Public Disclosure Authorized

Policy Research Working Paper 9941

Public Disclosure Authorized

Assessing the Impact of the 2017 PPPs

on the International Poverty Line and Global Poverty
Dean Jolliffe
Daniel Gerszon Mahler
Christoph Lakner
Public Disclosure Authorized

Aziz Atamanov
Samuel Kofi Tetteh-Baah
Public Disclosure Authorized

Development Data Group &

Poverty and Equity Global Practice
February 2022
Policy Research Working Paper 9941

Abstract
Purchasing power parity exchange rates (PPPs) are used to updated IPL of $2.15, the global extreme poverty rate in
estimate the international poverty line (IPL) in a common 2017 falls from the previously estimated 9.3 to 9.1 percent,
currency and account for relative price differences across reducing the count of people who are poor by 15 million.
countries when measuring global poverty. This paper assesses This is a modest change compared with previous updates
the impact of the 2017 PPPs on the nominal value of the of PPP data. The paper also assesses the methodological
IPL and global poverty. The analysis indicates that updating stability between the 2011 and 2017 PPPs, scrutinizes large
the $1.90 IPL in 2011 PPP dollars to 2017 PPP dollars changes at the country level, and analyzes higher poverty
results in an IPL of approximately $2.15—a finding that lines with the 2017 PPPs.
is robust to various methods and assumptions. Based on an

This paper is a product of the Development Data Group, Development Economics and the Poverty and Equity Global
Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution
to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at
http://www.worldbank.org/prwp. The authors may be contacted at djolliffe@worldbank.org, dmahler@worldbank.org,
clakner@worldbank.org, aatamanov@worldbank.org, and stettehbaah@worldbank.org.

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. 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.

Produced by the Research Support Team

 Assessing the Impact of the 2017 PPPs on the International Poverty
Line and Global Poverty1

Dean Jolliffe ⓡ Daniel Gerszon Mahler ⓡ Christoph Lakner ⓡ

Aziz Atamanov ⓡ Samuel Kofi Tetteh-Baah

JEL code: I32 - Measurement and Analysis of Poverty

Keywords: 2017 PPPs, global poverty, the international poverty line

1
The author ordering was constructed through American Economic Association’s randomization tool (confirmation
code: R-TVbwJ99ttd). Corresponding author: Christoph Lakner (clakner@worldbank.org). All authors are with the
World Bank. We would like to thank the staff of the International Comparison Program Global Office at the World
Bank, particularly Nada Hamadeh, Marko Rissanen, William Vigil Oliver, Maurice Nsabimana and Mizuki Yamanaka
for providing the PPP data and metadata, patiently answering our technical questions, and for their feedback on earlier
drafts. We are thankful for helpful comments and guidance from Laurence Chandy, Stefan Dercon, Francisco Ferreira,
Deon Filmer, Haishan Fu, Ana Revenga, Carolina Sanchez-Paramo, and Umar Serajuddin. Many thanks are also due
to several World Bank poverty economists and members of the World Bank’s Global Poverty Working Group
(GPWG) for their feedback, as well as to Judy Yang who was involved in the earlier stages of this project. We would
also like to thank Angus Deaton, Bob Allen, and members of the International Comparison Program’s Technical
Advisory Group for comments. Finally, we would like to thank the audience at the ECINEQ 2021 conference. This
paper is an expanded version of Atamanov et al. (2020), in which we analyzed the impact of the revised 2011 PPPs
and 2017 PPPs on global poverty. In this paper, we expand the analysis of the impact of the 2017 PPPs on the
derivation of global poverty lines, analyze whether the ICP methodology has stabilized between the 2011 and 2017
ICP rounds, and examine country-level changes in greater detail. We gratefully acknowledge financial support from
the Knowledge for Change program and from the UK government through the Data and Evidence for Tackling
Extreme Poverty (DEEP) Research Programme. The findings, interpretations, and conclusions expressed in this paper
are entirely those of the authors. They do not necessarily represent the views of the World Bank and its affiliated
organizations, or those of the Executive Directors of the World Bank or the governments they represent.
1. Introduction

The ambition to eliminate extreme poverty around the world by 2030 is at the forefront of
development policy. It is the first of the Sustainable Development Goals (SDGs), as well as the
World Bank’s Twin Goals. Given this emphasis on poverty within the development community,
it is important to accurately measure the number of poor people in the world and know where they
live. Some of the most challenging issues regarding the estimation of global poverty concern the
construction of the international poverty line (IPL)—which is derived from the national poverty
lines of some of the poorest countries in the world (Ferreira et al. 2016; Ravallion et al. 2009)—
and the comparability of living standards across countries. The principal data that allow for the
comparability of national poverty lines and different countries’ living standards are purchasing
power parities (PPPs). For this reason, the use of PPPs for the measurement of global poverty, and
the revisions to PPPs over time, have been the subject of considerable debate (Deaton and Heston
2010; Ravallion 2014; Kakwani and Son 2016; Deaton and Aten 2017; Inklaar and Rao 2017).

In May 2020, the International Comparison Program (ICP) published new 2017 PPPs.2 The 2017
PPPs reflect the most recent relative price differences across a wide range of countries around the
world. Differences between PPP rounds may not only reflect changes in real cost-of-living
differences across countries over time but also ICP methodological changes across rounds. For
example, the 2005 ICP round overstated price levels in most countries in the developing world,
particularly owing to its approach of linking regional prices through “ring” countries (Deaton and
Aten 2017). This contributed to increasing the estimated number of extremely poor people in the
world by 400 million compared to the earlier 1993 PPPs (Chen and Ravallion 2010).

An important question is whether the new 2017 PPP revisions reflect new information on relative
price differences around the world or are driven by changes in ICP’s methodology. With the
background of large revisions in earlier ICP rounds being driven by methodological changes, the
Atkinson Commission on Global Poverty (World Bank 2017) raised concerns that adopting future
ICP rounds for global poverty monitoring might “shift the goalposts” and undermine the credibility
of the World Bank’s efforts in monitoring global poverty until 2030, the target date for the poverty
elimination goal. In its response to the Commission, the World Bank left open the possibility that
the 2017 PPPs would be adopted if the ICP methods and price data collection had stabilized (World
Bank 2016).

2
The ICP also published revised 2011 PPPs in May 2020. This was the first time the ICP revised PPP estimates it had
already published. The revisions to the 2011 PPPs reflect changes made to national accounts expenditures that are
used as weights to aggregate elementary PPPs. The revised 2011 PPPs had only a small effect on the global poverty
rates, relative to the original 2011 PPPs (see Atamanov et al. (2020) and Appendix A for more details). In the rest of
this paper, we use the revised 2011 PPPs as the baseline, since they are currently being used for the global poverty
measures reported by the World Bank.

2
With this background, we analyze the 2017 PPPs to make three contributions to the literature on
global poverty measurement. First, we provide empirical evidence that is consistent with stability
between the 2011 and 2017 rounds when methodological stability is understood at the global level
(see Deaton and Schreyer (2022) for further evidence to that end). Yet for certain countries, we
find that the 2017 PPPs yield large differences to estimated poverty. In several cases, this is
because of country-level improvements in the data collection.

Second, we derive global poverty lines with the 2017 PPPs. We follow Jolliffe and Prydz (2016)
by calculating the IPL as the median national poverty line of low-income countries, which yields
a line of $2.15 in 2017 PPP, compared with $1.90 in 2011 PPP. We show that this value is robust
to varying the set of countries and set of poverty lines, and hence reliably reflects the typical
standard by which the poorest countries of the world judge their citizens to be impoverished.

In keeping with Atkinson’s concerns about not shifting the goalpost, we show that the $2.15 IPL
in 2017 PPP largely keeps the global poverty rate constant in 2010, 2012, and 2015, compared to
the 2011 PPPs. These years contained the latest available global poverty estimates when the World
Bank’s goal of ending extreme poverty was set, when the United Nations agreed on the SDGs, and
when Atkinson was writing his report, respectively.

Beyond the IPL, we also compute with the 2017 PPPs the higher poverty lines the World Bank
uses to monitor poverty in countries with a low incidence of extreme poverty—$3.20 and $5.50 in
2011 PPPs. These higher poverty lines reflect the median national poverty line of lower-middle-
income countries and upper-middle-income countries, respectively. With the 2017 PPPs, we
compute these lines to be approximately $3.65 and $6.85 per person per day. We also update the
Bank’s societal poverty line (SPL) originally defined as max($1.90, $1 + 50% of median
consumption) in 2011 PPPs to max($2.15, $1.15 + 50% of median consumption) in 2017 PPPs.

Third, we analyze the impact on the global and regional poverty estimates of the World Bank of
using these global poverty lines and the 2017 PPPs. The 2017 PPPs would slightly increase
historical estimates of extreme poverty and slightly decrease extreme poverty since 2014,
compared with the 2011 PPPs. Extreme poverty would decrease marginally by 0.2 percentage
points (pp) in 2017 with the 2017 PPPs. Compared to past adoptions of new PPP rounds, this is a
minor change. The global count of extreme poor reduces by 15 million, which is largely driven by
34 million fewer poor people in Sub-Saharan Africa, while poverty increases slightly in all other
regions. However, Sub-Saharan Africa remains the region with the largest share of extreme poor
in 2017; from 62% with the revised 2011 PPPs to 58% with the 2017 PPPs. Nigeria accounts for
almost half of the reduction in extreme poverty in Sub-Saharan Africa.

There is also a limited change in global poverty with the revised poverty line of $3.65 in 2017
PPP, as declines in Sub-Saharan Africa and the Middle East & North Africa are offset by increases
in East Asia & the Pacific and Europe & Central Asia. The revision of the $5.50 line (2011 PPP)

3
to $6.85 (2017 PPP) increases poverty markedly globally by 4.3pp. This increase is seen in all
regions with about two-thirds of the added poor living in China, India, and Indonesia alone. With
the updated parameters of the SPL, about 17.6 million more people would be living in societal
poverty in the world in 2017. Societal poverty slightly increases in all regions, except Sub-Saharan
Africa and the Middle East & North Africa.

The rest of the paper is organized as follows. Section 2 describes the data we use. Section 3 assesses
the stability of the 2017 PPPs, relative to the 2005 and 2011 PPPs. Section 4 proposes updates to
the IPL, the higher global poverty lines, and the societal poverty line in terms of the 2017 PPPs.
Section 5 documents the changes observed in global, regional, and country-level poverty estimates
with the 2017 PPPs. Section 6 discusses the drivers of the changes observed in the global poverty
estimates as well as the global poverty lines. Section 7 concludes.

2. Data

We utilize different types of data that are typically used in estimating global poverty. The data can
be categorized into five different headings: (a) consumption or income distributions, (b) national
poverty lines and national poverty rates, (c) CPIs, (d) PPPs and market exchange rates, and (e)
GDP per capita.

(a) Consumption or income distributions: We use micro or grouped data from 1,971 income
distributions in PovcalNet, which is the World Bank’s database for global poverty estimates.3 For
most countries, the distributions use micro data from nationally representative household surveys.
For a few countries, including China, the distributions are constructed from grouped data (i.e.,
quantiles of the distribution such as percentiles). The distributions cover 168 economies that
represent over 97% of the world’s population. Some economies have more surveys than others
depending on how often surveys are conducted. There is an average of 10 surveys for all economies
spanning the period between 1991 and 2017, but more advanced economies are more likely to
have annual distributions. For economies without annual data, annual poverty estimates are
produced by extrapolating forward or backward from surveys, or interpolating between surveys,
using national accounts growth rates and assuming no distributional changes (Prydz et al. 2019).
While this is a strong assumption to impose, it performs relatively well (Mahler et al. 2021) and
has the advantage that the same set of countries can be compared over time.

(b) National poverty lines and poverty rates: We estimate global poverty lines using national
poverty lines and national poverty rates covering a wide range of countries. We obtain a total of
1,438 country-year observations of national poverty rates from the World Bank’s Poverty and

3
PovcalNet is available at http://iresearch.worldbank.org/PovcalNet/home.aspx. PovcalNet includes surveys that
measure consumption or income. For simplicity, we refer to them interchangeably as income distributions. This paper
is based on the June 2021 vintage of PovcalNet.

