Welfare Dynamics in Rural Kenya and Madagascar

1
Welfare Dynamics in Rural Kenya and Madagascar
Christopher B. Barrett, Paswel P. Marenya, John
McPeak, Bart Minten, Festus Murithi, Willis
Oluoch-Kosura, Frank Place, Jean Claude
Randrianarisoa, Jhon Rasambainarivo and Justine
Wangila     November 14-15 2003 USAID BASIS CRSP
Pre-conference Workshop on Combatting Persistent
Poverty in Africa Cornell University, Ithaca, NY
2
Why is poverty so persistent in rural Africa?
The design of appropriate strategies to combat
persistent poverty depend on its origins. Is
poverty something all people naturally grow
out of in time (unconditional convergence)?
implies laissez- faire or general support for
economy. some people grow out of in time
(conditional convergence)? implies need for
cargo nets. some people can be trapped in
perpetually (poverty traps due to multiple
equilibria)? implies need for safety nets and
cargo nets.
3
Brief theoretical background The poverty traps
possibility
Figure 1 Welfare Dynamics With Multiple
Equilibria

Key nonlinear path dynamics with multiple stable
dynamic equilibria and at least one unstable
dynamic equilibrium (threshold)
4
Brief theoretical background The slow
convergence possibility
Figure 1 Welfare Dynamics With Unconditional
Convergence

Key unique, common path dynamics with a single
stable dynamic equilibrium
5
Brief theoretical background The conditional
convergence possibility
Figure 1 Welfare Dynamics With Conditional
Convergence

Key unique path dynamics with a single stable
dynamic equilibrium for distinct groups or
individuals
6
Why bother with the theory?

• These three alternative theoretical foundations
  for understanding persistent poverty carry very
  different policy implications.
• - safety nets as poverty alleviation?
• - need for targeting?
• - need for patience?
• So it seems desirable to try to get a firmer
  handle on the etiology of persistent poverty.
  Thats the objective of this paper.

7
Methodological challenges of studying welfare
dynamics

• Qualitative vs. quantitative methods?
• Within each (esp. w/n quantitative)
• which metric?
• what point of comparison (poverty line and
  poverty measure)?
• how to control for inflation, FX rate
  changes?
• how to control for (changing) household
  demography?
• how to control for attrition?
• Given methodological fragility of any given
  method, we try to triangulate using different
  quantitative methods and by integrating
  qualitative and quantitative methods.

8
Data and Site Description

• Household-level panel data constructed for 7
  sites, building on previous surveys (not
  nationally representative)
• Stratification Better-worse market access
• Wetter-drier agroecology

Antsirabe
1 Embu (n113) 5 Dirib Gombo (Marsabit)
(n30) 2 Madzuu (Vihiga) (n89) 6
Fianarantsoa (n58) 3 Siaya/Vihiga (n103) 7
Vakinankaratra (n94) 4 Ngambo (Baringo)
(n30) N 517
9
Qual-Quant Integration
Community focus group discussions, key informant
interviews, and in-depth, open-ended household
interviews (oral histories) based on sampling
from 8 cells/sub-cells of ultra-poverty
transition matrices
10
Economic Mobility and Poverty Dynamics
Ultra-Poverty Transition Matrices As measured
against 0.50/day per capita income poverty line
Kenya rural poverty line 0.53, Madagascar
poverty line 0.43
11
Estimated annual gross (net) poverty exit rates
Estimate using mobility transition probability
PRt mt PR0 Site Gross Net Dirib Gombo
0.0 (0.0) Madzuu 2.2
(1.0) Fianarantsoa 2.3 (0.7) Vakinanka
ratra 2.4 (-4.2) Ngambo 5.2 (4.1)
12
Caveat on income measures

• How robust is the transition matrix to changing
  definitions of poverty? Not very. Have to be
  careful!

Comparison of 2000-2002 Poverty Transition Rates
Using Different Measures Siaya-Vihiga Villages
13
Moving beyond headcount measures

• We want to know the directions and magnitudes of
  welfare change, not just discrete movements
  relative to an arbitrary poverty line.

