Safety nets, asset growth and poverty transitions: Any roles for safety nets to help lift the poor out of poverty?
1. ETHIOPIAN DEVELOPMENT
RESEARCH INSTITUTE
Safety nets, asset growth and poverty transitions:
Any roles for safety nets to help lift the poor out of poverty?
Evidence from Ethiopia
Guush Berhane, Mehari Hiluf Abay, & John Hoddinott
Presentation at The Institute of Social Studies
The Hague, The Netherlands,
June 6, 2017
1
2. Outline
Introduction – the PSNP in Ethiopia
Research questions
Data and Method
Descriptive tables
Estimation results
Conclusions
2
3. The PSNP in Ethiopia
3
Safety net-oriented Public Work programs have recently received
prominence as key social protection interventions in Africa.
The PSNP - as move from emergency aid to predictable (&
productive) social protection program in Ethiopia.
Provides multi-annual “predictable” transfers, as food, cash, or both
Two components of the PSNP – Public Works (PW) & Direct Support.
Primary objectives of the PSNP (PW):
Protect food insecure – stochastically income poor – households
from falling into – structural - poverty
By smoothing consumption, thereby preventing asset depletion,
when faced with negative economic shocks.
Prevent long-term consequences of short-term food inaccessibility;
4. 4
The PSNP covers drought-prone, low ag., potential areas of Ethiopia
5. The PSNP beyond protection … ?
Beyond protection, the PSNP expected to be useful in several ways:
Employment opportunity/wage income during slack season,
Build ‘community assets & village infrastructure’,
Trickle down effects – village & national – on economy
Reduce vulnerability and build resilience
Potentials for scaling up during slow onset disasters,
Avenue to improving nutrition outcomes.
5
11. 11
The ‘P’ in the PSNP inspired by global experiences
12. Impacts of the PSNP – a snapshot
Reduces food gap
- from 3.09 months/year in 2006 to 1.75 in 2014 (Berhane et al 2016)
Improves household level (calorie) availability and dietary diversity
But no effects on child nutrition outcomes
- child diet quality remains poor (Berhane et al 2017).
Reduces vulnerability and improves resilience
- takes no more than 2 years for beneficiaries but up to four years for non-
beneficiaries to recover (Knippenberg and Hoddinott 2016).
SWC infrastructures enhance crop yields by 2.8 percent (average).
Increases local incomes (multipliers)
- by 1 to 2.4 Birr per each Birr transferred (Flipski et al 2016).
Increases household income – spreads far beyond receipents.
- by nearly 6 percent in PSNP areas and by nearly 2 percent in non-PSNP areas.
Raises national GDP by about 1 percent (after covering costs)
- Total economic benefits exceed the total costs of PSNP
12
13. Research questions
Impacts on the dynamics of asset accumulation and structural
income poverty less known.
Specifically, we study:
1) whether and to what extent the PSNP has prevented poor
households from falling into structural asset poverty due to
recurrent shocks?
2) whether the PSNP – through its predictability effects –
enables households build their asset bases and, in the long-
term, help them escape out of structural poverty?
3) If so, would initial asset endowment levels matter?
We provide evidence on this understudied issue of link between
shocks, safety nets, asset accumulation dynamics and poverty.
13
15. Data - a panel of households (beneficiaries and non-beneficiaries)
68 woredas in Amhara, Oromiya, SNNP and Tigray; interviewed in
2006, 2008, 2010, 2012 and 2014;
Significant effort to minimize attrition:
3,670 hh in 2006; 3,091 hh in 2014,
Attrition rate of 15.8%; 1.7% p.a. (better than US Census Bureau).
Major effort to ensure comparability over time:
Survey fielded at approximately same time of year,
Core questions, enumerator training etc. not changed.
First survey occurred one year after PSNP began officially, but RECALL
information included in 2006 survey.
Data
15
20. Conceptual framework
We follow Barrette (2005) to model asset dynamics as:
Yit = Kit rit + µit + εit
𝑌𝑌𝑖𝑖𝑖𝑖 = Household income; 𝐾𝐾𝑖𝑖𝑖𝑖 = productive (human & physical) asset endowments;
𝑟𝑟𝑖𝑖𝑖𝑖 = Returns on assets; 𝜇𝜇𝑖𝑖𝑖𝑖 = Exogenous shocks.
Assume µit and εit are mean zero
Asset-based expected income – Asset Index – can be specified as:
E Yit = Ait = Kitrit
Total differentiation of this yields:
dYit = dKit rit + Kit drit
Asset-based expected income depends on changes in productive asset
holdings and changes in the rates of return on assets.