4
Equity Database and the Organization for Economic Development and Co-operation (OECD) (see
Table B.1 for more details). In addition, we use the national poverty lines of 15 poor countries that
Ferreira et al. (2016) estimated, which Ravallion et al. (2009) originally derived.

(c) Consumer price indices (CPIs): CPI data are required to adjust income distributions to obtain
a measure of household welfare in real terms. CPI data are also required to inflate national poverty
lines to the ICP reference year. We use the CPI data used in PovcalNet (Lakner et al. 2018). The
main source of CPI data is the International Monetary Fund’s International Financial Statistics
(IFS).4 For a few countries, including Bangladesh, Ghana, the Lao People’s Democratic Republic,
Malawi, and Tajikistan, PovcalNet uses inflation rates imputed from household survey data,
because they are deemed to be more reliable than CPIs from standard sources (Lakner et al. 2018).

(d) Purchasing power parity conversion factors (PPPs) and market exchange rates: When
estimating global poverty, both income distributions and national poverty lines are converted into
a common, internationally comparable currency unit using exchange rates. PPP conversion factors
are preferred to market exchange rates since market exchange rates only equilibrate the relative
prices of tradable goods across countries. PPP conversion factors incorporate the relative prices of
non-tradable services (e.g., getting a haircut) across countries, which tend to be less expensive in
developing countries, where labor costs tend to be lower (a phenomenon known as the Balassa-
Samuelson-Penn effect). PPP conversion factors are price indices that express the currency units
of one country in terms of the currency units of a reference country, typically the United States. In
other words, PPPs measure how much it costs to purchase a basket of goods and services in one
country compared to how much it costs to purchase the same basket of goods and services in the
United States. We obtain estimates of PPP and market exchange rates from the ICP. We use PPP
estimates from the 2005, 2011 (original and revised), and 2017 rounds. We use the PPPs for final
household consumption expenditure, which includes non-profit institutions serving households
(NPISHs).

For the 2011 PPP estimates, we do not use the official PPP estimates for six countries, namely the
Arab Republic of Egypt, Iraq, Jordan, Lao PDR, Myanmar, and the Republic of Yemen, following
Ferreira et al. (2016) and PovcalNet. There have been concerns over the official PPP estimates for
these countries due to large differences in the price changes implied by the 2005 and 2011 PPPs
and domestic CPI inflation (Ferreira et al. 2016; Atamanov et al. 2018).5 A seemingly unrelated
regressions (SUR) model—similar to what the ICP uses for countries without a PPP (so called
non-benchmark economies)—was then used to impute 2011 PPPs for these countries.6 We re-
estimate the imputed PPPs for these six countries, using revised 2011 PPPs and updated input data
(Atamanov et al. 2020) (see Table B.2 for more details). For 2017, we use the official 2017 PPPs

4
For this paper, we use monthly CPI series from the November 2020 IFS vintage averaged into annual CPI series.
5
We have verified this information with the revised 2011 PPPs (Atamanov et al. 2020).
6
Countries that collect prices to compute PPPs as part of the ICP exercise are referred to as benchmark countries. For
non-benchmark countries, there is no price collection and PPPs are imputed using a cross-country relationship.

5
for most economies, as published by the ICP. We use alternative 2017 PPPs for 14 exception
countries (see Section 3.B below and Table B.3 for more details).

Citing urban bias in the ICP price data collection, previous studies have imputed subnational
(rural/urban) PPPs for the three most populous countries in the developing world, namely China,
India, and Indonesia, when estimating global poverty (Chen and Ravallion 2008; 2010; Ferreira et
al. 2016). We follow these studies and impute rural and urban revised 2011 PPPs and 2017 PPPs
for these three countries, using the share of price collection outlets in urban areas, the ratio of urban
to rural poverty lines in the ICP reference year, and the national PPP as inputs (see Table B.4 for
more details).

(e) Gross domestic product (GDP) per capita: We use data on GDP per capita, denominated in
2011 PPP, 2017 PPP, and current USD, for the analyses done in this paper. We obtained these data
from the June 2021 vintage of the World Development Indicators (WDI).

3. Assessing the 2017 PPPs

A. Examining stability between the 2011 and 2017 PPPs at the global level

The ICP has estimated purchasing power parity (PPP) conversion factors for different reference
years—so far, 1970, 1973, 1975, 1980, 1985, 1993, 2005, 2011, and 2017. The ICP methodology
has evolved, resulting in improvements in the coverage and quality of PPP estimates.

Broadly speaking, there are five main components of the ICP methodology by which stability
between the 2011 and 2017 rounds can be assessed (more details are provided in Appendix C):
• First, the adoption of a common classification of expenditures across countries. For this,
the ICP relies on the final expenditures classification on GDP from the System of National
Accounts (SNA). While the 2005 and original 2011 rounds were based on the 1993 SNA,
the revised 2011 and 2017 rounds are based on the 2008 SNA. This means that there has
not been any drastic change in the classification of expenditures from the revised 2011
round to the 2017 round.
• Second, the selection of the basket of goods and services that will be priced. This basket is
based on the classification of expenditures mentioned above. Though the 2008 SNA were
used for both the 2011 and 2017 rounds, a few changes have been introduced into the ICP
expenditure classification at the basic-heading level.7 Since only aggregate PPPs (e.g.,
household final consumption or GDP) are relevant for our purposes, these changes at the

7
Basic headings in the ICP nomenclature refer to detailed expenditure categories containing similar item varieties.
For example, “Rice” basic heading contains several rice varieties. The changes at the basic-heading level are more
structural than substantive (e.g., “opening value of inventories” and “closing value of inventories” in the 2011 round
have been merged as “changes in inventories” in the 2017 round).

6
 sub-component level do not impact the PPPs we use, and hence this step does not influence
the stability of the 2011 and 2017 PPPs for poverty monitoring purposes.
• Third, the computation of the PPPs. While there was a major change in the method used
in linking prices within and between regions from the “ring” method in 2005 to the “Global
Core List” method in 2011, the computational steps followed to produce the final global
PPPs have remained unchanged since the 2011 round. The Country Product Dummy (CPD)
method is used to estimate PPPs at the basic-heading level, while the Gini, Eltetö and
Köves and Szulc (GEKS) method is used to estimate aggregated PPPs, which are both
transitive and base-country invariant.
• Fourth, the participating countries. The calculation of a country’s PPPs partially depends
on other countries’ PPPs. The number of participating countries declined from 199 to 176
between the 2011 and 2017 rounds, which is mostly explained by the dropping out of
several small Pacific Islands. However, since they were linked to the other participating
economies via three bridge-countries (i.e., selected countries which additionally collect
prices on a range of items from another region)—Australia (Eurostat-OECD), Fiji (Asia
and the Pacific), and New Zealand (Eurostat-OECD)—the inclusion of these islands did
not impact the global results.
• Fifth, the spatial coverage of ICP price collection. The spatial coverage may vary over
time within countries. In the 2005 round, for instance, price survey outlets were
predominantly in capital cities and urban areas (World Bank 2008). In India, 74% of prices
were collected from urban areas in the 2011 round, while 64% of prices were collected
from urban areas in the 2017 round. In OECD countries prices have been largely collected
from urban areas in both rounds. These differences do not necessarily imply overestimation
or underestimation of national prices, as the ICP requests countries to submit nationally
representative prices.8 Under the assumption that countries submit nationally
representative prices, changes to this component do not affect the stability of the 2011 and
2017 PPPs. The absence of documentation of what countries do makes it difficult to assess
whether this is indeed the case.

In addition to examining the ICP methodology by these five components, we can use the ratio of
actual to extrapolated PPPs to assess stability empirically (Deaton and Heston 2010; Deaton and
Aten 2017). PPPs are extrapolated from the past ICP rounds using a factor equal to domestic
inflation relative to US inflation, and compared to the actual PPP estimate. Deviations are expected
from such a comparison, because of conceptual and estimation differences between CPIs and PPPs.
Substantial, systematic deviations could, however, suggest issues with the PPPs, as shown with
previous rounds. For example, Deaton and Aten (2017) use this to show that the ICP overstated
prices in the 2005 round, especially in developing countries, which was reversed in the 2011 round.

8
National statistical offices apply temporal and spatial price indices to estimate nationally representative average
prices, although it is impossible to know whether these adjustments result in prices that are truly nationally
representative.

7
Figure 1 shows that the ratio of actual 2017 PPPs to 2017 PPPs extrapolated from the 2005 round
is systematically lower than 1, while the ratio of actual 2017 PPPs to 2017 PPPs extrapolated from
the revised 2011 round is more uniformly distributed around a mean of almost 1. The revised 2011
PPPs can better predict the 2017 PPPs. This is consistent with stability between the 2011 and 2017
PPPs, especially when compared with the 2005 PPPs.

Figure 1: Ratio of actual to extrapolated PPPs

Notes: This figure shows the ratio of actual to extrapolated PPPs as a function of GDP per capita. The actual PPPs are
household final consumption PPPs from the 2017 round. The extrapolations are based on official PPPs from the 2005
or revised 2011 rounds together with CPIs mainly from the IMF International Financial Statistics (IFS). Eurostat PPPs
have been used for OECD countries. Eurostat-OECD countries have a ratio of close to one, as they publish annual
PPP numbers, and therefore see only minimal revisions at new rounds of the ICP. Panel (A) shows 2005 PPPs
extrapolated to 2017. For the full sample of 143 observations, the ratio of actual to extrapolated PPPs has a mean of
0.822 with a standard deviation of 0.18. Panel (B) shows revised 2011 PPPs extrapolated to 2017. For the full sample
of 187 observations, the ratio of actual to extrapolated PPPs has a mean of 0.975 with a standard deviation of 0.11.
The difference in means between panels (A) and (B) is statistically significant.

In sum, we think that both when evaluating changes to the five components of how the PPPs are
calculated, and when evaluating the ratio of actual to extrapolated PPPs, the evidence speaks in
favor of the 2017 PPPs vis-à-vis the 2011 PPPs having stabilized at the global level.

B. Examining stability between the 2011 and 2017 PPPs at the country level

Though the ICP methodology may have stabilized and the changes between 2011 and 2017 are
smaller than earlier rounds, the changes in poverty at the country level could still be large. This
might result from different factors causing large changes in PPPs and/or CPIs. For example,

8
changes in PPPs can arise from: (i) geographical changes in ICP price collection (with an imperfect
adjustment to national average prices); or (ii) changes in the share of items in the ICP price survey
for which prices were actually collected (for example, Angola priced 18% of items in the 2011
round which almost doubled to 35% in the 2017 round). Furthermore, the 2017 ICP round has
benefited from technological advances that have reportedly improved data quality assurance across
all regions (World Bank 2020b). It is also important to note that the extrapolation of different PPPs
to a common year requires the use of national CPIs, which may have measurement issues, such as
following non-standard consumption classifications, or using expenditure weights and baskets that
are outdated or not nationally representative (Berry et al. 2019).

Ferreira et al. (2016) summarized the changes in PPPs and CPIs using the delta ratio. The delta
ratio is the ratio of the relative change in CPI between 2011 and 2017 to the relative change in PPP
between 2011 and 2017. It equals the ratio of the PPP-denominated mean income levels, and is
thus closely related to the change in poverty. 9 See Appendix E, section 9 for the derivation of the
delta ratio. The delta ratio is given by

𝐶𝑃𝐼2017 𝑃𝑃𝑃2017 𝑌𝑃𝑃𝑃2017

𝛿= ⁄ =
𝐶𝑃𝐼2011 𝑃𝑃𝑃2011 𝑌𝑃𝑃𝑃2011

where 𝐶𝑃𝐼2017 is the CPI in 2017, 𝐶𝑃𝐼2011 is the CPI in 2011, 𝑃𝑃𝑃2017 is the 2017 PPP, and
𝑃𝑃𝑃2011 is the (revised) 2011 PPP. 𝑌𝑃𝑃𝑃2017 is income in 2017 PPPs and 𝑌𝑃𝑃𝑃2011 is income in
(revised) 2011 PPPs.

