenarios we consider are the following: (i)Inputs: Applying fertilizers and pesticides to all of the land which currently does not use it; (ii) Extension services: All farms to receive technical assistance; (iii) Market access: All small- and medium-scale farms are simulated to sell output to intermediaries, while all large-scale farms bypass these intermediaries; in addition, all farms sell output, and remote small farms are given improved access to markets; (iv) Credit: All farms use credit; (v) Formal education: All farm operators are given at least five years of formal education; and (vi) Agricultural education: All farm operators are given at least three years o f agricultural education. 4.39 The results from these six different scenarios, estimated separately by farm size, are presentedinTable 4.9 (panels A, B, and C.The results can best be summarized as follows: The returns to any single policy change tend to be substantially bigger among small- and medium-scale farms than among large-scale farms. This is, in part, a reflection of current differences in "access" to the policy interventions we simulate: Many more large-scale farms already have credit, so makingcredit available to all large-scale farmers tends to have a relatively small effect on output inthis farm-size class. 101 Access to credit is, by far, the single policy intervention that would have the largest impact on the output of small- and medium-scale farms. Credit would raise the mean output per hectare by 15 percent or more in five provinces (Cotopaxi, Esmeraldas, Pastaza, Pichincha, and Sucumbios) among small-scale farms, and by 20 percent or more in more than half the provinces in the country among medium-scale farmers. Agricultural education of the farm-operator has a reasonably large impact, between 7 and 9 percent, on the output of small- and medium-scale farms, and a smaller impact, between 3 and 4 percent, on large-scale farms. The impact of general schooling, not included in the tables, tends to be very small. Note, however, that increasing the schooling of workers other than the farm operator could conceivably have a large impact on output; unfortunately, we have no data on this. Access to pesticides and fertilizers has a particularly large effect on the output of medium-scale farms, although this impact tends to vary significantly by province. In the provinces of Esmeraldas, Loja, and Morona Santiago, mean output per hectare would rise by more than 20 percent if all medium-scale producers were to use pesticides and fertilizers inproduction. Increasing market access would favor large-scale farms most. The impact of access to extension services i s generally small (between 3 and 5 percent), and roughly equal across provinces and farm sizes. Table 4.9: Access to credit and agriculturaleducation have the largestimpact on productivity A. Small farms k ofa1 Pesticides farms Baseline and fertilizers Extension Markets Credit Ag. Educatioi Increaseassociated with each policyreform (inUS$/hectare) 43.0 314.4 16.2 13.7 21.5 44.8 28.2 15.1 170.0 12.8 7.2 8.6 24.6 15.4 37.3 292.5 . 21.2 13.2 16.4 41.6 26.3 15.4 469.2 10.7 18.0 11.9 49.3 37.3 37.0 423.4 24.6 13.6 9.8 47.1 28.3 37.2 278.0 7.9 12.9 11.5 42.6 26.3 13.6 427.7 22.2 16.1 11.5 52.2 31.2 2.3 140.9 11.6 6.7 6.3 22.1 13.3 13.4 386.7 10.9 13.1 9.2 32.5 24.5 49.0 218.0 15.1 9.6 14.4 30.1 18.9 15.3 264.2 25.6 10.6 10.6 34.2 21.4 10.6 231.1 13.4 8.6 3.7 25.4 16.3 16.8 204.2 10.4 9.5 9.8 29.2 17.9 4.4 297.7 28.9 11.8 13.9 35.4 21.8 4.0 735.0 6.0 19.9 45.1 94.5 49.7 9.0 149.8 13.9 7.9 11.8 23.6 14.2 39.4 313.9 19.0 14.6 24.3 46.7 29.0 1.o 228.1 9.6 13.4 27.8 42.4 24.9 66.0 661.8 10.5 22.2 7.2 69.5 46.2 2.5 215.1 15.9 8.6 12.2 30.6 18.2 421.8 20.3 19.1 15.9 60.7 36.4 102 r------ + B. Mediumfarms 76 of all Pesticides rovince farms Baseline and fertilizers Extension Markets Credit Ag. Educatioi hcrease associated with each policy reform (inUS$/hectare) 38.9 238.3 33.3 12.5 17.3 56.8 21.1 43.8 139.8 20.7 6.9 3.1 29.0 11.3 42.2 275.6 29.3 13.2 9.6 69.4 24.8 42.6 432.8 15.3 18.5 4.7 82.1 33.5 47.7 208.0 26.2 9.5 4.8 46.6 17.8 44.1 195.5 12.1 10.3 5.1 45.8 17.4 34.0 313.0 28.1 9.9 3.1 43.9 17.0 11.7 130.9 34.2 5.8 2.9 25.0 9.3 41.6 348.8 17.5 13.8 2.7 37.3 22.8 29.8 180.7 20.1 8.2 6.4 37.6 14.5 38.6 194.0 41.8 9.3 7.0 41.6 15.5 38.1 221.8 14.8 9.4 1.4 33.5 14.6 32.6 151.9 19.9 7.1 2.3 28.8 11.9 9.5 233.9 66.3 10.6 7.6 45.8 17.0 8.3 157.0 30.1 5.9 3.4 36.5 13.0 10.7 190.1 22.2 4.7 5.8 46.2 17.2 27.0 288.2 50.4 11.2 12.8 49.7 20.4 8.8 128.3 19.1 6.3 2.1 29.0 10.3 28.2 557.3 14.6 23.5 2.3 96.2 44.7 11.1 146.6 26.5 7.1 5.0 31.6 11.4 I 210.3 28.9 7.7 .1.3 31.2 12.0 % of all Pesticides farms Baseline and fertilizers Extension Markets Credit Ag. Educatioi 18.1 580.6 42.3 17.1 28.4 27.7 16.1 41.1 190.8 9.6 10.5 13.3 10.7 6.9 20.5 678.2 35.7 22.8 33.9 28.8 21.6 42.0 707.1 8.2 30.3 52.0 29.4 23.0 15.3 249.5 9.6 9.7 17.1 11.0 7.5 18.8 1,440.8 109.9 35.8 80.2 53.5 31.6 52.4 784.6 9.9 26.0 22.5 39.0 26.0 86.1 140.9 7.4 6.4 8.0 7.8 4.8 45.0 444.9 6.4 18.2 18.8 16.9 14.5 21.2 506.0 31.9 16.6 22.8 21.0 9.3 46.1 234.9 14.4 13.1 19.0 13.4 8.6 51.2 428.6 3.9 13.5 16.6 17.8 12.1 50.5 154.1 6.0 7.8 11.9 8.3 5.4 86.0 281.8 19.2 15.8 21.6 16.5 10.1 87.6 141.0 6.1 6.9 8.2 8.3 5.0 80.3 69.2 3.4 3.8 5.5 4.0 2.3 33.6 2,682.4 226.9 57.8 82.3 70.9 52.8 90.1 83.3 5.0 4.5 6.8 4.9 3.0 5.8 843.5 13.4 32.4 58.6 34.6 29.6 86.4 186.8 10.8 9.9 14.7 10.6 6.8 303.1 6.5 12.8 12.6 16.4 9.8 Source: Authors' calculations basedon datafrom the 2001 Third Agricultural Census 103 4.40 Measures aimed at increasing agricultural productivity, the two most effective policies, should then include:. 0 Increased access to rural credit. Rural credit i s restricted by the difficulties of many producers, particularly those in small farms, to comply with the administrative and guarantee requirements of financial institutions. As a result, most of the existing credit i s informal, or provided by small savings and loans cooperatives. These cooperatives need to be strengthened, and so do other institutions with similar goals, such as women's credit groups (cajas solidarias). Credit regulation also needs to be modified to allow for the use of family assets, such as land and livestock as collateral. 0 Increased access to technical assistance and agricultural education. Agricultural technology exists only in specific areas, for specific export crops, while formal agricultural training and technical assistance are generally lacking, especially on small land holdings. Action to palliate this situation could be taken by supporting the National Institute for Training of Rural FarmWorkers (Capacitacion Campesina), operated by the Ministry o f Agriculture in a fairly decentralized fashion, as well as agricultural researchand development initiatives. 4.41 As a next step, we can combine the changesinoutput implied inthe simulation results above with the impact of changes in output on the per capita expenditures of households in the 1999 ECV. Recall that a one percent increasein farm output was associatedwith somewhere between a 0.16 and 0.34 percent increase in per capita expenditures of self-employed farmers. A simple back-of-the- envelope calculation suggests that an intervention which would raise the output of all farms by 20 percent-at the upper bound of the interventions discussed-would increase the per capita expenditures of self-employed farmers by between 3.2 and 6.9 percent. This, inturn, translates into a reduction of poverty of between 1.2 and 3.4 percentage points for self-employed farmers. This intervention would also have an effect on poverty among agricultural laborers, since a fraction of the increases in productivity are passed on as higher wages. Finally, as we argue below, increasing productivity in the on-farm sector i s likely to have positive spillover effects on the off-farm sector. But rural poverty in Ecuador is widespread and deep, and these simple calculations provide a sobering reminder that it will take considerable effort on many fronts, as well as a great deal of time, to bringa large fraction of households inrural areas inEcuador out of poverty. 4.42 We conclude this section, finally, with two words of caution. First, it is possible that providing some areas with a whole package of interventions-credit, technical assistance, agricultural inputs-may be more efficient and may have higher returns than providing all of the interventions ~eparately.~~Second, one should note that the estimates inTable 4.9 provides guidance of the relative benefits of different interventions in terms of their marginal impact on output, but not of the relative costs-direct, or indirect. For example, it i s not clear how much more or less expensive it would be to provide all small-scale farmers with credit or with an additional year of agricultural education. Moreover, costs to others-for example, the environmental costs associated with more use of fertilizers and pesticides-would also have to be estimated and taken into account. Nonetheless, Table 4.8 makes clear what policy interventions have the potential to increase output the most. Among these, access to rural credit particularly stands out. The Bank's forthcoming Ecuador Rural Development Strategy could provide recommendations on how best to estimate the costs of some of the proposed interventions, and how to design concrete programs and projects to implement them. 19 The calculations above shed no light on this question. Estimates of the Production Possibility Frontier results with interactions betweendifferentinterventions tendedto be very fragile, and we therefore do not report these. 104 POLICIESTO INCREASEACCESS TO LAND 4.43 In Ecuador, as elsewhere in Latin America, land is distributed very inequitably. The poor distribution of land reflects a historical and institutional legacy-much of it, going back to colonial times. In addition to this historical legacy, there are legal and economic barriers to a better functioning of landmarkets. A recent review of landlegislation inLatin America (FA0 2002) shows that Ecuador has one of the most rigid land markets in the region: It i s one of only two countries in Latin America (the other one being Honduras) in which there i s a blanket prohibition against sharecropping, and one of only a handful of countries which have legislation on the books which allows for expropriation of land-for example, if the land i s not used for its "social function" ("si la tierra no cumple con su fincion social"). Rigid land legislation and uncertain property rights both depress landrental and sales markets. 4.44 Poorly functioning land markets have serious costs-both interms of efficiency and, because wealthy landowners currently own much of the land, in terms of equity. Policies to improve the functioning of land markets in rural Ecuador are therefore critical. We discuss such policies below, separating them into those which have the potential to improve tenure security, and those which would help increase the number of transactions in land markets to a more efficient level. Of course, policies cannot be so neatly split-indeed, uncertainties about landtitling are one of the main reasons for the under-development of land rental markets. Still, we believe that the distinction of policies i s a useful one. As with those interventions which are meant to increase agricultural productivity, we leave the costing and design of specific policies as a subject for discussion in the forthcoming Ecuador Rural Development Strategy. Promoting tenure security 4.45 When property rights are uncertain, land owners will be reluctant to rent out landfor fear that they will not be able to recover it. Ifthey rent out land at all, they will favor close kin over strangers, and short-term contracts over long-term contracts. Uncertain title also has other, important costs: Land-owners may under-invest in their own land if their legal claim can be disputed, and they may be unable to use their land as collateral for credit. Potential land buyers, meanwhile, will be unwilling to purchase land for which they cannot be given legal title, while landrenters on short term contracts will under-invest in the land. Inadequate tenure security thus reduces investment in land, depresses its value, and makes it more likely that large landholders leave tracts of land idle while a large number of agricultural laborers are left landless-an outcome which i s both inefficient and regressive. Policies which would improve tenure.security include: 0 Removinglegal and other barriers to land titling. 0 Updatinglandregistries. 0 Making consistent legislation and removing uncertainty about the threat of landexpropriation. 0 Setting inplace an effective system for conflict resolution of landdisputes. Encouraging landtransactions 4.46 A recent World Bank document convincingly makes the case for flexible land markets: "Land transactions can play an important role by allowing those who are productive, but are either landless or own little land, to access land. Land markets also facilitate the exchange of land as the off-farm economy develops and, where the conditions for doing so exist, provide a basis for the use 105 of land as collateral in credit markets."" Indeed, inEcuador as in much of Latin America, the rigid nature of land markets i s a fundamental reason for the lack of access to land by the poor. Policies that would make landmarkets inEcuador more flexible include: Removing the blanket restriction against share-cropping and other rental restrictions. Governments in many countries have tried to limit sharecropping both because of concern about inefficiencies and the perceived "exploitative" nature of the sharecropping relationship. In practice, however, legal limitations on sharecropping and legal ceilings on land rental values have both reduced land access and equity-contrary to the professed goals. In Kenya, for example, it i s estimated that a ban on share-cropping has led to an efficiency loss of at least 10 percent.'l Drawingup standardcontracts for landrentals to reduce transaction costs. Removing transferability restrictions, so that properties can be legally passed down-for example, within a family-or sold. Evaluate the possibility of implementing a land tax. Land taxes can make holdings of idle, unused land much more expensive, and can therefore provide an incentive for land sale or rental. If small producers are more efficient than large-scale producers, land taxes can also have a positive impact on productivity. 4.47 Ecuador has one of the most unequal distributions of land in the world, and one of the most rigid codes regulating land sale and rental markets. In addition, there is considerable uncertainty about land ownership and title. Reducing tenure uncertainty and improving the functioning of land markets are clear policy imperatives. THERURALOFF-FARMSECTORINECUADOR 4.48 We conclude with a brief discussion of the opportunities and constraints inthe rural off-farm sector inEcuador, focusing on the degree to which there i s evidence of linkages between the on-farm and off-farm sectors.'* InEcuador, as elsewhere, the rural off-farm sector plays an important role as a means of income diversification for rural households. This i s critical because of the seasonal variation inagricultural incomes, as well as because of the highdegree of exposure of the agricultural sector to covariant shocks such as those caused by weather (floods, droughts) and, for export crops, to international commodity prices. In addition, it has often been suggested that there i s a virtuous cycle between agricultural intensification and the rural off-farm sector. This literature argues that agriculture can stimulate the off-farm sector through backward linkages-for example, when farmers need services such as plow production, machinery repair-as well as forward linkages-for example, where agricultural products need to be processed, milled, or canned. There could also be consumption linkages, as rising agricultural incomes stimulate the demand for goods and services produced in nearby towns. Finally, rising agricultural productivity could release labor from the on- farm sector. The expanding off-farm sector, in turn, could then lead to further agricultural intensification through lower inputcosts, profits re-invested inagriculture, and technological change. 4.49 What i s the evidence of such linkages between the on-farm and off-farm sectors? Empirical results suggest that there are considerable differences across countries in the extent to which a 8oWorld Bank,Land Policiesfor Growth and Poverty Reduction(May 2003, p. xxix). 81Collier (1989), citedin World Bank 2003, p. 119. 82This discussiondraws considerably on pre-existingwork, in particular on work on Ecuador by Lanjouw (1999), and a discussionof concepts and empiricalfindings from aroundthe world by Lanjouw, and Lanjouw (2001). 106 virtuous cycle actually materializes. In particular, some researchers have argued that the backward and forward linkages have been much stronger inEast Asian countries than in Latin America.83 We present some, albeit tentative evidence that higher agricultural productivity in Ecuador i s in fact associated with more employment and a lower incidence of poverty for households in the off-farm sector. 0 n 1 0 .2 Technical.4efficiency .6 .8 Source: Staff calculations basedondata from the 2001ThirdAgricultural Census and the 2001 PopulationCensus. Note: Householdsare assigned to employment categoriesby the employment of the household head. The size of the circles is proportional to total on-farm output in a canton.. ~~ 83See de Janvry and Sadoulet (1993). 107 Figure 4.11:Agricultural productivity andfood processing 0 .1 0 0 O O .. n 0 0 U 0 - 0 .2 Technical.4efficiency .6 .8 1 0 " 0 0 0 .2 Technical.4efficiency .6 .8 Source: Staffcalculationsbasedondata from the 2001Third Agricultural Census andthe 2001PopulationCensus Note: Householdsare assigned to employment categoriesby the employmentof the householdhead. The size of the circles is proportionalto totalon-farmoutput inacanton. . Figure 4.12: Agricultural productivity andthe non-agriculturaloff-farmsector 0 P 0 0 0 .2 Technical.4efficiency .6 .8 0 0 0 0 O Q 0 0 0 0I .2 I Technical.4efficiency I .6I .8 I Source: Staff calculations basedon data from the 2001 ThirdAgricultural Census and the 2001Population Census. Note: Householdsare assigned to employment categories by the employment of the household head. The size of the circles is proportionalto total on-farm output ina canton. 