### Lecture 2 (2up)

```The Global Social Challenge: Session II
(2013-400-SPP4205)
Richard Bluhm
February 17, 2014
How does poverty react to income growth and
changes in distribution?
Introduction
The questions we’ll try to answer in this session:
I
I
I
I
I
I
Is growth always good for the poor?
What is pro-poor growth?
How do we empirically measure pro-poor growth?
How much of poverty alleviation is due to growth?
How much of poverty alleviation is due to redistribution?
What influences pro-poor growth?
Let’s brainstorm: How would you define
pro-poor growth?
Two definitions from the literature
Absolute definition
I
I
Growth that reduces poverty. Period.
The growth process is pro-poor if and only if poor people
benefit in absolute terms, as reflected in an appropriate
measure of poverty (Ravallion 2004).
Relative definition:
I
I
Growth that disproportionately reduces poverty.
Growth is pro poor if any distributional shifts accompanying
economic growth favor the poor, meaning that poverty falls
more than it would have if all incomes had grown at the same
rate (Ravallion 2004).
Three sources of pro-poor growth
According to Kraay (2006) these are
1. rapid growth in average incomes
2. a high sensitivity of poverty to growth in average incomes; and
3. a poverty-reducing pattern of growth in relative incomes.
I
Which can actually also be broken up further into a)
distributional change and b) a high sensitivity of poverty to
distributional change.
This implies a decomposition equation with four components.
Poverty decompositions
I
At the micro-level, we can decompose poverty into changes in
incomes and changes in inequality up to some error (Datt &
Ravallion, 1992; Kakwani, 1993)
Ht − Ht−1 = ∆Ht =H(¯
yt /z, Lt−1 ) − H(¯
yt−1 /z, Lt−1 ) +
H(¯
yt−1 /z, Lt ) − H(¯
yt−1 /z, Lt−1 ) + ζt
I
where H is the poverty headcount ratio, y¯ is mean income, z
is the poverty line, and L denotes the Lorenz curve.
At the cross-national level, we typically don’t have each unit
record but only grouped data or just the Gini coefficient.
Calculus of poverty decompositions I
Let the poverty line (z) be fixed (absolute) and assume the poverty
headcount ratio is follows a two-parameter distribution, so that
H(¯
yt /z, Gt ) = H(¯
yt , Gt ) ≡ Ht . A first-order Taylor expansion gives
dHt
≈
Ht
∂Ht y¯t
∂¯
yt H t
|
{z
Growth
where
I dHt
Ht
d¯
I yt
y¯t
I
d¯
yt
∂Ht Gt dGt
yt
y¯ d¯
HG dGt
+
ε
+
= εH
t
t
y¯t
∂Gt Ht Gt
y¯t
Gt
} |
{z
}
Distribution
is the relative change in the headcount ratio
y¯
is the growth rate in mean incomes and εH
is the income
t
elasticity of poverty (recall: x% change for 1% change)
dGt
HG is the
Gt is the change in the Gini coefficient and εt
inequality elasticity of poverty
Calculus of poverty decompositions II
Same argument can be made with absolute changes and
semi-elasticities
∂Ht y¯t
dHt ≈
∂¯
yt 1
|
{z
Growth
where
I
I
I
d¯
yt
∂Ht Gt dGt
d¯
yt
dGt
+
= ηtH y¯
+ ηtHG
y¯t
∂Gt 1
Gt
y¯t
Gt
} |
{z
}
Distribution
dHt is the absolute change in the headcount ratio
H y¯
d¯
yt
is the income
is
the
growth
rate
in
mean
incomes
and
η
t
y¯t
semi-elasticity of poverty (recall: unit change for 1% change)
dGt
HG is the
Gt is the change in the Gini coefficient and ηt
inequality semi-elasticity of poverty
Putting it together
The Growth Elasticity of Poverty Reduction
9
Figure 1.2
Decomposition of change in distribution and poverty into growth and distributional effects
Back to Aart Kraay (2006)
For the absolute pro-poor definition and the 1.08\$ a day line,
Kraay finds
I
I
I
In the short run, about 2/3 of poverty reduction are due to
growth and about 1/3 is due to changes in distribution.
In the long-run, it’s even 97% due to growth!
