Global Poverty Monitoring Technical Note 13
How Much Does Reducing Inequality
Matter for Global Poverty?
Christoph Lakner, Daniel Gerszon Mahler, Mario Negre, Espen Beer
Prydz
June 2020
Keywords: Global poverty, inequality, inclusive growth, COVID-19, SDGs,
forecasting, machine learning
Development Data Group
Development Research Group
Poverty and Equity Global Practice Group
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
Abstract
The goals of ending extreme poverty by 2030 and working towards a more equal distribution
of incomes are part of the United Nations’ Sustainable Development Goals. Using data from
166 countries comprising 97.5% of the world’s population, we simulate scenarios for global
poverty from 2019 to 2030 under various assumptions about growth and inequality. We use
different assumptions about growth incidence curves to model changes in inequality, and rely
on a machine-learning algorithm called model-based recursive partitioning to model how
growth in GDP is passed through to growth as observed in household surveys. When holding
within-country inequality unchanged and letting GDP per capita grow according to World Bank
forecasts and historically observed growth rates, our simulations suggest that the number of
extreme poor (living on less than $1.90/day) will remain above 600 million in 2030, resulting in
a global extreme poverty rate of 7.4%. If the Gini index in each country decreases by 1% per
year, the global poverty rate could reduce to around 6.3% in 2030, equivalent to 89 million fewer
people living in extreme poverty. Reducing each country’s Gini index by 1% per year has a
larger impact on global poverty than increasing each country’s annual growth 1 percentage
points above forecasts. We also study the impact of COVID-19 on poverty and find that the
pandemic may have driven around 60 million people into extreme poverty in 2020. If the virus
increased the Gini by 2% in all countries, then more than 90 million may have been driven into
extreme poverty in 2020.
All authors are with the World Bank. Negre is also affiliated with the German Development Institute.
Corresponding author: clakner@worldbank.org. The authors wish to thank R. Andrés Castañeda, Shaohua Chen,
Francisco Ferreira, La-Bhus Fah Jirasavetakul, Dean Joliffe, Aart Kraay, Peter Lanjouw, Christian Meyer, Prem
Sangraula, Umar Serajuddin, and Renos Vakis, as well as two anonymous referees for helpful comments and
suggestions. The findings and interpretations in this paper do not necessarily reflect the views of the World Bank,
its affiliated institutions, or its Executive Directors. We gratefully acknowledge financial support from the UK
government from the child trust fund, Better Data and Methods for Tracking Global Poverty, in TF No. 072496
(and EFO No. 1340 – Measuring Poverty in a Changing World) and through its Strategic Research Program
(TF018888). This working paper is a substantially revised and updated version of Lakner et al. (2014) and a revised
version of Lakner et al. (2019). This paper estimates that COVID-19 is pushing 60 million people into
extreme poverty using a machine-learning algorithm to determine the fraction of growth in GDP per
capita that is passed through to income and consumption observed in household surveys. When
assuming that growth in GDP per capita passes through to income and consumption observed in
household surveys at the same rate across all countries, we find that COVID-19 is pushing around 70
million people into extreme poverty as reported here: https://blogs.worldbank.org/opendata/updated-
estimates-impact-covid-19-global-poverty.
The Global Poverty Monitoring Technical Note Series publishes short papers that document methodological aspects of
the World Bank’s global poverty estimates. The papers carry the names of the authors and should be cited accordingly.
The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not
necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its
affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Global
Poverty Monitoring Technical Notes are available at http://iresearch.worldbank.org/PovcalNet/
1
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
1. Introduction
Over the past two and a half decades, global extreme poverty has decreased rapidly. Since 1990,
the share of the world population living below the extreme poverty line of $1.90 per day has fallen
from 35.6% in 1990 to 10.0% in 2015 (World Bank, 2018). Against this backdrop, international
development actors, bilateral development agencies and countries themselves, have united
around a goal of ‘ending’ extreme poverty by 2030. This goal has been defined as complete
eradication (United Nations, 2014) or as reducing global extreme poverty to 3% of the world’ s
population (World Bank, 2014). Several bilateral development agencies such as DFID and USAID
have also made such goals central to their focus and mission. At the same time, the development
policy debate is increasingly paying attention to the level of inequality in countries around the
world (International Monetary Fund, 2014; Ravallion, 2001; World Bank, 2016). As a result, the
internationally agreed Sustainable Development Goals (SDGs) include both a goal to end poverty
(SDG1) and a goal to reduce inequality within countries (SDG10).
We simulate global extreme poverty until 2030 under different scenarios about how inequality
and growth evolve in each country. This serves to quantify the importance of reducing
inequalities vis-à-vis increasing growth in achieving the goal of eradicating extreme poverty.
Although previous papers have simulated poverty up to 2030, we offer four distinct
contributions. First, we use micro data for 150 countries and grouped data for an additional 16
countries, allowing for an unprecedented data coverage of 97.5% of the world’s population.
Second, we model the impact of distributional changes on future trajectories of global poverty by
changing a country’s Gini index. The Gini index is arguably the most frequently used measure of
inequality, and it makes for an intuitive way of modeling distributional changes which has direct
policy relevance and conceptual simplicity. Third, since there are infinitely many ways in which
a change in Gini indices can occur, we use different growth incidence curves to capture how
inequality reductions may occur in an intuitive manner. Fourth, addressing the criticism that
economic growth in national accounts is increasingly disconnected from income and
consumption as observed in surveys (Ravallion, 2003; Deaton, 2005; Pinkovskiy & Sala-i-Martin,
2016), we utilize a novel machine-learning algorithm to estimate the share of economic growth
passed through to income or consumption observed in surveys.
Our simulations suggest that the global poverty rate will remain around 7.4% in 2030 if growth
is distribution-neutral and follows World Bank forecasts until 2021 and country-specific historical
growth averages from then on. Under a scenario in which the Gini index of each country
decreases by 1% per year, the global poverty rate falls to 6.3% -- equivalent to 89 million fewer
people living in extreme poverty. Reducing each country’s Gini index by 1% per year has a larger
2
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
impact on global poverty than increasing each country’s annual growth rate 1 percentage point
above World Bank forecasts. Even under the most optimistic scenarios we consider – where the
Gini decreases 2% annually and the annual growth rate exceeds World Bank forecasts and
historical averages by 2 percentage points – the poverty rate in Sub-Saharan Africa would remain
around 20% in 2030 and the global target of 3% would not be met.
We also study how COVID-19 affects projections of global poverty. Our baseline scenario
suggests that the pandemic has driven 60 million people into extreme poverty in 2020. If all
countries’ Gini indices increased by 2% in 2020 due to the pandemic, then 94 million will have
been driven into extreme poverty. This is a larger effect than if all countries growth forecasts are
2 percentage points lower than anticipated in which case the pandemic is expected to drive 82
million people into extreme poverty. Hence, our finding that percentage changes in the Gini
matter more than percentage point changes in growth carries over to the estimated COVID-19
impacts. This finding, namely that a 1% decline in the Gini index implies greater poverty
reduction than a 1 percentage point increase in growth, of course does not imply that it will be
easier to implement inequality-reducing than growth-enhancing policies in practice. Political
economy reasons might well stand in the way of sustained decreases in inequality.
We simulate all changes in Gini indices at the national level, not globally. A pro-poor
distributional change as simulated in this paper implies a fall in within-country inequality, but can
be expected to have a more muted effect on global inequality, for which between-country
differences matter greatly (Anand & Segal, 2008; Lakner & Milanovic, 2016). One challenge with
modeling the impact of changes in the Gini index on poverty, is that there are infinitely many
possible distributional changes resulting in the same change in the Gini index. If the change in
the Gini index comes from redistributing resources from the wealthiest 1% to the middle class,
poverty may remain unchanged in countries with moderate to low levels of poverty. If the change
comes from instituting a basic income to all households, then a similar change in the Gini may
eliminate poverty. Our baseline results are based on a linear growth incidence curve (GIC), but
in a robustness check we use a convex GIC, which gives higher growth rates to the lowest
percentiles compared to the linear version. With the convex functional form, a 1% annual decrease
in the Gini index in all countries has about the same impact on global poverty as a 2 percentage
point higher annual growth in each country. In other words, the convex GIC further highlights
the importance of reducing inequality for ending extreme poverty.
The literature has adopted several alternative approaches to model distributional changes in
simulating global poverty trajectories. Some authors have imposed distribution-neutral growth,
thus ignoring any future changes in within-country inequality (Birdsall et al., 2014; Karver et al.,
2012; Hellebrandt and Mauro, 2015). Others have projected distribution-neutral growth but
3
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
chosen initial distributions with different levels of inequality (Ravallion, 2013; Edward and
Sumner, 2014). Other studies simulate additional distributional changes by extrapolating the
trend in the Q5/Q1 ratio (Edward and Sumner, 2014; Hillebrand, 2008; Higgins and
Williamson,2002), the Palma ratio (Chandy et al., 2013), or the income share of the bottom 40%
(Ncube et al., 2014). A previous version of this paper used differences in growth rates of the
bottom 40% and the mean to project poverty towards 2030 (Lakner et al. 2014), similar to Hoy
and Samman (2015).