4.50 Figures 4.10 through 4.12 graphically display the correlations between technical efficiency and employment shares (upper panels), and technical efficiency and poverty levels (lower panels) for households employed in agricultural services (Figure 4.10), food processing (Figure 4.11) and other 109 activities inthe rural off-farm sector (Figure4.12). The size of the circles i s given by the fraction of total agricultural output in the country that i s produced ina canton. Figures 4.10 and 4.11make clear that a more productive agricultural sector i s associated with more employment and less poverty in both agricultural services and in food processing. Figure 4.12 shows that there i s no significant relationship between technical efficiency and employment shares in the non-agricultural off-farm sector, although cantons with higher agricultural productivity also have lower poverty rates in the non-agricultural off-farm sector.84 4.51 Taken together, Figures 4.10 and4.11present evidence that "all good things come together": Cantons where the on-farm sector i s more efficient are also those in which the off-farm sector i s vibrant. However, these results should be interpreted with care-in particular, because the extent to which the causality flows from increasing agricultural productivity to the development of the off- farm sector (rather than inthe opposite direction) is not obvious. Also, there are likely to be ahost of omitted canton effects that cause at least part of the increase in technical efficiency as well as the lower poverty rates. Still, the results in Figures 4.10 through 4.12 do suggest some spillovers from the on-farm to the off-farm sectors. This has important policy implications. First, increasing agricultural productivity through the levers we have discussed above-credit, use of inputs in production-may well be the best way to stimulate the off-farm sector inrural areas. Second, there may be scope for some interventions in the off-farm sector-for example, raising general education levels, making credit or markets available for small-scale businesses-which, in turn, have positive spillover effects for the on-farm sector. 4.52 Finally, in Chapter 2 we showed that movements inemployment from the on-farm to the off- farm sector led to poverty reduction. How are those results then related to the discussion presented here? Very briefly: To the extent that increases in agricultural productivity release labor from the on- farm sector and, simultaneously, stimulate the demand for off-farm services, they cause employment to shift from the on-farm to the off-farm sector (as shown in Figures 4.10 to 4.12). This process will then lead to poverty declines for two reasons. First, we showed above that increases in agricultural productivity lead to lower poverty. Second, average income i s higher in the off-farm sector than in the on-farm sector, so that a change in the composition of employment that favors the second will translate into a decline in povertys5. This decline in poverty i s then correlated with a shift in employment from the on-farm to the off-farm sector. CONCLUSIONS 4.53 Poverty in rural areas in Ecuador i s widespread and deep. Low levels of income and consumption come hand in hand with other forms of deprivation: Low levels of education, poor health status, marginalization from economic and political processes. Many of the problems we observe now are the results of centuries of exploitation and neglect of rural areas in Ecuador-from colonial times onwards. It would be unreasonable to expect that all of these conditions can be reverted overnight. And yet there are at least some reasons to be hopeful-for example, from the increasing participation of the indigenous in national politics, and from some pockets of efficiency and high agricultural productivity (such as among producers of flowers or garden vegetables). The critical issue i s identifying those policies which are likely to have the largest impact on poverty. 84All of the correlations, except the correlation between technical efficiency and employment shares in the non- a ricultural, off-farm sector are significant at the 1percent level or better, both weighted and unweighted. "Notice that we should not expect declines inoff-farm sector wages to offset the positive effect associated with increased employment since this increase is demand-driven. 110 4.54 We have argued in this chapter that increasing productivity and improving the functioning of land and other markets inrural areas are critical policy areas. Specific interventions inthese areas- providing credit, inputs into agricultural production, promoting tenure security, facilitating the rental and sale of land-all have the potential to make important dents into poverty in both the on-farm and off-farm sectors inrural Ecuador. 111 5. SOCIALSERVICESAND THE POOR 5.1 The education and health status of a population i s a dimension of well-being in its own right. In addition, better education and health are strong predictors of higher incomes, and are therefore critical indetermining consumption or income-basedpoverty. 5.2 The Ecuadorian economy has undergone enormous transformations in recent years, and the social sectors have not been immune to these processes. Economic and fiscal instability have led to significant changes in the budgets allocated to these sectors, while changes in relative prices have changedthe private direct and opportunity costs of investing inhealth and education. 5.3 This chapter describes Ecuador's performance in terms of selected education and health outcomes, as well as the relationship between these outcomes and poverty. In addition, because social outcomes are the product of demand and supply factors, the chapter examines both utilization rates and financing of different programs and services, paying attention to their distributionover time and across households and space. 5.4 The main findings are: Ecuador under-performs interms of education and health outcomes relative to international standards, even after controlling for differences in development levels. However, there i s significant variation ineducation and health outcomes across provinces, which i s correlated with, but cannot be fully explained by poverty differences. Such variation has important policy implicationswhen thinkingabout how and where to invest (additional) social sectors moneys. Social expenditures have declined significantly over time, especially education and health expenditures. They also tend to be pro-cyclical so that the least resourcesare available when the need for them i s largest. Overall social spending i s progressive (i.e. it tends to benefit the poor relatively more than the rich), but there i s significant variation across different programs and services. Recent initiatives to improve the targeting of the two largest cashprograms (the Bono Solidaxio - now called Bono de Desarrollo Humano- and the gas subsidy) have the potential to significantly increasethe level of progressivenessof both programs while generating important fiscal savings -especiallyinthecaseofthelatter. 5.5 The rest of the chapter is organized as follows. The first section benchmarks Ecuador's performance ineducation and health against international standards and examines the extent to which different provinces deviate from the country's averageperformance. The secondsection describes the evolution, cyclicality and incidence of social expenditures, and discusses the policy implications associated with these patterns. The third section evaluates two recent initiatives to re-target the Bono Solidario and the gas subsidy. Finally, the fourth section discusses some policy recommendations derived from the analysis and concludes. 5.6 The chapter builds on previous work done by Vos et alia (2003), and by the World Bank (2000c and Fretes et alia, 2003). 112 SOCIAL OUTCOMES ECUADOR:COMPARATIVEPERSPECT~VE IN A 5.7 How well does Ecuador perform in terms of health and education outcomes compared to other countries? Are these outcomes similar across different regions and provinces inthe country? In this section we provide answersfor thesequestions. International comparisons of healthand educationoutcomes 5.8 We start with a benchmarking exercise that compares Ecuador to other countries after controlling for differences in development levels. For this purpose we follow World Bank (2003~). We run OLS regressions of selected health and education outcome indicators on (log) GDP per capitas6,and compare predicted and actual values for each one these different indicators to establish whether Ecuador under-performs (i.e. negative difference) or over-performs (Le. positive difference). 5.9 Accounting for data availability, we consider four indicators: (i) infant mortality, (ii) chronic malnutrition (wasting and stunting), (iii) (net) primary enrolment rates, and (iv) (net) secondary enrolment rates. Data on (i) and (ii) from the Pan-American Health Organization (PAHO), come while data on (iii) and (iv) come from World Bank databases. The sample contains all countries for which data was available. 5.10 We find that, as expected, higher income levels are correlated with better health and education outcomes (Figure 5.1 through 5.4). There is, however, a significant amount of cross- country variation that cannot be explained in terms of development levels - that is, some countries appear to be over-performers, and some others appear to be under-performers. 5.11 Ecuador appears to systematically under-perform in terms of health outcomes, and exhibits more mixed results in terms of education outcomes. Inparticular, primary enrolment rates are almost 10 percentage points higher, at 90 percent, than would have been predicted by Ecuador's level of development, while secondary education rates almost coincide with their predicted level. 5.12 These results, however, require some qualification and, in the case of education, need to be taken with caution. For instance, although still worrisome, the incidence of malnutrition among children under 5 today i s significantly lower than it was 15 years ago (26 percent, compared to 37 percent in 1986) - an improvement that cannot be explained by increasesinGDPper capita alone. 5.13 Regarding education, we must consider that, despite high enrolment, primary-school graduation rates and, particularly, education quality are extremely lows7 due to a multiplicity of factors, ranging from poor training of teachers to teacher absenteeism to low investment levels and funding (see Box 5.1 for a discussion on teacher absenteeism). Similarly, even though national secondary enrolment rates are neither too low nor too high, coverage i s very uneven across regions and, particularly, across indigenous and non-indigenous groupss8(see Box 5.4 for a discussion on social outcomes by ethnicity). 86GDPper capita is expressedinPPPterms to control for exchange rate volatility. 87Standardized Math and Language tests administered by the Ministry of Education in 1996, 1997, 1998, and 2000 to public and private school students in second, sixth, and ninth grades showed important deficiencies in both disciplines, particularly inrural areas and amongpublic school students. "Secondaryenrolmentratesare 22 percentamongindigenouschildren, comparedto 54 percentamongwhites (Lebn, 2003). 113 Figure 5.1: Infant mortality is highfor Ecuador's developmentlevel... 90 r 70 - u Y z Haiti : Bolivia a i?-6 0 B 5 0 - i 4 2 40- 0 2 3 0 - 3 3 20 - 10 - 0 I I I I Source: Authors' calculationsbasedon data from PAHO (2003). Figure 5.2: ...andso are wastingandstunting m - 4 - Hmynr w 0 7.29I I I 1 , , , 9.42 729I I 9.42 Logof Per CapitaGDP LcgofPer+ta GDP Source:Authors' calculationsbasedondatafromPAHO(2003). I'hemodel is estimatedusingpopulation-based weights. PanelA: Wasting rates amongchildrenunder5. PanelB: Stuntingrates amongchildrenunder 5. I14 Figure5.3: Ecuador over-performsin(net) primary enrollment.. . 100 St.guci Bobvia Br@l 70 7.5 I aI 8.5 I 9I 9.5 I Ln PerCapitaGDP (ppp) Source: Authors' calculationsbasedon datafrom the World Bank (2003). The model is estimatedusingpopulation-basedweights. Figure 5.4: ...and performs as predictedon (net) secondary enrolment. 80 St.&uci Arggntin Jawica Botjvia Mqico Vengzuel $ Cos@ Ri Pareuay -0 Grepda C 8 9) 40 Dornjnica v) GuaBmal 20 7.5 1 8I 8.5 I 9I 9.5 1 Ln Per CapitaGDP (ppp) Source: Authors' calculationsbasedon datafrom the World Bank. The model is estimatedusingpopulation-basedweights. 115 Box 5.1: Teacher absenteeisminprimaryschools During 2002-2003 atotal of 102primary schoolsemploying720 teachers were survey as part of a World Bank study on teacher absenteeism(Lhpez-Calixet alia, 2003). Schools were visitedinDecember 2002 andJanuary- February 2003 to measureteacher attendance. On average, teachers were in the classroom 75 percent of the time, although in 25 percent of those occasions they were not teaching. Teacher absenteeismrates, measuredas the fractionof unexplainedor unaccountedfor teacher absences,was about 16percent. Percentageof time the teacher was found... December January-February 2002 2003 Inthe classroomteaching 56.8 61.2 Inthe classroomnot teaching 14.8 14.3 Out of class on scheduledbreak 0.0 0.0 Out of class but in school premises 6.2 5.8 Doing administrativework 1.6 2.2 Cannot findabsent 18.5 15.1 Accompanying surveyor 1.9 1.2 Withincountry variation inhealthand education outcomes 5.14 We move now from cross-country comparisons to within country comparisons in order to explore the extent to which health and education outcomes across provinces differ from the national average, and, if so, the reasons why. We apply the same analytical principle used above - namely, that social outcomes are positively correlated with development levels -, but choose poverty rather than income as the main explanatory variable89. 5.15 Ingeneral we find that poorer provinces tend to fare worse than richer ones interms of both health and education. However, as with country-level comparisons, there i s a significant amount of across-province variation that i s not explained by poverty (or income). For instance, infant mortality rates vary from 30 percent in Los Rios to 60 percent in Carchi even though both provinces have similar poverty rates. Similarly, net primary enrolment rates are 10 percentage points higher in Tungurahua, at 93 percent, than in Esmeraldas, at 83 percent, despite no significant differences in poverty levels betweenboth provinces. 5.16 Interestingly, although uncorrelated with poverty, this variation does not seem entirely random. Infact, a closer look at Figures 5.5 and 5.6 reveals that provinces inthe Sierra tend to under- perform in terms of infant mortality and malnutrition, compared to provinces in the Costa, while provinces in the Costa (especially in the north) tend to under-perform in terms of education, particularly primary education, compared to provinces inthe Sierra. 89We couldnotdo this whencomparingcountries inLatinAmerica since comparablepovertyfigures for all countries are not available. 116 Figure 5.5: Provincesinthe Sierra under-perform inhealth, while... CAWHI 60 - COT@PAXI h m m -.-.-.-- m r $ 40 - v) E la -- 2r - la 20 - 0 - 193739 .6807 poverty Source: Authors' calculationsbasedon data from SIISEandthe 2001 PopulationCensus. Figure5.6: ...provincesin the Costaunder-perform ineducation 50I II 85-I 80 193739 6807 193739 6807 poverty poverty Source:Authors' calculations basedon datafrom SIISE andthe 2001 PopulationCensus. 5.17 Acknowledging that systematic differences in health and education outcomes exist across regions, and understanding why this i s the case i s extremely relevant if we are to design and implement effective policies. For instance, the fact that similarly poor provinces inthe Sierra and the Costa exhibit significantly different malnutrition rates implies that targeting schemes for nutrition programs that are exclusively basedon monetary poverty indicators may not be adequate". Instead, a more comprehensive criterion, that accounts for differences in food consumption per capita and, particularly, in dietary habit across regions (e.g. the traditional diet of the Sierra i s high in calories, Conceptually it is possiblethat malnutritionoutcomes are more correlated with structural poverty indicators, such as the Unmet Basic Needs(UBN)index - discussedinChapter 2 - or the SELBEN index - discussedlater on inthis chapter -, than with monetarypoverty indicators. This, however, remainsan open empirical question inEcuador. 117 but low inprotein, compared to that of the Costa), as well as for income differences, has the potential to substantially increase the impact of such programs (see Box 5.2 for a discussion on regional differences infood consumption per capita). Box 5.2: Analyzing regionaldifferences infood consumption per capita Using data from the 1998 Encuesta de Conditiones de Vida and the 2001 Population Census and applying a methodology similar to that described in Chapter 2 for the poverty map, Ecuadorian researchers working in collaboration with this team have prepared a map of food consumption per capita for Ecuador in2001. This map can then be usedto assess whether the value of food intakes indifferent areas of the country is below or above that of a pre-defined food basket". Although this is a somewhat crude approach to the issue of malnutrition, in the sense that it does not directly measure nutrient and caloric intakes, it provides a first approximation since we would expect insufficient food consumption and malnutrition to be related9*. We briefly summarize the main results of this work here and refer the interested reader to Larrea (2003). It is important to notice that, although we present the results at the regional level, similar figures are available at the canton-level from the mapping exercise: 0 Food consumption levels fall short of.the reference basket for 60 percent of the population. 0 Food consumption levels are lowest in the Oriente, and highest in the Costa. Food consumption levels are lower in rural than in urban areas, and they are especially low in the rural Sierra (rural areas in Chimborazo, Cotopaxi, Imbabura, Loja, Bolivar, Tungurrhua and Caiiar are particularly affected, in this order). 