Poverty and Income Distribution in Ethiopia:1994-2000 By Abebe Shimeles, PhD.

30
Poverty and Income Distribution in Ethiopia:1994-2000 By Abebe Shimeles, PhD

Transcript of Poverty and Income Distribution in Ethiopia:1994-2000 By Abebe Shimeles, PhD.

Poverty and Income Distribution in Ethiopia:1994-2000

By

Abebe Shimeles, PhD

Structure of the presentation

1. Objectives of the study

2. Methodlogical Issues

3. Data

4. Key Results

I. Objective of the Paper

Analyze the state of poverty and income distribution during a period of peace, intense reform, good peace and recovery (1994-1997) and major drought, external war, terms of trade deterioration (1997-2000) based on a panel data set from rural and urban areas.

Simulate the effects of potential policy interventions on poverty.

2. Methodlogical Issues

2.1.Poverty measurement: identificaiton and aggregation issues

Setting the poverty line Aggregating poverty among the poor population

2.2.Robustness of poverty estimates Semi-parametric Kernel Densities Non-parmetric dominance criterion

2.3. Poverty and Inequality Decompositions The roles of observed and unobserved household

characteristics and the residual (including measurement errors)

A model of poverty Regression based inequality decompositions

3. Data

Panel data for urban and rural areasNot nationally representative, but

represents major agro-climatic conditions and major urban centers

Sampling and non-sampling errorsAttrition Selecitivity bias (demographic and other

time-varying household characteristics)

4. Key Results

Poverty trends during 1994-2000Poverty decreased between 1994-1997 and

increased between 1997-2000 (Table 1)

Table 1: evolution of poverty and inequality in Ethiopia

Type of Welfare (Poverty) Measure 1994 1995 1997 2000Rural Areas (N=1216):Headcount ratio, per adult equivalent 48 40 29 41

-0.014 -0.014 -0.014 -0.014

Poverty Gap ratio, per adult equivalent 21 16 10 14

-0.5 -0.48 -0.46 -0.5

Squared Poverty Gap ratio, per adult equiv. 13.1 10.2 6.02 10.2

-0.5 -0.44 -0.34 -0.44

Gini Coefficient, per adult equivalent 49 49 41 51

(.8)* (1.3)* (1.6)* (2.0)*

Headcount ratio, per adult equivalent 34 32 27 39

-0.015 -0.014 -0.014 -0.02

Poverty Gap ratio, per adult equivalent 13 11.4 9.6 14.5

-0.21 -0.2 -0.19 -0.24

Squared Poverty Gap ratio, per adult equiv. 6.5 5.6 4.7 7.5

-0.45 -0.42 -0.39 -0.48

Gini Coefficient, per adult equivalent 43 42 46 49

(1.3)* (1.0)* (2.0)* (2.3)*

Urban Areas(N=927)

Robustness of poverty trends

Semi-parametric kernel density estimates (Figures 1 and 2)

Non-parametric dominance criterion (Figures 3-6)

Figure 1: Kernel density Estimates for Rural Households: 1994-2000

0.1

.2.3

.4.5

dens

ity

0 2 4 6 8consumption

1994 19951997 2000

Figure 2: Kernel Density Estimates for Urban HHs: 1994-2000

0.2

.4.6

de

nsi

ty

0 2 4 6 8consumption

1994 19951997 2000

Decomposition of poverty : the role of unobserved household characteristics

Modelling poverty

itii k

jitkitk

K

k kitkit uXXXc ln

0iEu

22uiEu

0itE 22 itE 0),( itkitXCov

ksomeforuXCov ikit 0),(

Variables

Rural areasHousehold demographics Farming systemsAccess to marketSize of landRainfallMajor crops producedOff-farm activity, etc.

Variables (contd)

Urban areasHousehold demographicsOccupationEthnic backgroundAssets

Dealing with endogenity of regressors

Random-effects is preferred to fixed-effects if regressors are strictly exogenous. Hausman-specification test can be used to test if the two are equivalent. If not,

Instrumental variable methods(Hausman-Taylor random-effects model) is recommended to deal with endogenity.

Contd.

In our case, the random-effects specification was rejected for rural as well as urban regressors.