Two main approaches are used in the literature to project poverty forward which can produce
quite different results (Dhongde and Minoiu, 2013; Edward and Sumner, 2014). First, scenarios
based on historical survey growth rates (e.g. Yoshida et al., 2014). Second, scenarios derived from
national accounts either through growth models (Birdsall et al., 2014; Hillebrand, 2008), or
projecting historical or forecasted growth rates into the future (Karver et al., 2012, Chandy et al.
2013). We base our projections on forecasted growth rates from the World Bank’s Global
Economic Prospects (GEP) until 2021 (the last year for which growth forecasts are available at the
time of writing), and country-specific historical growth rates (from 2008-2018) to project forward
from 2022 to 2030. Our projections adjust for differences between growth from household survey
and national accounts.
Several other papers have estimated the impact of COVID-19 on global poverty. In particular,
Sumner et al. (2020) explore what happens if all countries’ growth rates decline a fixed amount
while Laborde et al. (2020) estimate the impact using a general equilibrium model. We estimate
the impact using household survey data and two vintages of growth projections for 166 countries,
allowing us to compare COVID-19 poverty projections with counterfactual projections from right
before COVID-19.
We model distributional changes and growth rates in GDP independently of each other.
Although the famous Kuznets Hypothesis (Kuznets, 1955) would predict that higher growth in
low-income countries would tend to increase inequality, the empirical support for this hypothesis
is weak. Ferreira and Ravallion (2009), for example, find no correlation between growth and
changes in inequality in the developing world.
The paper is structured as follows. Section 2 describes the conceptual framework for the
simulations, while Section 3 describes the data and our method for implementing the simulations.
Section 4 presents the results on global and regional poverty for different growth and inequality
scenarios, while Section 5 presents robustness checks by using different growth incidence curves,
poverty lines, poverty measures, and passthrough rates. Section 6 discusses the reasons why
poverty reduction might be slowing down while Section 7 concludes.
4
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
2. Conceptual Framework
In this paper, we model how changes in each country’s Gini index impact poverty towards 2030.
One challenge with modeling the impact of changes in the Gini index on poverty is that there are
infinitely many possible distributional changes resulting in the same change in the Gini index. To
conceptualize this, we use GICs, and in particular focus on two functional forms of the GIC.1 Let
������������ be the mean income of percentile group ������ (e.g. the bottom 1%) in the initial period. Final mean
income ������������ ∗ can be expressed as
������������ ∗ = ������������ (1 + ������������ ) (1)
where ������������ is the growth rate associated with this percentile group. We define the GIC as the plot
of ������������ against the percentile group (�����?������ ) in the initial period.
An intuitive and convenient way to allow the Gini to change is through a tax and transfer scheme
introduced by Kakwani (1993) and further discussed by Ferreira and Leite (2003). This scheme
involves an increase of everyone’s income at a rate ������ together with a tax and transfer scheme
which taxes everyone at a rate �����? and gives everyone an equal absolute transfer. As pointed out
by Ferreira and Leite (2003), this is a type of Lorenz-convex transformation. They show that the
transformed Lorenz curve is given by ������(�����?)∗ = ������(�����?) + �����?(�����? − ������(�����?)), where ������(�����?) is the original
Lorenz curve, which is a function of the percentile �����? , and ������(�����?)∗ is the post-transfer Lorenz curve.
This transformation can be obtained by moving every point on the Lorenz curve upwards by an
amount proportional to its vertical distance to the equidistribution (45-degree) line. The
transformed Gini index can be readily obtained as ������������������������(������)∗ = (1 − �����?)������������������������(������). In other words, the
tax rate imposed, �����?, is equivalent to the percentage change in Gini observed, ������ , such that �����? = −������ .
This direct link between the tax-and-transfer scheme and the change in the Gini makes it a
convenient way to model changes in the Gini. We can express the final incomes as a function of
the initial income, mean income, and changes in the Gini:
������������ ∗ = (1 + ������)[(1 − �����?)������������ + �����?������], (2)
where ������ is the mean income in the initial period. Using (2) and (1), it can be shown that the
corresponding GIC takes the following form:
1
������������ = (1 − �����?)(1 + ������) − 1 + [�����?(1 + ������)������] (3)
������������
1In Ravallion and Chen (2003), the GIC shows the growth rate of the income at a given percentile (e.g. the 10th
percentile) between the initial and final period. In contrast, we compute the growth rate in the mean of a particular
percentile group.
5
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
This GIC is a convex, decreasing function (when �����? > 0) along the percentile groups. It attributes
high growth rates at lower percentiles, while it becomes flatter at higher percentiles. It is
decreasing throughout, meaning that the growth rate will be lowest for the richest percentile
groups.
Another way of simulating a change in the Gini index uses a linear GIC. Such a GIC takes the
following form:
������������ = ������ − �����������?������ (4)
Substituting (4) into (1), we can obtain the following expression for the income of percentile group
������ in the final period
������������ ∗ = (1 + ������)������������ − ������������������ �����?������ (5)
This linear GIC can be obtained by taxing everyone in proportion to both their income and rank
– the poorest person is taxed at a rate of ������ and the tax increases proportionally with the rank –
combined with a transfer where every person receives a share ������ of their income. Unlike the
convex GIC, whose central parameter is directly related to percentage changes in the Gini index,
there is no functional relationship between the percentage change in the Gini index, ������ , and the
parameters of the linear GIC.
To illustrate how the convex and linear GICs could look like in practice, we use the welfare
distribution for Cote d’Ivoire from 2018 from PovcalNet. From 2018 to 2019 the World Bank’s
Global Economic Prospects (GEP) suggest that real GDP per capita in Cote d’Ivoire grew by 4.3%.
Figure 1 explores how this growth can be distributed if inequality stays unchanged or if the Gini
increases or decreases by 1% (ignoring for the moment whether only part of this growth is passed
through to the consumption observed in surveys). The initial Gini in Cote d’Ivoire was 41.5,
meaning that a 1% drop (������ = −0.01) would bring the Gini to 41.1, while a 1% increase (������ = 0.01)
would bring it to 41.9.
Lowering the Gini index by 1% does not have to impose a large cost (in terms of reduced growth)
on the top of the distribution. Because of the larger income share of the top of the distribution,
the reduction in the growth rate of the wealthiest individuals necessary to ensure that the bottom
grows substantially faster than the mean is relatively small. For example, in the case of Cote
d’Ivoire, a convex growth incidence such that the Gini decreases by 1% means that households at
the 10th percentile grow 2.5 pp faster than the mean, yet only reduces the growth at the 90th
percentile by 0.5 pp.
6
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
Figure 1: Different growth incidence curves compatible with same change in the Gini index
Note: Growth incidence curves (GICs) drawn using data for Cote d’Ivoire from PovcalNet for 2018 under different
assumptions about how much inequality changes, and in what manner inequality changes. The mean is assumed to
grow at 4.3%, according to data from the GEP.
In our baseline simulations we use linear GICs for three reasons: First, it is probably the simplest
realistic pro-poor GIC that can be constructed. Second, it constitutes a relatively conservative pro-
poor distributional change, in contrast to the convex GIC, which may provide a too optimistic
picture of how reducing inequality affects poverty. Finally, in contrast to the convex GIC, it can
easily be implemented for increasing Gini indices as well. A challenge with using convex GICs is
that certain large increases in the Gini can only be implemented if the poorest households attain
a negative income level. In those cases, the best solution may be to constrain the income levels to
be zero, implying that the Gini does not increase as much as desired.
Nonetheless, the convex GIC has some advantages: First, it intuitively relates to public policy, as
it represents the outcome of a simple tax and transfer scheme. Second, it is analytically related to
changes in the Gini index, allowing for a direct link with the measure of distributional change we
are looking at. In contrast, for the linear GIC, we are forced to use an algorithm that iteratively
changes the slope of the GIC until it matches the desired ������ . Third, it is directly linked to
differences in growth rates of the bottom 40% and the overall distribution, also called shared
7
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
prosperity, which is the first target of the SDG on inequality.2 For these reasons, we will use
convex GICs as a robustness check.
Figure 2: Empirically Observed Growth Incidence Curves
(a) Approximately linear GICs
(b) Approximately convex GICs
(c) GICs following other patterns
Note: Empirically observed growth incidence curves using the surveys in World Bank (2020) .