0 A moderate improvement can be observed over time, when food consumption levels in 2001 are compared to those of 1990. Table B.5.2.1 Foodconsumptionby regionand area - - Consumptionbelowbasicbasket Gap Severity (%I Nacional 61.0 22.9 11.5 Rural 78.1 32.9 17.4 Urban 50.6 16.8 8.0 Costa 57.2 19.7 9.4 Rural 74.8 27.2 12.6 Guayaquil 44.2 13.8 6.3 Urban 54.4 18.7 9.4 Sierra 64.2 25.5 13.3 Rural 81.7 37.4 20.9 Quito 44.0 11.6 4.5 Urban 57.2 20.9 10.3 Oriente 71.3 32.4 18.4 Rural 73.7 34.0 19.3 Urban 65.3 28.0 16.0 Galapagos 37.7 10.2 4.1 Source: Larrea (21 3). "Thereference food basket used in the analysis corresponds to that developed in World Bank (1996) for the purposeof '*Worki defininga extremepovertyline, once its costs hasbeenupdatedusing 1999prices. s currentlyunderway to produce a chronic malnutritionmap usingthese data. 118 5.18 A similar exercise aimed at improving our understanding of the factors underlying the observed differences in primary enrolment rates between the Sierra and the Costa should be undertaken, and such differences should be taken into account in targeting (primary) education expenditures (see Box 5.3 for a discussion on targeting of education spending). Box 5.3: Targetingof EducationSpending I Regional and provincial education budgets have traditionally been allocated following historical values and trends. As part o f the social sector reform program supported by the Programmatic Human Development Reform Loan (The World Bank, 2003), the Government of Ecuador i s currently considering the use of an alternative formula to allocate education spending across different areas of the country. This formula, developed in the spirit o f expenditure capitation schemes, incorporates information on: (i)size of the schooling-age population, (ii) current enrolment rates, as well as (future) improvements in enrolment, and (iii) poverty levels. Simulations of budget allocations show that this scheme will generate significant redistribution of resources across provinces. LEVEL, COMPOSITION, CYCLICALITY AND INCIDENCE OF SOCIAL SPENDING 5.19 How has the level and composition of social expenditure in Ecuador changed over time? I s social expenditure higher when it i s needed most? Who benefits from social expenditures? In this section, we attempt to provide some insights into these questions by analyzing changes in the level, composition and cyclicality of social expenditure over time, as well as its incidence across households and provinces. Moreover, since expenditure i s only one of many factors behind health and education outcomes, when possible we complement this information with data on utilization rates of different services93. 5.20 As we mentioned in the introduction to this chapter, the discussion presented here draws heavily on previous work done by Vos et alia (2003) and Fretes et alia (2003). We strongly encourage interested readers to consult these sources since we only offer a glimpse of their results here. 5.21 This section, however, is not limited to the contents of these studies. It complements the discussion on social expenditure incidence across households, presented in Vos et alia (2003), with new results on incidence across provinces using the poverty map developed for this report and described in Chapter 2. Given that, as we saw in the previous section, there are important geographic differences in social outcomes, and that a large proportion of social expenditure, especially education spending, i s distributed according to geographic criteria (most normally following historical trends), this analysis i s highly relevant for policy. 93The assumptionis that socialexpenditurewill capture supply-sidefactors, while data on utilization will capture demand- side factors. This, of course, is not entirelyaccuratesince often utilization ratesare low because the servicei s not available (e.g. secondaryenrolment rates are low in areas where there are no secondary school), and availability is itself a function of expenditure. We hope, however, that it providesafirst approximation. 119 I As with Box5.4: Educationand healthamong the indigenousand Afro population provinces, we find significant variation in social outcomes and access to social services across different ethnic groups. This box briefly reviews some of these differences. Education outcomes are worse and enrolment rates lower among the Afro and, particularly, the indigenous populations than among whites. According to the 2001Population Census, the illiteracy rate stands at 26 and 13 percent for indigenous and Afro adults, respectively, compared to 10 percent for whites. Similarly, the average indigenous adult has completed 3.7 years of education and the averageAfro adult has completed 5.6 years of education, while this number i s 6.6 for whites. Finally, enrolment rates at all levels are lowest among indigenous children andhighest among white children (Table B.5.4.1). Higher-than-expected given education outcomes, their income level, among the Afro population are mainly due to the concentration of this groups in urban areas, while poor education outcomes and low access rates among the indigenous population are to some extent the result of the concentration of this group in rural areas and the low quality and coverage of bilingual education. Table B.5.4.1: Enrolment rates are lowest among the indigenous populationand highest among whites Primary Secondary Tertiary Enrolment rates (%) Indigenous 85 22 3 Afro 83 31 6 White 92 54 19 Total 89 44 14 Unfortunately we do not count with equivalent data (i.e. disaggregated by ethnicity) for health outcomes or access to health services. We know, however, from our analysis of health outcomes across provinces, that infant mortality and malnutrition rates are particularly high in the rural Sierra, where the indigenous population i s concentrated. In addition it i s unreasonable to assume that, as with education, access to health services is most likely lowest among the indigenous population, especially those in ruralareas, and somewhat higher among the Afro population inurban areas. The Seguro Social Campesino, a health-insurance scheme for the rural poor, i s a pro-poor program (see incidence analysis below) that has undoubtedly contributed to increase access to basic health services among the indigenous poor living in rural areas. The coverage of the program, however, i s small limiting the scope of its impact. Finally, the lack of capability in the current system to accommodate cultural differences may have played an important role in keeping the indigenous population away from health centers. Trends inSocial Spending: Level, Composition and Cyclicality 5.22 Social expenditure levels are low inEcuador, both in terms of GDP and on a per-capita basis, as a result of continuous declines over the past two decades. Ecuador spent approximately 6 percent of GDP (or US$60per capita) on the social sectors during the 1990s, compared to a Latin American average o f 12 percent (or US$550 per capita)94. The per-capita figure improves slightly, to approximately US$lO, when Social Security i s considered. This i s misleading, however, since 94Figures are in 1997US$ terms. ECLAC (2002), and Vos et alia (2003) for Ecuador. Spending by the central government only. Expenditure does not include Social Security. 120 pensions are paid only to those retiring from the formal employment and unlikely to reach the needy. In addition, social expenditure has fallen significantly over time and, although there evidence of some recent recovery, it currently stands below 1980slevels (Table 5.1). Table 5.1: Socialexpenditure (as a percentageof GDP) hasfallen dramatically over time. 1973 1979 1981 1984 1988 1992 1996 1998 2000 2002 1 ~ ~ Total 3.8 4.6 6.3 4.9 4.7 5.2 3.8 3.4 3.6 4.5 Education 3.2 3.5 4.8 3.7 3.2 3.8 2.5 2.4 1.7 2.4 Health 0.5 1.0 1.3 1.1 1.3 1.1 0.8 0.7 0.6 1.2 Social Assistance 0.1 0.1 0.2 0.1 0.2 0.3 0.5 0.2 1.3 1.o Bono Solidario 0.0 0.8 0.4 Other 0.1 0.1 0.2 0.1 0.2 0.3 0.5 0.2 0.5 0.6 5.23 Overall reductions in social expenditure were not distributed equally across different sectors, leading to an effective transfer from the health and education sectors to social assistance - particularly after the inclusion of the Bono Solidario program in the Ministry of Social Welfare portfolio in 2000. Health and education spending declined from 1.3 and 4.8 of GDP in 1981, to 0.6 and 1.7 respectively in 2000, while funding for social assistance climbed from 0.2 to 1.3 of GDP in the same period. Although these trends were partially reversed in 2002, spending levels inhealth and education continue to fall below historical levels. 5.24 As we discussed in Chapter 1, social expenditure appears to be fairly volatile across time, as it responds to economic, fiscal and political cycles. This is worrisome since procyclical social expenditure, by injecting relatively more resources into the social sectors at times when individuals are more capable of providing for themselves, and financially starving these same sectors at times when individuals may be cutting down their spending on health and education, runs the risk of being ineffective and of leaving the neediest unprotected. In addition, volatile spending compromises the continuity of social programs, and dampensthe effectiveness of long-term social investments. 5.25 Inorder to increase their effectiveness, social expenditure levels needto recover at least back to historical levels, and volatility in the social sectors budget needs to be reduced to guarantee the continuity and effectiveness of social programs. Several tools could be usedfor this purpose: 0 Improvement of budget management process. Often resources allocated to key social programs and functions are not only insufficient, but disbursed irregularly within the budget cycle due to cash-flow problems at the level of the Treasury, severely compromises program effectiveness. Alternative fiscal instruments, such as fiduciary contracts like the one currently operating for the Bono de Desarrollo Humano, need to be put in place in order to mitigate this problem and guaranteethat funds for (a selected set of) social programs are available when required. 0 Improvement inthe function of the oil-stabilization fund, as discussed above. 121 Incidence of Social Spending: Does Social Spending Benefit the Poor? 5.26 In this section we use incidence analysis to determine the share of social expenditure that accrued to different households (household incidence) and different provinces (geographic incidence). Because households and provinces vary in terms of their income and poverty levels, we relate shares in social expenditure to these variables and examine to what extent spending i s progressive (Le. improves the consumption levels of the poor relatively more than those of the rich, and hence reduces consumption inequality) or regressive (Le. improves the consumption levels of the richrelativelymore than those of the poor, and henceincreasesconsumption inequality). 5.27 As simple as this sounds, incidence analysis and, in particular, the interpretation we give to the results it generates are sometimes not straightforward. For instance, the level of progressiveness of a particular program or expenditure item i s a function of a multiplicity of factors. First, not all programs are designed exclusively for poor households or areas. The number and distribution of the poor, the extent of a program's coverage (Le. targeted or universal) and the difference in take-up rates across poor and non-poor individuals can then affect the program's level of progressiveness. Second, even if a program i s targeted to the poor, its level of progressiveness will depend on how effective the targeting mechanism and its implementation are. As a result, the higher the leakages, the larger the loss in progressiveness.Finally, the cost of providing a particular service may vary across groups or areas. If provision to non-poor households or areas i s more expensive than provision to poor ones, more money may be spent on groups or provinces that are not necessarily worse off, making spending lessprogressive. 5.28 In addition, average incidence analysis (the type that is usually performed) has some important shortcomings. First, i t i s possible for the characteristics of the population covered by a particular program to change over time as the program expands or contracts or as the total population itself changes, so that the characteristics of the marginal beneficiary of the program are different from those of the average beneficiary. Incidence analysis performed at a particular point intime does not allow us to distinguish between the marginal and the average beneficiary. Second, incidence informs us about how resources are spent rather than about the real impact that those resources have. Finally, analysis incidence does not provide us with a counterfactual that we can compare the actual situation against. 5.29 While we incorporate the ideas put forward above regarding progressiveness into our discussion on incidence, we do not have access to the data required to perform marginal incidence analysis and so our results are subject to the criticism above. 5.30 Household-level incidence of social expenditure. Total social expenditure i s progressive, although it cannot be considered pro-poor since a significant amount of resources accrues to households at the top of the income distribution. Using the 1999 ECV, Vos et alia (2003) show that social expenditure improved the distribution of consumption, decreasing the Gini coefficient by 3 percentage points. In absolute terms, however, the poorest quintile received 12 percent of social expenditure, compared to 27 percent for the richest quintile. 122 Quintiles Consumption Program 1 Program2 Program3 Equity 1 5 10 50 1 20 2 9 20 20 6 20 3 13 20 10 13 20 4 30 20 10 15 20 5 43 30 10 65 20 I FigureB.5.5.1: Distributionof consumptionandprogrambenefitsby quintile (% of total cumulative) - I 0 1 2 ConsumptionQuintiles Poorestto Richest - 3 4 5 ] Consumption Program 1 Program2 -Program 3 -Equity] 123 Table 5.2: Most social programsare progressive,and some are also pro-poor tertiary education and - the gassubsidyare the exceptions Deciles Consumption per capita Primary Secondary Tertiary Poorest 10% 2 20 5 0 2 3 15 10 3 3 4 14 12 3 4 5 12 12 10 5 6 11 13 4 6 7 9 12 12 7 9 8 11 13 8 11 5 11 15 9 16 4 10 20 Richest 10% 37 2 4 20 Seguro Social Servicios del Deciles Consumption per capita Campesino MSP IESS Poorest 10% 2 17 10 1 2 3 9 9 3 3 4 15 8 2 4 5 19 15 5 5 6 5 12 11 6 7 7 11 10 7 9 11 11 10 8 11 10 13 12 9 16 4 8 20 Richest 10% 37 1 3 26 Deciles Consumption per capita Bono Solidario Gas subsidy Poorest 10% 2 12 3 2 3 14 5 3 4 16 7 4 5 13 8 5 6 13 9 6 7 12 10 7 9 8 12 8 11 8 13 9 16 3 16 Richest 10% 37 1 17 ource: Vos et alia (2003). Figuresrepresentshare of decile intotal consumptionor programexpenditure. 1 , 124 5.31 When each program i s analyzed separately, we find significant variation in terms of how progressive or pro-poor different programs are. The main results obtained by Vos et alia (2003) can be summarized as follows (Table 5.2): 0 Education spending: Primary education i s bothprogressive and pro-poor, while secondary education i s progressive, and tertiary education is regressive.The pro-poor nature of primary education i s the result of almost universal coveragecombined with the fact that poor households tend to have more children of primary-school age, while the regressive nature of tertiary education can be explained interms of higher enrolment rates among better-off students combined with highunit costs. 0 Health spending: Health services providedby the Seguro Social Campesino (insurance scheme for the rural poor) are both progressive and pro-poor, while service providedby the Ministry of Health are progressive, and services providedby the IESS (insurance for the formally employed) are regressive.Interms of facilities, services providedby local ambulatories are pro-poor and progressive, while servicesprovidedby health centers and hospital are progressive (not shown). 0 Cashprograms: The Bono Solidario (a program createdto compensate the poor for the elimination of energy subsidies- see Box 5.6 for a more detailed description of the program), which account for 65 percent of social assistanceexpenditure, is pro-poor and progressive, while the gas subsidy i s slightly progressive. This is, however, misleading since a large fraction of the poor do not use gas for cooking purposes and thus do not benefit from the program. Moreover the subsidy i s distributed on a container-basis rather than a household-basis, hencebenefitingthose who consume relatively more gas (i.e. richer households). Box 5.6: The BonoSolidario A briefhistory - The Bono Solidario, created in 1998, was conceived as an unconditional cash-transfer to mothers in poor families, the elderly and the disabled in compensation for the termination of the gas subsidy. Even though the gas subsidy was eventually never phased out, elimination of the Bono was politically unfeasible once it was on its way, and so the program has survived up to the present. Selection of beneficiaries was done through voluntary registration. Verification of the eligibility criteria by program administrators at that time was poor, and this led to important leakages of funds to non-poor households (approximately 30 percent o f all program beneficiaries in 1999 belong to the richest 40 percent o f the population). The program grew fast and reached 1.6 millionbeneficiaries in January 1999.Concems about poor targeting led to a series of depurations of the beneficiary database, which brought the total number down to 1.2 million by 2002. The Bono Solidario i s financed with general revenues and is administered centrally by an office inthe Ministry of Social Welfare. Its budget in 2002 was US$lSO million, or about 0.75 percent of GDP, which makes it the second largest social assistanceprogram. As part the social sector reform program supported by the Programmatic Human Development Reform Loan (see World Bank 2003 for details), the Bono Solidario is currently experiencing a radical transformation. The two main elements of this transformation are: (i) re-targeting using the SelBen (the proxy-means test developed in Ecuador to target social programs) and, (ii) making transfers conditional on schooling and health behavior at the level of the household. Section 5.3 in this chapter evaluates the effects of the first measure, while an impact evaluation of the education and health conditionality i s currently underway with support from the World Bank. 5.32 Geographic incidence of social expenditure. We next examine the geographic distribution of social expenditure, and correlate this distribution with poverty, as well as with health and 125 education outcomes. This exercise i s complementary to that of household-level incidence for two important reasons. First, the poor are not uniformly distributed across the country so that it i s possible for a program to be well-targeted at the household level (Le. no leakages) and not at a regional level (Le. deficit in coverage of poor householdsinpoor areas). We will see below that this i s the case of the Bono Solidario. 5.33 Second, social outcomes vary significantly across provinces and appear to be correlated with poverty (Section 5.1). Inaddition social sector budgets are allocated at the provincial level. Incidence analysis can then be informative as to whether resources are being channeled to and have an impact inareas that needthemmost 5.34 We measure expenditures at the province level using information from the 2001 Fiscal Budget, while poverty and social outcomes data correspond to those discussedin the first section. We present the results in the form of concentration curves, where the horizontal axis measures the cumulative share of the population ineach province, ordered from poorest to richest, and the vertical axis shows the cumulative share of per capita spending in education, health and social assistance. The results can be summarized as follows: Education spending: Both primary and secondary spending are concentrated on poor provinces, so that education spending per capita i s higher in provinces with low enrolment rates (Figure 5.7). Healthspending: The distribution of health spending i s almost identical to the equity line, so that, contrary to what we observed ineducation, health spending per capita i s similar across provinces and does not follow apro-poor province pattern (Figure 5.8). Bono Solidario: As with health, the distribution of spending on this program i s close to the equity line, so that per-capita transfers from the program are similar across different areas (Figure 5.9). Figure 5.7: Education spendingper capita is higher inpoor provinces,while ... 