Since the PPPs are expressed in the US dollar (i.e., the numeraire), for the United States the delta
ratio boils down to the relative change in CPI between 2011 and 2017 (i.e., 𝛿 = 1.0897 or an
inflation rate of 8.97%). For Angola, with a delta ratio of 1.80, the same bundle of goods and
services in Angola that would cost $1.00 in 2011 (converted by 2011 PPP into LCU) would cost
$1.80 in 2017 (converted by 2017 PPP into LCU). Thus, the Angolan Kwanza has appreciated in
PPP terms relative to the US dollar, and it is less expensive to buy the same bundle of goods and
services in Angola, hence poverty will reduce.

Large differences in the movements in PPPs and CPIs—outlier delta ratios—indicate large
changes in poverty when moving to the 2017 PPPs. Ferreira et al. (2016) identified countries with
delta ratios of two standard deviations from the mean as outliers. If the delta ratio for a country is
an outlier, we do not know if this is caused by the 2011 PPP, 2017 PPP and/or the CPI. Put
differently, a large change in the poverty rate between the revised 2011 and 2017 PPP does not
necessarily mean the 2017 PPP-based poverty rate is wrong and the 2011 PPP-based poverty rate
is more appropriate.

9
The delta ratio can also be expressed as the inverse of the actual to extrapolated PPPs multiplied by the US inflation
between the two rounds.

9
Even if the data collection and methodology stabilized between any two rounds, the delta ratio
would still deviate from the U.S inflation rate since CPIs and PPPs do not measure the same thing
(Krijnse Locker and Faerber 1984; Dalgaard and Sørensen 2002; McCarthy 2013).

Figure 2 shows the delta ratio between 2011 and 2017 PPPs and indicates the mean and cut-offs
outside of which a country is considered as an outlier. The mean delta ratio is 1.13, which
incidentally is also the relative change in the IPL, which we derive in a later part of the paper (i.e.,
$2.15 in 2017 PPP compared with $1.90 in 2011 PPP). This means that extreme poverty decreases
in countries with a delta ratio higher than 1.13 and increases in countries with a delta ratio less
than 1.13.

Figure 2: Real changes in PPP-adjusted dollars between 2011 and 2017 PPPs (delta ratio)

Notes: The horizontal axis is on a log scale. In contrast to Figure 1, this graph includes only benchmark countries in
both 2011 and 2017 ICP rounds. Panel (A) uses PPPs and CPIs currently used for global poverty monitoring, and
panel (B) uses official PPPs (from the ICP) and CPIs (from IFS). Official 2017 PPPs are used in both cases.

Panels (A) and (B) use slightly different PPPs and CPIs, which results in somewhat different
outliers to be identified. We use PPPs and CPIs for global poverty monitoring in panel (A), while

10
panel (B) uses those published by the ICP and IFS.10 The delta ratio is slightly higher on average
with greater dispersion in panel (A) than in panel (B). Taken together, there are 14 outlier
countries: Angola, Belize, Bolivia, Central African Republic, Egypt, Ghana, Guinea, Iraq, Jordan,
Liberia, Nigeria, São Tomé and Príncipe, Suriname, and Trinidad and Tobago.

As done with the 2011 PPPs (Ferreira et al. 2016), for these outlier countries we have conducted a
series of additional checks to decide whether to use the official PPPs or imputed PPPs for global
poverty monitoring.11 Apart from the delta criterion, we have also included a PPP residual criterion
to identify outlier countries. We define the PPP residual as the log difference between published
and predicted PPP (using the ICP’s model for non-benchmark country PPP estimation), and select
as an outlier any country whose residual is more than two standard deviations from the mean. The
residual criterion is added to address the concern that the delta criterion fails to unambiguously
identify whether there are issues with PPPs or CPIs. We select Egypt, Suriname, Central African
Republic, Djibouti, and Sudan based on the residual criterion (three of which were already selected
based on the delta criterion). We also include as outliers the countries whose revised 2011 PPPs
were imputed by Ferreira et al. (2016) (Egypt, Iraq, Jordan, Lao PDR, and Myanmar).

For all countries selected as outliers, we consider a range of indicators that could provide some
evidence on whether official or imputed 2017 PPPs are more appropriate for these countries. 12 To
examine whether the quality of CPIs rather than PPPs might be the reason why a country is an
outlier, we assess the CPI expenditure classification used, the reference year of the CPI weights,
and the spatial coverage of the expenditure data used in deriving CPI weights (Berry et al. 2019).
To examine the quality of the PPPs, we look at the changes in the share of items priced between
the 2011 and 2017 rounds.13 We also compare 2017-PPP-based poverty rates with correlates of
poverty that do not rely on PPPs, such as the multidimensional poverty index (MPI) from the
Oxford Poverty and Human Development Initiative (OPHI) (Alkire et al. 2020) and age
dependency ratio from the World Development Indicators (WDI). Last but not the least, we
incorporate the assessments of the country poverty economists—World Bank staff specialized in
the measurement of poverty in a country—who provide country-specific nuances. Appendix D has
the details of all the steps we followed to identify and deal with outlier countries.

10
The differences are explained by the revised 2011 PPPs which are imputed for exception countries (Egypt, Iraq,
Jordan, Lao PDR, and Myanmar) and survey-based CPIs used for some countries (such as Ghana). The Republic of
Yemen, which also has an imputed 2011 PPP, drops from the analysis since it is not included in the 2017 ICP.
11
Deaton (2001) similarly argues that with a new ICP round, the poverty estimates should be subjected to detailed,
local scrutiny.
12
We impute 2017 PPPs using the seemingly unrelated regressions (SUR) model the ICP uses to predict PPPs for
non-benchmark economies, while excluding from the regression those economies which are subjected to greater
scrutiny. With the revised 2011 PPPs, a slightly different model was used, which particularly addressed concerns in
the Middle East & North Africa region (e.g., by including a conflict dummy in the model) (Atamanov et al. 2018).
Since this region does not stand out this time around, we decided to use the ICP model without any adjustments.
13
Unfortunately, we cannot assess the changes in the geographical coverage of ICP price data collection between
rounds, as we lack (reliable) metadata from the 2011 round for all regions.

11
Given this analysis, we decided to deviate from the official PPPs for eight countries (Belize, Egypt,
Guinea, Iraq, Nigeria, São Tomé and Príncipe, Sudan, and Trinidad and Tobago). For these
countries, we use the geometric average of official and imputed 2017 PPPs. In the absence of a
definitive conclusion on which PPP is appropriate, we believe using the averages of the official
and imputed PPPs for these eight countries provides the most robust PPPs.14 We use extrapolated
2017 PPPs for a few other exception countries that do not have official 2017 PPPs, including
Kiribati, Nauru, the Syrian Arab Republic, Tuvalu, the República Bolivariana de Venezuela, and
the Republic of Yemen.15

We conduct sensitivity analysis to understand the extent to which the alternative PPPs influence
the IPL and extreme poverty estimates. The IPL remains robust to alternative PPPs and global and
regional trends are virtually unaffected by the alternative PPPs (see Figure D.4). Global extreme
poverty would decline by 0.4pp (instead of 0.2pp) in 2017 if the official 2017 PPPs were used
throughout (see Table D.7). This change is largely driven by Sub-Saharan Africa (particularly
Nigeria) and the Middle East & North Africa (particularly Egypt).

4. Derivation of global poverty lines

National poverty lines form the basis of the construction of the World Bank’s global poverty lines.
A national poverty line typically expresses the minimum amount of expenditure expected to cover
the basic needs of a person, usually including food, clothing, and other non-food items. By relying
on national poverty lines to guide the determination of global poverty lines, we rely on countries’
own judgments of what it means to be poor. National statistical offices are responsible for deriving
the national poverty lines in most low- and middle-income countries.16 National poverty lines are
expressed in domestic currencies, and PPPs are used to convert different currencies into a common,
comparable unit. PPPs are therefore instrumental in setting global poverty lines.

The IPL has historically been derived as a summary measure (e.g., mean or median) of the national
poverty lines of some of the poorest countries in the world. For example, the IPL of $1.25 per
person per day was the mean of PPP-adjusted national poverty lines of 15 of the poorest countries
in the world, as judged by countries’ household final consumption expenditure per capita around
2008 when the 2005 PPPs were released (Ravallion et al. 2009). When the 2011 PPPs were
released in 2014, the same 15 national poverty lines were used, but now converted to 2011 PPPs

14
Using the geometric average is similar to the ICP’s approach for interpolating between ICP rounds, where forward
and backward extrapolations are averaged (Inklaar and Rao 2020). Furthermore, invoking a Bayesian interpretation,
Chen and Ravallion (2010) take the geometric average between mean consumption in surveys and national accounts.
Recently, Ravallion suggested taking a weighted average of the revised 2011 and 2017 PPPs (Ravallion 2020).
15
We extrapolate from the revised 2011 PPPs used for global poverty monitoring, meaning that for the Republic of
Yemen we extrapolate from imputed revised 2011 PPP.
16
Low-capacity national statistical offices often do so in collaboration with the World Bank. In such cases, the goal
of the Bank is to assure that the national poverty lines reflect countries’ own normative judgments and are based on
the latest advances in poverty estimation.

12
yielding an IPL of $1.88, rounded to $1.90 (Ferreira et al. 2016; Ferreira et al. 2015). In this paper,
we maintain the principle that the IPL should be constructed as the typical poverty line of the
poorest countries in the world but, following Jolliffe and Prydz (2016), make four changes to the
methodology, addressing four issues with the current derivation of the IPL.

First, the national poverty lines of the 15 poor countries were reported in different units, e.g., some
are in adult equivalent terms and others in per capita terms, creating incomparability in the final
PPP-adjusted national poverty lines (Jolliffe and Prydz 2016). Second, the 15 countries were
selected based on data showing that, for countries with per capita household final consumption
expenditure at the level of these countries, national poverty lines were uncorrelated to per capita
household final consumption expenditure (Ravallion et al. 2009). But later evidence with more
data found that national poverty lines are positively correlated to per capita household final
consumption expenditure at all levels (Jolliffe and Prydz 2016). Hence, increasing the sample to
more than 15 countries might induce greater statistical support. Third, using national poverty lines
constructed primarily in the 1990s for setting global poverty lines for measuring poverty today
would rely heavily on CPI series over a long period of time from countries with typically weak
statistical capacity. This would make the IPL sensitive to revisions made to historical CPI data
from the 15 poor countries. Finally, using the mean of the national poverty lines might make the
final IPL vulnerable to outliers, compared to the median.

In an attempt to address these four issues, we follow the ‘harmonized poverty line’ approach of
Jolliffe and Prydz (2016) to derive the IPL in 2017 PPPs. Rather than converting national poverty
lines using PPPs and CPIs, the harmonized poverty line approach matches national poverty rates
with consumption/income distributions in PovcalNet, which are already expressed in the same
units; per capita PPP terms.17 For each country, the percentile of the PovcalNet distribution that
corresponds to the reported national poverty rate is found (in other words, we query the inverse
cumulative distribution function). This yields national poverty lines expressed in per capita PPP
terms for all countries. Jolliffe and Prydz (2016) selected surveys conducted as close as possible
to the 2011 ICP reference year, such that only one survey per country is used for deriving the IPL.
Here we will do the same with 2017 as the reference year, thus minimizing the reliance on CPI
series to move national poverty lines to the 2017 ICP reference year. We again follow Jolliffe and
Prydz (2016) and use all low-income countries rather than 15 countries, thus increasing statistical
support, and take the median poverty line instead of the mean to make the IPL robust to outliers.

In what follows, we describe the results of using this methodology to update the IPL. We show
that the IPL is robust to choosing a broader set of countries, a broader set of national poverty lines,
and using an equivalent poverty line—a line that retains the global poverty rate in a pre-specified

17
There are thus a practical and a conceptual reason for this approach. First, the national poverty rates are readily
available in WDI while the national poverty lines are not. Second, the national poverty lines may be in different units
and using the harmonized lines avoids this problem.