A worthwhile question to ask is whether these GICs are observed empirically. Using the World
Bank’s Global Shared Prosperity Database (World Bank, 2020), which provides a list of 90 recent
growth spells with a comparable welfare aggregate in household surveys that lie about 5 years
apart, we can explore how GICs for these countries look in practice. Figure 2 shows examples of
2The percentage change in the Gini index and the shared prosperity premium (the difference in growth of the bottom
������
40% and the mean, denoted ������) are related as follows: ������ = 0.4 , where ������40 is the income share of the bottom 40%.
(1+������)( −1)
������40
Hence, for a given income share of the bottom 40% and overall growth rate, there is a linear relationship between the
size of the tax rate, the percentage change in the Gini index, and the shared prosperity premium. For more details on
the formal relationship between the convex growth incidence curve and shared prosperity see the appendix of the
earlier version of this paper (Lakner et al. 2014).
8
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
GICs that look approximately linear and convex, and GICs that follow different shapes. Based on
these patterns, we believe there are enough empirical examples of the two types of GICs that we
will focus on in this paper to make them relevant.
An alternative to using a theoretically defined GIC would be to impose one that has been
observed in practice, e.g. for the same country, a best performer in the region, etc., as done in
World Bank (2015). Yet, this does not provide a sense of the magnitude of the distributional
change required, which our paper attempts to specify. It is also challenging for the many
countries that lack comparable data over time, preventing historical GICs to be created.
3. Data and Methodology
a. PovcalNet
To predict poverty in 2030, we rely on the surveys used in PovcalNet, which contains the World
Bank’s official country-level, regional, and global estimates of poverty.3 Most of the data in
PovcalNet comes from the Global Monitoring Database, which is the World Bank’s repository of
multitopic income and expenditure household surveys used to monitor global poverty.
PovcalNet contains more than 1900 surveys from 166 countries covering 97.5% of the world’s
population. The data available in PovcalNet are standardized as far as possible but differences
exist with regards to the method of data collection, and whether the welfare aggregate is based
on income or consumption. By relying on the PovcalNet database, we ensure consistency with
the official numbers used by the World Bank and United Nations for monitoring poverty,
inequality and related goals.
For 150 of the countries, housing 69% of the world’s population, micro data are available. For an
additional 8 economies (Australia, Canada, Germany, Israel, Japan, South Korea, Taiwan, and the
United States), or 9% of the world’s population, grouped data of 400 bins are available. For the
purposes of these projections, we treat the bins as microdata. Finally, for China and seven other
countries constituting about 19% of the world’s population, only decile or ventile shares and the
overall mean are available. Aside from China, this concerns Algeria, Guyana, Suriname,
Turkmenistan, Trinidad & Tobago, Venezuela, and the United Arab Emirates. For these countries,
we follow PovcalNet and fit a General Quadratic Lorenz curve and a Beta Lorenz curve, choosing
the one that gives the best fit, and use it to recover a full distribution.4
3 Data from PovcalNet can be accessed at http://iresearch.worldbank.org/PovcalNet/povOnDemand.aspx or directly
through Stata or R (Castaneda et al. 2019).
4 Shorrocks and Wan (2008) suggest that a lognormal functional form fits better. Minoiu and Reddy (2014) show that
for global poverty estimates a parametric Lorenz curve should be preferred to estimating kernel densities.
9
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
b. Growth scenarios
Our starting point in each country is the welfare distribution the World Bank uses to measure
poverty for the country in 2018, which is the latest year with poverty estimates at the time of
writing. These welfare distributions are based on (often) extrapolated distributions from
household surveys. The median year of data for these estimates is 2016, but the range spans from
1992 to 2018.5
To project poverty forward, a commonly used strategy is to rely on historical annualized growth
rates. COVID-19 makes this a quite unattractive option due to the high likelihood of an increase
in poverty in 2020. From 2018 to 2021, we therefore use the growth projections from the June 2020
edition of the World Bank’s Global Economic Prospects (GEP) to account for the impact of
COVID-19 on economic activity. 6 2021 is the last year for which growth projections are available.
Beyond that, one could use the annualized growth in the forecasting period, or the last growth
rate of the forecasting period to project forward towards 2030. COVID-19 makes this an
unattractive option as well due to extreme growth rates observed in both 2020 and 2021.
Therefore, beyond 2021 we use three different scenarios based on historical growth rates: that
each country grows according to its annualized growth rate from national accounts for the last 5,
10 or 20 years for which we have data (1998-2018, 2008-2018, 2013-2018). The simulations relying
on the 20-year historic growth rates may be optimistic, as Rodrik (2014) suggests that the rapid
growth experienced by emerging economies in recent decades is unlikely to persist indefinitely
and that convergence will slow down in coming decades.
Our preferred source of historical growth data is growth in real GDP per capita from national
accounts, as reported in the World Development Indicators (WDI). When such data are not
available for the whole period, we complement it with growth data used by PovcalNet for
monitoring global poverty. Most of the added sources are from the Maddison Project Database
(Prydz et al., 2019).
c. The relationship between growth in national accounts and surveys
A challenge with using growth rates in GDP per capita to project poverty forward is that prior
evidence has shown that only a fraction of growth observed in national accounts is passed
5 If countries do not have survey data for 2018, PovcalNet extrapolates their latest survey to 2018 using growth in GDP
per capita or Household Final Consumption Expenditure per capita assuming distribution-neutrality (Prydz et al.,
2019). The only country for which an extrapolated estimate is not available in 2018 is India. For India, we follow the
extrapolation approach used for the other countries to generate an estimate of the distribution in 2018.
6
For the economies not in GEP, we use growth forecasts from IMF’s World Economic Outlook. Syria does not have
growth projections towards 2021 in either of these sources. In this case, we use the regional average growth forecast of
the Middle East and North Africa region to project forward .
10
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
through to growth observed in household surveys (Ravallion, 2003; Deaton, 2005; Pinkonvskiy &
Sala-i-Martin, 2016). Estimating this fraction across our entire sample is fairly straightforward.
One would simply regress annualized growth in survey means on annualized growth in real GDP
per capita, under the constraint that the intercept is zero, ������������������������������������������ = ������ ∗ ������������������������/�����?�����������?������������������ + ������ , and use ������
as the fraction of growth in GDP per capita that is passed-through to welfare observed in surveys.
Using 1429 spells with comparable household survey data suggests that ������ = 0.85. Each spell
relies on two adjacent comparable surveys from the same country with welfare measured in the
same way, either income or consumption (World Bank, 2019).
Yet there is no reason to believe that ������ is constant across different contexts. It may differ by
geographical region, by income level, by whether income or consumption is used, over time, etc.
Although interactions for these additional covariates can easily be accommodated in the
equation, it is not clear which variables should be used to define the interactions and using all
possible interactions will likely overfit the data. Applying a selected number of interactions is
common practice in adjusting between household survey growth and national accounts growth
rates (see for example Birdsall et al., 2014; Chen and Ravallion, 2010; Chandy et al., 2013; and
Corral, 2020), but it is not entirely clear on what basis to select the variables to be included.
To circumvent this issue, we apply a machine learning algorithm, model-based recursive
partitioning, to determine when there is reason to believe that the passthrough rate varies in
different contexts (Zeileis et al. 2008). This algorithm can take as input all potential variables that
might matter for the passthrough rate. In our case, as input variables we use geographical region
(we use two versions, the official World Bank geographical regions, and the regions from
PovcalNet, where most high-income countries form a separate region), a dummy for whether
consumption or income is used, mean consumption, median consumption, the Gini index,
population, GDP/capita, and the year of the survey. The algorithm is a variant of classification
and regression trees, pioneered by Breiman et al. (1984), and works in the following manner:
1. Run the regression ������������������������������������������ = ������ ∗ ������������������������/�����?�����������?������������������ + ������ on all relevant data.
2. Add interactions between ������������������������/�����?�����������?������������������ and each of the input variables separately, and
conduct Wald tests indicating whether the interaction coefficient(s) are statistically
significant.
3. If the lowest p-value of these interaction coefficients (after adjusting for multiple
hypothesis testing) is less than 0.05, then the variable with the lowest p-value is chosen as
a splitting variable. If the lowest p-value is greater than 0.05, no split is made, and the
algorithm stops (suggesting that there is no evidence in favor of passthrough rates
differing by context).
4. Split the sample into two using the splitting variable. If the splitting variable is not binary,
meaning there is more than one way of splitting the sample into two, all possible splits
11
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
are tried out (respecting monotonicity for continuous and ordered variables), and the split
that results in the greatest rejection of equality of the passthrough rates is chosen, and the
sample is split into two. Splits are only made if at least 10 observations will be in each
subsample.
5. The algorithm is repeated from the beginning by applying it to observations in each of the
two subsamples separately.
Figure 3 and Table 1 show the results of model-based recursive partitioning using our data at
hand. There is significant evidence in favor of the data type mattering for passthrough rates.