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Cum provinces, most poor lo leas1 poor, weighted by population Source :Authors' calculations basedondatafrom the 2001 Fiscal Budget, the 1999ECV andthe 2001 PopulationCensus. 126 Figure 5.8 ...health spending per capita is similar across provinces. 1 as I 0.8 a7 0 6 0.5 a4 0.3 a2 a i 0 a i 0.2 0.3 0.4 0.5 0.6 0.7 a8 0.9 1 Cum provinces, most poor to leas1 poor, Welghledby populallon Source :Authors' calculationsbasedon data from the 2001FiscalBudget,the 1999ECV andthe 2001Population Census. Figure 5.9 The Bono Solidario tends to reach the poor innon-poor areas rather than the poor inpoor areas 1 0.9 0.8 0.7 0.6 /t8ono 0 0.5 a 45 degree 0.4 0.3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Cum parroquias,poorestto eastpoor, weightedbypopulation Source: Authors' calculationsbasedon datafrom the 2001FiscalBudget,the 1999ECV, and the 2001Population Census. 127 5.35 Outcomesandspending: Puttingthe resultstogether. What havewe then learnedfrom the previous two exercises? How are outcomes and expenditures related? We briefly review the results and draw some conclusions, by sector, before movingon to the next section. Primary and secondary education spending appears to be progressive at the household level and pro-poor at the province level. However, enrolment rates in poor households and provinces are significantly lower than the national average. We propose three potential (and not mutually exclusive) explanations for this apparent contradiction. First, poor households are larger than non-poor households and have more children of schooling age. Second, the unit costs of primary and secondary education provision are higher in poorer areas, according to Vos et alia (2003)95. These differences reflect (i) variation in the level of urbanization (i.e. it i s more costly to provide education inrural areas), and (ii) variation in teacher salaries (Le. more experienced teachers and rural teachers are paid higher salaries) across provinces. Third, spending efficiency and effectiveness could be lower in poorer areas. For instance, the quality of teachers in (remote) rural areas i s generally perceived to be lower than that of teachers inurban areas. However, since pay i s not related to performance, differences inquality do not translate into differences inwages. Whether we live in a high unit cost world or a low efficiency world has important implications for policy making. Where unit costs are high, improvements in supply and alternative schooling systems could be tested out (e.g. distance learning, alternative schedules, etc); where quality i s low, increasesininputs, and training and performance-basedpay for teachers could be useful. The distribution of health spending at the household level depends on the source of funding and the service provider, while there seems to be little variation in spending per capita across provinces. Inother words, there i s significant user sorting across the different health sub-systems, while funding for each system i s not a function of geographic factors. This implies that efforts to improve health outcomes among the poor should take into account the organization of the sector and be sub-system based. They should also promote sub-system integration to maximize coverage and eliminate inequalities inservice quality. Spending on the Bono i s pro-poor at the household level (Table 5.2), but neutral at the province level (similar per-capita transfers across regions), implying that the program has been relatively more successful in reaching poor households in non-poor areas (i.e. urban areas) than in poor ones (i.e. rural areas). This will change, however, as a result of the current initiative to re-target the Bono Solidario, as we discuss below. RE-TARGETINGSOCIAL PROGRAMS: ANEVALUATIONOFTOOLS AND PROJECTS. 5.36 Ecuador spends a significant amount of its social sectors budget on social assistance programs, but does not benefit as much as it should out of this investment. About 1.5 percent of GDP i s devoted to programs whose main objectives are, in principle, to alleviate poverty and to assist individuals and families to cope with the main adverse effects of negative income shocks. The social safety net, however, suffers from numerous problems that often render it an ineffective tool to fulfill its goals. 5.37 Poor targeting and, as a consequence, high leakage of resources to non-poor households are among the most significant of these problems, and the Government of Ecuador seems determined to correct them. In particular, it i s currently re-targeting or planning to re-target a number of social programs, ranging from feeding and nutrition programs, to the Bono Solidario to the polemic gas 95 Vos et alia (2003) estimate that the province-level unit cost of provision varies between US$78 and US$153.5 for primary education, andbetweenUS$170 and US$723 in secondaryeducation. 128 subsidy. For this purpose two different targeting instruments are available: the poverty map prepared for this report and the SelBen, a welfare index (see Box 5.7). 5.38 In this section we evaluate the re-targeting proposals for the Bono Solidario (now rebom as the Bono de Desarrollo Humano) and the gas subsidy, analyzing the impact that these will have on expenditure redistribution and, in the case of the gas subsidy, on fiscal revenue. These two programs are the biggest cash programs, jointly accounting for 3-3.5 percent of GDP, and they both will rely on the SelBen as their targeting instrument. Box 5.7: The SelBen An evaluation - The SelBen96, a welfare index that ranks households according to their demographic and structural characteristics, has been selected by the Government of Ecuador as the main targeting instrument for social programs -i.e. social programs will give priority to households in the first and second SelBen quintiles. Given the key role that it envisions for the SelBen, the Government of Ecuador recently commissioned an independent evaluation of the index. This evaluation covered various topics, including (i) methodology, (ii) coverage of potential target population (Le. the poor and the vulnerable), and (iii) correlation between the SelBen and measuresof (monetary) poverty. The evaluation was based on the construction of a counterfactual population based on a survey collected among a representative sample of households residing in areas covered by the SelBen. In addition, all those households in the survey that had been interviewed for SelBen were matched to their original SelBen results in order to check for accuracy in the original (self-reported) questionnaires. The main results from the evaluation can be summarized as follows: Approximately 83 percent of the target population (poor households) has already been interviewed and assigneda SelBen score. There are, however, some regional differences in terms of coverage. While coverage is almost 85 in the urban Sierra, it i s only 45 percent inthe Oriente. Withinareas already covered, SelBen was successful inidentifyingthe poor (on target), and almost exclusively the poor (cost efficient). The degree of agreement between the information on household characteristics, collected through self- reporting in SelBen and through direct observation inthe evaluation survey, ranges from very high(97 percemt of all cases) to high(55 percent of all cases), depending on the characteristic. When comparing the SelBenwith consumption-based welfare measures, 13 percent o f those who would be considered poor accordingto SelBen (Le. first and secondquintiles), would not be so according to the selectedpoverty line (inclusion error). Similarly, 23 percent of those considered poor according to the poverty line, where not so according to SelBen (exclusion error). These differences reflect the fact that structural and monetary poverty are not identical concepts. Although these percentages are not extremely high, it is important to consider the implications associatedwith bothtypes of error when usingSelBen to target different programs (see the discussion on the gas subsidy in Section 5.3 for a relevant example). Finally, since the evaluation took place, almost a year after the first round o f the SelBen was completed and almost 3 years after this round started, changes inhouseholdcharacteristics over time were examined. Most households did not experience significant changes in their assets, and those that did were not in the first and second SelBen quintiles. This implies that high-frequency, costly updates o f the SelBen are probably not required, and confirms that the current 5-year update horizon seems to be appropriate. Source: Cely (2002). 96SelBen stands for Sistema de Identificacidn y Seleccidn de Beneficiaries de Programas Sociales, or system for identificationof social programbeneficiaries. 129 Usingthe SelBento re-targetthe Gas Subsidy A Simulation- 5.39 The subsidy to cooking gas, which amounts to almost 3 percent of GDP, has a long and contentious history. Being an expensive program, different administrations over the past years have tried to eliminate it (last in 1998, which led to the creation of the Bono Solidario), and failed due to significant social unrest and protest. 5.40 The current value of the subsidy i s $3.4 per liter of liquid gas -the difference between the price paid by consumers, $1.6, and the current market price of $5. The subsidy is paidevery time gas i s purchased, independent of household characteristics (i.e. the subsidy has universal coverage). Since gas use i s low among the poor and, particularly, among the indigenous poor, the subsidy is extremely regressive and does not favor ethnic minorities (see Table 5.2 above, andTable 5.3). Table 5.3: Gas useis low amongthe indigenouspopulation. I Ethnicity (self reported) I Number of People Percentageof total Percentageusinggas I Indigenous 824,189. 6.8 42.9 Black 267,196 2.2 89.5 White 1,257,466 10.45 92.9 5.41 It must be noticed, however, that, even if the fraction of the subsidy accrued to the poor is small, it i s still significant compared to their consumption levels (4 percent of poor households net- of-subsidies private consumption) so that complete elimination of the subsidy could lead to important welfare declines for the poor if not accompanied with other compensatory measures. 5.42 The Government of Ecuador i s currently considering to re-target the gas subsidy using the SelBen. Using the 1999 ECV, we evaluate this re-targeting scheme against the status quo and the total elimination of the subsidy, assessingits impact on incidence and on fiscal revenue. We calculate the SelBen index for households in the survey, following the SelBen methodology, and use information on consumption of gas and all other goods to determine the impact of changes in the subsidy. We impose the following two assumptionsinorder to simplify the simulations: i.Therearenobehavioralresponsesintheconsumptionofgastochangesinthesubsidy.Instead households adjust their consumption of all other goods to accommodate the decline in overall expenditure associated with the elimination of the subsidy. Inother words, a household that inthe past received US$10 in the form of the subsidy and spent US$lOOO in all goods but gas, will spendUS$990 after the subsidy i s eliminated97. ii.Thepovertylineremainsatthesamelevelforeveryscenario. 5.43 Re-targeting using the SelBen i s a pro-poor proposal, which improves significantly on the status quo (Figure 5.6). Under this scheme, most of the spending would be directed to households at the bottom of the income distribution - that is, poor households. In particular, the poorest quintiles ''Cuestaet alia (2002) allow for behavioral responses in simulating the re-targeting of the gas subsidy using the Bono Solidario beneficiary database as the targeting tool, and find negative labor supply effects associatedwith the subsidy. 130 would receive 44 percent of the program's resources, compared to 15 percent under the status quo. In addition, re-targeting could generate significant savings of up to 76 percent of the total current subsidy (or approximately US$275 million) - the equivalent to 60 percent of the 2003 health sector budget or more than 4 times the investment budget inthe education sector. 5.44 Two caveats are worth mentioning, however. First, since the correlation between the SelBen index and monetary poverty i s not perfect (see Box 5.6) there will be some poor households that will not receive the subsidy and some non-poor households that will receive it (inparticular, households inthe top halfof the consumption distributionwill receive about 10percent of the total amount of the subsidy)'*. Second, since a large fraction of poor and indigenous households does not use gas, this measure would leave their welfare unaffected. Figure5.10: Re-targeting the gas subsidy usingthe Semenis pro-poor and progressive. -SubsidyAccordingloSelben 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 8 0.9 1 C"" POpVlatii Source: Authors' calculationsbasedon datafrom the 1999ECV. 5.45 Incontrast, complete elimination of the subsidy would have a negative impact on those poor households that currently consume gas, if not accompanied by compensatory measures. It would also produce a somewhat counterintuitive improvement in the distribution of consumption, given its regressive nature, equalizing household consumption downwards. Finally, it would generate the largest possible fiscal savings (100percent of the total subsidy). 5.46 Leavingpolitical economy considerations aside, however, it i s important to remember that, to the extent that it i s a price subsidy, the gas subsidy creates distortions. Hence, an alternative scenario where the gas subsidy was eliminated and the savings this produced were targeted to the poor in a non-distorting way would be preferable to both re-targeting and elimination alone. A policy option for the government to consider. Notice that poverty here refers to monetary poverty, as measuredinChapter 2. 131 5.47 Finally, the results presented here should be understood as a first approximation, while a more comprehensive evaluation of the re-targeting proposal should allow for changes in the consumption of gas in response to changes in the structure of the subsidy. In addition, these results assume that the implementation of the targeting scheme is perfect (Le. all households in the first and second SelBen quintile that consume gas receive the subsidy and no households outside those quintiles do), something that should not be taken for granted and that will very much depend of the particular implementation strategy the government chooses. Usingthe SelBento re-target the Bono de Desarrollo Humano (BDH) 5.48 We argued above that, even though the Bono Solidario was overall pro-poor at the household level, this was not the case at the geographic level. That is, the Bono Solidario was more likely to reach poor households in non-poor areas than poor householdsinpoor areas. Moreover, the program suffered from important leakages (see Box 5.5). 5.49 Almost by definition, re-targeting the program using a welfare index such as the SelBen will reduce overall leakages to non-poor households. Usingthe 2001Population Census, we evaluate if it will also improve the program's capacity to reach the poor, at a geographical level, compared to the status quo, and find that this i s indeed the case (Figure 5.?). Figure 5.11: The BDHwill be geographically more progressive after re-targeting with the SelBen. 1 0.9 0.8 0.7 0.6 0.5 0.4 0 3 0.2 0.i 0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 I -45 degree -Selben Bono1 Source: Authors' calculationsbasedon datafromthe SelBen andthe 2001PopulationCensus. 132 I Box 5.8: Ecuador andthe Millennium Development Goals We assess here the likelihood that Ecuador will achieve the poverty, education and health-related Millennium Development Goals (MDGs).These goals are: 0 Eradicate extreme poverty and hunger: Halve the proportion of people living on less than one dollar a day. 0 Achieve universal primary education: Ensure that boys and girls alike complete primary schooling. 0 Reduce childmortality: Reduce by two-thirds the under-five mortality rate. 0 Improve maternal health: Reduceby three-quarters the maternal mortality ratio. Inparticular, we use the SimSip program, developed by the World Bank, to predict the value of these outcomes in 2015 under two different growth scenarios: (i) the country continues to growth at an annual rate of 2.5 percent, the average of the last 10 years, or (ii) the country grows at an annual growth of 4 percent. For each goal, we present the simulation results next a crude estimate of the intermediate landmarks that would guarantee success in achieving it. These landmarks are generated through linear interpolationusingthe actual value of the outcome andthe value that correspond to the MDG. We find that, although a higher growth rate will get Ecuador closer to achieving the MDGs, it i s unlikely that the country will be able to do, even with a growth rate of GDP that almost doubles its historical average. This implies that relying on growth alone will not suffice and specific interventions in the social sectors will be needed inorder to achieve these goals. Table B.5.8.1 Ecuador is unlikely to meet the poverty, education and health MGDs Extreme MDG levels Net Primary I Year Poverty enrolment MDG levelUnder-5 mortality MDGlevel GDPannual growth of 2.5 percent 2005 26.6 23.25 90.8 93.3 31.6 27.8 2010 25.1 18.6 91.3 96.6 28.2 19.8 2015 23.3 13.95 91.9 100.0 25.1 11.8 I GDP annualgrowth of 4.0 percent 2005 23.3 22.4 91.0 93.3 30.2 27.1 2010 20.0 17.9 91.7 96.6 26.6 19.3 2015 16.9 13.45 92.4 100.0 22.4 11.6 CONCLUSIONS 5.50 Social outcomes in Ecuador are generally worse than we would predict given the country's level of development. We have argued in this chapter that poor performance i s most likely the product of low, and highly volatile, social expenditure levels, as well as of, in some cases, poorly targeted expenditures. 