20
21. 1. Construct an asset-based expected income
Follow Adato et al. (2006) to construct asset index using a livelihood
regression model:
𝑌𝑌𝑖𝑖𝑖𝑖
= �
𝑗𝑗
𝛽𝛽𝑗𝑗 𝐾𝐾𝑖𝑖𝑖𝑖𝑖𝑖 + �
𝑗𝑗,𝑘𝑘
𝛽𝛽𝑗𝑗𝑘𝑘 𝐾𝐾𝑖𝑖𝑖𝑖𝑖𝑖 𝐾𝐾𝑖𝑖 𝑖𝑖𝑖𝑖 + 𝛽𝛽ℎ 𝐻𝐻𝑖𝑖𝑖𝑖 + 𝜓𝜓𝑣𝑣 + 𝐷𝐷𝑡𝑡 + 𝛿𝛿𝑣𝑣𝑣𝑣 + 𝜀𝜀𝑖𝑖𝑖𝑖
• 𝒀𝒀𝒊𝒊𝒊𝒊= Scaled consumption expenditure, defined as a ratio of annual
consumption expenditure (per adult equivalent) to national poverty line.
• 𝑲𝑲𝒊𝒊𝒊𝒊𝒊𝒊 = vector of key asset variables 𝑗𝑗 owned by household 𝑖𝑖 at time 𝑡𝑡,
• Note: all items j are interacted with each other (k) to allow the marginal return
of assets to vary with the levels of other assets
• 𝑯𝑯𝒊𝒊𝒊𝒊 = vector of household characteristics,
• 𝝍𝝍𝒗𝒗, 𝑫𝑫𝒕𝒕, 𝜹𝜹𝒗𝒗𝒗𝒗 = dummies to control for district, time, & interaction terms.
21
22. 22
Using the predicted asset index, estimate the impact of PSNP on
household asset growth.
Use the following reduced-form growth model that allows for
transitional dynamics:
𝛥𝛥𝛥𝛥𝛥𝛥 𝐴𝐴𝑖𝑖𝑖𝑖 = 𝑋𝑋𝑖𝑖𝑖𝑖−1 𝛽𝛽 + 𝑙𝑙𝑙𝑙𝐴𝐴𝑖𝑖𝑖𝑖−1 𝜂𝜂 + 𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒊𝒊𝒊𝒊−𝟏𝟏 𝛾𝛾 + 𝛼𝛼𝑖𝑖 + 𝜇𝜇𝑖𝑖𝑖𝑖
• 𝛥𝛥𝑙𝑙𝑙𝑙𝐴𝐴𝑖𝑖𝑖𝑖 = asset growth 𝑙𝑙𝑙𝑙𝐴𝐴𝑖𝑖𝑖𝑖 − 𝑙𝑙𝑙𝑙𝐴𝐴𝑖𝑖𝑖𝑖−1 ,
• 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑖𝑖𝑖𝑖−1 = program participation or PW payment level.
• 𝑋𝑋𝑖𝑖𝑖𝑖−1= vector of observable household- and village-level
characteristics.
• 𝛼𝛼𝑖𝑖 = time-invariant unobservable
2. Estimate asset growth equations using predicted index
23. 23
Average asset index and poverty by PW status and year
Survey round
PW beneficiaries Non-beneficiaries
Asset
index
Asset
poor
(%)
Asset
index
Asset
poor
(%)
2006 0.76 79 0.87 71
2008 0.63 91 0.68 86
2010 1.06 57 1.13 49
2012 1.27 37 1.30 36
2014 1.41 26 1.57 23
Annual growth rate: 2006-
2014
10.4 % 10.1 %
24. 24
Static asset based poverty classification (2006)
Income
Poor
Income
non-poor
Total
Asset poor
Structurally, or
Chronically poor
Stochastically
non-poor
79 % 21 % 100 %
Asset non-poor Stochastically poor Structurally non-poor
33 % 67 % 100 %
25. 25
Asset poverty dynamics in the five rounds (2006 – 2014)
Number of times below
asset poverty line
Percentage of households:
PW beneficiaries
(%)
Non-beneficiaries
(%)
Never asset poor 3 5
Once asset poor 12 14
Twice asset poor 22 24
Three times asset poor 30 29
Four times asset poor 22 19
Always asset poor 11 9
26. 26
Panel data fixed-effects (FE) model to remove time-invariant
heterogeneities;
Combined with a time-trend (or random trend model) component to
mitigate some of the endogeneities due to time-varying
heterogeneities:
We thus estimate an empirical model of the following form:
𝛥𝛥𝛥𝛥𝛥𝛥 𝐴𝐴𝑖𝑖𝑖𝑖 = 𝑋𝑋𝑖𝑖𝑖𝑖−1 𝛽𝛽 + 𝑙𝑙𝑙𝑙𝐴𝐴𝑖𝑖𝑖𝑖−1 𝜓𝜓 + 𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒊𝒊𝒊𝒊−𝟏𝟏 𝛾𝛾 + 𝒈𝒈𝒊𝒊 + 𝜇𝜇𝑖𝑖𝑖𝑖−1
• t = individual-specific trend, g = trend parameter.
• where 𝒈𝒈𝒊𝒊 = 𝒈𝒈𝒊𝒊 𝒕𝒕 - 𝒈𝒈𝒊𝒊(t-1)
• Given A is in log form, then 𝒈𝒈𝒊𝒊 the average growth rate over a period.