Observations using income have a passthrough rate of 1.01, while observations using
consumption have a passthrough rate of 0.72. With a p-value of 0.041, we can reject that the
coefficient is identical for the two subgroups at a 5% level. For observations using consumption,
there is no variable which significantly yields different passthrough rates. For the observations
using incomes, the median matters for determining the passthrough rate. Cases with a median
less than 172 USD per person per month in 2011 PPPs (or 5.7 per day) have a passthrough rate of
2.11 while observations with a median above this threshold have a passthrough rate of 0.87, and
so forth. Table 1 contains more details on the Wald tests, the splits conducted, and the associated
passthrough rates. Two-thirds of all cases are predicted to have a passthrough rate between 0.72
and 0.86.
Figure 3: Decision tree of passthrough rates
Note: Results of using model-based recursive partitioning to determine when passthrough rates differ in various
contexts. The figure should be read from the top down. The circles show the variable for which passthrough rates differ
significantly and the p-value associated with the Wald test. The square boxes show the resulting regression plot and
the fitted line.
12
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
Table 1: Details on decision tree algorithm
p-values from Wald tests
Node Obs. ������ World Povcal
Data- Popula- Head-
Gini Median Mean GDP Bank Net Year
type tion count
region region
1 1429 0.85 0.04 0.54 1.00 0.86 0.08 0.98 0.91 0.99 0.97 0.99
2 540 0.72 ----- 1.00 0.96 0.92 0.52 1.00 1.00 0.99 0.79 0.99
3 889 1.01 ----- 0.27 0.00 0.14 0.59 1.00 0.59 0.54 0.98 0.02
4 84 2.11 ----- 1.00 0.84 1.00 0.94 0.00 0.00 0.87 1.00 0.98
5 805 0.87 ----- 0.02 0.82 0.71 0.91 1.00 0.70 0.63 0.81 0.06
6 298 0.46 ----- 0.86 1.00 1.00 1.00 1.00 0.90 1.00 0.93 0.36
7 507 1.08 ----- 0.11 1.00 1.00 0.37 0.18 0.00 0.63 0.70 1.00
8 87 1.51 ----- 0.99 1.00 1.00 1.00 ----- ----- 0.99 1.00 1.00
9 420 0.86 ----- 1.00 0.91 1.00 1.00 1.00 1.00 1.00 01.00 0.98
Note: The table shows the number of observations in each node ((sub)sample) of the tree, and the passthrough rate for
observations in each node. The columns to the right show the p-values (adjusted for multiple hypothesis testing) from
the tests exploring if passthrough rates vary by the variable in question in each particular node. Elements in bold show
the p-values that govern the splits in the tree. “---�? indicates that no test can be conducted since there is no variation in
the input variable in question in the particular subsample. In node 4 the region variables are significant but no splits
are made since the desired splits would leave less than 10 observations in one of the subsamples.
Using model-based recursive partitioning is only one machine learning method amongst many
that endogenizes the interactions to include. We find this method attractive because it is
specifically designed to test whether a parameter of interest differs by subgroups, it relies on
statistical tests, and it is easy to visualize. A shortcoming of this method is that its coarseness
means that small changes in the underlying data could change the predictions. In section 5.2 we
discuss our choice in more detail, show robustness checks using the lasso and a constant
passthrough rate across all observations, and compare the out-of-sample performance.
d. Inequality scenarios
We consider five different scenarios for changes in the Gini index; that it changes by -2%, -1%,
0%, 1% and 2% per year beginning in 2019. If a country starts with a Gini index of 0.40 in 2019
(which is close to the median Gini of the latest survey for each country), under our five different
scenarios, it would end up with a Gini of 0.32, 0.36, 0.40, 0.45 and 0.50 in 2030, respectively.
Evaluating the plausibility of these Gini changes requires comparable data across countries over
time. Utilizing the comparability database associated with PovcalNet (World Bank 2019), we can
recover 8,322 comparable spells. Figure 4 shows the annualized percentage change in the Gini
13
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
index from these spells, as a function of spell length, where each spell has been given a weight
equal to the inverse of the number of spells by country-spell length.7
Figure 4: Observed annualized changes in the Gini
Note: Distribution of observed annualized changes in the Gini index using the 8,322 comparable spells from the
surveys available in PovcalNet. Each spell is weighted by the inverse of the number of spells by country-spell length.
The figure reveals that Gini changes tend to be smaller the longer the spell length, suggesting that
large changes in the Gini are difficult to sustain over long periods of time. For spell lengths of 11
years, which are equivalent to the 2019-2030 spell length we look at in this paper, annual declines
of 1 percent per year are just below the 75th percentile while annual declines of 2 percent per year
are around the 95th percentile of the distribution of changes in the Gini index. Thus, both of these
seem plausible in a historical perspective. An annual increase in the Gini index of 1 percent is
around the 5th percentile and is therefore also plausible. Annualized increases of 2 percent,
however, have not been seen sustained over a 11-year period.
e. Estimating global poverty
Armed with growth rates, passthrough rates, and changes in the Gini index, using the linear or
convex growth incidence curve, we can project the welfare distribution in each country towards
2030. To project the distribution, we use the povsim simulation tool (Lakner et al. 2014).
7
For the passthrough rate analysis in the previous section, we recovered many fewer comparable spells (1429) since
we only looked at adjacent surveys for a particular country. Here we also consider surveys that are comparable even
if they are not adjacent (meaning other surveys were carried out in between). We are applying weights in order to get
a balanced sample of countries at each spell length, to the extent possible. Still, since most countries do not have
comparable surveys corresponding to all spell lengths, the set of countries at each spell length varies.
14
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
In order to derive global poverty rates, a few more pieces are needed. First, we need consumer
price indices (CPI) and purchasing power parity (PPP) exchange rates to convert the national
welfare aggregates into constant USD that have been adjusted for international price differences.
To that end, we rely on the data used by PovcalNet. Most CPIs are from the IMF’s International
Financial Statistics, while most PPP exchange rates are from the International Comparison
Program (PPPs for household final consumption expenditure).8 More details on the price data
used are available in Lakner et al. (2018) and Atamanov et al. (2018). Second, we need population
data to aggregate poverty estimates across regions and globally. We use country-level population
projections from the World Bank.9 Finally, to arrive at regional and global poverty rates, we also
need estimates for the 2.5% of the world for which we have no distributional data. In these cases,
we follow the aggregation method used by Chen and Ravallion (2010) and deployed by
PovcalNet, which assumes regional poverty rates for countries without a poverty estimate.
4. Results
This section presents the results from the simulations described above. First, we show poverty
nowcasts to 2020 in an attempt to quantify the impact of COVID-19 on global poverty, and the
relevance of assumptions about inequality and growth for quantifying this impact. Second, we
project poverty towards 2030, both at the global and regional level, and explore what would
happen if growth or inequality changes in a positive or negative direction. Unless otherwise
specified, we focus on the international poverty line at $1.90 per person per day in 2011 PPPs.10
a. Nowcasting poverty: The impact of COVID-19, growth and inequality
Figure 5 shows nowcasts of poverty for 2020 (as well as projections to 2021) utilizing the growth
forecasts from the June 2020 edition of the World Bank’s GEP. As the crisis is still unfolding at
the time of writing (June 2020), there is considerable uncertainty with regards to the growth
impact and the impact of the pandemic on within-country inequality. Therefore, Figure 5 also
displays scenarios where the growth forecasts differ by -2, -1, 1 or 2 percentage points (compared
to the GEP baseline) as well as scenarios where the Gini coefficient changes by -2, -1, 1 or 2 percent
(using a linear GIC). In order to quantify the impact of the virus on global poverty, we compare
these projections with the projections we would obtain using growth forecasts from the World
Bank’s GEP published in January 2020, which predates the global spread of COVID-19. Of course,
8 We use the original 2011 PPPs as published in December 2014. Revised 2011 PPPs were published in May 2020 but
they have not been adopted for global poverty monitoring at the time of writing. Atamanov et al. (2020) show that the
impact of the PPP revisions on the global poverty estimates is very small.
9 These are available at https://datacatalog.worldbank.org/dataset/population-estimates-and-projections.
10 See Ferreira et al. (2016) for a description of how the $1.90 international poverty line has been derived.
15
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
other factors may have also worsened (or improved) countries’ growth outlooks over these six
months but it is safe to say that most of the changes in the forecasts are due to COVID-19.
Figure 5: Impact of COVID-19 on global poverty under different growth and Gini scenarios
Note: Projected global poverty rate measured at $1.90 per person per day in 2011 PPPs assuming countries exceed or
fall behind the growth projections from the GEP by 1 or 2 percentage points (left panel), or follow the GEP projections
exactly but reduce/increase their Gini index by 1 or 2% (right panel). Inequality scenarios are based on linear growth
incidence curves.