133 5.5 1 Social expenditure levels needto recover at least back to historical levels, and volatility inthe social sectors budget needs to be reduced to guarantee the continuity and effectiveness of social programs. Several tools could be used for this purpose, including the creation of a stabilization fund that receives resources when fiscal revenues are high, andprovides resources when they are low. For this kindof tool to be effective, it needs to receive enough resources duringgood times (i.e. the rule that triggers diversion of resources to the fund should be realistic) and to be protected from potential alternative uses. 5.52 Inaddition, social moneysshould be spenteffectively. Important differences ineducationand health outcomes across regions suggest that there i s potential for improvement. This, however, should not be understood to depend exclusively on the level of resources in each sector. Instead, policies aimed at increasing spending efficiency should be considered seriously. In the case of education, this involves significant improvements in provision and quality, especially in rural areas. In the case of health, it requires expansion of coverage and a higher level of integration among different sub-systems andproviders. 5.53 Regarding social assistance programs, the objectives and target population of different programs should be stated clearly, and resources allocated accordingly. Inthe past, targeting of most social programs has been poor and resource leakages high. Current initiatives to re-organize and re- target several social programs are steps inthe right direction. 134 BIBLIOGRAPHY Aguirre, Carolyn Espinosa. 2001. "Effects of InternationalMigrationon the Labor Supply of the sending households: The case of Ecuador." Araujo, CaridadandPeterLanjouw. 2003. "Constructing panelsof poverty maps: an exercisefor Ecuador." Processed. Asociaci6n Latinoamericanade EducacidnRadiof6nica-ALER, Caritas, Espaiia, Radio Comunitario Espaiia, ConferenciaEpiscopal Ecuatoriana-CEPAS,Fondo EcuatorianoPopulorum Progressio- FEPP, Instituto Latinoamericanode InvestigacionesSociales. 2003a. NO3 Cartillas Sobre Migracibn. "Causas del recienteprocesoemigratorio ecuatoriano." .2003b. NO4 Cartillas Sobre Migraci6n. "Verdades y mediasverdades de la migraci6n." .2002a. Nol.Cartillas SobreMigracibn. "Las remesasde 10s emigrantesy sus efectos en la economiaEcuatoriana." .2002b. NO2Cartillas SobreMigraci6n. "El trabajo domestic0en lamigraci6n." BancoCentralDelEcuador.2002a. "Informaci6n EstadisticaMensual.'' Direcci6n Generalde Estudios. Quito, Ecuador. .2002b. Catdogo. PublicacionesEconbmicas,Quito, Ecuador. .2002c. "Una Propuestade Plan EstratCgicode Desarrollo de Largo Plazoparael Ecuador." CD- ROM. .2002d. 2001. Cuadernosde Trabajo. "Las remesasde ecuatorianosen el exterior." .2002e. "Memoria Anual." CD-ROM. Barro, R.J. andX. Sala-I-Martin. 1995. Economic Growth.Mc Graw-Hill. Bebbington,Anthony. 1999. "Capitals andcapabilities: aframework for analyzing poorest viability: rural livelihoods andpoverty." WorldDevelopment vol. 27 (12):2021-2044. Bebbington, Anthony andThomas Perreault. 1999. "Social Capital, Development, and accessto Resourcesin Highland Ecuador." Economic Geography v75, n4 (October): 395-418. Beckeman, Paul. 2003a. "Dollarization and Semi-dollarization inEcuador." The World Bank, Washington, D.C. ,2003b. 2001. Policy ResearchWorking Paper#2643, World Bank, Washington, D.C. Beckerman, PaulandAndres Solimano (editors). 2002. Crisis and Dollarization in Ecuador. The World Bank, WashingtonD.C. BergoeingR., P. Kehoe, andR. Soto. 2002. "Policy-Driven Productivity inChile and Mexico inthe 1980 and 1990." American Economic Review, 92 (2): 16-21. 135 Binswanger, Hans P., Klaus Deininger and Gershon Feder. 1995. "Power, Distortions, Revolt and Reform inAgricultural LandRelations." InJ. Behrman andT.N. Srinivasan, eds. Handbookof DevelopmentEconomics VolumeIII.ElsevierPublishing. Boye, Francois. 2001. "Oil and Macroeconomic Fluctuations inEcuador." OPECReview v25, n2:145-71. Brysk,Alison. 2001. "Indian Market: the Ethic Faceof Adjustment inEcuador." InMiltonJ. Esman and RonaldJ. Herring, eds. Carrots, sticks and ethnic conflict: Rethinking development assistance. Ann Arbor: University of Michigan Press. Calvo, G, and C.A. Vegh. 1992. "A Currency substitution indevelopingcountries: an introduction." Revista de Andisis Econdmico,7 (1): 1-28. Carroll, Thomas F. 2003. "Construyendo Capacidades Colectivas." FortalecimientoOrganizativo de las federaciones campesinas-indigenasen la Sierra Ecuatoriana. RISPERGRAF, Quito, Ecuador. CEDIME. 2003. "Etnicidad, Pobreza y Exclusi6n ( Ecuador) Estudio Comparativo rural-urbano, indigena no indigena." Banco Mundial. Cely, Natalie S. 2002. "AnAlisis de la ejecuci6n de 10s programas sociales prioritarios y del instrumento de focalizacidn selben 2001-2002." Banco Interamericano de Desarrollo, Washington, DC. Collier, Paul. 1989. "Contractual Constraints on Labour Exchange in Rural Kenya." International Labour Review 128 (6), pp. 745-68. ComunidadAndina. 2003. "Estrategia parael Desarrollo Estadistico de la Comunidad Andina." Propuestade financiamiento del FondoFiduciario de Banco Mundial para el Desarrollode la Capacidad Estadistica. COMUNIDEC. 1996. Poblaci6nIndigenay Pobrezaen Ecuador. Cuesta, Jose', Juan Ponce y Mauricio Le6n. 2002. "Efectos de gasto social fiscal en la generaci6n de ingresos en el Ecuador." Institute of Social Studies and Sistema Integradode Indicadores Sociales del Ecuador, Quito Ecuador. DeFerranti, David, Guillermo Perry, Francisco FerreiraandMichael Walton. 2003. Inequality in Latin America and the Caribbean: Breaking with History?. The World Bank, Washington D.C. Deininger, Klaus, and Pedro Olinto. 2000. "Asset distribution, inequality, and growth." Policy Research Working Paper 2375. World Bank,Washington D.C. Diaz Maya, Milton. 2002. "Las pequeiias cooperativas rurales: surge un nuevo actor econbmico, CIRIEC-Espaiia." Revistade Econdmica Publica, Social y Cooperativa ~0,1143, special November issue:85-105. Dowrick, S. AndJ. Quigguin. 1994. "International Comparisons of Living Standards and Tastes: A Revealed Preference Analysis." American EconomicReview,84 (1): 332-41 Easterly, W., M.Kremer, L.Pritchett, andL.Summers. 1993. "Good Policy or Good Luck?Country Growth Performance and Temporary Shocks." Journal of Monetary Economics,Vol. 32. ECLAC. 2002. "Anuario Estadisticode Ame'rica latina y el Caribe 2002.'' Santiago, Chile. 136 Elbers, Chris, Jean Olson Lanjouw and Peter Lanjouw. 2003. "Micro-Level Estimationof Poverty and Inequality." Econometric71: 1, pages355-364. Elbers, Chris, and Peter Lanjouw. 2001. "Intersectoral Transfer, Growth, and Inequality inRural Ecuador." World Development v29, n3: 481-96. Encalada, E. et .al. 1998. "Pobreza Indigena y Negra en Ecuador." Inter-American Development Bank, Washington, DC. Fiess, Norbert and Dorte Verner. 2003. "Oil, Agriculture, and the Public Sector Linking Intersector Dynamics inEcuador." WorkingPaper #3094, The World Bank, Washington, D.C. FOMIN. 2003. "Receptores de Remesasen Ecuador. UnaInvestigaci6ndel Mercado." Fondo Multilateral de Inversiones Banco Interamericano de Desarrollo. Quito, Ecuador. Food and Agricultural Organization. 2002. "Arrendamiento de Tierras en AmCrica Latina: Una Altemativa de Acceso a la Tierra para 10s Pobres Rurales?" Unpublishedmanuscript. Foster, George, Tzintzuntzan. 1967.Mexican peasants in a changing world. Little Brown, Boston. Fretes-Cibils, Vicente, Marcel0 Giugale and JosC RobertoL6pez-Calix (editors). 2003. Ecuador: An Economicand SocialAgenda in the New Millennium. The World Bank, Washington, D.C. Gallego, F. and R. Soto. 2001. "Evoluci6n del consumo y compras de bienes durables en Chile, 1981- 1999". Estudios de Economia28 (2) :309-338. Glassman, Amanda and CCsar Patricio Bouillon. 2002. "Conceptual issues inthe design of a safety net in a dollarized economy." Prepared for IBDSeminar on Dollarization inEcuador: Policies to Ensure Success, Washington, DC. Gollin, D. 2001 "Getting Income Shares Right." Journal of Political Economy, vl10, n2:458-74 Halac, Marina, and Sergio L.Schmukler. 2003. "Distributional Effects of Crises: The Role of Financial Transfers." Processed. Heckman, J. and Carmen Pages. 2000. "The Cost of Job Security Regulation: Evidencefrom Latin American Labor Markets." Economia, Vol. 1, No. 1, Fall. Latin American andCaribbean Economic Association. Hentschel, Jesko and William F. Waters. 2002. "Rural Poverty inEcuador: Assessing Local Realitiesfor the Development of Anti-poverty programs." World Development v30,nl (33-47). Hentschel, Jesko, Water, William, Vandever Webb, Ann K. 1996. "Rural Poverty inEcuador -A Qualitative Assessment." Working Paper Series #1576, The World Bank. Hidalgo, A. Roberto. 2002. "Las pequeiias y medianas industriasen Ecuador." InPeres Nbiiez, Wilson, (coor). NUCEPAL. Las pequeiiasy medianasempresas industriales enAmkrica Latina y el Caribe.MCxico, D.F. Hodrick, R.And E.Rescott. 1997. "Postwar U.S. Business Cycles: An EmpiricalInvestigation". Journal of Money, Credit and Banking, 29. 137 IMF. 2003a. "Ecuador: FirstReview Underthe Stand-By Arrangement andRequestsfor Modifications and waiver of Nonobservance and Applicability of Performance Criteria" -Staff Report; Staff Statement; PressReleaseon the ExecutiveBoardDiscussion; and Statement by the Executive Director of Ecuador. InternationalMonetary Fund, Country ReportNo. 03/248. .2003b. "Ecuador: SelectedIssuesand Statistical Appendix." InternationalMonetary Fund, Country Report No. 03/91. .2000. "Ecuador: SelectedIssues and Statistical Annex." InternationalMonetary Fund, Country ReportNo. 00/125. INEC. 2003a. Politica Estadistica Institucionalde INEC. Propuesta. INEC, Quito, Ecuador. .2003b. Lineamientos de laPoliticaNacionalde Estadistica de Ecuador. Propuesta. INEC, Quito, Ecuador. .2003c. Sistema Integrado de Encuestasde Hogares (SEH). Propuesta. INEC, Quito, Ecuador. .2001a. "Encuesta de Empleo, Desempleo y Subempleo en el k e a Urbanay Rural del Ecuador." Instituto Nacional de Estadistica y Censos. .2001b. "VI Censo de Poblacidny V de Vivienda 2001." Presentacidn del Economista Carlos Cortkz Castro, Director General del INEC, Quito, Ecuador. .2001c. ENEMDUR2001. Encuesta de Empleo, Desempleo y Subempleo, Area Urbana .2000. ENEMDUR2000. Encuesta de Empleo, Desempleo y Subempleo, k e a Urbana .1999a. ENEMDUR1199.EncuestadeEmpleo, Desempleoy Subempleo, AreaUrbana .1999b.EncuestaCondiciones deVida-CuartaRonda. Quito, Ecuador. . 1998.Encuestade Empleo, Desempleo Subempleo, k e aUrbana. Inter-American Development Bank(BID).2003. "Las Remesas de Emigrantes entre Espaiia y Latinoamkrica." Estudio elaborado por CECA, Caja Murcia, Caja de Ahorros ElMonte de Sevilla y SADAI. Artes Grgicas Palermo, S.L. InternationalLabor Organization. 2001. "Empleo y proteccih social en Ecuador." Janvry, Alain, and Elizabeth Sadoulet. 1993. "Rural Development inLatin America: Re-linking Poverty Reduction to Growth." InMichaelLipton and Jacques van der Gaga, eds. Including the Poor. The World Bank, Washington D.C. Katz, Lawrence F.and Kevin M.Murphy. 1992. "Changes inRelative Wages, 1963-1987: Supply and Demand Factors." Quarterly Joumal of Economics 107(1):35-78. Kristensen, Nicolai, Wendy Cunningham andClaudia Sepulveda. 2003. "MinimumWages inLAC: Do They Matter?Evidencefrom 20 Countries." The World Bank, Washington D.C. Processed. Kyle, D. 2000. Transnationalpeasants: migrations, networksand ethnicity in Andean Ecuador, Baltimore and London: Johns Hopkins University Press. 138 Lanjouw, Peter. 1999. "Rural Nonagricultural Employment andPoverty inEcuador." Economic Developmentand Cultural Change48, pp. 91-122. Lanjouw, Jean, Olson. 1997. "Poverty Comparisons with No compatible Data. Theory and Illustrations." Processed. Lanjouw, Jean Olson and Peter Lanjouw. 2001a. "How to Compare Apples and Oranges: Poverty Measures basedon Different Definition of Consumption." Review of Income and Wealth47:1, pages 25-42. .2001b."The Rural Non-FarmSector: Issuesand Evidencefrom DevelopingCountries." Agricultural Economics 26, pp. 1-23. . 1997. "Poverty comparison with non-compatible data." Policy ResearchWorkingPaper 1709, the World Bank, Washington, D.C. Larrea, Carlos. 2003. "Mapas Parroquiales de Consumo de Alimentos por Habitante." Processed. .2002. "Social and Economic Effects of Dollarization inEcuador". Presentedat the Conference "Cuba's Dual Economy: Lessons from other CountriesConfronting the Issuesof Convertibility andDollarization" Department of ForeignAffairs and InternationalTrade, Ottawa, Canada. Larremendy, P. 2003. "Indigenous Networks: Politics and development interconnectivity among the Shuar in Ecuador." PhD, Thesis University of Cambridge. Unpublished. Le&, Gustavo. 2003. "Etnicidad Pobreza y Exclusidn. Estudio comparativo Urban0 Ruralhdigena -No Indigena." Banco Mundial. L e h , M. 2003. "Etnicidad y Exclusi6n en el Ecuador: una mirada a partirdel Censo de Poblaci6nde 2001."Draft version, unpublished. .2002. "La medicidnde lapobreza en elEcuador: mdtodos y fuentes." Sistema Integrado de Indicadores Sociales del Ecuador (SIISE), PanamB. Processed. Le6n, M, and Norbert Schady. 2003. "Metodologia para la estimacion monetariade la produccion agricola." Processed. Leon, M.andM.P. Troya. 2000. "Mecanismos de transmisi6n de la crisis y Estrategas de ajuste de 10s hogares pobres en Ecuador." Documento de Trabajo #6, SIISE. Leon, M.And R. Vos. 2000. "Ecuador: el impacto de la crisis en la pobreza." Documento de Trabajo #8, SIISE. Loayza, Norman, Pablo Fajnzylber andCCsar Calderdn. 2002. "Economic Growth inLatin America and the Caribbean." The World Bank Regional Studies Program. Washington D.C. Processed. MacIsaac, D.and MartinRama. 1997. "Determinants of hourly earnings inEcuador: The role of labor market regulations." Joumal of Labor Economics, vol. 15, No. 3. Marconi, Salvador R. (editor). 2001.Macroeconomia y economia politica en dolarizacidn.Docutech- UPS, Quito, Ecuador. 139 Martinez, A. and G. Herrera. 2002. "Informe de InvestigacidnGCneroy Migracidn den la regidn Sur." FLACS0.Draft. Unpublished. Martinez, L.2000. "Caracterizacidn de la situacidn de la tenencia y regularizacidn de latierra." Quito IICA Febrero. Megill, DavidJ. 2002. "Diseiio de la muestra maestraparael sistema integrado de encuestas de hogares del INEC-Ecuador, y diseiio de la muestra para la encuestanacional de ingresos y gastos de hogaresurbanos." InternationalPrograms Center, U.S. Census Bureau. Processed. Melo, A. 2003. "La Competitividaddel Ecuador en la Erade la Dolarizacidn: Diagnostic0y Propuestas." Banco Interamericano de Desarrollo. Processed. Morales, M.2003. "Informe: Patrimonio Territorial y Tenencia de la Tierra." Fundacidn HLS/USAID/Fundacidn Lexis, Octubre. Narayan, E. and Petesh, P. (editors). 2002. Voicesof thepoorfrom many lands. OxfordUniversity Press and the World Bank. Newman, Constance, Pilar Larreamendy and Ana MariaMaldonado. 2002. "Mujeres y Floricultura: Cambios y consecuenciasen el hogar." CONAMU, Producciones digitales Abya-Yala, Quito, Ecuador. Newman, Constance. 2002. "Gender, Time Use, and Change: The Impact of the Cut Flower Industry in Ecuador." The World Bank Economic Review, vol. 16, No. 3: 375-396. North, LisaL.and John D.Cameron. 2000. "Grassroots-based Rural Development Strategies: Ecuador inComparative Perspective." WorldDevelopmentv28, n10(October): 1751-66. NU.CEPAL. 1999. "Efectos macroecondmicosdel fendmeno ElNiiio de 1998: suimpactoenlas econom'as andinas." OrganizacidnPanamericanade la Salud - Pan American Health Organization-PAHO. 2001. Ecuador" . Technical Reports Series No. 5, Washington D.C. "Desigualdades en el acceso, us0 y gasto con el agua potable en AmCrica Latina y el Caribe- Parandekar, Suhas D. 1999. "Protecting the Poor inEcuador: Priorities and Options for the Bono Solidario." World Bank, Washington DC. Processed. Ponce, Juan, JosC Cuestay Mauricio Ledn. 2002 "El subsidio a1gas: Algunas simulaciones para la toma de decisiones." Processed. Prescott, E. 1998. "Hended: A Theory of Total Factor Productivity." InternationalEconomic Review, 39 (August)525-552. PRODEPINE. 2002. "Informe Componente Tierras y Aguas, (10-1998 a 12- 2002)." Unpublished. Ramadas, Krishnan, Bernadette Ryan and Quentin Wodon. 2002. "SimSIP Goals: Assessing the Realismof Development Targets." World Bank. Processed. 140 Roy, Atrayee Ghosh and Hendrik Van den Berg. 2000. "Are PetroleumExports an Engine for Growth? Time-Series Evidencefor Five Oil Exporters." Journal of Energy and Development v26, nl (autumn): 55-69. Sala-I-Martin, X. 1996. "Regional cohesion: evidence and theories of regional growth", European Economic Review, 40: 1325-1352. SAnchez-PAramo, Carolina and Norbert Schady. 2003. "Off and Running?Technology, Trade, and the RisingDemandfor Skilled Workers inLatinAmerica." Policy ResearchWorkingPaperNo. 3015. The World Bank, Washington, D.C. Secretm'a General de la Comunidad Andina (compilador). 2001. "La Dolarizacidn en Ecuador. Efectos sobre el Comercio Andino." Comunidad Andina, Quito, Ecuador. SelBen. 2000. "Metodologia de construcci6n del indice de bienestar para potenciales beneficiariosde programas sociales." Sistema de identificacidny seleccidn de beneficiarios de programas sociales, Quito, Ecuador. SICA/Banco Mundial. 2003. "El Productor Agropecuario y su Entomo." 111Censo Nacional Agropecuario. Ecuador. www.sica.gov.ec. SIISE. 2002a. "La exclusidn social en el Ecuador: 10s indigenas y la educacidn." www.siise.gov.ec. Quito, Ecuador. .2002b. "Los indicadores sociales.Lamigraci6nintemacionalreciente: algunos interrogantes." www.siise.gov.ec. Quito, Ecuador. ,2002~."La pobrezaurbana en Ecuador." 1998-1998. Mitos y Realidades. .2002d. "El desarrollo social en la ddcadade 1990." Los logros y desafios del Ecuador frente a 10s compromisos de L a Cumbre Mundialsobre Desarrollo Social y la Cumbre Mundial a favor de la Infancia. www.siise.gov.ec. Quito, Ecuador. .2001. "Los niiios y niiias ahora." Una seleccidn de indicadores de su situacidn a inicios de la nueva ddcada. Resultados de la "Encuesta de medicidn de indicadores de la niiiez y 10s hogares." Quito, Ecuador. .2000. "Validacidn de lapropuesta metodoldgica parael establecimiento del `Sistema de identificacidn de beneficiarios de 10s programas sociales' (SISBEN)." Repliblica del Ecuador - Ministerios de FrenteSocial, Quito Ecuador. Smith, StephenC. 2002. "Village Bankingand Maternal and Child Health: Evidence from Ecuador and Honduras." WorldDevelopment v30, n4 (April):707-23. Solimano, Andrds. 2003. "Remesas a 10s Paises Andinos: Tendencias, Costos e Impact0 Econdmico." PowerPoint presentation. UNICEF. 2000. "La escolarizacidn bAsicaen el Ecuador." hdice, Unidad de Informacidny AnAlisis de la Secretm'a TCcnica del Frente Social (SIISE) Ministerios de Frente Social -INEC-UNICEF. 141 UnitedNations. 