13
reference year (Kakwani and Son, 2016).18 We also update the two higher lines Jolliffe and Prydz
(2016) derived (see Section 4.A below for more details).

A. Global poverty lines based on harmonized national poverty lines

Jolliffe and Prydz (2016) estimate a median harmonized national poverty line of $1.91 in 2011
PPPs from 33 low-income countries. This finding provided evidence in support of the IPL of $1.90
per person per day. They also estimated $3.21 and $5.47 as the median harmonized national
poverty lines for 32 lower- and 32 upper-middle-income countries, respectively. These lines,
rounded to $3.20 and $5.50, have been adopted by the World Bank as additional global poverty
lines. Using the same approach to update the global poverty lines with the 2017 PPPs would ensure
consistency in the Bank’s methodology in monitoring global poverty. When we update the analysis
of Jolliffe and Prydz (2016) with the revised 2011 PPPs, the median lines become $1.85, $3.21,
and $5.65 (see Table 1).19 The World Bank decided to keep the global poverty lines unchanged
with the revised 2011 PPPs, including the $5.50 line (World Bank 2020a), and we follow that
approach here when we use the revised 2011 PPPs.

Table 1: Updating global poverty lines with harmonized national poverty lines
Income (A) Original 2011 PPP (B) Revised 2011 PPP (C) 2017 PPP
classification Median Mean N Median Mean N Median Mean N
Low 1.91 2.23 33 1.85 2.22 33 2.15 2.42 28
Lower-middle 3.21 3.88 32 3.21 3.89 32 3.63 3.95 54
Upper-middle 5.47 5.62 32 5.65 5.64 32 6.85 7.05 37
High income 21.70 21.19 29 21.70 21.31 29 24.36 23.36 38
Observations 126 126 157
Notes: In panels (A) and (B), we use the harmonized poverty lines Jolliffe and Prydz (2016) derived. Panel (A)
replicates the results of Jolliffe and Prydz (2016). Panel (B) uses revised 2011 PPPs to update the analysis. Panel (C)
is based on new survey data together with changes in PPPs to 2017 PPPs and changes in CPIs between 2011 and 2017
(see Appendix E, section 5). Countries are categorized according to the World Bank’s income classification in the
years the surveys were conducted. In panel (C), the line that is closest to 2017 is selected, one for each country. If
harmonized national poverty lines are available for 2016 and 2018 but not 2017, 2018 is selected.

For the 2017 PPPs, we use new harmonized poverty lines to estimate global poverty lines. We
select for each country one survey that was conducted in 2017 or the closest year. Eighty-four
percent of selected surveys were conducted after 2011 and 43% were conducted in 2017. More
than three-quarters of selected surveys were conducted within three years from 2017. Compared
with Jolliffe and Prydz (2016), new countries have been included in the analysis of the harmonized
national poverty lines (e.g., Algeria, Bangladesh, and Egypt), and some of the countries in the low-

18
There are alternative methods that could be used to derive global poverty lines than the ones we use. Global poverty
lines, for example, could be estimated using PPPs that equilibrate the least-cost diets across countries (Bai and Masters
2020; Allen 2017).
19
See Appendix E, section 3 for the derivations of the harmonized national poverty lines, section 4 for how these lines
are updated with the revised 2011 PPPs, and section 5 for the updating to the 2017 PPPs.

14
income category moved up to the lower-middle-income category (e.g., Ghana, Kenya, Pakistan,
São Tomé and Príncipe, and Tajikistan). We categorize countries into low, lower-middle, upper-
middle, and high-income countries based on the World Bank’s income classification in the years
the surveys were conducted.

Based on this approach, the median poverty line for low-income countries, or the international
poverty line (IPL), is estimated to be $2.15 (2017 PPP). The higher lines using data from lower-
middle-income countries and upper-middle-income countries are estimated to be $3.65 and $6.85,
respectively, in 2017 PPP. We use these lines with the 2017 PPPs in the rest of the paper. Figure
3 illustrates the global poverty lines graphically (updating a figure in Ferreira and Sánchez-Páramo
(2017)).

Figure 3: Global poverty lines using harmonized national poverty lines

Notes: This chart is based on the data summarized in Table 1, panel (C). The national poverty lines are the harmonized
national poverty lines, expressed in 2017 PPP per person per day. GDP is also given in 2017 PPP per person per day.
The global poverty lines are the median poverty lines for low-income countries (LIC), lower-middle-income countries
(LMIC), upper-middle-income countries (UMIC), and high-income countries (HIC), rounded to the nearest 5 cents.
Income classifications are determined using a slightly different measure—GNI per capita with the Atlas method (see
the footnote in the main text)—hence GDP per capita does not perfectly order countries according to their income
classification. There are fewer observations of GNI per capita, which is why GDP per capita is used here.

15
In line with one recommendation of the Atkinson Commission on Global Poverty—to express the
global poverty lines in local prices—in Appendix F we list the four global poverty lines with the
2011 PPPs and 2017 PPPs in 2020 local currency units for each country with available data.

B. Robustness check: Using different sets of poorest countries

One issue with the approach above is that it is not clear that the group of low-income countries is
the appropriate set of poor countries. Given the positive slope between national income and
national poverty lines, one faces a trade-off between including more countries in the calculation—
and hence increasing statistical support—and focusing on only the poorest countries. To check the
robustness of the $2.15 IPL derived above, we order countries by GDP per capita in 2017 PPP and
plot the cumulative median of harmonized poverty lines in 2017 PPPs, sequentially increasing the
pool of countries starting with the country with the lowest GDP per person in 2017 (see Table B.5
for details). Figure 4 shows that using anywhere between 11 and 41 poorest countries, the median
poverty line rounds to $2.15. This suggests that using the median national poverty line of low-
income countries is robust to using a broad range of poorest countries.

Figure 4: Cumulative mean and cumulative median of harmonized national poverty lines

Notes: The harmonized national poverty line is in 2017 PPP-adjusted dollars per person per day. The number of
countries indicates the count of countries ranked by GDP per capita in 2017 PPP-adjusted dollars from the poorest to
the richest. For example, 20 indicates the poorest 20 countries. This chart is based on the data summarized in Table 1,
panel (C) but the number of countries is capped at 100 to better illustrate the IPL.

16
 C. Robustness check: Using different sets of harmonized national poverty lines

The global poverty lines we derive are different from the $1.90, $3.20, and $5.50 lines because the
PPPs have changed, but also because of changes to national poverty lines, the set of countries with
national poverty lines, and the income classification of these countries. In Appendix B, we show
that when we use the same harmonized national poverty lines as Jolliffe and Prydz (2016), we still
get an estimated international poverty line of $2.15 (see Table B.6). When we pool together the
over 1,300 harmonized national poverty lines that we newly derived, we obtain an IPL of $2.16
(Table B.6). One might worry about using old harmonized national poverty lines given that the
statistical capacity of most countries is increasing over time and given that using old poverty lines
relies on many years of CPI data. If we restrict our sample to poverty lines at most 10 years old
with respect to the PPP year, we arrive at an IPLs $2.14 or $2.16 depending on whether we use the
pooled sample or only the poverty line closest to 2017 for each country (see Tables B.6 and B.7).

D. Robustness check: Using equivalent poverty lines

We use equivalent poverty lines, a concept introduced by Kakwani and Son (2016) at the country
level, as an alternative method to derive the IPL. The idea behind this method is to estimate the
IPL with the 2017 PPPs by choosing a line that retains the global poverty rate in a pre-specified
reference year. For example, the equivalent poverty line could be set to keep constant the global
poverty rate in 2010, the year that had the most recent global poverty estimate when the World
Bank’s goal of ending extreme poverty was set. This method is attractive in light of the Atkinson
Commission Report, which was concerned with future ICP rounds shifting the goalposts (World
Bank, 2017). With this approach, the parameter of choice is then the year in which poverty should
be held constant, for which several options appear reasonable. For example, this could be 2012,
which is the year with the most recent global poverty estimate when the UN agreed on the SDGs.
It could also be 2015, the year that had the most recent global poverty estimate when Atkinson
wrote his report. Finally, we include the 2017 reference year in the analysis, which is the year with
the latest global poverty numbers at the time of writing. See Appendix E, section 6 for the
derivation of equivalent poverty lines.

Table 2 presents the results of the analysis described above.20 The equivalent IPL that retains the
global poverty rate rounds to $2.15 (2017 PPP) regardless of which of the four reference years is

20
The poverty rates in Table 2 are based on the revised 2011 PPPs, even though the World Bank and UN goals to
eradicate extreme poverty were set with reference to the $1.25 IPL in 2005 PPP. We use the revised 2011 PPPs for
two reasons. First, the problems with the 2005 ICP round are well documented (Deaton and Aten 2017; Inklaar and
Rao 2017). Second, Atkinson wrote his report based on the $1.90 IPL in 2011 PPP with the acknowledgment that the
adoption of the 2011 ICP round might have already shifted the goalposts.

17
selected.21 Our preferred year for evaluating the equivalent IPL is 2017, which is both the latest
ICP reference year and the year with the latest global poverty data. The equivalent IPL for 2017 is
$2.173.
Table 2: Equivalent international poverty line, 2017 PPP
Region Year Headcount, % (Revised 2011 PPP) Equivalent IPL (2017 PPP)
World 2010 16.02 2.119
World 2012 12.89 2.128
World 2015 10.14 2.161
World 2017 9.27 2.173
Note: This table shows equivalent international poverty lines that keep global poverty headcounts constant in pre-
specified reference years.

E. Robustness check: Using the approach of Ravallion et al. (2009)

For completeness, let us also state that if we were to use the 15 national poverty lines selected by
Ravallion et al. (2009) and further applied by Ferreira et al. (2016), convert them to 2017 PPPs,
and take the average, we would end up with an IPL of $2.08 (see Table B.9. in Appendix B for
details). As argued above, this is not our preferred approach to estimate the IPL.

F. Updating the societal poverty line

Just as the value of the international poverty line and the higher lines reflecting typical assessments
of basic needs in middle-income countries are expressed in PPP dollar terms, so too is the World
Bank’s societal poverty line (SPL). The SPL was introduced by the World Bank (2018) in response
the Atkinson Commission on Global Poverty (2017, recommendation 16) recommendation to
report a measure of poverty that combines fixed and relative elements of poverty. The SPL,
estimated for each country separately, was initially parameterized as: max(IPL, $1 + 50% of
median consumption), with the fixed parameters (i.e., the IPL and the intercept) expressed in 2011
PPPs (Jolliffe and Prydz 2021). Of the two parameters in PPP terms, the IPL is updated to $2.15
in 2017 PPPs, leaving only the intercept of $1 in need of an update. We examine two
methodological approaches to consider how best to update this value. First, we follow an approach
similar to how we have updated the higher lines by replicating the methodology used to determine
these values except now using the 2017 PPPs and more recent data. As with the IPL and the higher
poverty lines, we consider different samples and approaches.

Following Jolliffe and Prydz (2021) we regress harmonized national poverty lines on median
values of consumption (or income, depending on the welfare aggregate used in the country).

21
Table B.8 reports a slight variant of the equivalent poverty lines. The estimated international poverty line in this
case is derived as the weighted sum of the equivalent regional poverty lines that hold constant regional poverty rates
in a particular reference year. The weights correspond to the regional shares of the global poor in the reference year.
This approach suggests an IPL between $2.16 and $2.24 with the 2017 PPPs.