With the new distribution-neutral forecasts, global poverty is projected to increase from 8.2% in
2019 to 8.7% in 2020, or from 630 million people to 675 million people. Compare this with the
projected decline from 8.0% to 7.7% over the same time period using the previous GEP forecasts.
The slight change from 8.2% to 8.0% for 2019 happens because the new vintage of the GEP also
revised 2019 growth rates for some countries. Taking this into account, it means that COVID-19
is driving a change in our 2020 estimate of the global poverty rate of about 0.8 percentage points
— (8.7%-8.2%)-(7.7%-8.0%). Another way to put this is that the estimates suggest that COVID-19
will push 60 million people into extreme poverty in 2020, or equivalently that the number of
extreme poor increases by 10%.11 This marks the first time since the East Asian Financial Crisis of
1997-1998 that the global poverty rate is increasing.
11 These results, like all results in this section, use model-based recursive partitioning to determine passthrough rates.
If one were to use the average passthrough rate observed historically (0.85, see section 5b), then the number of added
poor due to COVID-19 would be around 70 million. For more analysis on the impact of COVID-19 under this
assumption, see here: https://blogs.worldbank.org/opendata/updated-estimates-impact-covid-19-global-poverty.
When using IMF’s WEO GDP forecasts to estimate the impact of COVID-19 on poverty, together with an average
passthrough rate across all countries, we find that COVID-19 is pushing about 50 million people into extreme poverty,
see here for details: https://blogs.worldbank.org/opendata/impact-covid-19-coronavirus-global-poverty-why-sub-
saharan-africa-might-be-region-hardest
16
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
If growth in 2020 in all countries is two percentage points lower than GEP projections, COVID-
19 would increase global poverty by 1.1 percentage point and the number of poor would increase
by 82 million. If inequality increases by 2% in 2020 in all countries, then global poverty would
increase by 1.2 percentage points and the number of poor would increase by 94 million.12 The
latter scenario would imply that the global progress on ending extreme poverty would be set
back by three years. On the other hand, if all countries decrease their Gini coefficient by 2% in
2020, which could happen if countries successfully implement and expand social protection
programs, then the number of people pushed into extreme poverty due to COVID-19 would be
cut in half from the baseline, to around 30 million. Taken together, these results suggest that a
given percentage change in inequality matters more for global poverty than a similar percentage
point change in growth rates.
b. Global and Regional Trajectories to 2030
Turning to poverty projections to 2030, we are faced with the problem that growth forecasts end
in 2021. Figure 6 presents our simulated trajectories for the global poverty rate to 2030 for three
different distribution-neutral growth scenarios: that countries beyond 2021 follow their growth
patterns of the past 5, 10 or 20 years.
All scenarios put the global poverty rate in 2030 in the range of 7-8%. The scenario using historical
growth rates from 1998-2018 is slightly more optimistic for some regions due to the high growth
rates at the turn of the century. It is important to stress that these projections are not a prediction
of what poverty will look like in 2030. Rather, they represent a hypothetical scenario of what
would happen if all countries from 2022 onwards grow in accordance to what has occurred in the
past. In Latin America & the Caribbean, using growth rates from the past 5 years results largely
in a stagnation of poverty, while the other two scenarios decrease poverty substantially towards
2030. In the Middle East & North Africa, all scenarios yield increasing poverty rates towards 2030.
The global poverty rate is largely driven by Sub-Saharan Africa, which in all three scenarios has
poverty rates above 30% in 2030, while the other regions of the world have rates below 15% (the
vertical axes differ across regions in Figure 6).
Next, we look at how changing inequality or the growth rates impact global poverty. We focus
on the scenario that uses annualized growth rates from 2008-2018 beyond 2021, and simulate the
change in poverty if each country’s annual growth rate is 1 or 2 percentage points higher than the
historical rate. We use the 2008-2018 scenario as our baseline since it is the intermediate scenario
in terms of how optimistic it is for the future. In addition, we consider simulations if each
12
Earlier we showed that annual increases of 2% in the Gini index do not persist for long periods. Yet, such changes
are observed from year to year.
17
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
country’s Gini index decreases or increases by 1% or 2% per year using linear growth incidence
curves. Results are shown in Figure 7.
Figure 6: Global and regional distribution-neutral poverty projections to 2030
Note: Projected global and regional poverty rates measured at $1.90 per person per day in 2011 PPPs assuming
distribution-neutrality under three different growth scenarios (after 2021): countries follow their growth patterns of
the past 5 years, the past 10 years, or the past 20 years.
Figure 7: Simulations of global poverty under different growth and Gini scenarios
Note: Projected global poverty rate measured at $1.90 per person per day in 2011 PPPs assuming that countries exceed
or fall behind the growth projections from the GEP (until 2021) and 2008-2018 historical growth rates (from 2022) by 1
or 2 percentage points annually (left panel), or follow the growth rates exactly but reduce or increase their Gini index
by 1 or 2% annually (right panel).
18
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
Decreasing the Gini index by 1% annually in each country has a larger impact on poverty than
increasing growth 1 percentage point above forecasts, and in general the projections are quite
sensitive to changes in the Gini index. Under the same growth scenario, the global poverty rate
could be 6% or 9% with 1% annual decreases or increases in the Gini index, respectively. This
does not speak to whether reducing inequality every year in countries is politically feasible, only
that doing so would matter more for reducing global poverty than boosting growth rates (when
comparing percentage changes in the Gini with percentage point changes in growth rates).
Changes in the Gini index are particularly relevant for Sub-Saharan Africa, where the poverty
rate fluctuates between 30% and 40% with 1% annual decreases or increases in the Gini index
(Figure 8a). Due to rapid projected population growth, only the scenarios that lower inequality
are expected to decrease the number of poor in Sub-Saharan Africa (Figure 8b). Since the
inequality-reducing scenarios rapidly reduce poverty in other regions (with the exception of the
Middle East and North Africa), the share of the global poor that live in Sub-Saharan Africa
increases under these scenarios. Around 85% of the global poor would reside in Sub-Saharan
Africa by 2030 if all countries experience a fall in inequality.
Figure 8: Poverty Projections in Sub-Saharan Africa
Note: Projected poverty rates in Sub-Saharan Africa measured at $1.90 per person per day in 2011 PPPs assuming that
countries follow the growth projections from the GEP and 2008-2018 historic growth, under five different scenarios
about how inequality will change in each country.
The combinations of scenarios changing the Gini index and making the growth rate higher or
lower than GEP projections allows for the creation of iso-poverty curves. These curves,
introduced by Ferreira and Leite (2003), show combinations of inequality changes and growth
changes resulting in the same level of poverty, as shown in Figure 9. The flatness of the curves
illustrates the relative role of growth and inequality in shaping poverty rates.
19
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
Figure 9: Global and country-specific iso-poverty curves, 2030
Note: The figure shows different combinations of changes in the Gini index and in growth scenarios that result in the
same poverty rate globally and for four selected countries. The flatter the curves, the more growth matters relative to
reducing inequality.
At a global level, the curves are steeper than 45 degrees, suggesting that reducing the Gini index
by one percent with a linear GIC is more impactful than exceeding growth forecasts with a one
percentage point. This pattern varies greatly by country. For countries with low poverty rates,
the picture is mostly the same, and changing the Gini generally has a greater effect than exceeding
20
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
growth forecasts. For countries with high poverty rates, the opposite occurs. In these cases, where
the initial poverty rate may be above 50%, inequality-reducing growth might even increase the
poverty rate, as the ones on the margin of being poor will have resources transferred to the very
bottom of the distribution. In the Central African Republic, for example, in certain scenarios
increasing the Gini lowers poverty.
These conclusions are tied to the set-up we have explored. If we used higher poverty lines, other
countries would present a similar pattern to that seen in the Central African Republic.
Conversely, if we use measures of poverty that account for the depth and severity of poverty,
improving the conditions of the bottom of the distribution will unambiguously reduce poverty.
That inequality is more important than growth may also be influenced by our choice of growth
incidence curve. In the next section we will explore the robustness of the results to the choice of
alternative poverty lines, poverty measures, and GICs. We will also look at how sensitive our
projections are to our passthrough rate calculations.
5. Robustness checks
a. Poverty measure, poverty line and growth incidence curve
Our results thus far have used a linear GIC. This placed a limit on the simulated growth rates for
the poorest individuals. If a convex GIC is used instead, the bottom of the distribution experiences
large shifts in their welfare. To check the sensitivity of our results to our choice of GIC, we
implement the changes using a convex GIC as well. The resulting global iso-poverty curve is
shown in panel (b) of Figure 10. Compared to our original iso-poverty curve, reproduced in panel
(a), using a convex GIC increases the impact of Gini changes on poverty reduction, as shown by
the iso-poverty curves generally becoming steeper. Now a 1% annual reduction in the Gini
matters as much as exceeding growth forecasts by 2 percentage points annually, as both bring the
global poverty rate in 2030 to about 6%.