The HT method was employed to address endogenity. Results showed that the HT and Fixed effects specification are equivalent. So, HT is the preferred model of consumption.

Table 2: observed vs predicted poverty

1994 1995 1997 2000

Rural AreasHeadcount ratio due to observable and unobservable individual effects(%)

47 36 22 34

Headcount ratio due to “random-shocks” (%) 0 4 7 6Overall poverty (%) 47 40 29 40Contributions of predicted poverty in actual poverty (%) 100 98 76 85

Correlation between actual and predicted poverty at HH level (%) 51%

Urban AreasHeadcount ratio due to observable and unobservable individual effects(%)

32 26 19 34

Headcount ratio due to “random-shocks” (%) 1 6 8 4Overall poverty (%) 33 32 27 38Contributions of predicted poverty in actual poverty (%) 97 92 82 96Correlation between actual and predicted poverty at HH level (%) 57.4

Table 3: some policy simulations

1994 1995 1997 2000

Rural AreasLand Reform (redistribution) -1 -4 -5 -1Increase in the size of per capita land by 75% -2 -4 -4 -3Doubling of per capita land ownership -3 -4 -5 -4Increase in the “returns” to land by 20% -1 -3 -2 -2Increase in the “returns” to land by 50% -3 -7 -5 -5Improvement in infrastructure by 30% -6 -7 -8 -6Improvement in infrastructure by 50% -9 -10 -11 -9Urban AreasReduction in unemployment -1.5 -1 -2 -2Access to primary education (universal primary education) -9 -11 -10 -10

Table 4: Decomposition of inequality: rural areas

Gini Coefficient of Variation

Household size -6.3 4.38Land per capita 8.5 6.9land per capita squared 1.2 -1.4Mean age of the household 2.6 2.4Age of the head of the household -0.64 0.2Access to market 6.02 3.64Household durables 8.24 4.4Location 38.38 28.38Residual 42 51Total 100 100

TAble 5: Decomposition of inequality: urban areas

Ginihousehold size -8.2 5.7

household head completed primary 9.6 5.8

age hhh -1 0.6

Asset 10.9 7.3

Unemployment 0.31 0.2

Ethnicity 1.3 0.7

Occupation 18 2.4

Location 9.1 14.4

Residual 60 63

Total 100 100

Coefficient of Variation

Summary and conclusions

This paper analysed the state of poverty and income distribution in rural and urban Ethiopia during 1994-2000. Poverty declined from 1994 to 1997, and then increased in 2000.

This finding is consistent with major events that took place in the country: peace and stability, reform and economic recovery during 1994-1997, then, drought, war with Eritrea and political instability during 1997-2000.

To examine the robustness of these results, we used stochastic dominance criteria and model based decompositions of poverty and inequality.

Poverty trends were unchanged regardless of where one sets the poverty line.

.

contd

In addition, the paper attempted to look at the relative contributions of observed and unobserved household characteristics, and the residual, which includes random shocks and measurement error to observed poverty.

contd

This decomposition is useful to get a sense of how much of the observed poverty is due to persistent differences in household characteristics, and random transitory shocks that includes simple measurement errors.

From our results, we found that the contribution of the residual in observed poverty is in the range of 4%-27% in rural areas and 3%-18% in urban areas, which is reasonably low given the commonly held assumption that transitory factors account for much of observed poverty than persistent household characteristics.

Contd..

Part of the reason is that most of the omitted variables that could affect permanent attributes of a household are captured through the household-specific error term. In addition, attempt was also made to control for the effects of these error terms on observed regressors by using valid instruments in estimation. Perhaps this feature makes this paper interesting as it made an attempt to grapple with the often-ignored aspects of poverty measurement.

contd

The rest of the paper reported simulation results as well as inequality decompositions using standard methods

The results revealed that in rural areas, poverty responds quite strongly to improvements in infrastructure and increased size of land or its productivity, while in urban areas educational expansion could reduce poverty significantly

contd

Decomposition of inequality revealed that in rural areas 65% of overall inequality was due to location differences, access to market, size of land, dependency ratio in the household, and age of the household.

In urban areas, 49% of inequality was attributed to differences in education, occupational categories, and household durables. The results therefore imply that inequality is caused mainly by structural factors with the possibility that it may persist over time before significant decline can be observed

Thank you!!