18
Jolliffe and Prydz subsample by country type to try to exclude countries with relative poverty lines.
As a preferred alternative to this, we have collected metadata on type of poverty line from most
countries and subsample on all countries that have absolute poverty lines. (The original objective
of selecting SPL parameters was to demonstrate an empirical relationship between absolute
poverty lines and level of economic wellbeing in a country.) Table 3 shows the estimated intercept
and slope coefficients when considering three such specifications. We first consider a sample
(n=119) where we have one observation for each country with an absolute poverty line. For each
country, we select the observation which is closest in time to 2017 (and refer to this as our circa-
2017 sample). We then consider the subsample of all (n=815) country-year observations, allowing
multiple observations per country capturing some within-country variation over time in their
values. We use two different weighting schemes, where in both cases we ensure that all countries
are weighted equally, but in one specification, we weight more heavily the data points that are
closer in time to 2017 (see table notes for details). Across these three specifications, the intercept
ranges from 1.15 to 1.17 (in 2017 PPPs). While that variation is not substantial, it’s noteworthy
that these are very imprecisely estimated parameters with 95 percent confidence intervals that are
almost $1 in width (e.g., $1.15 ± $0.49 for the circa-2017 sample). For the circa-2017 sample, the
standard error is $0.25 but this drops only a few cents when considering sample sizes that are
nearly seven-fold larger.22

Table 3: Updating the societal poverty line with new survey data and 2017 PPPs
Description Intercept Slope N Delta ratio
One line per country – circa-2017 sample 1.15 0.47 119 1.15
(0.249) (0.043) (0.014)
Pool all lines – uniform weights 1.16 0.52 815
(0.213) (0.043)
Pool all lines – higher weights for lines closer to 2017 1.17 0.49 815
(0.235) (0.044)
Notes: Using the same model specification in Jolliffe and Prydz (2021) and the new data used in updating the absolute
poverty lines in this paper (Table 1, panel C), the parameters are updated with the 2017 PPPs. This table excludes all
national poverty lines that are defined as relative lines (which is typical of high-income countries). For the circa-2017
sample, every country with an absolute poverty line is included but only for the year that is closest to the 2017 ICP
benchmark year. In the larger samples (n=815), the weighting schemes differ. In one scheme, every observation has a
weight equal to the inverse of the total number of observations for each country, so that each country has a sum of
weights equal to 1. In the other scheme, we weight the observations to give greater importance to those close to 2017
and to give equal weight to each country: In the first step, we assign a survey conducted in 2017 a weight of 1, a
survey conducted in 2016 or 2018 a weight of 1/2, a survey conducted in 2015 or 2019 a weight of 1/3, and so on.
Next, we rescale the weights to ensure that the sum of weights for each country is again equal to 1. For example, if a
country has three harmonized poverty lines for 2016, 2017, and 2018, the corresponding weights will be 0.25, 0.5,
and 0.25. All parameter estimates are statistically significant at the 1% level. Standard errors of the estimates are in
parentheses.

The magnitude of the confidence intervals is concerning and leads us to consider a complementary
approach to updating the intercept. We use this complementary approach to assess the sensitivity

22
This relatively minor decline in standard errors is due to the significant intra-country correlation of the error term
(and correcting for this with the sandwich variance estimator).

19
of the estimated intercept. Part of the reason why the confidence intervals around the intercept are
so large is because the intercept is an out-of-sample prediction. The intercept essentially represents
a minimum cost of basic needs under which no person could survive, no matter how poor a
particular country is. No country provides a direct estimate of this value; we can only infer it if we
impose modelling assumptions. This is conceptually very different from how we approached
updating the IPL which reflects a median value over a large sample of national poverty lines and
is a well-supported, within-sample estimate. The alternative method considered is to update the
intercept based on the delta ratio, first discussed in Section 3.B. The delta ratio, unique to each
country, directly converts 2011 PPP values for each country into 2017 PPP values. For the
subsample of countries that use absolute poverty lines (i.e., those countries used in this analysis),
Table 3 reports an average value of the delta ratio of 1.15. If using this estimation approach to
update the intercept, the revised intercept would be $1.15 in 2017 PPPs.

Based on these findings, we assert that an appropriate updating of the SPL in 2017 PPPs is
parameterized as: max(IPL, $1.15 + 50% of median consumption). We base this inference on three
main points coming from the two approaches. The first is that both methods result in estimated
intercepts that fall within the narrow range of ($1.15, $1.17), with two of the four estimates being
$1.15. Second, while the confidence intervals from the estimates based on replicating the
methodology of Jolliffe-Prydz are massive, the confidence interval based on the delta method is
very tight ($1.15 ± $0.03) indicating significantly greater precision from this estimation approach.
And, finally, we note that all of the estimated intercepts, when rounded to the nearest five cents
(as we do for the other poverty lines), are equal to $1.15.

5. Global, regional, and country-level poverty estimates

In this section, we compare global, regional, and country-level poverty estimates over time with
the 2017 PPPs, relative to the revised 2011 PPPs. We analyze how using the 2017 PPPs would
change the incidence and geographic distribution of poverty at the IPL and at the higher global
poverty lines.

Figure 5 shows regional and global trends in extreme poverty between 1991 and 2017 with the two
sets of PPPs. Extreme poverty is measured as the share of the population living on less than $1.90
or $2.15 a day expressed in revised 2011 PPP or 2017 PPP, respectively. At the global level, the
revised 2011 PPPs and 2017 PPPs induce relatively small changes to extreme poverty.

Between 1991 to 2017, extreme poverty in the world falls from 36.05% to 9.27% with the revised
2011 PPPs, and from 37.46% to 9.07% with the 2017 PPPs. The 2017 PPPs slightly increase
historical estimates of extreme poverty and slightly decrease extreme poverty since 2014. Regional
trends in extreme poverty are similar for both revised 2011 and 2017 PPPs. There is progress
against poverty across regions, except for the Middle East & North Africa where conflict and
fragility reverse the progress made (World Bank 2020a).

20
 Figure 5: Global and regional trends in extreme poverty

Note: This figure shows the trends in extreme poverty at the global and regional levels. The international poverty lines
used are $1.90/day (revised 2011 PPP) and $2.15/day (2017 PPP).

Table 4 reports the changes in global and regional poverty levels in 2017, the most recent year.
Overall, extreme poverty in the world decreases marginally by 0.2pp with the 2017 PPPs. Extreme
poverty more than doubles in Europe & Central Asia (but from a low level), and slightly increases
in all other regions, except Sub-Saharan Africa. Extreme poverty reduces substantially in Sub-
Saharan Africa (34 million fewer poor people), driving down the global count of the extreme poor
by 15 million. Sub-Saharan Africa still has the largest share of millions of extreme poor, though it
falls from 62% with the revised 2011 PPPs to 58% with the 2017 PPPs.

To understand the regional trends further, Table 5 shows the ten countries with the largest absolute
changes in extreme poverty.23 The change observed in Europe & Central Asia is driven by
Uzbekistan, where poverty increases by 22pp (equivalent to 7 million more poor people).24 About

23
We use the 2017 estimates for all countries. As explained above, when there is no survey in 2017, these numbers
are estimated using growth rates from national accounts and assuming no distributional changes.
24
This is the first time the ICP provides a PPP estimate for Uzbekistan based on actual prices. The price data collection
was based on an experimental approach, meaning that prices were collected for only actual individual consumption
and individual consumption expenditure by households, but not for all the components included in GDP. The original
and revised 2011 PPPs were imputed by the ICP like other non-benchmark countries. The 2017 PPP estimate is
deemed more credible, but potentially not comparable to estimates from previous rounds. This might explain why
Uzbekistan has large changes in poverty.

21
half of the change in extreme poverty in Sub-Saharan Africa is driven by Nigeria, where poverty
falls by 8pp (equivalent to 15 million fewer poor people).

Table 4: Changes in extreme poverty in 2017 at the global and regional levels
Poverty Poverty Change in
Change in Change in
Region rate, % rate, % millions of
poverty, pp poverty, %
(2011 PPP) (2017 PPP) poor
World 9.27 9.07 -0.20 -2.21 -15.36
East Asia & Pacific 1.41 1.82 0.41 28.98 8.45
Europe & Central Asia 1.30 2.77 1.48 113.93 7.26
Latin America & Caribbean 3.77 4.07 0.31 8.15 1.93
Middle East & North Africa 6.34 6.36 0.02 0.36 0.09
Rest of the World 0.68 0.68 0.00 0.12 0.01
South Asia 9.65 9.71 0.06 0.58 1.01
Sub-Saharan Africa 41.18 37.94 -3.25 -7.88 -34.10
Note: This table compares extreme poverty between the revised 2011 PPPs and 2017 PPPs at the global and regional
levels.

Table 5: Changes in extreme poverty in 2017 at the country level, top 10 countries
Change Change Change
Poverty rate,
Poverty rate, in in in
Country Region % (2011
% (2017 PPP) poverty, poverty, millions
PPP)
pp % of poor
Nigeria SSA 41.36 33.41 -7.96 -19.23 -15.19
Uzbekistan ECA 12.70 34.35 21.65 170.53 7.01
Congo, Dem.
SSA 72.65 64.42 -8.24 -11.34 -6.70
Rep.
Indonesia EAP 4.46 6.55 2.09 47.01 5.54
Angola SSA 45.07 26.50 -18.57 -41.20 -5.54
Kenya SSA 35.11 27.50 -7.61 -21.67 -3.82
Ethiopia SSA 24.66 21.16 -3.50 -14.18 -3.72
Ghana SSA 12.34 24.68 12.33 99.91 3.59
Pakistan SAS 4.02 5.44 1.42 35.35 2.95
China EAP 0.35 0.55 0.19 55.05 2.69
Notes: This table compares extreme poverty between the revised 2011 PPPs and 2017 PPPs at the country level. The
top 10 countries with the largest absolute changes in extreme poverty are shown in this table. Countries are ranked in
descending order of absolute changes in millions of poor. When a country does not have a survey in 2017, recent
surveys are extrapolated or interpolated using growth rates in national accounts data, particularly GDP or household
final consumption expenditure (HFCE).

22
 Figure 6: Global and regional trends in poverty at $3.20

Note: This figure shows the trends in poverty at the global and regional levels at the following poverty lines: $3.20/day
(revised 2011 PPP) and $3.65/day (2017 PPP).

Figure 7: Global and regional trends in poverty at $5.50

23
Notes: This figure shows the trends in poverty at the global and regional levels at the highest global poverty line. The
lines used are $5.50/day (revised 2011 PPP) and $6.85/day (2017 PPP).

Changes in poverty at the global level are also relatively small when the $3.65 line is used, as
regional changes offset each other (see Figure 6). Poverty increases in East Asia & the Pacific, and
Europe & Central Asia, and Latin America & the Caribbean, while poverty decreases in the Middle
East & North Africa and Sub-Saharan Africa. In 2017, global poverty changes by 0.6pp, or 43
million more poor people, with the 2017 PPPs. The largest changes in millions of poor in 2017 are
observed in East Asia & the Pacific (43 million more poor people), and Sub-Saharan Africa (24
million fewer poor people). The change in East Asia & the Pacific is mainly driven by China and
Indonesia, while the change in Sub-Saharan Africa is mainly driven by Nigeria. See Tables B.10
and B.11 for a full set of results.

Using a $6.85 line (2017 PPP) relative to a $5.50 line (2011 PPP) would lead to increases in
poverty markedly in all regions, including Sub-Saharan Africa (see Figure 7). With the 2017 PPPs,
about 321 million more people in the world would be considered poor in 2017 by the standards of
upper-middle-income countries. About 129 million of them live in China and 64% of them live in
China, India, and Indonesia alone. See Tables B.12 and B.13 for a full set of results.

Figure 8: Global and regional trends in societal poverty

Notes: This figure shows the trends in societal poverty at the global and regional levels. With the revised 2011 PPP,
the societal poverty line (SPL) used is given as max($1.90, $1.00 + 0.5 * median consumption/income). With the 2017
PPP, the societal poverty line (SPL) used is given as max($2.15, $1.15 + 0.5 * median consumption/income).

24
If the societal poverty line were strongly relative (without any fixed or absolute elements), as in
the OECD where the poverty line is 60% of the median, revisions to PPP data would not change
global and regional poverty trends. The global welfare distribution would change by the same
factor as the societal poverty line. Unlike the OECD poverty line, the SPL has two terms expressed
in PPPs (i.e., the IPL and the intercept) whose revision drives the changes observed in societal
poverty trends when moving to the 2017 PPPs (see Figure 8). In fact, the changes in societal
poverty trends with the 2017 PPPs are small and similar to the changes observed in extreme
poverty with the 2017 PPPs. Societal poverty slightly increases globally and, in all regions, except
Sub-Saharan Africa and the Middle East & North Africa. About 17.6 million more people would
be living in societal poverty in the world in 2017 when moving from the revised 2011 to 2017
PPPs. A few large countries drive this relatively small change, including China and Indonesia from
East Asia & the Pacific, Nigeria and the Democratic Republic of Congo from Sub-Saharan Africa,
India and Pakistan from South Asia, Iraq and the Arab Republic of Egypt from the Middle East &
North Africa, and Uzbekistan from Europe and Central Asia (see Tables B.14 and B.15).