The contribution of growth decreases as the poverty measure
becomes more bottom sensitive.
This is not set in stone. It’s an analysis of the historical
contributions, which could have involved a higher contribution
from distribution. The story is a little more intricate. . .
Annualized Change in Log Gini (∆ ln Git )
Changes in (within) inequality and changes in incomes
0.1
0.0
-0.1
-0.25
0.00
0.25
Annualized Change in Log Mean Income (∆ ln y¯it )
I
0.50
Basically no trend; also over time, trend rate is mildly positive,
so Kraay (2006) cannot find a large contribution of inequality.
Properties of elasticities and semi-elasticities
Elasticities:
I
I
y¯
The income elasticity is weakly negative (εH
t ≤ 0) and
decreasing in average income but increasing in inequality
The inequality elasticity is weakly positive (εHG
≥ 0) and
t
increasing in average income but decreasing inequality
Semi-elasticities:
I
I
The income semi-elasticity is weakly negative (ηtH y¯ ≤ 0). It
first decreases (increases) then increases (decreases) in
incomes (inequality). Bounded by zero on each side.
The inequality semi-elasticity is weakly positive (ηtHG ≥ 0). It
first increases (decreases) then decreases (increases) in
incomes (inequality). Bounded by zero on each side.
The shape of poverty elasticities
4
0
-1
Gt
0.25
0.35
0.45
0.55
0.65
0.75
0.85
0.95
3
-2
εˆH y¯
εˆHG
2
-3
1
-4
-5
0
2-2
2-1
I
I
20
y¯t /z
21
22
23
2-2
2-1
20
Strictly (decreasing) increasing in incomes.
Bounded on the left, unbounded on the right.
y¯t /z
21
22
23
The shape of poverty semi-elasticities
Gt
0.0
0.25
0.35
0.45
0.55
0.65
0.75
0.85
0.95
0.4
0.3
ηˆH y¯
ηˆHG
-0.2
0.2
-0.4
0.1
0.0
-0.6
2-2
2-1
I
I
20
y¯t /z
21
22
23
2-2
2-1
20
y¯t /z
21
22
Very non-linear when we look at absolute changes.
Bounded on the left and right.
Estimates of income elasticities
1981–1989
EAP
ECA
LAC
MNA
SAS
SSA
-0.991
(0.030)
-4.358
(0.555)
-2.284
(0.243)
-2.176
(0.203)
-0.548
(0.053)
-0.831
(0.027)
Time period
1990–1994 1995–1999
-1.029
(0.033)
-2.892
(0.309)
-2.374
(0.257)
-2.116
(0.188)
-0.629
(0.048)
-0.437
(0.039)
-1.237
(0.055)
-2.700
(0.277)
-2.425
(0.271)
-2.024
(0.168)
-0.810
(0.030)
-0.436
(0.040)
2000–2004
2005–2010
-1.139
(0.043)
-2.846
(0.304)
-2.349
(0.258)
-1.966
(0.161)
-1.024
(0.032)
-0.592
(0.035)
-1.578
(0.101)
-3.304
(0.384)
-2.985
(0.366)
-2.501
(0.246)
-1.192
(0.046)
-0.632
(0.033)
23
Estimates of inequality elasticities
1981–1989
EAP
ECA
LAC
MNA
SAS
SSA
0.732
(0.105)
3.219
(0.510)
1.687
(0.186)
1.607
(0.197)
0.405
(0.093)
0.614
(0.087)
Time period
1990–1994 1995–1999
0.760
(0.101)
2.136
(0.307)
1.753
(0.198)
1.563
(0.198)
0.464
(0.097)
0.322
(0.055)
0.914
(0.113)
1.994
(0.283)
1.791
(0.199)
1.495
(0.196)
0.598
(0.095)
0.322
(0.060)
2000–2004
2005–2010
0.841
(0.108)
2.102
(0.296)
1.735
(0.189)
1.452
(0.185)
0.756
(0.107)
0.437
(0.066)
1.165
(0.144)
2.440
(0.353)
2.205
(0.269)
1.847
(0.253)
0.880
(0.127)
0.467
(0.069)
2000–2004
2005–2010
-0.583
(0.042)
-0.225
(0.015)
-0.355
(0.026)
-0.463
(0.043)
-0.572
(0.036)
-0.440
(0.015)
-0.552
(0.051)
-0.134
(0.010)
-0.194
(0.013)
-0.313
(0.024)
-0.585
(0.044)
-0.459
(0.015)
Estimates of income semi-elasticities
1981–1989
EAP
ECA
LAC
MNA
SAS
SSA
-0.568
(0.034)
-0.031
(0.008)
-0.374
(0.028)
-0.405
(0.034)
-0.418
(0.023)
-0.532
(0.024)