1986. "National Accounts Statistic: Main Aggregates andDetailedTablets,1984. New York" UniversidadTCcnica particular de Loja. CD-ROM Vos, Rob. 2002a. "Dollarization, real wages, fiscal policy and social protection: Ecuador's policy trade- offs." Paper prepared for IDBConference "Dollarization inEcuador: Policies to Ensure Success." Washington, DC. ,2000b. "Economic Liberation, Adjustment, Distribution and Poverty inEcuador, 1988-99." Working Paper, Institute of Social Studies, The Hague. Vos, Rob, Jos6 Cuesta y Arjun S. Bedi. 2003. "LQuiCn se beneficia del gasto social en el Ecuador? Desafios para mejorar la equidad y la eficienciadel gasto social." Institutode Estudios Sociales de L a Haya. Quito, Ecuador. Vos, Rob and M.Lebn. 2003. "Dolarizacibn, dinamicade exportaciones y equidad, LComo compatibilizarlasen el cas0 de Ecuador?' Informes del SIISE No. 5. Vos, Rob and Niek de Jong. 2003. "Trade liberalization and poverty inEcuador: A CGE Macro- Microsimulation Analysis." Economic SystemsResearch v15 n2 :211-32. Vos, Rob, MauricioLedn y Rend Ram'rez. 2002a. "Politica Social y Tendencias en el Gasto Social: Ecuador 1970-2002." Institute of Social Studies and Sistema Integrado de Indicadores Sociales del Ecuador, Quito Ecuador. .2001b. "Politicas Sociales y Tendenciasdel Gasto Social: Ecuador, 1970-2002." Vos, R.M.Velasco and E. De Labastida. 1999. "Economic and Social Effects of ElNiiio inEcuador, 1997-1998." Inter-American Development Bank, Sustainable Development Department, Technical Papers Series, POV-107 World Bank. 2004. Doing Businessin 2004. Understanding Regulation. A copublicationof the World Bank andOxford Universtiy Press, Danvers, MA. Forthcoming. .2003a. World Development Indicators. CD-ROM. .2003b. Land Policiesfor Growthand Poverty Reduction. OxfordUniversity Pressfor the World Bank. .2003c. Closing the Gap in Education and Technology.The World Bank, Washington D.C. .2003d. "Proposed Programmatic Human Development Reform Loan Ito the Republic of Ecuador." Report No. 25791-EC, the World Bank, Washington, D.C. .2002a. "Millennium DevelopmentGoals: Ecuador Country Profile." Devdata.worldbanbk.org. .2002b. "Crisis and Dollarization in Ecuador, Stability, Growth, and Social Equity." The World Bank, Washington D.C. .2000a. Ecuador Gender Review: Issues and Recommendations. AWorld Bank Country Study, Washington, DC. 142 .2000b. "World DevelopmentReport2001."AttackingPoverty, WashingtonD.C .2000c. "Ecuador, Crisis, PovertyandSocialServices." Report#01992@EC,WashingtonD.C . 1999."Ecuador: Consultationswith thePoor." WB-PREMGroup. Washington, DC. .1996.Ecuador Poverty Report. A World BankCountry Study, WashingtonD.C. Younger, Stephen D. 1999. "The RelativeProgressivityof SocialServices inEcuador." Public Finance Review v27, n3:310-52. 143 ANNEXES ANNEX1.ECUADOR INTHEANDEANCONTEXT We briefly compared selectedeconomic and social indicators for Ecuador and its Andean neighbors. Growth, InflationandEmployment Compared to its neighbors in the Andean region ,Ecuador performed poorly duringthe 1990s in terms of real GDP growth (Figure 1). While the region grew at an average annual rate of 2.7 percent between 1990 and 1999, Ecuador lagged behind with a more modest 1.6 percent (these figures become 3.0 percent and 2.6 percent respectively when 1999, a crisis year in Ecuador, is excluded from the calculation). Low real GDP growth, combined with a steady populationincreaseduringthe decade, translated into small gains in real GDP per capita, which increased by 5.7 percent up to 1998 compared to 10.6 percent for the Andean region inthe same period, andthen decreasedby 9 percent as a consequence of the 1999crisis. Fortunately Ecuador recovered from the effects of the crisis during 2001 and 2002, when the country's real GDP growth rate systematically surpassedthat of its neighbors. By the end of 2001 real GDP was back to pre-crisis levels and real GDP per capita was 5 percent higher than in 1999, with further improvements during2002. Average inflation in Ecuador was at the level of other countries in the Andean region until 1997, but increased significantly after that. In particular the inflation rate climbed up to 36, 52, and 96 percent in 1998, 1999 and 2000 respectively, from 30 percent in 1997, and did not decrease back to regional levels until2002 (Figure A.1.1). FigureA.l.l: RealGDPandinflationintheAndeanRegion 8.0% I 120.0 6.0% \ - 100.0 -5 4.0% 60.0 -z B 2.0% Y a 0) c 0.0% 60.0 .-0 E -- B e B = -2.0% - 0 40.0 -4.0% 20.0 -6.0% I/-Ecuador - Inflation] U 0.0 Source: Authors' calculationsbasedon data from the World Bank. RealGDP in 1995 $US.Andean regionGDP growth rates andinflationrate are GDP-weighted. 144 Ecuador exhibited low labor force participationrates compared to other countries inthe Andean region in 1990, especially regardingwomen, and the situation had not changed significantly in 2001 (Table A.1.1). That is, the country i s far from fully utilizing its labor resources, which in turn may have had a negative impact on growth. The distribution of overall employment across sectors appears to be closer to that of Venezuela than to that of any other country in the region due to the relative weight of the oil sector (included under `Agriculture and Mining'). The gender distribution of employment, however, is similar across all countries with men being relatively more present in agriculture, mining and agriculture, while women tend to be employed inthe service sector (Table A.1.2). Labor Force Participation Rate Labor ForceParticipation Rate All Female 1990 2001 1990 2001 Bolivia 72.1 72.5 36.9 37.9 Colombia 67.1 70.1 35.9 38.9 Ecuador 61.8 64.0 24.8 28.3 Peru 58.7 61.6 27.5 31.6 Venezuela 64.1 66.5 31.3 35.1 Table A.1.2 Employmentby Sector Agriculture Manufacturing Services All Male Female All Male Female All Male Female Bolivia 1990 L 1.2 1.9 0.2 25.1 34.6 11.5 72.8 62.8 87.3 1996 2.1 2.2 2.0 28.8 39.8 15.8 69.0 58.0 82.1 Colombia 1990 1.4 1.9 0.6 30.9 34.7 25.0 67.7 74.3 63.4 2000 1.1 1.6 0.5 25.5 30.2 19.9 73.3 79.5 68.1 Ecuador 1990 6.9 10.3 2.5 26.8 30.0 16.5 66.3 59.7 81.0 1998 7.3 10.5 2.3 21.4 26.4 13.6 71.2 63.0 84.0 Peru 1991 0.9 1.2 0.4 24.5 30.1 15.6 74.6 68.7 84.0 1999 5.8 7.9 3.3 18.7 24.8 11.1 75.5 67.3 85.6 Venezuela 1990 13.4 18.5 2.2 25.3 29.6 15.5 61.2 51.7 82.2 1997 10.8 15.6 1.5 23.8 29.2 13.4 65.1 54.9 84.9 I Source: World Development Indicators,The World Bank (2003). 145 Poverty and Social Indicators Table A.1.3 presents poverty numbers for all five countries and the Latin American region as a whole from various sources. However, the absence of a common methodology to measure poverty makes it extremely hard to compare poverty levels. Trends, in contrast, can be to some extent compared. It can then be observed that the second half of the 1990s was not a good time for the Andean region. Poverty increased in all five countries due to the economic crisis that affected the area in the late 1990s, and this increase was particularly severe inEcuador. Incontrast overall regionalpoverty levels decreasedby about 2 percentagepoints (or 5 percent) inthe same period. Table A.1.3 Poverty Level inthe Andean Region - I 1 1994 1995 1996 1997 1998 1999 2000 Change (%) A Ecuador Peru 53.5" 49.0 a 54.8 2.4 Venezuela 36.4e 44.6e 47.0e 48.5 e 33.2 L A C 36.7 34.4 34.9 -4.9 146 ANNEX2. ECONOMIC GROWTH ECONOMIC AND VOLATILITY -METHODOLOGICAL ISSUES. This annex describes important methodological and technical issues regarding the results presented in Chapter 1. In particular, it discusses (i) use of growth accounting to identify the sources of economic the growth, and(ii) the construction of an econometric modelfor long-term labor demand. Identification of Sources of Growth using Growth Accounting FollowingBergoeing et al. (2002), we employ the aggregate Cobb-Douglas production function: Y, =A,KplL:-" where Ytis output (value added), K,i s capital, L,i s labor, andA,i s total factor productivity (TFP), A, = KplL;-" y, In order to calculate A,, given Y, and L, ,we need to choose a value for a and generate series for Kt . Estimates of l-a, the share of labor compensation in GDP valued at factor prices (GDP at market prices minusindirecttaxes), basedon data fromthe NationalAccounts are implausibly small (0.39) compared to those available for developed countries (around 0.6-0.7) due to the fact that measured labor compensation in developing countries fails to account for the income of most self-employed and family-business workers, who make up a large fraction of the labor force (Gollin, 2002). Given this, and the observation that a value of 0.35 for l-awould imply unrealistically high rates of return to capital, we decide to make l-a equal to 0.6 for the purpose ofthe analysis. We then construct a capital stock series accumulating investment, I,over time as follows: K,+1 =(1-6)K,+ 4+l where 6, the depreciation rate, equals 5 percent, and KO(1950 is assumedto be t=O) equals 3. This latter assumption i s inconsequential since the initial value i s mostly depreciated by 1970, the startingpoint of my analysis. Using a and I?,we can then calculate TFP (A,) for every period. Figure A.2.1 plots the cyclical components of TFP after removing the long-run trend usingthe Hodrick andPrescott filter. 147 Figure A.2.1: The cyclical component of TFP I 2 I We know have all the elements we need to calculate the contribution of labor, capital and TFP to economic growth. In order to do this, we first take natural logarithms on the Cobb-Douglas production function presented above, andrearrangeterms to obtain: where Nt i s the number of hours available for work by working-age persons. With some further manipulation we decomposethe changeinreal GDPper capita over the period t to t +s into: S S S S According to this expression, GDP per capita growth is a function of (i) changes in TFP, (ii) changes in the capital-output ratio, and (iii) inhours worked per working-age person.99 changes 99Since total hours worked are not available, we use effective employment (l-p)Lf,where p is the unemployment rate, as an approximation. 148 On a balanced growth path, output per worker and capital per worker grow at the same rate, and the capital-output ratio and hours worked per working-age person are constant. On such a path, the growth accounting would attribute all growth to changes in TFP. The growth accounting exercise presented in Chapter 1 measures the contributions of labor and capital to the extent that they deviate from balanced- growth behavior (i.e. it measures the contribution of changes in the investment rate and changes in work effort). The results are reported on Table 1.1. An econometric modelof long-run labor demand We estimate an econometric model of the long-run demand for labor according to the principles of the standard neoclassical model (i.e. the representative firm maximizes profits -or minimizes costs- by choosing an adequatecombination of labor and capital) as follows: where W is the real wage and C K i s the real cost of capital. Estimation of this model requires -in addition to the capital series constructed above- a series for the cost of capital. Following Martinez et al. (2001), we use an approximation of the form: where r, i s the real interest rate, 6 i s the depreciation rate, and (PKt+l-PKJ/PK, is the expected benefit of holding one unit of capital (capital gains). The estimation results are reported below. Table A.2.1: Estimationof long-run demandfor labor (1980-2002) Constant GDP Real cost of Wage Capital CapitaYGDP R ADF 12.79 0.33 -0.15 0.01 Model 1 (0.43) (0.02) (0.03) (0.02) 0.88 -4.27 0.68 -0.15 Model 2 (0.67) (0.06) (0.03) I Source: Authors' calculations basedondata from the World Bank and the Ministry of FinanceinEcuador. Note: critical value for the ADF test is -3.45. The elasticity of the demand for labor with respect to value added in model 2 i s around 0.7, which i s consistent with estimates for other economies (Martinez et al., 2001). Likewise, the wage elasticity of around -0.15 and the capital ratio elasticity of 0.18 are in line with existing estimates for Chile. We perform stability tests - such as CUSUM and CUSUM-q- and find that these relationships are stable and do not present structural breaks (Figure A.2.2). 149 Figure A.2.2 Stability tests for structural breaks inthe long-rundemandfor labor - 0 . 4 ! , , , , , , , I I 1 1986 1988 1990 1992 1994 1996 1998 2000 -CUSUM-----5%Significance -CUSUMofSquares-----5%Significance Source: Authors' calculations based on data from the World Bank and the Ministry o f Finance in Ecuador. For this to be a reasonablerepresentation of the behavior of long-term labor demand, econometric theory requires that the various variables of the model are integrated of order 1and co-integrated. The unit root nnd rn-intpmntinn tests nresented helnw cnnfirm thnt this i s the case (Tables A 3 3 2nd A 3 ?I ~ Labor GDP Real wage cost of capital CapitaVGDP Consumption Firstunit root 0.17 -2.3 1 1.02 -2.45 -1.21 -0.74 Secondunit root -5.04 -5.28 -3.52 -3.30 -5.06 -5.09 Table A.2.3 Co-integrationtests Eigen value Trace Statistic 5% critical Max Eigen 5% critical value value value None 0.73 32.18* 29.68 27.18* 20.97 At most 1 0.18 4.99 15.41 4.23 14.63 i 150 ANNEX3. COMPARABLESMALL-AREA ESTIMATESOFPOVERTY: TECHNICALNOTE This annex discusses the methodology used to establish comparability across the 1990 and 2001 poverty maps presented inChapter 2. Comparableconsumptionaggregates from 1994 and 1999ECVs. We use the 1994 and 1999 ECVs and the 1990 and 2001 Population Censuses to construct the 1990 and 2001 poverty maps. For these two maps to be comparable, 1990 and 2001 poverty measures need to be based on comparable consumption aggregates. However, the consumption and expenditure modules in both 1994 and 1999 ECVs, on which the models for consumption are based, are different (i.e. different recall periods and level of disaggregation for certain goods or groups of goods). In order to solve this problem, we decomposed total expenditures in 1994 and 1999 as follows: Total expenditures = Comparableexpenditures Non-comparableexpenditures + where comparable expenditures (CE) refers to those goods in the consumption aggregate that were defined exactly the same in the 1994 and 1999 surveys. Non-comparable expenditures (NCE) i s the differencebetween Total expenditures (TE) andComparable expenditures. Table A.3.1 describesthe elements ineach of these components. Expenditures in housingand in water are treated as NCE. This i s becausethe values of housing and water expenditures were not directly asked in the survey for all of the households, and were imputed. Specifically, since a large number of households do not report the rental value of their home, we predicted their expenditure in housing using the parameters from a regression on the dwelling characteristics and other wealth indicators of those who did report their housing expenditures. Similarly, the water expenditures of those connected to the network were imputedfrom a regression on household characteristics of those who buy water from street vendors. Table A.3.1: Comparableand non-comparableexpenditures Comparable Non-comparable 1994per capita Food items (84%)" Fooditems (16%)* :onsumption +Consumer goods (46%)* +Consumer goods (54%)" +Educationand energy (26%)" +Durables +Education+Water** and energy (74%)* +Other basic services +Housing** =TOTAL COMPARABLE =TOTAL NON-COMPARABLE (63% of total expenditure) (37% of total expenditure) 1999per capita Food items (68%)" Food items (32%)" :onsumption, +Consumer goods (41%)* +Consumer goods (59%)" ?PIadjusted +Education andenergy (45%)" +Education and energy (55%)* + +Other Durables (91%) +Durables (9%)* basic services ++Housing** Water** =TOTAL COMPARABLE =TOTAL NONCOMPARABLE (55% of total exoenditure) (45% of total exoenditure) 151 Following Lanjouw and Lanjouw (1997), the headcount ratio does not change as the consumption definition changes from CE to TE when the followingassumptionshold: 1. CE andNCE are increasing in TE. 2. The budget share of CE declines as TE increase. Figures A.3.1 and A.3.2 explore whether these assumptions hold for 1994 and 1999 respectively. FigureA.3.1: Relationshipbetweencomparable and total expenditure -1994 TE 1.6e+06 1.2e+06 + w $ 600000 400000 100000 200000 30dOOO 400000 CE FigureA.3.2: Relationshipbetweencomparableand total expenditure -1999 TE 1.Oe+07 w + w^ 0 5.0e+06 1.Oe+06 2.0e+06 3.0e+06 CE 152 The shape of the distribution of points in both figures shows that CE and NCE are increasing in TE (assumption 1). In addition, the distance between the two lines increases as they move away from the origin (assumption 2). Comparable measuresof poverty in 1994and 1999 To construct comparable measuresof poverty for 1994and 1999, we start from the poverty figures of the 1996World Bank Ecuador Report, i.e. a headcount of 0.35 for 1994. Then, we follow these steps: 1. Findthe value of CE94that would correspond to a headcount ratio of 0.35 in 1994. Call it ZcE,94 or CE poverty line. 2. Average the value of TE94 in 1994 in the neighborhood of the household where CEg4= ZCE, Call itZTE, or 1994TEpoverty line (44,715 sucres of 1994). 94. 94 3. Usingthe consumer price index, deflate ZCE,94 to its value in 1999. Call it ZcE,99. 4. Average the value of TE99in 1999 in the neighborhood of the household where CEg9= ZcE,99. Call it ZTE, or 1999 TEpoverty line (221,304 sucres of 1999). 