6. Discussion

A. Poverty in poorer versus richer countries

Compared to the 2011 PPPs, the 2017 PPPs imply lower price levels in relatively poor countries
and higher price levels in relatively rich countries. In fact, in 2017, the share of the world’s
population living in poverty in low-income countries systematically decreases while the share of
the world’s poor living in upper-middle-income countries systematically increases at all three
global poverty lines (see Figure B.1). The largest change occurs at the highest line, where the share
of people in upper-middle-income countries considered as poor by the standards of these countries
increases from 15% to 19%. Some of this change is driven by upward revisions in the national
poverty lines of these countries (see Section 6.C below for more details).

We can make sense of the global changes in poverty and the shifting shares of poverty across
income groups by investigating (1) the changes in poverty at the country level, and (2) the
aggregation of country-level poverty estimates. The global pattern we observe could be explained
by the former if the real changes in welfare when moving to the 2017 PPPs (i.e., the delta ratio
discussed in Section 3.B) are systematically higher for low-income countries than for middle-
income countries. As the IPL/LIC line increases by 13% from $1.90 to $2.15, extreme poverty
increases in countries whose delta ratios fall short of 1.13 and decreases in countries whose delta
ratios exceed 1.13. Figure B.2 shows that the average delta ratio by income group is decreasing in
income, and the delta ratio for low-income countries on average exceeds 1.13. These results
suggest that low-income countries are slightly richer with the 2017 PPPs, hence slightly shifting
the concentration of extreme poverty away from the poorest countries.

25
The global pattern we observe could also be driven by the fact that populous countries carry a
larger weight when aggregating country-level poverty estimates. Figure B.2 also plots delta ratios
that are population weighted. For low-income countries, the delta ratio is even higher when
population weighted, while the reverse holds for upper-middle-income countries. This suggests
that the observation that poverty is shifting to relatively rich countries is partly driven by a few
populous countries in the world. With their extreme poverty levels decreasing with the 2017 PPPs,
Ethiopia and the Democratic Republic of Congo are top populous countries driving the change in
extreme poverty in low-income countries. At the LMIC line, the changes in poverty in the populous
countries offset each other; for example, the poverty counts increase in Indonesia by 13 million
and decrease in Nigeria by 14 million. At the highest line, poverty increases in China, Brazil, the
Russian Federation, and Mexico with the 2017 PPPs, driving down the delta ratio for upper-
middle-income countries markedly when population weighted (see Figure B.2).

B. Extreme poverty in Sub-Saharan Africa

This sub-section presents additional analysis to better understand the changes in Sub-Saharan
Africa, which shows the largest changes in extreme poverty as a result of the 2017 PPPs. Since
Sub-Saharan Africa accounts for nearly two-thirds of the global extreme poor, it is not surprising
that the region drives the global changes. At the same time, it is important to understand if there
are any systematic issues with the new (or old) round of PPPs in the region.

On the whole, the 2017 ICP round has benefited from improved data quality assurance across all
regions (World Bank 2020b). We investigate the quality of PPPs by comparing the residuals
between published and predicted estimates of price level indices (PLIs) across regions. We predict
PLIs from country-level characteristics (e.g., GDP per capita, export and import shares, and
dependency ratio), following the official ICP model used to estimate PPPs for non-benchmark
economies (for the model specification and more details, see Appendix D). Sub-Saharan Africa
does not have systematically different residuals (Figure B.3), suggesting that there is no evidence
of a worse PPP quality in the region.

Additionally, we consider the quality of CPIs for Sub-Saharan African countries relative to other
regions. We draw on metadata covering 196 countries to assess the quality and coverage of official
CPI data used in the global poverty estimates (Berry et al. 2019). We identify and analyze three
relevant indicators, including the CPI expenditure classification, the reference year for the weights,
and the spatial coverage of the expenditure weights (also see Section 3.B and Appendix D). Most
countries in the world (77%) follow the standard Classification of Individual Consumption
According to Purpose (COICOP) classification, with the Sub-Saharan Africa region not being very
different (71%). On average, the CPI weights reference year for countries in Sub-Saharan Africa

26
is 2010.96, which is close to the world’s average of 2011.85. CPI expenditure weights with
national coverage in Sub-Saharan Africa (73%) is also similar to the world (79%).25

In summary, it does not appear that the substantial reduction in extreme poverty in Sub-Saharan
Africa is related to PPP or CPI quality issues. While there are obviously changes in rankings within
the region, African countries are actually quite evenly split between increasing and decreasing
poverty: Extreme poverty is lower with the 2017 PPPs in 24 countries and higher in 21 countries,
with most countries being quite stable (see Figure B.4). However, most of the large countries see
a decline with the 2017 PPPs. In particular, Nigeria, the Democratic Republic of Congo, Angola,
Kenya, and Ethiopia are driving the regional result. Nigeria alone accounts for about half of the
change in the millions of extremely poor people in the region.

C. Drivers of changes in the global poverty lines

In this sub-section, we try to understand the drivers of the revisions to the global poverty lines.26
We pay particular attention to the large change at the highest line from $5.50 to $6.85.

The observed changes in the global poverty lines are not only due to the change in PPPs but also
three other factors: (a) the national poverty lines used, (b) income group classifications, and (c) the
set of countries with national poverty lines and welfare aggregates available. The global poverty
lines are obviously also a function of the method used to derive them. Though we have changed
the method by which we derived the IPL—we now use the harmonized national poverty line
approach to update the IPL, similar to what is used for the higher lines— we can abstract from this
methodological change since this approach also resulted in a $1.90 poverty line with the 2011
PPPs (Jolliffe and Prydz 2016).

When deriving the IPL we already showed that it was robust to using the exact same poverty lines
as Jolliffe and Prydz (2016). In Table B.14 we try to gauge whether any of the other three factors
are driving changes to the poverty lines. To isolate the various factors, we update the harmonized
national poverty lines previously used by Jolliffe and Prydz (2016) with the 2017 PPPs while
keeping everything else constant, and sequentially account for each of the three factors above.27
First we revisit our result of maintaining the exact same sample of national poverty lines but update
the PPPs from the 2011 round to the 2017 round. As we argued in Section 4.A, this generates an
IPL of $2.15, meaning that our revision to the IPL can entirely be explained by the new PPPs,
regardless of the other factors. For the poverty line typical of lower-middle-income countries, we
arrive at a poverty line of $3.68, quite close to the final one when accounting for all changes

25
These differences are not statistically significant. These indicators are available in Appendix D, Table D.2 for the
outlier countries identified in Section 3 of the paper.
26
We are thankful to Francisco Ferreira and Benoît Decerf for encouraging us to delve deeper into this point.
27
This decomposition is path dependent, meaning that the order in which we account for the various factors matters.
Our results remain qualitatively the same regardless of which order we use.

27
($3.63), meaning that nearly all of the updates to that line can be explained by the changes to PPPs.
For the line typical of upper-middle-income countries, revising only the PPPs brings the line to
$6.32, quite far from the final line ($6.85) meaning that the PPP changes cannot explain all of the
increase of that line.

Next, we account for the fact that the countries Jolliffe and Prydz (2016) used for their analysis
might since have published more recent national poverty lines. This has small impacts on the IPL
and the lower-middle-income line, but increases the upper-middle-income line notably, from $6.32
to $7.15. Thus, the relatively high increase in the upper-middle-income line is partially driven by
real upward shifts in the national poverty lines of upper-middle-income countries. Part of this can
be explained by some of these countries now being high-income countries, but even when
removing those countries, the line still sits at $6.95 (see Table B.16 for more details). This might
be because poverty in upper-middle-income countries is more likely to reflect the concept of
relative poverty, which changes over time with consumption patterns and incomes, rather than
prices, which are more relevant for absolute poverty (Smeeding 2016). The fact that we now have
national poverty lines and welfare aggregates for more countries than Jolliffe and Prydz (2016)
only has a minimal impact on the value of the global lines.

Even for countries whose national poverty lines are based on concepts of absolute poverty, we
know that as countries get wealthier, they tend to increase the real value of their national poverty
line (Ravallion 1998; Jolliffe and Prydz 2016). Between the release of the 2011 PPPs and 2017
PPPs, on average countries got wealthier. This means that we would expect the national poverty
lines close to 2017 to be higher in real terms than the national poverty lines close to 2011. How is
this consistent with our finding that the real value of the IPL and lower-middle-income line is
purely driven by PPPs and not by changes to the real value of national poverty lines? The answer
lies in the use of income groups to define the global poverty lines. Suppose a poor country grew
between the release of the 2011 PPPs and 2017 PPPs and as a consequence adopted a higher
national poverty line. Suppose further that the country graduated from being a low-income country
to a lower-middle income country as a result of this growth. Then it is not obvious that the increase
in the real value of their national poverty line will increase the median national poverty line of
low-income countries or lower-middle income countries. As long as the median income level
within each income group has not changed between 2011 and 2017, then—even if countries’
national poverty lines increase as they grow—it does not need to be the case that the global poverty
lines we derive on expectation increase in real value.

28
7. Conclusion

In this paper, we analyze the impact of the 2017 PPPs on three important issues related to global
poverty monitoring.

First, we analyze the stability of the 2017 PPPs relative to the 2011 PPPs, the previous round of
PPPs. The adoption of previous ICP rounds has resulted in large revisions to global poverty
estimates, especially the 2005 round that added 400 million more people to the estimated number
of extremely poor people in the developing world (Chen and Ravallion 2010). Against this
background, the Atkinson Commission on global poverty argued that future ICP rounds should
not be adopted until 2030, the target date for the SDGs (World Bank 2017, also see Deaton 2001
and Klasen et al. 2016 for a similar argument made earlier). We show in this paper that the ICP
methodology has remained fairly stable between the 2011 and 2017 rounds. While changes at the
country level could still be important, we find no evidence of substantial broad-based changes
which characterized earlier rounds and motivated the Atkinson recommendation.

Second, we estimate global poverty lines with the 2017 PPPs. The international poverty line (IPL)
we derive to measure extreme poverty is $2.15 per person per day in 2017 PPP. We employ the
harmonized national poverty lines technique (Jolliffe and Prydz 2016) to arrive at this line, and
show that it is consistent with an equivalent poverty lines approach (Kakwani and Son 2016). As
such, the IPL we derive keeps the global poverty rate largely at the levels which were observed
when the SDGs were set. Therefore, the 2017 PPPs do not significantly shift the goalposts. We
estimate the line that is more typical of lower-middle-income countries (i.e., $3.20 in 2011 PPP)
to be $3.65 in 2017 PPP, and the line common in upper-middle-income countries (i.e., $5.50 in
2011 PPP) to be $6.85 in 2017 PPP. The Bank’s societal poverty line, which was originally defined
as max($1.90, $1 + 50% of median consumption) with the 2011 PPPs, is also updated to
max($2.15, $1.15 + 50% of median consumption) with the 2017 PPPs.

Third, we analyze the impact of these global poverty lines and the 2017 PPPs on the global poverty
counts. While the changes at the upper-middle-income line are larger, the 2017 PPPs have small
implications for extreme poverty and poverty at the lower-middle-income line. Between 1991 to
2017, extreme poverty in the world falls from 36.05% to 9.27% with the revised 2011 PPPs, and
from 37.46% to 9.07% with the 2017 PPPs. This is equivalent to a decrease in the estimated
number of poor people in the world by 15 million in 2017 (or 0.2pp). This is a very small change
compared to prior PPP rounds, which is consistent with the methodological stability over the last
two ICP rounds.