Next, we use different poverty measures. The headcount ratio, which all our results thus far were
based upon, is insensitive to the distributional differences among the poor, i.e. it does not value
how far below the poverty line the poor fall. Distributional differences among the poor may be
particularly important to consider in countries with high poverty rates, where an inequality-
reducing simulation may transfer resources from the marginally poor to the very poor. When
using poverty measures that account for the depth and severity of poverty, FGT1 and FGT2
(Foster et al., 1984), the iso-poverty curve become slightly steeper, meaning that changes in the
Gini have an even larger impact on poverty reduction, relative to higher growth (panel c and d).
21
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
Finally, we use higher poverty lines. Specifically, we use the poverty lines of $3.20 and $5.50,
which are official higher poverty lines of the World Bank (panel e and f). These lines are
constructed to reflect typical national poverty lines in lower- and upper-middle income countries,
respectively (Jolliffe & Prydz, 2016). With the $3.20 line, a 1% annual decline in the Gini still has
a larger impact on global poverty than exceeding growth forecasts by 1 percentage point per year,
while at the $5.50 line, they are about equally important for reducing poverty. This is despite the
fact that almost half of the world lived below $5.50 in 2015 (World Bank, 2018).
Figure 10: Global iso-poverty curves, 2030 under different assumptions
Note: The figure shows the global iso-poverty curve in 2030 under our baseline assumptions (panel a), and under five
different robustness checks. Panel b uses convex GICs rather than linear GICs (all other panels use linear GICs). Panel
c and d use different poverty measures, the poverty gap and the squared poverty gap, respectively, and panel e and f
use higher poverty lines, than the $1.90.
22
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
The impact of reducing the Gini index varies with the initial poverty rate, as well as the shape of
the GIC, the income distribution, and the measure of poverty. The impact of reducing the Gini by
1% is plotted against initial poverty levels in Figure 11 for linear GICs using three different
measures of poverty. The figures are drawn for the change in predicted poverty from 2019 to 2020
assuming zero growth to abstract from differences in growth rates across countries.
Figure 11: Impact of reducing Gini by 1%, by poverty measure and poverty level
Note: Figures show the one-year change in poverty measures assuming the Gini decreases by 1%, zero per capita growth
and a linear GIC.
The initial level of poverty matters for the impact of a fall in the Gini index on the poverty rate.
The relationship takes a U-shape where the reduction in the poverty rate at first increases with
the initial poverty rate, attains its maximum impact with poverty rates of about 40%, and then
decreases (panel a). For very high poverty rates, reducing the Gini increases poverty. Hence, there
may be a certain tradeoff between decreasing the poverty rate and decreasing inequality for very
poor countries. Panels b and c show that reducing inequality almost unambiguously decreases
both the poverty gap and the squared poverty gap even for high initial headcount ratios. This
indicates that the tradeoff is rather about maximizing the reduction in the headcount ratio or the
poverty gap – the latter corresponding to a stronger focus on the poorest of the poor.
b. Passthrough rates
Our results were based on using model-based recursive partitioning to estimate the fraction of
growth in GDP per capita that is passed through to growth in welfare observed in household
surveys. This method is only one method amongst many to estimate passthrough rates. We
23
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
employed this method because it is designed to test whether a parameter of interest – here the
passthrough rate – differs by subgroups. That said, other machine-learning methods could be
adapted to this setting as well by including interactions between GDP per capita and relevant
variables. In contrast to many other machine-learning approaches, model-based recursive
partitioning has the advantage of being quite transparent in the sense that it is a sequence of
statistical tests which can be visualized. Its coarseness on the other hand, implies that the fit might
not be optimal. Here we compare the results and predictive performance using another common
machine learning method, the lasso.13
For the right-hand side of the lasso regression we use the same variables as before, but we convert
the median, mean, GDP per capita, and population to log terms. This does not matter for model-
based recursive partitioning since the splitting point for any of these variables would have been
the same in log and non-log terms. Given that we are interested in the share of GDP per capita
passed through to consumption, we include these variables as interactions with GDP per capita.
Table 1 shows the results from the lasso, with ������ selected to minimize the out-of-sample mean-
squared error from a 10-fold cross validation, and one standard deviation less this optimal level.
Table 2 Output from lasso regression
�����������������?������ �����������������?������ ������������������������ 1 ������������
Variable Coefficient Coefficient
Growth in GDP/capita -0.1911 0.8506
∗ [�����������������������������������������?������ = �����������������?������������������] 0.3540 --
∗ ������������������������ (0-100) 0.0162 --
∗ ln (�����?�����������?������������������������������������������) (in millions) 0.0595 --
∗ ℎ�����������������������?������������������������ rate (0-100) 0.0004 --
∗ [������������������������ ������������������������ & �����������������?�����������? = ������������������] -0.0273 --
∗ [�����?������������������ℎ �����������������������������������?������ = ������������������] -0.0167 --
∗ [������������������������ℎ ������������������������ = ������������������] 0.0200 --
∗ [�����������������������������?������ ������������������ ������������������������������������������ ������������������������ = ������������������] 0.3114 --
∗ [������������������������������������ ������������������������ & �����?������������������ℎ �����������������������������?������ = ������������������] -0.2834 --
∗ [������������ℎ������������ ℎ������������ℎ �����������������?������������������ = ������������������] -0.0928 --
Note: �����������������?������ refers to the ������ that minimizes the out-of-sample error from a 10-fold
cross validation, while �����������������?������ ������������������������ 1 ������������ refers to one standard deviation less this
optimal level. The last three rows refer to PovcalNet regions while the three
other region rows refer to World Bank regions.
13
The lasso is a regular OLS regression but with an added penalty that the sum of the coefficients cannot exceed a
specified number. This helps to constrain the coefficients and reduce them to zero for the least informative variables,
thus assuring that only the most important variables are included in the model. The penalty size is governed by a
parameter, ������, which is often selected to minimize the out-of-sample error, or as one standard deviation less of the value
that minimizes the out-of-sample error in an attempt to err to the side of parsimony and lower the risk of overfitting
(Friedman, Hastie, and Tibshirani 2009).
24
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
When selecting the ������ that minimizes out-of-sample performance, 10 variables matter for the
passthrough rate. For example, with the optimal ������, the passthrough rate for a country that uses
consumption, with a Gini of 40, a population of 10 million, a poverty rate of 20%, located in South
Asia is predicted to have a passthrough rate of 0.62 (= −0.1911 + 0.0162 ∗ 40 + 0.0595 ∗ ln (10) +
0.0004 ∗ 20 + 0.0200). When being slightly conservative and choosing the ������ that is one standard
deviation less than this, no variables matter for determining the passthrough rate, which then
becomes 85% for all cases. We will refer to this as using a global passthrough rate below.
Compared with model-based recursive partitioning, both methods place a large emphasis on the
datatype and the Gini and give Europe and Central Asia high passthrough rates. Model-based
recursive partitioning makes use of the median, while the lasso uses the headcount, both of which
are likely contain much of the same information. Important differences lie in the emphasis on the
population size in the lasso, and the big negative coefficient on Middle East & North Africa in the
lasso, which is not (directly) picked up by the model-based recursive partitioning.
We can perform cross validation to compare how well the three methods perform out of sample.
Using the root mean square error, all three methods give an error between the predicted and
actual annualized growth rates in mean consumption of 0.067 and the mean absolute deviation
for all three methods is 0.040. The median absolute deviation is 0.025 for the global passthrough
rate and model-based recursive partitioning but 0.024 for the lasso, suggesting a slight advantage
to the lasso, but certainly not a difference that is statistically significant. The large difference
between the median absolute error and mean absolute error suggests that a lot of the difficulty
has to do with predicting the passthrough rate for spells with extreme growth rates. In general,
the out-of-sample errors combined with large differences between the two lasso models as well
as the fact that in the model-based recursive partitioning we were close to making no splits at all
(the first split has a p-value of 0.041) suggests that the noise to signal ratio is very large.14
The passthrough rate method matters at the regional level but less so at the global level. Figure
12 shows the global and regional projections using our benchmark passthrough rate method as
well as projections using the lasso and projections using a global passthrough rate (lasso with ������
one standard deviation less the optimal value). Using model-based recursive partitioning gives
slightly more pessimistic estimates at the global level with a forecasted global poverty rate of
7.4% in 2030 in contrast to 7.2% with the lasso model and the global passthrough rate.
These results are consistent with Castaneda Aguilar et al. (2019) who use more sophisticated machine learning
14
methods and many more features to try to predict changes in mean consumption. They find that just using growth in
GDP per capita performs nearly as well as selecting from hundreds of features.
25
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
Figure 12: Distribution-neutral poverty projections with different passthrough rate methods
Note: Projected global and regional poverty rates measured at $1.90 per person per day in 2011 PPPs under three
different scenarios about how growth in GDP per capita is passed through to welfare observed in household surveys.