99 153 Table A.3.2: Comparableand Totalexpendituresin1994 and 1990 Incidenceof Incidenceof poverty Std. Err Numberof poor poverty Std. Err Numberof poor National 0.35 0.02 3,791,308 without Oriente 0.34 0.03 3,585,494 0.51 0.03 5,425,333 Costa 0.32 0.03 1,693,159 0.48 0.03 2,682,207 Sierra 0.36 0.03 1,892,335 0.54 0.03 2,743,126 Oriente 0.58 0.05 205,814 Quito 0.34 0.04 404,493 0.27 0.02 338,808 Guayaquil 0.31 0.03 536,375 0.39 0.03 719,434 Urbancosta 0.22 0.03 374,544 0.51 0.03 1,194,705 Urbansierra 0.20 0.02 263,113 0.51 0.03 909,722 Rural costa 0.43 0.05 782,240 0.55 0.04 768,068 Ruralsierra 0.46 0.03 1,224,729 0.73 0.02 1,494,596 Urbanoriente 0.22 0.05 12,462 Ruraloriente 0.65 0.05 193,352 Total Expenditures National 0.36 0.04 3,942,495 without Oriente 0.36 0.04 3,725,125 0.52 0.03 5,558,597 Costa 0.35 0.04 1,830,409 0.51 0.03 2,821,880 Sierra 0.36 0.03 1,894,716 0.54 0.03 2,736,7 17 Oriente 0.61 0.05 217,370 Quito 0.28 0.03 338,434 0.23 0.03 293,946 Guayaquil 0.27 0.03 467,810 0.36 0.03 665,258 Urbancosta 0.26 0.03 447,914 0.54 0.03 1,260,590 Urbansierra 0.19 0.02 256,539 0.50 0.04 884,472 Ruralcosta 0.50 0.05 914,685 0.64 0.03 896,03 1 Ruralsierra 0.48 0.04 1,299,743 0.76 0.03 1,558,299 r Urbanoriente 0.22 0.05 12,280 Ruraloriente 0.69 0.05 205,090 4391 5824 strata 8 11 463 640 Population 10,845,99 1 10,664,678 Source: Authors' Ci ulations basedon data from the 1994ECV andthe 1999ECV. 154 The first three of columns of the table show the incidence of poverty, its standarderror, and the number of poor from the 1994 ECV, while those in the last three columns show the figures from the 1999 ECV. Since no data was collectedfor the Oriente in the 1999 ECV, we present national figure with and without the Oriente for 1994. Moreover, in order to test for the validity of the assumptions imposed on TE and CE we compare the poverty figures generatedby bothandfind them to be statistically indistinguishable. Other data limitations Collection period of the 1999 ECV: As the 1999 ECV was collected over a period of 11 months and of high inflation, and since we have information on the month in which household was interviewed, the figures in Table B 2 were constructed with a deflated 1999 consumption aggregates so they are all in September 1999units. Geographic comuarabilitv of the data sources: As observed earlier, the evolution of poverty seems to present important differences between urban and ruralareas. However, underlying these differences there i s also the process of urbanization that rearranged the distribution of population along the territory. There i s one more issue that i s problematic when comparing the 1994 and 1999 ECVs. In addition to the comparability problems due to differences in the definition of consumption, there are comparability problems coming from changes in the definition of "rural" used in the survey design". Although the 1994 and the 1999 ECVs surveys are both designed based on the 1990 Population Census, by expanding the samples, we find that according to the 1994 sample, 56% of the population i s urban, while according to the 1999 sample, 68% of the population i s urban. These figures correspond to our definition of urban and rural, which is based on the administrative organization of Ecuador incantons and parishes (Le. the divisio'n politica administrativa elaborated by INEC) and not on any pre-determined population-based cutoff criteria according to which the distinction between urban and rurali s based on a particular size of the town in terms of population. Poverty maps The two Ecuador poverty maps were constructed following the methodology proposed by Elbers, Lanjouw and Lanjouw"', which has been used by the World Bank in several countries since the mid- 1990s. Infact, Elbers, Lanjouw, and Lanjouw constructed the first-period poverty map used in this study. Ina first stage, the methodology developed by these authors proposes to, with the household survey (in our case the 1994 and 1999 ECVs), estimate equations of per capita household expenditure at the stratum level. The explanatory variables in this equation must be defined in exactly the same manner in the household survey and in a population census that i s closest to the survey interms of time. The disturbance term in the per capita household expenditure equation consists of a cluster-specific term as well as of a household-specific term and i s modeled to control for spatial autocorrelation between neighboring locations as well as for heteroskedasticity in its household component. In its second stage, the methodology uses the parameters from the estimation based on the survey data to impute welfare measures on the census population (in our case, the 1990 and 2001 population censuses). This exercise i s done at the level of the different strata for which the survey i s designed to be representative. It allows producing different measures of poverty and inequality as well as their standard errors at very small levels of disaggregation. For the case of Ecuador, this level will be the parish.Moreover, for four of the largest cities in the country for which INEC was able to re-construct neighborhoodfrom the census tracts (Quito, Guayaquil, Cuenca, and Loja), we can go one level below and present neighborhood-level estimates of poverty and inequality"'. loo The 1994LSMS definition of rural included only small rural towns but not disperse rural settlements.These were only included in the sample starting in 1995. lo' Chris Elbers, Jean 0.Lanjouw, and PeterLanjouw, Micro-Level Estimation of Poverty and Inequality, Econometrica,71:1, pages 355-364, January 2003. lo'Itis important to notethat these neighborhoodsdo not correspondexactly to the urban parishes. 155 As the poverty map methodology implies deriving structural relationships between household and neighborhood assets and welfare, optimally, one would wish that the household census and the population survey corresponded to the same time period. At least, it is desirable that they correspond to a period when few structural changes took place. Unfortunately, this i s not exactly the case for Ecuador. For the first period of the poverty map, the best data available are a 1990 population census and the 1994 ECV, while for the second period we use a 2001 populationcensus and a 1999ECV. The use of different-year census and household surveys for the construction of the poverty map implies we are assuming that the returns to assets changed little over the period. It i s always problematic to make this assumption ina country that has faced many shocks of different kindslike Ecuador didinthe 90s. To illustrate this, Figure A.3.3 depicts the evolution of GDP per capita over the decade. The large dots mark the periods where censuses and household surveys were collected. While the change in GDP per capita i s larger between 1990 and 1994 than between 1999 and 2001 (Le. 6.6% versus 3.5%), 1999 i s the year of the plunge of the economy and where the crisis was at its worst. FigureA.3.3: Realper-capitaGDP 1990-2002 20.0 s 0 K 19.5 f u) 19.0 c IC 18.5 18.0 90 91 92 93 94 95 96 97 98 99 00 01 02 Year The Ecuadorian government has not collected a more updated household survey after the 1999 ECV so there was no alternative data source we could use in the poverty map. This also means that even if the crisis altered significantly the returns to household assets, there i s no instrument available for us to assess the magnitude of the change. However, as most of the variation in the imputation of poverty estimates comes from the census-level variables, we expect these estimatesto be closer to the "real" poverty figures that would correspond to the census year than to those of the survey year. Table A.3.3 compares the poverty estimates coming from the ECVs to the imputations from the poverty maps. Again, since the 1999 ECV did not include the Oriente region, it was not possible to apply the poverty map methodology and obtain poverty estimates for this region in 1999 and 2001. 156 Table A.3.3: Poverty rates I PERIOD2 ~ I PERIOD1 1994ECV 1990Census 1999ECV 2001Census HC Std.Err HC Std.Err Diff HC Std. Err HC Std. Err Diff National 0.363 0.036 0.410 0.020 ~ wlo Oriente 0.355 0.035 0.403 0.019 0.521 0.028 0.451 0.024 Quito 0.283 0.034 0.222 0.021 0.234 0.029 0.185 0.020 .I Guayaquil 0.272 0.026 0.382 0.018 *** 0.363 0.026 0.337 0.027 UrbanCosta 0.257 0.034 0.258 0.015 0.537 0.029 0.464 0.021 I Urban Sierra 0.194 0.024 0.213 0.017 0.495 0.035 0.459 0.022 Rural Costa 0.502 0.053 0.505 0.025 0.637 0.026 0.587 0.026 Rural Sierra 0.484 0.036 0.528 0.019 0.765 0.025 0.663 0.028 Source: Authors' alculationsbasedon datafrom the 1994ECV andthe 1999ECV. ~~ Note: Inthe povertymaps, Galapagosis classifiedas part of ruralcosta. *** ifECV andcensus figures are significantly different at 95%. While these figures are the "best possible" estimates of poverty because they are constructed using the most complete definition of the consumption aggregates for each year, they correspond to different consumptionaggregates and thus, are not useful to make statements on poverty changes over the decade for the same reasons discussed earlier. In order to produce poverty maps that could be compared over time, we used the same logic described earlier and re-estimated the consumption equations using - as dependent variables - stratum-level comparable expenditure (CE) for each of the two household surveys. With the parameters from this estimation, we imputed consumption using the census data and produced comparable poverty maps with the 1990 and 2001population censuses. Since the explanatory variables that entered in the consumption equations used to impute household expenditure were the same in both estimations where total expenditure (TE) and comparable expenditure (CE)were the dependent variables, it is not surprisingthat they resulted in similar rankings of the parishes in terms of poverty. Usingperiod-2 data (Le. the 1999 ECV andthe 2001 populationcensus), Figure B4 As we can see, they are close to each other, although - especially for parishes where poverty is higher - illustrates the differences in the parishlevel incidence of poverty imputed from models using CE and TE. the headcount imputed using TE as dependent variable seems systematically larger than the one imputed usingCE as dependentvariable. However, when we do a similar exercise with period-1data (Le. the 1994 ECV and the 1990 population census), we find the imputed headcounts at the parish level usingthe two methodologies are less than two standarddeviations away from each other. 157 FigureA.3.4: Incidenceof povertyimputedwiththe 2001populationcensususingcomparableand total expendituresatthe parishlevel 1 0.9 0.8 0.7 0.6 0.5 Headcount- CE 0.4 0.3 TableA.3.4 :Comparablepoverty rates 1990-2001 1990Census 2001Census Totalexpenditures Totalexpenditures Comparableexpenditures HC Std. Err HC Std. Err HC Std. Err National 0.410 0.020 w/o Oriente 0.403 0.019 0.45 1 0.024 0.452 0.023 Quito 0.222 0.021 0.185 0.020 0.243 0.016 Guayaquil 0.382 0.018 0.337 0.027 0.386 0.028 UrbanCosta 0.258 0.015 0.464 0.021 0.464 0.013 Urban Sierra 0.213 0.017 0.459 0.022 0.467 0.029 Rural Costa 0.505 0.025 0.587 0.026 0.504 0.017 Rural Sierra 0.528 0.019 0.663 0.028 0.617 0.034 UrbanOriente 0.192 0.020 Rural Oriente 0.598 0.026 I I 158 Table A.3.4 presents comparable poverty figures for the two poverty maps. Due to the fact that the differences found inthe 1990 headcounts when usingTE and CE as dependent variables were negligible, we stay with the map based on the total expenditure, or the best possible definition of expenditure. However, since this was not the case for the imputationsusingthe 2001 census, for this period we report headcount ratios based on both TE and CE'03.And therefore, it is the first and third columns of Table 5 that have comparable poverty figures from the two poverty maps. As was mentioned earlier, probably due to costs reasons, the sample of households in the 1999 ECV did not include any observations from the Orienteregion. Although the Oriente is not densely populated and i s small in relative population terms (it comprises about 4.4% of the national population, according to the 2001 Population Census), it i s a region with very unique economic, social, and institutional characteristics. Unfortunately, the lack of data does not allow us to estimate separate structural relations between welfare and household characteristics for the Orienteregion for the 2001 poverty map. Intryingto resolvethisproblem, afirst optionthat weexploredwasto apply oneofthe modelsestimated for the other survey strata (for example, rural Coast or rural Sierra) to the Orientecensus data in order to recover poverty figures. Since we do have Oriente survey and census data for 1990, we tested the accuracy of this alternative, but the results were not satisfactory. The rankingof the parishes in terms of poverty basedon other stratum models differed significantly from the one coming from a model estimated usingOrientedata. A second option to produce poverty estimates for the Oriente region was to predict poverty not from household-level attributes, but from parish-level attributes that could be constructed from the census. Usingthe parish-level estimates of poverty for the strata for which we did have survey and census data, we estimated the structural relationship between average parish per capita expenditure (and the headcount ratio) and parish-level variables. As explanatory variables, we used things such as the distribution of employment across sectors, the age-group distribution of population, availability of basic services and infrastructure, and the percentage of parish-level population who identified themselves as indigenous. Again, it was possible to validate the exercise usingfirst-period data for which we hadboth census (1990) and survey (1994) information available and the validation results were not satisfactory. The 1990 data showed that the rural Sierra was the better stratum in predicting poverty in the rural Oriente. Infact, the correlation between the "true" rural Oriente headcount numbers and the predicted numbers (based on the rural Sierra model) was .77 and the correlation between the rankings of parishes based on "true" and predicted numbers was .71. Similar results were obtained from predicting average per capita expenditure. As a result of the lack of data, we were not able to produce poverty figures for the Orienteregionin2001. The lesson learned from this exercise i s conclusive: in the estimation of poverty figures, it i s not possible to substitute for household survey data. lo3 Although at the parishlevel, differences in the 2001 poverty map imputed from TEand the one imputed from CE were statistically significant for a number of cases, it is worth noticing, however, that inTable 5 the strata-level figures are less than two standard deviations from one another. 159 ANNEX 4. MEASURING ANDMONITORINGPOVERTY, SOCIAL OUTCOMESAND PROGRAMS. We briefly discuss here the mechanismcurrently in place to measure and monitor both poverty and social outcomes inEcuador, as well as some ongoing initiatives to improve the effectiveness and accountability of social programs. Measurementand monitoringof poverty The main tool available to policy makers and researchers interested in measuring poverty in Ecuador i s the Encuesta de Condiciones de Vida (ECV) -the Ecuador version of the Living Standards Measurement Survey. Four such surveys have been collected during the past 10 years, in 1994, 1995, 1998 and 1999. However, as the proximity in the survey dates indicates, their collection has often responded to the availability of funding rather than to a more structured strategy to monitor poverty, which explains why a new post-crisis, post-dollarization survey has not yet been collected. As discussed in Box 2.2, the lack of a recent survey makes it impossible both to validate the poverty map exercise presentedinthis report and to produce more up-to-date poverty figures. The Statistical Institute of Ecuador (INEC) has recently presented a new strategy that includes a reorganization plan for all its household-based surveys. This plan proposes to unify the existing surveys under a single tool with a core questionnaire, basedon the Encuesta de Empleo, Desempleo y Subempleo (labor force survey) and a series of rotatingmodules. One of these modules would then be an income-and- expenditure module that would, potentially, allow for the periodical construction of comparable poverty figures. There are, however, two important problems with this proposal. First, the income-and-consumption module that i s currently being considered i s based on the Encuesta de Ingresos y Gastos (an income and expenditure survey used to adjust the composition of the urban CPI), rather than on the ECV. As we discussed in Annex 2, poverty measures are sensitive to the consumption aggregate used in their construction so that measures based on different aggregates are not comparable. Going ahead with INEC's proposal would then imply that the new poverty figures could not be compared to those produced previously usingthe ECV. Second, the LNEC plans to administer this survey quarterly, covering only the Sierra and the Costa and, within these, only urban areas in three out of the four quarters in the year. This implies that, for the poverty numbers to be representative of both urban and ruralareas inthe Sierra and the Costa the income- and-expenditure module would have to be administered during the quarter when both are covered. Moreover, since the Oriente will not be covered, poverty figures will not be nationally representative. We already discussedinChapter 2 the void inknowledge that such a decision would generate. We then strongly encourages the INECto reconsider their current strategy (i) to use the ECV income-and expenditure module to measureconsumption and, hence, poverty, and (ii) to include the Oriente in at least one of the quarterly surveys done each year. Only in this way will poverty numbers be consistent over time and nationally representative. Monitoring socialoutcomesandprograms Social outcomes, such as enrolment rates or infant mortality have traditionally been measuredusingeither the ECVs or the Population Census, of which there have been two in the last 15 years - 1990 and 2001. The plan put forward by the INEC proposes to use a series of modules whose objective i s to provide a more extensive set of measures at more regular intervals. This i s an excellent initiative, especially to the extent that this information can be linked to policy variables and to information on incidence of social programs so as to allow for the evaluation of their impact. 160 Being able to correlated outcomes to policies and programs i s key since most social programs in the country are poorly targeted and have never been subject to a formal evaluation. This, however, is starting to change since the Government of Ecuador has recently launched an ambitious reform plan for social programs whose essential pillars are improved targeting and impact evaluation. The main programs currently under transformation are the Bono de Desarrollo Humano (former Bono Solidario) and the feeding and nutrition programs, andplans exist to re-target the energy subsidies inthe near future. The tool being used for re-targeting i s the SelBen (see Box 5.6 for details), a household-level welfare index, and the target population the first and second SelBen quintiles of the household population. At the same time, impact evaluations are being conducted for these programs with technical and financial support from the World Bank. In sum, we are witnessing important advances regarding the availability of information on social outcomes and the adoption of policies to increasethe effectiveness and accountability of social programs. The Social Sector Technical Secretariat Finally, the social Ministries receive technical supported for their day-to-day operations from the Secretaria Tecnica del Frente Social (or technical secretariat, STFS hereafter) and, particularly, from the Sistema Integrado de Indicadores Sociales del Ecuador (the technicalunit within the STFS). The STFS and the SIISE have officially been appointed as the institutions responsible of processing information on and producing indicators for social outcomes and social programs on a regular basis, as well as of implementing the various impact evaluations of social programs. Their role, hence, will be crucial inensuring the success of the current reforms. 