Time period
1990–1994 1995–1999
-0.573
(0.036)
-0.214
(0.015)
-0.348
(0.025)
-0.422
(0.037)
-0.458
(0.019)
-0.354
(0.020)
-0.585
(0.046)
-0.260
(0.020)
-0.334
(0.024)
-0.447
(0.042)
-0.526
(0.022)
-0.353
(0.020)
Estimates of inequality semi-elasticities
1981–1989
EAP
ECA
LAC
MNA
SAS
SSA
0.419
(0.053)
0.023
(0.007)
0.276
(0.046)
0.299
(0.041)
0.309
(0.056)
0.393
(0.050)
Time period
1990–1994 1995–1999
0.423
(0.053)
0.158
(0.015)
0.257
(0.043)
0.311
(0.040)
0.338
(0.055)
0.261
(0.037)
0.432
(0.055)
0.192
(0.019)
0.247
(0.043)
0.330
(0.041)
0.389
(0.052)
0.261
(0.040)
2000–2004
2005–2010
0.431
(0.054)
0.166
(0.017)
0.262
(0.045)
0.342
(0.044)
0.423
(0.054)
0.325
(0.042)
0.408
(0.053)
0.099
(0.012)
0.143
(0.029)
0.231
(0.025)
0.432
(0.055)
0.339
(0.043)
What does this mean for poverty reduction?
Elasticities:
I
I
Depend on the initial level of income and inequality
Very popular measure of the sensitivity of poverty but
somewhat misleading: richer countries appear to become ever
better at reducing (absolute) poverty
Semi-elasticities:
I
I
I
Depend on the initial level of income and inequality
Much more useful to think of the percent of population lifted
out of poverty, more policy relevant!
Highlight the non-linearity much better (are more intuitive?)
Different returns to each component depending on location on the
curves!
A strangely relative perspective
If we define poverty as relative not absolute poverty, and a rising
tide lifts all boats, then growth should lead to a one-to-one
increase in incomes of the poor.
Dollar and Kraay (2002) “Growth is good for the poor” JEG (very
controversial study!)
I
I
I
I
Test the relationship between growth of income shares of the
poorest quintile and GDP per capita growth
Strongly confirm one-to-one relationship even after accounting
for reverse causality
Find no other systematic relationships, except a weak negative
effect of government size and high inflation
Say very little about initial locations and changes in inequality
Incomes of the poor and average incomes
Is this robust? What do we learn?
Robust?
I
I
There are many flaws with they data used but . . .
Dollar, Kleineberg and Kraay (2013) repeat the analysis with
better data; they find the same.
What do we learn?
I
I
I
I
By looking at relative poverty they conflate the effects of
growth and inequality
It only proves that growth is distribution-neutral it does not
show that growth is pro-poor on average
Should have been titled “growth does not affect inequality”
Has nothing to do with absolute poverty reduction!
Better definitions of pro-poor growth
Why don’t we simply compute the growth rates of each quintile/
percentile over time?
Ravallion and Chen (2003) propose the following:
Let yt (p) the income of the pth quantile, then the growth rate in
income of the pth quantile is gt (p) = (yt (p)/yt−1 (p)) − 1 (for one
period).
L0t (p)
(γ + 1)
L0t−1 (p) t
slope L0t (p)) and
More generally, let gt (p) =
− 1 where where Lt (p)
is the Lorenz curve (with
µt is the mean and
γt = (µt /µt−1 ) − 1 is the growth rate in the mean.
Very easy to estimate in Stata if you have at least two household
surveys for the same country. Visually intuitive (next slide).
GIC for China 1990-1999: is this pro-poor?
GIC for China 1993-1996: or, is this pro-poor?