6. Discussion
Our projections suggest that getting close to the 2030 goal of ending extreme poverty will be
unlikely. This contrasts with several of the scenarios in Ravallion (2013) that have global poverty
below 3% in 2030. When comparing our forecasts with the historical trend, the pessimism of our
results might seem counterintuitive: Global poverty has decreased by almost 1 percentage point
per year from 1981 to 2018, so how come we project a decrease in global poverty of about 1
percentage point in total from 2018 to 2030? If the historical trend continued linearly, one would
find that global poverty would reach 0% well before 2030 (Figure 13, based on Ravallion (2020)).
Some of the divergence is due to COVID-19, yet even if the historical trend only were to continue
from 2021, global poverty would still reach 0% by 2030. What explains this discrepancy?15
One answer can be found by looking at another simple forecast: Using the annualized global
15One possible answer to this question which we will not consider here relates to incomparable estimates of poverty.
Global poverty declined by around 4% as a result of China introducing imputed rent into its welfare aggregate (and
switched to an integrated nationwide survey) (World Bank, 2016), and similar strong declines are expected when India
goes from measuring consumption with a uniform reference period to a modified mixed recall period (World Bank
2018). If changes to welfare aggregates systematically increase welfare, which the two examples above suggest and
which would be the case if countries increase their capacity to include more elements in their consumption aggregate
over time as they get wealthier, then our forecasts will overestimate poverty reduction.
26
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
mean growth from household surveys from 1981 to 2018 and applying this to the global
distribution from 2018 onwards. Between 1981 to 2018, global mean consumption as observed in
household surveys increased by 1.4% per year.16 If we continue this towards 2030, we only get to
a global poverty rate of 6% -- far from the linear projection and close to our baseline projection
(Figure 13).
Figure 13: Alternative projections to 2030
Note: Historical global poverty rates from PovcalNet (where the PovcalNet lining up method has been used for years
where PovcalNet does not report a global poverty rate), and three projected global rates to 2030 assuming, (1) that the
historical trend continues linearly, (2) that the historical growth in global mean consumption as observed in household
surveys continues towards 2030, and (3) our baseline projections from earlier. Poverty line is $1.90 per person per day
in 2011 PPPs.
This suggests that a non-negligible share of the global population remain quite far from the $1.90
threshold. This is consistent with the finding from Ravallion (2016), that the consumption floor
has risen little over the past decades, or equivalently, that there has been little progress in
improving the welfare for the very poorest globally. Ravallion (2020) shows that historically it
has been difficult for countries who have made remarkable progress in terms of eliminating
poverty to go from a poverty rate of 3% to no poverty. All of this helps explains why reducing
inequality rather than speeding up growth might be the most efficient way of getting these people
to surpass the extreme poverty threshold.
Another (connected) reason for the deceleration of global poverty reduction is that the growth
rate among the poorest countries has declined over the past decades (World Bank 2018). This
16The mean consumption observed in household surveys is taken from the lined-up global distributions in PovcalNet.
When countries do not have surveys in 1981 and 2018, this lining-up relies on extrapolating distributions based on
growth from national accounts, so in practice the mean growth rate we recover is based on a mix of growth observed
in household surveys and growth observed in national accounts.
27
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
mostly boils down to growth in Sub-Saharan Africa being lower than growth was in China and
India when the latter countries accounted for the bulk of the world’s poor. With China and India
making up a smaller share of the world’s poor than in the past decades, progress needs to come
from other places in order for the speed of poverty reduction to continue. Current growth
forecasts do not suggest that this will happen.
In fact, when plotting country-level poverty rates in 2000 and 2015 (Figure 14a), the largest
changes come from countries ranked in the 10th to 60th percentile of the global distribution in terms
of $1.90 poverty. The population making up the poorest 10% of countries saw relatively small
declines in headcount rates over this time period. Particularly, countries in long-term conflict and
fragility have not managed to reduce poverty over the past decades (Corral et al. 2020). As our
projections assume that this pattern continues, and as the scope for countries in the middle of the
global distribution to further reduce poverty diminishes, this projects a leveling off in the speed
of reduction of global poverty rates.
Figure 14: The world population ordered by country-level poverty rates
(a) Poverty rates (b) Share of global poor
Note: The figure ranks the global population according to the poverty rate of their country of residence and plots this
against the poverty rates and the cumulative share of global poor.
This means that poverty is becoming more concentrated in a small number of countries (Figure
14b).17 In 2000, half of the world’s poor could be found in countries making up 25% of the global
population, while in 2015, half of the world’s poor could be found in countries making up 10% of
the world’s population. If current trends continue, this number could fall to 5% by 2030, and
countries making up 80% of the world’s population will be almost free of extreme poverty while
17 We are thankful to R.Andrés Castañeda for making this point very clear to us.
28
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
the remaining 20% will have poverty rates in high double digits.18 The inability of 80% of the
world’s population to contribute significantly to global poverty reduction and the projected lack
of progress in many of the countries in the bottom 20% explain why our forecasts suggest small
gains over the next decade.
7. Conclusion
Using a global database covering 97.5% of the world’s population, this paper shows that under
assumptions of distribution-neutral growth, the World Bank’s goal of achieving less than 3%
extreme poverty by 2030, as well as the Sustainable Development Goal of complete eradication
of poverty, will be difficult to reach by 2030. It also shows that these goals become more viable
by reducing inequalities. Conversely, regressive distributional changes can severely limit the way
in which growth contributes to poverty reduction.
Motivated by the Sustainable Development Goal 10 on inequality, we modeled inclusive growth
in terms of lowering the Gini index in every country. The poverty impact of more inclusive
growth defined in this way is different across countries and depends on the initial level of
poverty, the shape of the distribution, and the growth incidence curve used. At high levels of
initial poverty, reducing the Gini index could lead to a decrease in the pace of poverty reduction
in the short term compared with a distribution-neutral growth scenario. In other words, for a
country with a high headcount ratio, the welfare of the marginally poor may be growing slower
when lowering the Gini than in a distribution-neutral scenario. In such cases the poorest of the
poor still receive a growth premium and thus the poverty gap and severity are reduced.
One of the contributions of the paper is to quantify these effects using plausible distributional
changes. A 1% annual decline in each country’s Gini index is shown to have a bigger impact on
global poverty than if each country experiences 1 percentage point higher annual growth rates
than expected. Making growth more pro-poor as simulated in this paper does not impose a large
cost on the rest of the distribution. Because of the large income share of the top of the distribution,
the reduction in the growth rate of the wealthiest individuals necessary to ensure that the bottom
grows substantially faster than the mean is relatively small. In other words, the distributional
changes simulated in this paper are technically feasible and highlight that pro-poor growth is
crucial for reaching the poverty goals set by the global development community.
18As with all global poverty patterns, a lot of these patterns are driven by India and China, yet even when reproducing
the figures without these two large countries, a qualitatively similar pattern emerges.
29
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
8. References
Anand, S., and P. Segal. What Do We Know About Global Income Inequality? Journal of Economic Literature,
46(1), 57-94, 2008.
Atamanov, A., D. M. Jolliffe, C. Lakner and E. B. Prydz, “Purchasing Power Parities used in Global Poverty
Measurement“, World Bank Group Global Poverty Monitoring Technical Note, no. 5., September
2018.
Atamanov, A., Lakner, C., Mahler, D. G., Tetteh Baah, S. K., & Yang, J. The Effect of New PPP Estimates on
Global Poverty: A First Look. Global Poverty Monitoring Technical Note, 12. Washington, D.C.:
World Bank Group.
Ahluwalia, M. S., “Income Inequality: Some Dimensions of the Problem.�? In H. Chenery, M. S. Ahluwalia,
C. L. G. Bell, J. H. Duloy and R. Jolly (eds.), Redistribution with Growth, The World Bank, Oxford
University Press, 1974.
Birdsall, N., N. Lustig and C. J. Meyer, “The Strugglers: The New Poor in Latin America?�? World
Development, 60, 132-146, 2014.
Breiman, L., Friedman, J., Stone, C., and Olshen, R. (1984). Classification and Regression Trees. Taylor &
Francis, Belmont.
Castaneda Aguilar, R. A., C. Lakner, E. B. Prydz, J. A. Soler Lopez, R. Wu, and Q. Zhao. 2019. “Estimating
Global Poverty in Stata: The Povcalnet Command.�? Global Poverty Monitoring Technical Note; 9.
Washington, D.C.: World Bank Group.
Castaneda Aguilar, R. A., D. G. Mahler, and D. Newhouse. 2019. “Nowcasting Global Poverty.�? Paper
presented at the Special IARIW-World Bank Conference “New Approaches to Defining and
Measuring Poverty in a Growing World,�? Washington, DC, November 7–8, 2019.