Annual average percent change in income, by site
and resurveying interval
14
Filtered vs. unfiltered income change regressions
Unfiltered Y Ar eR U eT eM
(2) dY dA r eR Adr deR deT deM
(4) includes measurement error negative
bias Filtered EY Ar U
(3) EdY EdAr AEdr (5) omits
true stochastic component of income positive
bias Regress dY on Y, EdY on EY, or both to
bracket?
15
Site-specific filtered and unfiltered income
change regressions It clearly makes a difference

16
Qualitative Evidence on Economic Mobility and
Poverty Dynamics

• Complementary qualitative work strongly suggests
  multiple equilibria/poverty traps or conditional
  convergence, depending on who one asks, the poor
  or the nonpoor
• (1) The nonpoor describe conditional
  convergence. Individual attributes (work ethic,
  drunkenness, etc.) lead to long-term differences.
  
• (2) The poor describe poverty traps, emphasizing
  the role of asset shocks in explaining observed
  mobility. Health shocks (chronic or transitory
  disease, death and resulting funeral expenses)
  especially important, but also livestock shocks
  (e.g., raids or disease) and land shocks (e.g.,
  costly disputes). Higher-return activities
  beyond their reach given entry barriers to
  remunerative livelihoods.

17
Summary of Findings on Economic Mobility and
Poverty Dynamics

• Considerable persistence of ultra-poverty with
  low rates of net exit from poverty
• Poverty deepest where agroecology and markets
  least favorable (remote rural areas or less
  favored lands)
• Stochastic component of income appears
  substantial
• Whether by quantitative methods, with some effort
  to filter out measurement error, or by
  qualitative methods, the hypothesis of
  unconditional convergence seems hard to support.
  Not at all clear whether the conditional
  convergence or poverty traps hypotheses, or
  both, best explain these data.

18
Does Economic Immobility Signal Poverty Traps?

• 1) Risk-taking and consumption smoothing behaviors

Wealth-dependent risk management among northern
Kenya pastoralists
19
Does Economic Immobility Signal Poverty Traps?

• 1) If income variability increases with wealth,
  so should returns on assets. Indeed, the
  income-herd size relation exhibits increasing
  returns, consistent with risk-based poverty traps

20
Does Economic Immobility Signal Poverty Traps?

• 2) Barriers to entry into higher-return
  activities
• - educational attainment and rationing
• - lack of credit and liquid savings (negligible
  credit access) limited capacity to enter
  higher-return businesses or even to buy livestock
  
• - pastoralist mobility depends on herd size
• expected result is nonlinear asset dynamics,
  with rapid accumulation beyond key thresholds

21
Does Economic Immobility Signal Poverty Traps?
The asset data appear consistent in the Kenya
sites with multiple equilibria, but in the
Madagascar sites, conditional convergence seems
to fit better.
Herd Dynamics in Northern Kenya
Asset Index Dynamics
22
Does Economic Immobility Signal Poverty Traps?
Same with the income data. Multi-modal income
distribution in Madzuu.
2002 Income Distribution in Madzuu

• Consistent with qualitative evidence
• Importance of non-farm salaried employment,
  incl. to agricultural intensification
• Fragility of non-poor status, esp. to health
  shocks

23
Does Economic Immobility Signal Poverty Traps?
But unimodal distribution in Madagascar
reflective more of conditional convergence with
significant geographic grouping.
Implied dynamic real income equilibria Vakinankar
atra 0.61 Fianarantsoa 0.33 Latter
seems a geographic poverty trap
24
Conclusions and Policy Implications
1) Evidence seems to point rather strongly
against the unconditional convergence hypothesis
on which neoliberal reforms of the 1990s were
implicitly based, which implies all households
naturally grow out of poverty in time. 2)
Qualitative and quantitative evidence
inconclusive as to whether the conditional
convergence, poverty traps or both hypotheses are
accurate. Cross-sectional differences at
community level appear consistent with the
conditional convergence hypothesis. 3)
Geographic targeting would thus seem useful in
targeting interventions intended to stimulate
growth and higher equilibrium productivity/income
levels.
25
Conclusions and Policy Implications
4) Evidence on within-community differences
associated with conditional convergence appears
weaker, although the nonpoor certainly claim this
to be true. 5) There certainly seem cases in
Kenya the evidence is less supportive in
Madagascar of poverty traps, multiple dynamic
equilibria between which households can shift.
This places a premium on effective, pre-set
safety nets to keep people from falling into
lower-level equilibria.
26
Misaotra! (Thank you!)