Estimation: FE after first-differencing; or OLS after twice differencing
2. Estimation method – identification strategy
28. 28
The PSNP PW contributes to asset growth …
Controlling for individual-specific trends and several other
household and village level factors, compared to non-
participants’, participants’ asset was 5 percent more likely to
grow every year.
Specifically, a 100 Birr increase in the level of transfers
increases asset growth by about 2 percent a year.
This is after controlling for drought related shocks that
decreased asset growth by about 6 percent points.
29. 29
FE
Model
Individual
Trend
Model
FE
Model
Individual
Trend
Model
(1) (2) (3) (4)
Public Work participation 0.027** 0.050***
(0.013) (0.015)
Level of PW payment /100 0.017*** 0.015***
(0.002) (0.002)
Lagged asset (Log) -0.134*** -0.592*** -0.130*** -0.585***
(0.011) (0.011) (0.011) (0.011)
Drought -0.079*** -0.060*** -0.079*** -0.059***
(0.011) (0.011) (0.011) (0.011)
R2 0.58 0.47 0.58 0.48
F 497*** 243*** 505*** 249***
No. Fes 2196 2196
No. observations 8784 4392 8784 4392
The PSNP PW contributes to asset growth …
30. 30
More interestingly, we find that effects of safety nets do
not appear to be (initial) asset endowment neutral (non-
linearity).
In terms of long-term asset growth, the structural asset
poor are more likely to benefit from safety nets than the
non-structural asset poor (by about 6 percent as
compared to only 0.7 percent, respectively).
Impact varies by initial asset endowments …
31. 31
Household-specific trend model estimates of the impact of PW on
asset growth by initial welfare status
Initially structurally
poor
Initially structurally
non-poor
(1) (2) (3) (4)
Public Work participation 0.057*** 0.033
(0.018) (0.023)
Level of PW payment /100 0.019*** 0.007***
(0.002) (0.003)
R2 0.48 0.49 0.50 0.50
F 189*** 195*** 70*** 71***
No. observations 3260 3260 1132 1132
32. 32
3) Asset sensitivity to drought and the drought mitigating
Impact of PW payments by welfare status
Further, we find that the effect of drought is strong among the
poor, suggesting they are more sensitive to drought than the non-
poor group of households
Among the structurally poor, the effect of drought is about 10
percent
PSNP PW payment has a mitigating effect on drought shocks for
the poor households
33. 33
Individual trend model results of sensitivity to drought and the drought mitigating
Impact of PW payments by welfare status
(1) (2)
Drought*Structurally Poor -0.069*** -0.095***
(0.013) (0.016)
Drought*Structurally non-Poor -0.028 -0.035
(0.019) (0.022)
Log PW payments*Structurally Poor 0.010***
(0.003)
Log PW payments*Structurally non-Poor 0.009**
(0.003)
Drought*Log PW payments*Structurally Poor 0.012***
(0.005)
Drought*Log PW payments*Structurally non-Poor 0.004
(0.007)
R2 0.473 0.477
F 218.477 181.037
Observations 4392 4392
35. 35
Household Asset Structures & Holding Motives
Other than for productive purposes, households keep assets for
precautionary (saving or insurance) motives.
Livestock often used as store of value for a bad day
Livestock takes large part of asset composition.
Three potential reasons why the non-poor tend to slow down saving in
such assets (and perhaps invest in other forms?):
Timeliness and predictability of PSNP increased
PSNP reduce distress asset sales
Grazing area has become increasingly scarce, also due to PSNP PW
that often lead enclosures!
Two oxen (drought animals) and a cow often considered optimal;
observed shift away from quantify to quality – e.g., hybrid cows
36. 36
1. Distress sale of assets
Percent of households reporting distress sales of livestock by PSNP (Public
Works) status
0%
5%
10%
15%
20%
25%
30%
2006 2008 2010 2012 2014
non-psnp psnp
37. 37
PW payments decreased likelihood of distress asset sales
Dummy of
distressed sales
of assets
(1)
Dummy of distressed
sale of livestock
(2)
Level of PW payment per adult/100 -0.007** -0.009***
(0.003) (0.003)
R2 0.10 0.17
F 31.748 55.442
Observations 4392 4392
39. 39
Conclusions (tentative)
1) Beyond its protective roles, the PSNP has contributed to household asset
growth – lifting – households out of structural (asset-based) poverty.
2) This is important given results sustain after controlling for reported
drought effects.
3) Effects are however sensitive to initial asset endowments –increases larger
for households with poorer initial asset endowments.
4) Such heterogeneity, although subject to further scrutiny, may be possible if
the PSNP plays as an insurance into the livestock holding dynamics.
Households may have responded to the increasingly higher and predictable
transfer levels by saving less on assets for precautionary motives, but more
so for productive purposes (& until a max. asset threshold is reached).
This may be interacted with increasingly decreasing grazing area (increases
in cost of keeping an additional livestock) – triggering trade of between
quality and quantity assets hold.