29
References

Alkire, S., U. Kanagaratnam, and N. Suppa. 2020. “The Global Multidimensional Poverty Index
(MPI).” 49. OPHI MPI Methodological Notes. Oxford: Oxford Poverty and Human
Development Initiative, University of Oxford.
Allen, Robert C. 2017. “Absolute Poverty: When Necessity Displaces Desire.” American
Economic Review 107 (12): 3690–3721. https://doi.org/10.1257/aer.20161080.
Atamanov, Aziz, R Andres Castaneda Aguilar, Tony H M J Fujs, Reno Dewina, Carolina Diaz-
bonilla, Daniel Gerszon Mahler, Dean Jolliffe, et al. 2020. “March 2020 PovcalNet Update:
What’s New.” Global Poverty Monitoring Technical Note 11.
Atamanov, Aziz, Dean Jolliffe, Christoph Lakner, and Espen Beer Prydz. 2018. “Purchasing
Power Parities Used in Global Poverty Measurement.” Global Poverty Monitoring Technical
Note 5.
Atamanov, Aziz, Christoph Lakner, Daniel Gerszon Mahler, Samuel Kofi Tetteh-Baah, and Judy
Yang. 2020. “The Effect of New PPP Estimates on Global Poverty: A First Look,” Global
Poverty Monitoring Technical Note 12.
Bai, Yan, and William A. Masters. 2020. “Retail Food Prices at Purchasing Power Parity Exchange
Rates: A First Look at Aggregate ICP 2017 Data.” World Bank Data Blog. 2020.
https://blogs.worldbank.org/opendata/retail-food-prices-purchasing-power-parity-exchange-
rates-first-look-aggregate-icp-2017.
Berry, Francien, Brian Graf, Michael Stanger, and Mari Ylä-Jarkko. 2019. “Price Statistics
Compilation in 196 Economies.” IMF Working Papers 19 (163).
https://doi.org/10.5089/9781513508313.001.
Chen, Shaohua, and Martin Ravallion. 2008. “China Is Poorer than We Thought, but No Less
Successful in the Fight against Poverty.” World Bank Policy Research Working Paper
4621.Washington, DC.
———. 2010. “The Developing World Is Poorer than We Thought, but No Less Successful in the
Fight against Poverty.” The Quarterly Journal of Economics, no. November: 1577–1625.
Dalgaard, Esben, and Henrik Sejerbo Sørensen. 2002. “Consistency between PPP Benchmarks and
National Price and Volume Indices.” In OECD Meeting of National Accounts Experts.
Stockholm, Sweden.
Deaton, Angus. 2001. “Counting the World’s Poor: Problems and Possible Solutions.” World Bank
Research Observer 16 (2): 125–47. https://doi.org/10.1093/wbro/16.2.125.
Deaton, Angus, and Bettina Aten. 2017. “Trying to Understand the PPPs in ICP 2011: Why Are
the Results so Different.” American Economic Journal: Macroeconomics 9 (1): 243–64.
https://doi.org/10.1257/mac.20150153.
Deaton, Angus, and Alan Heston. 2010. “Understanding PPPs and PPP-Based National Accounts.”
American Economic Journal: Macroeconomics 2 (4): 1–35.
Deaton, Angus, and Paul Schreyer. 2022. “GDP, Wellbeing, and Health: Thoughts on the 2017
Round of the International Comparison Program.” Review of Income and Wealth 68(1): 1-15.
https://doi.org/10.1111/roiw.12520.
Ferreira, Francisco H. G., and Carolina Sánchez-Páramo. 2017. “A Richer Array of International
Poverty Lines.” World Bank Blog: Let’s Talk Development. 2017.
https://blogs.worldbank.org/developmenttalk/richer-array-international-poverty-lines.
Ferreira, Francisco H.G., Shaohua Chen, Andrew Dabalen, Yuri Dikhanov, Nada Hamadeh, Dean
Jolliffe, Ambar Narayan, et al. 2016. “A Global Count of the Extreme Poor in 2012: Data

30
 Issues, Methodology and Initial Results.” Journal of Economic Inequality 14 (2): 141–72.
https://doi.org/10.1007/s10888-016-9326-6.
Ferreira, Francisco H.G., Dean Jolliffe, and Espen Beer Prydz. 2015. “The International Poverty
Line Has Just Been Raised to $1.90 a Day, but Global Poverty Is Basically Unchanged. How
Is That Even Possible?” World Bank Blog: Let’s Talk Development. 2015.
https://blogs.worldbank.org/developmenttalk/international-poverty-line-has-just-been-
raised-190-day-global-poverty-basically-unchanged-how-even.
Inklaar, Robert, and D S Prasada Rao. 2017. “Cross-Country Income Levels over Time: Did the
Developing World Suddenly Become Much Richer?” American Economic Journal:
Macroeconomics 9 (1): 265–90.
Inklaar, Robert, and Prasada Rao. 2020. “ICP PPP Time Series Implementation.” Washington,
DC. https://thedocs.worldbank.org/en/doc/f32e966db6ec404699b3381b32e2e589-
0050022021/original/2-01-RA-Item-01-ICP-PPP-Time-Series-Implementation-Rao-and-
Inklaar.pdf.
Jolliffe, Dean, and Espen Beer Prydz. 2016. “Estimating International Poverty Lines from
Comparable National Thresholds.” Journal of Economic Inequality 14 (2): 185–98.
https://doi.org/10.1007/s10888-016-9327-5.
———. Jolliffe, Dean, and Espen Beer Prydz. “Societal Poverty: A Relative and Relevant
Measure.” The World Bank Economic Review 35, no. 1 (2021): 180–206.
https://doi.org/10.1093/wber/lhz018.
Kakwani, Nanak, and Hyun H. Son. 2016. “Global Poverty Estimates Based on 2011 Purchasing
Power Parity: Where Should the New Poverty Line Be Drawn?” Journal of Economic
Inequality 14 (2): 173–84. https://doi.org/10.1007/s10888-016-9322-x.
Klasen, Stephan, Tatyana Krivobokova, Friederike Greb, Rahul Lahoti, Syamsul Hidayat, and
Pasaribu Manuel. 2016. “International Income Poverty Measurement: Which Way Now?”
The Journal of Economic Inequality, 199–225. https://doi.org/10.1007/s10888-016-9324-8.
Krijnse Locker, H., and H. D. Faerber. 1984. “Space and Time Comparisons of Purchasing Power
Parities and Real Values.” Review of Income and Wealth 30 (1): 53–83.
https://doi.org/10.1111/j.1475-4991.1984.tb00477.x.
Lakner, Christoph, Daniel Gerszon Mahler, Minh C Nguyen, Joao Pedro Azevedo, Shaohua Chen,
and Dean Jolliffe. 2018. “Consumer Price Indices Used in Global Poverty Measurement.”
Global Poverty Monitoring Technical Note 8.
Mahler, Daniel Gerszon, R. Andres Castaneda Aguilar, and David Newhouse. 2021. “Nowcasting
Global Poverty.” Policy Research Working Paper 9860. World Bank, Washington, DC.
McCarthy, Paul. 2013. “Extrapolating PPPs and Comparing ICP Benchmark Results.” In
Measuring the Real Size of the World Economy, 473–505.
https://doi.org/10.1596/9780821397282_ch18.
Prydz, Espen Beer, Dean Jolliffe, Christoph Lakner, Daniel Gerszon Mahler, and Prem Sangraula.
2019. “National Accounts Data Used in Global Poverty Measurement.” Global Poverty
Monitoring Technical Note 8.
Ravalllion, Martin. 1998. “Poverty Lines in Theory and Practice.” LSMS Working Paper 133. The
World Bank, Washington, DC.
Ravallion, Martin. 2014. “An Exploration of the International Comparison Program’s New Global
Economic Landscape.” National Bureau of Economic Research Working Paper 20338.
https://doi.org/10.3386/w20338.
———. 2020. “Book Review of Measuring Poverty around the World.” The Journal of Economic

31
 Inequality 18 (1): 131–36. https://doi.org/10.1007/s10888-019-09430-w.
Ravallion, Martin, Shaohua Chen, and Prem Sangraula. 2009. “Dollar a Day Revisited.” World
Bank Economic Review 23 (2): 163–84. https://doi.org/10.1093/wber/lhp007.
Smeeding, Timothy M. 2016. “Poverty Measurement.” In The Oxford Handbook of the Social
Science of Poverty, edited by David Brady and Linda M. Burton. Oxford.
World Bank. 2008. Global Purchasing Power Parities and Real Expenditures: 2005 International
Comparison Program. Washington, DC: World Bank.
———. 2015. Purchasing Power Parities and the Real Size of World Economies: A
Comprehensive Report of the 2011 International Comparison Program. Washington, DC:
World Bank.
———. 2016. “Monitoring Global Poverty: A Cover Note to the Report of the Commission on
Global Poverty, Chaired by Prof. Sir Anthony B. Atkinson.” Washington, DC: World Bank.
http://pubdocs.worldbank.org/en/733161476724983858/MonitoringGlobalPovertyCoverNot
e.pdf.
———. 2017. Monitoring Global Poverty: Report of the Commission on Global Poverty.
Washington, DC: Washington, DC: World Bank.
———. 2018. Poverty and Shared Prosperity Report 2018: Piecing Together the Poverty Puzzle.
Washington, DC: Washington, DC: World Bank.
———. 2020a. Poverty and Shared Prosperity Report 2020: Reversing Reversals of Fortune.
Washington, DC: Washington, DC: World Bank.
———. 2020b. Purchasing Power Parities and the Size of World Economies: Results from the
2017 International Comparison Program. Washington, DC: World Bank.
https://doi.org/10.1596/978-1-4648-1530-0.

32
 The Price of Poverty: The Impact of the 2017 PPPs on the
International Poverty Line and Global Poverty28

Dean Jolliffe ⓡ Daniel Gerszon Mahler ⓡ Christoph Lakner ⓡ

Aziz Atamanov ⓡ Samuel Kofi Tetteh-Baah

Online Appendix

Appendix A. Revised 2011 PPPs

In May 2020, the ICP released new PPP estimates for the 2017 reference year, and also revised
PPP estimates originally published for the 2011 round. The 2011 PPPs were revised primarily to
account for revisions to national accounts expenditures. The ICP expenditure classification for the
original 2011 round was based on the 1993 System of National Accounts. For the 2017 round, the
ICP adopted the new 2008 System of National Accounts, which was also used to revise the original
2011 PPPs. The underlying price data were largely unchanged in the revised 2011 round. For non-
benchmark economies—countries that do not collect prices to compute PPPs as part of the ICP
exercise—new input data were used to impute PPPs. For details on the revisions, see World Bank
(2020) and Tetteh-Baah et al. (2020). Using the harmonized poverty line approach (Jolliffe and
Prydz 2016), the revised 2011 PPPs do not change the international poverty line of $1.90 or the
$3.20 line (the $5.50 line slightly increases by $0.15) (Atamanov et al. 2020). Table A.1 updates
the IPL with the revised 2011 PPPs using the approach by Ravallion et al. (2009). Table A.1 shows
how revisions to the historic CPIs affect the IPL; when we take the national poverty lines
denominated in the original 2011 PPPs in Ferreira et al. (2016) as given and consider only the
revisions to the 2011 PPPs, the IPL remains at $1.90.

Compared to the original 2011 PPPs, the revised 2011 PPPs increase extreme poverty in the world
in 2017 by 0.24pp, representing 18.0 million more poor people (see Tables A.2). This result is
mainly driven by the two poorest regions—Sub-Saharan Africa (11 million more poor people) and
South Asia (11 million more poor people). These regional numbers are in turn largely driven by
Nigeria (5 million poorer) and India (11 million poorer). Table A.3 shows the top 10 countries
with the largest absolute changes in extreme poverty in 2017.

28
The author ordering was constructed through American Economic Association’s randomization tool (confirmation
code: R-TVbwJ99ttd). Corresponding author: Christoph Lakner (clakner@worldbank.org). All authors are with the
World Bank. We gratefully acknowledge financial support from the Knowledge for Change program and from the
UK government through the Data and Evidence for Tackling Extreme Poverty (DEEP) Research Programme. The
findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not
necessarily represent the views of the World Bank and its affiliated organizations, or those of the Executive Directors
of the World Bank or the governments they represent.