161 ANNEX 5. NOTWANTINGTO REINVENT THEWHEEL: THEECUADOR POVERTYASSESSMENT AND OTHERRECENT WORK ONECUADOR. This report has benefited significantly from existing work on macroeconomic, social, labor and poverty issues in Ecuador, generated both by national and international authors and institutions. Some of that work i s briefly discussed below, paying special attention to the differences and complementarities between these documents and the work presented in this report. The discussion is organized around five topics (macroeconomic developments, poverty, urban and rural economies, and social expenditures), so as to better related it to the structure of the report. Macroeconomic developments: The latest publications by the IFM (2003a and 2003b) provide a good review of the overall macroeconomic environment and the changes observed in the last few years, focusing on growth and fiscal policy, the evolution of investment, the banking system and the role of the oil industry-all of it, inthe context of a dollarized economy. Interms of particulartopics, the role of the oil sector and the dollarization seemto bethe two that have attracted the most attention in the recent literature. The first one has been studied indepth by Roy (2000) and Boye (2001), who analyze the advantages and disadvantages of the oil sector having such a primary role as the country's economic engine. On the issue of dollarization, the World Bank (2002) provides a comprehensive examination of the inherent opportunities and risks associatedwith the process, and its potential impact on social outcomes, based on the Ecuadorian and other relevant international experiences. Specifically on trade issues, the volume edited by Marconi(2001) studies the effects of dollarization on the country's capacity to compete and the Andean commercial integration initiatives, while work by Vos and Le6n (2003) examines the impact of the dollarization on export industries andequality. The work presented on this report focuses on the relationship between dollarization and poverty, paying special attention to their connection through labor markets and changes in prices. Measuring poverty: The two most recent World Bank reports on poverty in Ecuador date from 1996 and 2000 (World Bank 1996 and 2000?).Inaddition, significant work on the evolution and determinants of (urban) poverty has beendone by Vos and Le6n (2000), and Larreaand Sanchez (2002). The IntegratedSystem of Social IndicatorsinEcuador (SIISE), the technicalunitof the Ministry of Social Welfare, both collects and disseminates a series of social and poverty indicators on a regular basis, and operates as a researchgroup, having published in recent years numerous reports and studies based on this information. Many of these studies are cited throughout the report. The work discussed on Chapter 1 adds to this work in that it provides a dynamic analysis of rural and urbanpoverty and its determinants for the period 1990-2001 at the level of the canton. Urbanlabor markets andpoverty: Work by the L O (2001a and 2001b), the IMF(2000), MacIsaac and Rama (1997), and Heckman and Pages (2000) has explored the impact of social protection, labor market flexibility and wage determination mechanisms on labor market outcomes, such as employment and wages. The IDB (2004) also provides a good overview of labor markets in the region, putting the Ecuadorian case in context. Chapter 3 explores in detail the relationship between labor market outcomes and poverty, as well as existingconstraints to formal employment creation. Farmand off-farm sectors and ruralpoverty: The relationship between the farm and off-farm sectors and their capacity to generate employment and income is explored Elbers and Lanjouw (2001), while Hentschel and Waters (2002) and Kyle (2000) analyze the different strategies used by rural households to cope with shocks and poverty and find these include temporary migration, increased female and child labor and decreasedconsumption. 162 Chapter 4 focuses on the determinants of (local) agricultural productivity, and its relationshipto poverty, anddraws onexisting work with regardsto the off-farm sector. Social expenditures: This i s a topic that have received much attention in Ecuador in recent years. Work by Younger (1999), UNICEF (2000) and, in particular, SIISE (2003) has focused on the issues of progressiveness, efficiency, targeting, and impact of social expenditures, as well as on the relationship between social policy and the country's political economy. Chapter 5 of this report relies heavily on such work and contributes to the discussion with a detailed assessment of the links between social expenditures and poverty using new geographical disaggregated information that allows evaluation of relationships between targeting mechanisms and social outcomes. Other work: Finally a recent publication by the World Bank (2003?) covers an array of topics ranging from trade tojudiciary reform, to urban development, to health. 163 DATAAPPENDIX. Table DA.l: 1999Characteristics, by poverty status (mean and standard errors) Urban Rural Non - Poor Poor Non Poor - Poor Observations(Number of households) Costa 1317 689 402 398 Sierra 1085 488 439 1006 PopulationSize (expanded to the individual-level) Costa 2,251,554 1,925,848 510,973 896,031 Sierra 1,864,545 1,178,418 479,009 1,558,299 Composition of expenditure (proportions) Food Costa 0.444 0.579 0.574 0.632 Sierra 0.371 0.518 0.456 0.591 Consumer goods Costa 0.291 0.192 0.261 0.193 Sierra 0.341 0.222 0.331 0.217 Housing Costa 0.115 0.125 0.074 0.086 Sierra 0.119 0.133 0.077 0.091 Energyand education Costa 0.113 0.086 0.062 0.071 Sierra 0.125 0.104 0.090 0.080 Durables Costa 0.035 0.016 0.025 0.011 Sierra 0.044 0.019 0.044 0.015 Other Costa 0.001 0.002 0.003 0.007 Sierra 0.001 0.003 0.001 0.006 Dwelling characteristics Per capita number of rooms Costa 0.798 0.421 0.781 0.433 Sierra 0.961 0.462 0.949 0.476 Per capita number of bedrooms Costa 0.490 0.220 0.458 0.259 Sierra 0.575 0.282 0.506 0.285 Per capita number of bathrooms Costa 0.345 0.185 0.293 0.163 Sierra 0.398 0.212 0.358 0.188 Type of dwelling(proportions) House Costa 0.601 0.489 0.653 0.575 Sierra 0.470 0.349 0.673 0.548 Apartment Costa 0.238 0.063 0.042 0.009 Sierra 0.344 0.144 0.099 0.010 Room Costa 0.024 0.050 0.011 0.006 Sierra 0.113 0.143 0.057 0.024 164 Urban Rural Non- Poor Poor Non Poor - Poor Mediagua" Costa 0.120 0.283 0.149 0.154 Sierra 0.067 0.311 0.157 0.337 lanchor hut Costa 0.017 0.114 0.145 0.256 Sierra 0.006 0.054 0.014 0.081 Material of the roof :ement Costa 0.222 0.036 0.057 0.010 Sierra 0.482 0.210 0.217 0.063 isbestos Costa 0.194 0.043 0.113 0.113 Sierra 0.256 0.196 0.195 0.085 inc Costa 0.549 0.869 0.762 0.737 Sierra 0.085 0.244 0.162 0.339 ile Costa 0.006 0.002 0.019 0.056 Sierra 0.174 0.307 0.415 0.444 itraw Costa 0.002 0.030 0.036 0.074 Sierra 0.001 0.041 0.011 0.065 Material of the walls :ement Costa 0.881 0.599 0.675 0.446 Sierra 0.870 0.671 0.743 0.429 rdobe Costa 0.004 0.002 0.004 0.021 Sierra 0.110 0.262 0.172 0.347 food Costa 0.024 0.043 0.107 0.164 Sierra 0.014 0.040 0.059 0.136 :ane Costa 0.081 0.333 0.203 0.320 Sierra 0.000 0.001 0.006 0.009 Matei 11of the floor food Costa 0.011 0.000 0.003 0.000 Sierra 0.439 0.120 0.189 0.032 iles Costa 0.289 0.031 0.051 0.004 Sierra 0.092 0.017 0.070 0.003 :ement Costa 0.477 0.487 0.503 0.362 Sierra 0.248 0.390 0.393 0.224 ,oil Costa 0.020 0.069 0.045 0.072 Sierra 0.036 0.290 0.088 0.451 165 Non Poor - Urban Rural Poor Non Poor - Poor Public utilities 'ipedwater insidedwelling Costa 0.651 0.235 0.245 0.180 Sierra 0.771 0.280 0.506 0.073 'ipedwater out of dwelling, inside lot Costa 0.294 0.599 0.500 0.513 Sierra 0.202 0.584 0.418 0.771 'ipedwater out of dwelling and lot Costa 0.055 0.166 0.255 0.307 Sierra 0.027 0.135 0.076 0.156 later source: public network Costa 0.861 0.656 0.492 0.450 Sierra 0.977 0.864 0.855 0.803 later source: well Costa 0.035 0.124 0.318 0.252 Sierra 0.005 0.031 0.056 0.061 :onnectedto sewerage network Costa 0.536 0.209 0.058 0.008 Sierra 0.883 0.535 0.506 0.123 ionnectedto electricitynetwork Costa 0.985 0.955 0.863 0.745 Sierra 0.991 0.942 0.952 0.842 las phone Costa 0.452 0.091 0.090 0.029 Sierra 0.564 0.098 0.334 0.054 Sanitation , lastoilet Costa 0.928 0.752 0.689 0.450 Sierra 0.975 0.738 0.821 0.541 las latrine Costa 0.046 0.124 0.171 0.294 Sierra 0.005 0.055 0.060 0.109 las shower Costa 0.585 0.146 0.213 0.067 Sierra 0.815 0.312 0.592 0.157 Cooking and fuel lasexclusive room . ~cooking r Costa 0.753 0.566 0.648 0.678 Sierra 0.896 0.752 0.858 0.758 :ookswith gas Costa 0.947 0.877 0.867 0.743 Sierra 0.950 0.71 1 0.780 0.361 :ookswith wood Costa 0.019 0.108 0.112 0.248 Sierra 0.032 0.268 0.202 0.615 Homeownership iwner Costa 0.647 0.681 0.727 0.748 Sierra 0.564 0.500 0.730 0.798 166 Non - Poor Urban Rural Poor Non Poor - Poor lenter Costa 0.172 0.100 0.088 0.028 Sierra 0.315 0.322 0.135 0.037 Householddemographics :emalehead Costa 0.202 0.197 0.167 0.092 Sierra 0.178 0.160 0.133 0.131 Lgeof head Costa 45.453 45.836 46.484 47.480 Sierra 44.992 43.040 44.666 46.824 louseholdsize Costa 4.802 6.673 4.794 7.230 Sierra 4.486 5.852 4.436 6.302 lead issingle/widow/separateddivorced Costa 0.224 0.207 0.224 0.115 Sierra 0.226 0.165 0.176 0.137 lead ismarriedhasa partner Costa 0.776 0.793 0.776 0.885 Sierra 0.774 0.835 0.824 0.862 `0 of memberswho are male Costa 0.486 0.504 0.500 0.528 Sierra 0.502 0.502 0.490 0.488 b of memberswho are maleages 0-10 Costa 0.114 0.157 0.108 0.168 Sierra 0.111 0.176 0.102 0.162 b of memberswho are maleages 11-20 Costa 0.104 0.121 0.100 0.122 Sierra 0.104 0.114 0.104 0.114 b of memberswho are maleages 21-30 Costa 0.082 0.080 0.097 0.076 Sierra 0.102 0.075 0.088 0.067 b of memberswho aremaleages 31-40 Costa 0.070 0.058 0.066 0.057 Sierra 0.069 0.056 0.063 0.040 `0 of memberswho are maleages41-50 Costa 0.057 0.043 0.053 0.044 Sierra 0.045 0.030 0.055 0.037 b of memberswho are maleages >=51 Costa 0.058 0.044 0.076 0.060 Sierra 0.072 0.050 0.078 0.068 b of memberswho are female Costa 0.514 0.496 0.500 0.472 Sierra 0.498 0.498 0.510 0.512 b of memberswho are female ages 0-10 Costa 0.104 0.150 0.133 0.167 Sierra 0.087 0.159 0.123 0.164 167 Urban Non - Poor Poor Non Poor - Rural Poor % of memberswho are female ages 11-20 Costa 0.109 0.117 0.096 0.096 Sierra 0.099 0.100 0.101 0.113 % of members who are female ages 21-30 Costa 0.086 0.075 0.089 0.068 Sierra 0.102 0.086 0.083 0.065 % of memberswho are female ages 31-40 Costa 0.086 0.067 0.053 0.049 Sierra 0.066 0.059 0.059 0.046 % of memberswho are female ages 41-50 Costa 0.062 0.038 0.042 0.039 Sierra 0.065 0.034 0.053 0.042 0.004 0.003 0.006 0.003 % of memberswho are female ages >=51 Costa 0.067 0.048 0.087 0.052 Sierra 0.079 0.060 0.091 0.082 Ethnicity % ages 6 and older who speak only Spanish Costa 0.960 0.991 0.996 0.982 Sierra 0.906 0.852 0.916 0.739 % ages 6 and older who speak only Quichua Costa 0.000 0.000 0.000 0.000 Sierra 0.001 0.025 0.001 0.028 % ages 6 and older who speak Spanish and Quichua Costa 0.002 0.006 0.000 0.000 Sierra 0.015 0.116 0.051 0.232 Householdswhere a member speaks Quichua or Shuar Costa 0.003 0.009 0.000 0.001 Sierra 0.030 0.172 0.093 0.317 Education Head is illiterate Costa 0.044 0.217 0.108 0.201 Sierra 0.032 0.132 0.044 0.283 Grades completed by head Costa 9.761 5.474 6.528 4.217 Sierra 10.050 5.683 7.647 3.479 Head'sspouse is illiterate Costa 0.032 0.130 0.059 0.190 Sierra 0.026 0.157 0.071 0.342 Grades completed by head'sspouse Costa 7.424 4.437 5.616 3.734 Sierra 7.257 4.174 5.476 2.423 %of membersage 6 and older with no schooling Costa 0.023 0.059 0.047 0.066 Sierra 0.028 0.079 0.045 0.152 168 Urban Non - Poor Poor Non Poor - Rural Poor % of members age 6 and older with some primary schooling Costa 0.310 0.467 0.506 0.586 Sierra 0.309 0.484 0.461 0.567 %of members age 6 and older with some secondaryschooling Costa 0.360 0.264 0.274 0.144 Sierra 0.335 0.213 0.258 0.093 %of members age 6 and older with some post-secondaryschooling Costa 0.200 0.033 0.049 0.015 Sierra 0.226 0.030 0.113 0.008 %of members age 6 and older with a university degree Costa 0.060 0.006 0.006 0.002 Sierra 0.079 0.005 0.031 0.001 Fertility(among women ages 15-49) Number of live births Costa 2.421 4.079 2.673 4.709 Sierra 2.030 3.695 2.276 4.030 Employment(for members ages 10and older) Household head is employed Costa 0.889 0.845 0.881 0.895 Sierra 0.888 0.884 0.946 0.945 Head'sspouse is employed Costa 0.439 0.383 0.399 0.418 Sierra 0.501 0.551 0.623 0.690 %of memberswho are economically inactiveand in housework Costa 0.077 0.097 0.106 0.104 Sierra 0.054 0.042 0.036 0.036 Yoof members who are economically inactive and are students Costa 0.129 0.103 0.074 0.072 Sierra 0.138 0.090 0.093 0.038 Yoof members who are employed Costa 0.512 0.420 0.524 0.453 Sierra , 0.534 0.468 0.600 0.577 Yoof members with occupation: boss or director Costa 0.043 0.018 0.045 0.014 Sierra 0.047 0.014 0.074 0.018 Yoof memberswith occupation: independent worker Costa 0.121 0.117 0.133 0.107 Sierra 0.105 0.137 0.116 0.140 %of memberswith occupation: government employee Costa 0.051 0.008 0.027 0.006 Sierra 0.065 0.014 0.048 0.006 %of memberswith occupation: privateemployee (includesagriculturallaborers) Costa 0.222 0.201 0.185 0.177 Sierra 0.237 0.169 0.215 0.154 %of memberswith occupation: familyworker Costa 0.065 0.067 0.121 0.140 Sierra 0.071 0.122 0.132 0.248 Yoof members in the formal sector Costa 0.155 0.063 0.078 0.037 Sierra 0.193 0.054 0.143 0.031 169 Urban Non - Poor Poor Non Poor - Rural Poor K of membersin the informal sector Costa 0.356 0.357 0.445 0.415 Sierra 0.341 0.414 0.457 0.546 K of members in the agricultural sector Costa 0.035 0.091 0.198 0.259 Sierra 0.039 0.142 0.250 0.407 K of membersin the non-agriculturalhigh-productivitysector Costa 0.352 0.191 0.218 0.101 Sierra 0.357 0.173 0.248 0.085 K of members inthe non-agriculturallow-productivitysector Costa 0.125 0.138 0.107 0.093 Sierra 0.138 0.152 0.103 0.085 iousehold head is in the agricultural sector Costa 0.087 0.228 0.388 0.592 Sierra 0.075 0.256 0.411 0.650 iousehold head is in the non-agriculturallow-productivity sector Costa 0.246 0.322 0.213 0.213 Sierra 0.262 0.330 0.209 0.193 iousehold head is inthe non-agricultural high-productivity sector Costa 0.667 0.450 0.399 0.195 Sierra 0.663 0.414 0.380 0.157 iousehold head is in the formal sector Costa 0.364 0.191 0.165 0.087 Sierra 0.431 0.166 0.312 0.075 iousehold head is in the informalsector Costa 0.636 0.809 0.835 0.913 Sierra 0.569 0.834 0.688 0.925 source: Authors' calculations basedon data from the 1999ECV 170 : 1 4 e C i i E c c e C 1 C i 2 5a P P I "I I-- - b 4 S % Z Z 4 % 3 Z % ; f % f % B$m m m v b v e m m m m N m ~ m m N m m m m m m m o ~ m w m - m ~ ~ ~ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 \I II I, I I oc) t- 3 Table DA.4: Neighborhood-level poverty inGuayaquil, Quito, Loja and Cuenca I Number of Incidence of poverty Poverty Gap Severity of Poverty Gini Neighbhorhood Persons Coeff. Std.Err. Coeff. Std.Err. Coeff. Std.Err. Coeff. Std.Err. Guayaquil 1,936,731 0.34 0.03 0.11 0.01 0.05 0.01 0.43 0.01 Ayacucho 40,054 0.48 0.04 0.17 0.02 0.08 0.01 0.38 0.01 Bolivar 791,700 0.3 1 0.03 0.1 0.01 0.05 0.01 0.44 0.01 Carbo 25,281 0.26 0.03 0.08 0.01 0.04 0 0.39 0.01 Febres Corder0 4,990 0.1 0.02 0.03 0.01 0.01 0 0.42 0.02 GarciaMoreno 7,223 0.17 0.02 0.05 0.01 0.02 0 0.42 0.01 Letamendi 8,543 0.13 0.02 0.04 0.01 0.02 0 0.42 0.01 Nuevede Octubre 7,170 0.2 0.02 0.07 0.01 0.03 0.01 0.43 0.02 Olmedo 14,634 0.24 0.03 0.08 0.01 0.04 0.01 0.4 0.01 Roca 9,000 0.24 0.02 0.08 0.01 0.04 0.01 0.41 0.01 Rocafuerte 9,446 0.2 0.02 0.06 0.01 0.03 0 0.42 0.02 Sucre 332,571 0.38 0.03 0.13 0.01 0.06 0.01 0.38 0.01 Tarqui 98,876 0.34 0.03 0.11 0.01 0.05 0.01 0.38 0.01 Urdaneta 54,327 0.25 0.03 0.08 0.01 0.03 0 0.38 0.01 Ximena 11,695 0.18 0.02 0.05 0.01 0.02 0 0.39 0.01 Chongdn 9,499 0.44 0.04 0.16 0.02 0.08 0.01 0.44 0.01 Pascuales 503,142 0.37 0.03 0.13 0.01 0.06 0.01 0.41 0.01 Otros 8,580 0.69 0.03 0.33 0.03 0.2 0.02 0.45 0.03 Quito 1,386,414 0.19 0.02 0.05 0.01 0.02 0 0.45 0.01 ElCondado 54,665 0.26 0.03 0.07 0.01 0.03 0 0.43 0.01 CarcelCn 38,689 0.15 0.02 0.04 0.01 0.02 0 0.42 0.01 ComitCdel Pueblo 38,215 0.24 0.03 0.07 0.01 0.03 0 0.4 0.01 Poncean 51,463 0.14 0.02 0.04 0.01 0.02 0 0.42 0.01 Cotocollao 31,770 0.13 0.01 0.04 0 0.01 0 0.42 0.01 Cochapamba 44,042 0.23 0.02 0.07 0.01 0.03 0 0.44 0.01 Concepcidn 36,459 0.07 0.01 0.02 0 0.01 0 0.42 0.01 Keneddy 69,833 0.11 0.01 0.03 0 0.01 0 0.43 0.01 SanIsidro del Inca 29,718 0.23 0.02 0.06 0.01 0.03 0 0.44 0.01 Jipijapa 34,705 0.1 0.01 0.03 0 0.01 0 0.45 0.01 Iiiaquito 43,024 0.05 0.01 0.01 0 0 0 0.47 0.01 Rumipamba 30,794 0.05 0.01 0.01 0 0.01 0 0.47 0.01 BelisarioQuevedo 45,311 0.13 0.02 0.04 0.01 0.02 0 0.45 0.01 MariscalSucre 14,536 0.06 0.01 0.01 0 0.01 0 0.47 0.01 San Juan 58,534 0.2 0.02 0.06 0.01 0.02 0 0.43 0.01 Itchimbia 34,372 0.15 0.02 0.04 0.01 0.02 0 0.43 0.01 Puengasi 48,042 0.18 0.02 0.05 0.01 0.02 0 0.41 0.01 CentroHistdrico 47,780 0.24 0.02 0.07 0.01 0.03 0 0.43 0.01 LaLibertad 28,943 0.25 0.03 0.07 0.01 0.03 0 0.4 0.01 Chilibulo 46,123 0.22 0.02 0.06 0.01 0.03 0 0.4 0.01 San Bartolo 59,649 0.17 0.02 0.04 0.01 0.02 0 0.38 0.01 LaMagdalena 31,855 0.14 0.02 0.04 0.01 0.02 0 0.41 0.01 Chiimbacalle 44,173 0.17 0.02 0.05 0.01 0.02 0 0.41 0.01 179 I Number of Incidenceofpoverty PovertyGap Severity of Poverty Gini Neighbhorhood Persons Coeff. Std.Err. Coeff. Std.Err. Coeff. Std.Err. Coeff. Std.Err. La Ferroviaria 64,275 0.25 0.03 0.07 0.01 0.03 0 0.4 0.01 LaArgelia 45,907 0.26 0.03 0.07 0.01 0.03 0 0.39 0.01 Solanda 76,888 0.16 0.02 0.04 0.01 0.02 0 0.38 0.01 La Mena 35,877 0.19 0.02 0.05 0.01 0.02 0 0.39 0.01 Chillogallo 41,093 0.24 0.03 0.07 0.01 0.03 0 0.39 0.01 LaEcuatoriana 41,107 0.25 0.03 0.07 0.01 0.03 0 0.38 0.01 Quitumbe 38,863 0.23 0.03 0.07 0.01 0.03 0 0.39 0.01 Turubamba 31,260 0.29 0.03 0.08 0.01 0.04 0.01 0.38 0.01 Otros 48,449 0.31 0.03 0.09 0.01 0.04 0.01 0.39 0.01 Cuenca 271,280 0.28 0.02 0.1 0.01 0.05 0 0.47 0.02 San Sebastihn 28,424 0.31 0.02 0.11 0.01 0.05 0.01 0.48 0.02 Bellavista 24,699 0.29 0.02 0.1 0.01 0.05 0.01 0.46 0.02 Batik 18,953 0.31 0.03 0.11 0.01 0.05 0.01 0.48 0.05 Yanuncay 33,965 0.34 0.02 0.12 0.01 0.06 0.01 0.46 0.02 Sucre 17,251 0.19 0.02 0.06 0.01 0.03 0 0.45 0.02 Huaynachpac 14,831 0.18 0.02 0.06 0.01 0.03 0 0.45 0.03 GilRam'rez Dhvalc 8,307 0.24 0.02 0.08 0.01 0.04 0.01 0.49 0.05 ElSagrario 8,390 0.23 0.02 0.07 0.01 0.03 0.01 0.47 0.02 SanBias 10,772 0.18 0.02 0.06 0.01 0.03 0 0.44 0.02 ElVecino 27,918 0.32 0.02 0.11 0.01 0.05 0.01 0.45 0.02 Caiiaribamba 11,648 0.18 0.02 0.06 0.01 0.03 0 0.44 0.02 Totoracocha 23,241 0.24 0.02 0.08 0.01 0.04 0 0.44 0.02 Monay 15,610 0.27 0.02 0.09 0.01 0.04 0.01 0.43 0.02 Machhgara 12,530 0.35 0.03 0.12 0.01 0.06 0.01 0.47 0.03 Hermano Miguel 13,132 0.38 0.03 0.13 0.01 0.07 0.01 0.45 0.02 Otros 1,609 0.58 0.04 0.24 0.03 0.13 0.02 0.44 0.03 Loja 139,298 0.37 0.02 0.14 0.01 0.07 0.01 0.47 0.02 El Valle 20,614 0.38 0.03 0.14 0.01 0.07 0.01 0.47 0.02 Sucre 42,287 0.35 0.02 0.13 0.01 0.06 0.01 0.46 0.02 ElSagrario 15,477 0.23 0.02 0.08 0.01 0.04 0 0.47 0.03 San Sebastih 37,670 0.29 0.02 0.1 0.01 0.05 0.01 0.45 0.02 Otros 23,250 0.63 0.02 0.27 0.01 0.15 0.01 0.43 0.01 Source: Authors' calculations basedon data from the 1999ECV andthe 2001 PopulationCensus. 180