Chandy, L., N. Ledlie, and V. Penciakova, “The Final Countdown: Prospects for Ending Extreme Poverty
by 2030,“ Global Views Policy Paper 2013-04, The Brookings Institution: Washington DC, 2013.
Chen, S. and M. Ravallion, “The Developing World is Poorer than We Thought, But No Less Successful in
the Fight Against Poverty, “The Quarterly Journal of Economics, 125(4), 1577-1625, 2010.
Corral, P., A. Irwin, N. Krishnan, D. G. Mahler, and T. Vishwanath. Fragility and Conflict: On the Front
Lines of the Fight against Poverty. Washington, DC: World Bank, 2020.
Deaton, Angus. “Measuring poverty in a growing world (or measuring growth in a poor world).�? Review
of Economics and Statistics 87(1), 1-19, 2005.
Dhongde, S. and C. Minoiu, “Global Poverty Estimates: A Sensitivity Analysis,�? World Development, 44, 1-
13, 2013.
Edward, P. and A. Sumner, “Estimating the Scale and Geography of Global Poverty Now and in the Future:
How Much Difference Do Method and Assumptions Make?“ World Development 58, 67-82, 2013.
Ferreira, F. and P. Leite, “Policy Options for Meeting the Millennium Development Goals in Brazil: Can
Micro-Simulations Help?“ Economia Journal of the Latin American and Caribbean Economic Association
0 (Spring 20), 235-280, 2003.
Foster, J., J. Greer and E. Thorbecke, “A Class of Decomposable Poverty Measures,�? Econometrica, 52(3),
485-97, 1984.
GMD (Global Monitoring Database), Global Solutions Group on Welfare Measurement and Capacity
Building, Poverty and Equity Global Practice, World Bank, Washington, DC.
Hastie, T., R. Tibshirani and J. Friedman. The Elements of Statistical Learning: Data Mining, Inference, and
Prediction. Springer Science & Business Media, 2009.
Hellebrandt, T. and P. Mauro, “The Future of Worldwide Income Distribution,�? Working Paper Series 15 -
7, Peterson Institute for International Economics, 2015.
30
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
Higgins, M. and J. G. Williamson “Explaining Inequality the World Round: Cohort Size, Kuznets Curves,
and Openness,�? Southeast Asian Studies 40(3), 2002.
Hoy, C. and E. Samman. “What If Growth Had Been as Good for the Poor as Everyone Else?�? Report (May),
Overseas Development Institute, London, 2015.
Hillebrand, E., “The Global Distribution of Income in 2050,�? World Development, 36(5), 727-40, 2008.
International Monetary Fund, “Fiscal Policy and Income Inequality,�? IMF Policy Paper. Washington, 2014.
Jolliffe, D. and E. Prydz., “Global Poverty Goals and Prices: How Purchasing Power Parity Matters,�? Policy
Research Working Paper Series 7256, The World Bank: Washington DC, 2015.
Jolliffe, D. and E. Prydz., “Estimating International Poverty Lines from Comparable National Thresholds.�?
The Journal of Economic Inequality 14 (2): 185–98. 2016.
Kakwani N., “Poverty and Economic Growth with Application to Côte d’Ivoire,�? Review of Income and
Wealth, 39(2), 121-139, 1993.
Karver, J., C. Kenny and A. Sumner, “MDGs 2.0: What Goals, Targets and Timeframe?,�? CGD Working
Paper, Center for Global Development: Washington DC, 2012.
Kraay, A. “When Is Growth Pro-Poor? Evidence from a Panel of Countries,�? Journal of Development
Economics, 80(1), 198-227, 2006.
Laborde, D., W. Martin, and R. Vos. "Poverty and food insecurity could grow dramatically as COVID-19
spreads." 2020.
Lakner, C., D. G. Mahler, M. C. Nguyen, J. P. Azevedo, S. Chen, D. M. Jolliffe, E. B. Prydz and P. Sangraula,
“Consumer Price Indices used in Global Poverty Measurement“, World Bank Group Global
Poverty Monitoring Technical Note, no. 4., 2018.
Lakner, C, and B. Milanovic. Global Income Distribution: From the Fall of the Berlin Wall to the Great
Recession. The World Bank Economic Review, 30(2), 203-232, 2015.
Lakner, C., M. Negre, and E. B. Prydz, “Twinning the Goals: How Can Promoting Shared Pros perity Help
to Reduce Global Poverty?,�? Policy Research Working Paper Series 7106, The World Bank, 2014.
Minoiu, C. and S. Reddy, “Kernel Density Estimation on Grouped Data: The Case of Poverty Assessment,�?
Journal of Economic Inequality, 12(2), 163-89, 2014.
Ncube, M., Z. Brixiova and Z. Bicaba, “Can Dreams Come True? Eliminating Extreme Poverty in Africa by
2030,�? IZA Discussion Paper Series No. 8120, 2014.
Palma, J. G., “Homogeneous Middles vs. Heterogeneous Tails, and the End of the ‘Inverted�?U’: It’s All
About the Share of the Rich,�? Development and Change, 42(1), 87-153, 2011.
Pinkovskiy, M., and X. Sala-i-Martin. “Lights, Camera… Income! Illuminating the national accounts-
household surveys debate.�? The Quarterly Journal of Economics 131(2), 579-631, 2016.
PovcalNet: The Online Tool for Poverty Measurement Developed by the Development Research Group of
the World Bank, http://iresearch.worldbank.org/PovcalNet.
Prydz, E. B., D. M. Jolliffe, C. Lakner, D. G. Mahler, P. Sangraula, �?National Accounts Data Used in Global
Poverty Measuremen,�? Global Poverty Monitoring Technical Note; no. 8. Washington, D.C. :
World Bank Group. 2019.
Ravallion, M., “Growth, Inequality and Poverty: Looking Beyond Averages,�? World Development, 29(11),
1803-1815, 2001.
Ravallion, M., “Measuring Aggregate Welfare in Developing Countries: How Well Do National Accounts
and Surveys Agree?�? The Review of Economics and Statistics, 85(3), 645-652, 2003.
Ravallion, M., “A Global Perspective on Poverty in India,�? Economic and Political Weekly, 43, 33-37, 2008.
Ravallion, M., “Do Poorer Countries have less Capacity for Redistribution?�? Journal of Globalization and
Development, De Gruyter, vol. 1(2), 1-31, 2009.
Ravallion, M., “How Long Will It Take to Lift One Billion People Out of Poverty?�? The World Bank Research
Observer 28(2), 2013.
Ravallion, M., “Are the world’s poorest being left behind?�? Journal of Economic Growth, 21, 139-164, 2016.
31
GLOBAL POVERTY MONITORING TECHNICAL NOTE 13
Ravallion, M. “SDG1: The Last Three Percent.�? Center for Global Development Working Paper No. 527. 2020.
Ravallion, M. and S. Chen, “Measuring Pro-Poor Growth,�? Economic Letters 78(1), 93-99, 2003.
Rodrik, D., “The Past, Present, and Future of Economic Growth,�? Challenge 57(3), 5-39, 2014.
Shorrocks, A. and G. Wan., “Ungrouping Income Distributions,�? Working paper 2008/16, UNUWIDER,
2008.
Sumner, A., C. Hoy, and E. Ortiz-Juarez. "Estimates of the Impact of COVID-19 on Global Poverty." UNU-
WIDER Working Paper 43. 2020.
United Nations, “Open Working Group proposal for Sustainable Development Goals,�? United Nati ons:
New York, 2014.
World Bank, “Prosperity for All/Ending Extreme Poverty: A Note for the World Bank Group Spring
Meetings 2014,�? Washington DC, 2014.
World Bank, “A Measured Approach to Ending Poverty and Boosting Shared Prosperity: Concepts, Data,
and the Twin Goals,�? Policy Research Report, The World Bank: Washington DC, 2015.
World Bank, “Poverty and Shared Prosperity 2016: Taking on Inequality,�? The World Bank: Washington,
DC, 2016.
World Bank, “Poverty and Shared Prosperity 2018: Piecing Together the Poverty Puzzle�?, The World Bank:
Washington DC, 2018.
World Bank. “Comparability Over Time at the Country Level for International Poverty Measures.�? Data
Catalog, Washington DC, 2019.
World Bank, 2020: Global Shared Prosperity Database. https://datacatalog.worldbank.org/dataset/global-
database-shared-prosperity. 2020.
Yoshida, N., H. Uematsu and C. E. Sobrado, “Is Extreme Poverty Going to End? An Analytical Framework
to Evaluate Progress in Ending Extreme Poverty,�? Policy Research Working Paper 6740, The
World Bank: Washington DC, 2014.
Zeileis, A., T. Hothorn, and K. Hornik. “Model-Based Recursive Partitioning.�? Journal of Computational and
Graphical Statistics 17(2), 492-514, 2008.
32