The pro-poor rate of growth
Based on the graph we can come up with many relative definitions
(e.g. growth of the poor must be higher than the rest). Ravallion
and Chen (2003) define pro-poor growth as the mean growth rate
of the poor divided by the poverty headcount ratio:
PPGt =
Z
Ht−1
gt (p)dp/Ht−1
0
I
I
I
absolute interpretation: it’s the growth rate in the mean
scaled up or down according to whether the distributional
changes were pro-poor
relative interpretation: higher than mean growth = pro-poor,
lower than mean growth = not pro-poor
turns out the PPGt is just the negative change in the Watts
index of poverty divided by the initial headcount
GIC with Stata
* get required software
net install gicurve, replace ///
* get example data
local url1 "http://siteresources.worldbank.org/"
local url2 "INTPGI/Resources/342674-1223471357039"
use "‘url1’/‘url2’/ugahh05.dta", clear
save ugahh05, replace
use "‘url1’/‘url2’/ugahh02.dta", clear
* compute GIC (normally use pw, here aw needed!)
gicurve using ugahh05 [aw=iwe], var1(welfare) ///
var2(welfare) yperiod(3) np(200) ci(500 95) ginmean ///
hcindex(38.82) title("Uganda 2002/03-2005/06") ///
ytitle("Growth rate") legend(off) xline(38.82)
Results
−5
0
Growth rate
5
10
15
Uganda 2002/03−2005/06
0
20
40
60
80
100
Percentiles
I
Pro-poor growth 4.44, growth in mean 3.61, z is at 38.82
Extensions and other measures
Kakwani (2000) defines pro-poor growth as “any distributional
shift accompanying economic growth that favors the poor”.
Basically targets inequality. This has some problems:
I
I
In a contraction the rich may be hit harder, does that qualify
as pro-poor growth?
In a growth spurt everyone may gain but the rich a little more,
is that not pro-poor?
Son (2003) proposes a poverty growth curve (PGC). Growth is
pro-poor, if the generalized Lorenz curve (Lt (p) × µt ) shifts up.
Kakwani and Son (2008) discuss the poverty equivalent growth
rate (PEGR). It’s “the growth rate that will result in the same level
of poverty reduction as the present growth rate if the growth
process had not been accompanied by any change in inequality”.
Is growth typically pro-poor?
I
Kraay (2006) and many other have shown: yes, in the absolute
sense. Growth is the main driver behind poverty reduction.
I
However, in the relative sense growth is typically
distribution-neutral. Neither pro-poor, nor pro-rich.
I
The (semi)elasticity perspective gives a complicated picture:
I
I
I
In extremely poor countries growth takes precedent. Period.
Reductions in inequality matter a lot for high-inequality
countries and have a double dividend (direct effects and
indirect effects through growth).
Very different picture depending on whether we target relative
changes in the headcount or changes in the percent of
population that is poor.
I
Fact 1:
I
I
Fact 2:
I
I
Economic growth tends to be distribution-neutral in developing
countries, in that inequality increases about as often as it
decreases in growing economies.
Measures of absolute poverty tend to fall with economic
growth in developing countries.
Fact 3:
I
The higher the initial level of inequality or the greater the
increase in inequality during the growth spell, the higher the
rate of growth that is needed to achieve any given
(proportionate) rate of poverty reduction.
What influences pro-poor growth?
Most of the variation in changes in poverty can be
attributed to growth in average incomes, suggesting that
policies and institutions that promote broad-based
growth should be central to the pro-poor growth agenda.
Most of the remainder of the variation in changes in
poverty is due to poverty-reducing patterns of growth in
relative incomes, rather than differences in the sensitivity
of poverty to growth in average incomes. Cross-country
evidence provides relatively little guidance as to the
policies and institutions that promote these other sources
of pro-poor growth. (Kraay 2006)
Conclusions on ‘pro-poor growth’
In the long-run:
I
I
I
Growth matters more than anything else for poverty reduction.
We need to understand the sources of sustained growth (and
absence of crises).
Can we have high long-run growth with high inequality?
In the short-run:
I
I
I
Growth is a blunt instrument. Reductions in inequality will
almost always help.
(Semi)elasticity perspective shows that effect of growth
depends on inequality.
Huge unused potential from true/relative pro-poor growth.
Short discussion: Tutorial prep - Are growth
policies really pro-poor?