33
With a poverty line of $3.20 a day, global poverty increases by 0.29pp, or 21.5 million. Poverty
falls in East Asia & the Pacific (10 million), which is offset by large increases in South Asia (21
million) and Sub-Saharan Africa (10 million). These changes are largely driven by India,
Indonesia, and Nigeria. See Tables A.4 and A.5 for the full set of results.

With a poverty line of $5.50 a day, global poverty slightly increases by 0.07pp, or 5 million.
Poverty again falls in East Asia & the Pacific (12 million), which is offset by a similar increase in
poverty in South Asia (12 million). These changes are largely driven by Indonesia and India,
respectively. See Tables A.6 and A.7 for the full set of results.

Table A.1: Updating the $1.90 IPL with revised 2011 PPPs
Poverty Line, Poverty Line, Poverty Line, Poverty Line,
Original 2011 Update PPP Update CPI Update both
Country
PPP only only PPP and CPI
(1) (2) (3) (4)
Ethiopia 2.03 1.98 2.02 1.97
Ghana 3.07 3.11 3.01 3.04
Gambia, The 1.82 1.81 1.82 1.81
Guinea-Bissau 2.16 2.08 2.16 2.08
Mali 2.15 2.13 2.15 2.13
Mozambique 1.26 1.24 1.33 1.30
Malawi 1.34 1.33 1.29 1.28
Niger 1.49 1.48 1.49 1.48
Nepal 1.47 1.47 1.38 1.38
Rwanda 1.50 1.47 1.51 1.49
Sierra Leone 2.73 2.64 2.10 2.04
Chad 1.28 1.29 1.36 1.36
Tajikistan 3.18 3.35 3.18 3.35
Tanzania 0.88 0.88 0.88 0.88
Uganda 1.77 1.77 1.71 1.72
Mean 1.88 1.87 1.83 1.82
Notes: Column (1) reproduces the national poverty lines from Ferreira et al. (2016). When only PPPs are updated, the
IPL still rounds to $1.87 (see column (2)). Since the derivation by Ferreira et al. (2016), some of the CPIs used to
convert the national poverty lines to 2011 prices have been updated. When CPIs are updated, or when both CPIs and
PPPs are updated, the IPL rounds to $1.80 (columns (3) and (4)). This is mainly driven by CPI revisions in Ghana,
Malawi, Sierra Leone, and Tajikistan. See Appendix E, section 1 for the derivations.

34
 Table A.2: Global and regional changes in poverty in 2017 at $1.90/day
Poverty Poverty Change in Change in
Change in
Region rate, % rate, % poverty, millions of
poverty, %
(original) (revised) pp poor
World 9.04 9.27 0.24 2.65 17.98
East Asia & Pacific 1.61 1.41 -0.20 -12.27 -4.08
Europe & Central Asia 1.36 1.30 -0.06 -4.74 -0.32
Latin America & Caribbean 3.73 3.77 0.03 0.94 0.22
Middle East & North Africa 6.43 6.34 -0.09 -1.45 -0.36
Rest of the World 0.68 0.68 0.00 0.00 0.00
South Asia 9.02 9.65 0.63 7.00 11.33
Sub-Saharan Africa 40.12 41.18 1.06 2.65 11.18
Notes: This table compares extreme poverty between the original and revised 2011 PPPs at the global and regional
levels. Extreme poverty is measured as the share of the population living below the international poverty line, which
is $1.90/day.

Table A.3: Countries with the largest changes in millions of poor in 2017 at $1.90/day
Poverty rate, Poverty rate, Change in Change in Change in
Country
% (original) % (revised) poverty, pp poverty, % millions of poor
India 9.71 10.55 0.85 8.71 11.32
Nigeria 38.65 41.36 2.71 7.02 5.18
Indonesia 5.71 4.46 -1.25 -21.92 -3.31
Ethiopia 23.14 24.66 1.52 6.56 1.62
Angola 41.18 45.07 3.89 9.44 1.16
Congo, Dem.
72.00 72.65 0.65 0.91 0.53
Rep.
Cameroon 21.73 23.66 1.93 8.87 0.47
Egypt, Arab
2.58 2.97 0.39 15.03 0.37
Rep.
Côte d'Ivoire 24.02 25.49 1.47 6.14 0.36
Myanmar 2.01 1.36 -0.65 -32.21 -0.35
Notes: This table compares extreme poverty in 2017 estimated with the original and revised 2011 PPPs at the country
level. The global poverty line used is $1.90/day. The top 10 countries with the largest absolute changes in millions of
poor are shown here. Countries are ranked in descending order of absolute changes in millions of poor.

35
 Table A.4: Global and regional changes in poverty in 2017 at $3.20/day
Poverty Poverty Change in Change in
Change in
Region rate, % rate, % poverty, millions of
poverty, %
(original) (revised) pp poor
World 23.96 24.25 0.29 1.20 21.52
East Asia & Pacific 8.92 8.43 -0.49 -5.53 -10.19
Europe & Central Asia 4.87 4.63 -0.23 -4.78 -1.14
Latin America & Caribbean 9.26 9.31 0.05 0.56 0.33
Middle East & North Africa 17.99 18.30 0.32 1.76 1.21
Rest of the World 0.89 0.89 0.00 0.00 0.00
South Asia 42.26 43.43 1.16 2.75 20.85
Sub-Saharan Africa 66.34 67.34 1.00 1.50 10.47
Notes: This table compares poverty in 2017 estimated with the original and revised 2011 PPPs at the global and
regional levels. The poverty line used is $3.20/day.

Table A.5: Countries with the largest change in millions of poor in 2017 at $3.20/day
Poverty rate, Poverty rate, Change in Change in Change in
Country
% (original) % (revised) poverty, pp poverty, % millions of poor
India 43.65 45.17 1.52 3.48 20.34
Indonesia 27.25 24.64 -2.62 -9.60 -6.93
Nigeria 69.34 71.74 2.41 3.47 4.59
Egypt, Arab
22.22 24.79 2.57 11.56 2.48
Rep.
Ethiopia 59.11 61.33 2.22 3.76 2.36
Myanmar 19.35 14.95 -4.40 -22.72 -2.35
Angola 63.88 68.03 4.15 6.50 1.24
Iraq 15.12 12.19 -2.94 -19.41 -1.10
Cameroon 42.15 44.58 2.43 5.76 0.60
Bangladesh 47.85 47.49 -0.36 -0.75 -0.58
Notes: This table compares extreme poverty in 2017 estimated with the original and revised 2011 PPPs at the country
level. The poverty line used is $3.20/day. The top 10 countries with the largest absolute changes in millions of poor
are shown here. Countries are ranked in descending order of absolute changes in millions of poor.

36
 Table A.6: Global and regional changes in poverty in 2017 at $5.50/day
Poverty Poverty Change in Change in
Change in
Region rate, % rate, % poverty, millions
poverty, %
(original) (revised) pp of poor
World 43.47 43.53 0.07 0.15 4.93
East Asia & Pacific 28.21 27.62 -0.59 -2.10 -12.27
Europe & Central Asia 12.89 12.63 -0.27 -2.07 -1.31
Latin America & Caribbean 22.87 22.97 0.09 0.41 0.60
Middle East & North Africa 42.85 43.08 0.23 0.54 0.88
Rest of the World 1.29 1.29 0.00 0.27 0.04
South Asia 77.80 78.48 0.68 0.87 12.20
Sub-Saharan Africa 85.77 86.23 0.46 0.53 4.80
Notes: This table compares poverty in 2017 estimated with the original and revised 2011 PPPs at the global and
regional levels. The poverty line used is $5.50/day.

Table A.7: Countries with the largest change in millions of poor in 2017 at $5.50/day
Poverty rate, % Poverty rate, Change in Change in Change in
Country
(original) % (revised) poverty, pp poverty, % millions of poor
India 78.64 79.52 0.87 1.11 11.69
Indonesia 58.66 55.79 -2.87 -4.90 -7.60
Myanmar 60.81 54.30 -6.52 -10.71 -3.48
Nigeria 90.61 91.80 1.19 1.31 2.27
Egypt,
66.76 69.11 2.34 3.51 2.26
Arab Rep.
Iraq 53.11 47.59 -5.52 -10.40 -2.07
Russian
2.53 3.82 1.29 51.11 1.87
Federation
Iran,
Islamic 11.30 13.24 1.94 17.16 1.56
Rep.
Argentina 9.07 10.96 1.89 20.83 0.83
Angola 83.56 86.29 2.73 3.27 0.81
Notes: This table compares poverty in 2017 estimated with the original and revised 2011 PPPs at the country level.
The poverty lines used is $5.50/day. The top 10 countries with the largest absolute changes in millions of poor are
shown here. Countries are ranked in descending order of absolute changes in millions of poor.

37
Appendix B. Additional tables and figures

Table B.1: Country-year observations of national poverty rates

Source All Matched into PovcalNet
Poverty and Equity database 1,315 1,315
OECD 123 66
Total 1,438 1,381
Notes: 1. Most of the national poverty rates were obtained from the World Bank’s Poverty and Equity database. These
data can be found in the following series from https://databank.worldbank.org/source/poverty-and-equity: Poverty
headcount ratio at national poverty lines (% of population), including noncomparable values (SI.POV.NAHC.NC).
We downloaded the series on July 11, 2021.
2. The remaining data on national poverty rates were obtained from the Organization for Economic Co-operation and
Development (OECD). We downloaded the series PVT6A: Poverty rate after taxes and transfers, Poverty line 60%
(i.e., share of population living on below 60% of median disposable income) on December 19, 2020 from the OECD
website: https://stats.oecd.org/Index.aspx?DataSetCode=IDD#. We use the OECD data only for countries with
missing data in the Poverty and Equity database, or where the OECD time series is longer.

Table B.2: Exception 2011 PPPs for global poverty monitoring

Original Original Revised Revised
Country
(official) (imputed) (official) (imputed)
1. Egypt. Arab
Rep. 1.80 2.78 1.71 2.87
2. Iraq 573.42 1003.80 477.56 939.22
3. Jordan 0.32 0.45 0.33 0.44
4. Lao PDR 2914.85 3325.20 3124.08 3248.44
5. Myanmar 275.83 320.60 278.39 296.14
6. Yemen, Rep. 82.09 111.30 76.77 109.53
Source: Atamanov et al. (2020), Table A.1.
Notes: See main text, Atamanov et al. (2020), Atamanov et al. (2018), and Appendix C for details. Whenever the
(revised) 2011 PPPs are used in the main text, the PPPs in the last column are used.

38
 Table B.3: Exception 2017 PPPs for global poverty monitoring
Revised 2017 PPP 2017 PPP 2017 PPP Note
Country 2011 PPP (official) (imputed) (exception)
(1) (2) (3) (4) (5)
1. Belize 1.17 1.48 1.16 1.31 Average
2. Egypt, Arab Rep. 2.87 3.41 6.88 4.84 Average
3. Guinea 2599.89 3213.98 3696.76 3446.93 Average
4. Iraq 939.22 555.39 619.35 586.50 Average
5. Nigeria 83.58 112.10 136.44 123.67 Average
6. São Tomé and Príncipe 10.49 10.76 12.14 11.43 Average
7. Sudan 1.46 5.38 7.77 6.46 Average
8. Trinidad and Tobago 4.52 4.21 5.16 4.66 Average
9. Kiribati 1.07 0.98 Extrapolated
10. Nauru 1.21 1.38 Extrapolated
11. Syrian Arab Republic 22.26 151.78 Extrapolated
12. Tuvalu 1.17 1.25 Extrapolated
13. Venezuela, RB 2.94 315.91 Extrapolated
14. Yemen, Rep. 109.53 255.68 Extrapolated
Notes: The PPPs in columns (1) and (4) are used in this paper. The geometric averages of official and imputed 2017
PPPs (column 4) are used for the exception countries. For countries without official 2017 PPPs, extrapolated PPPs are
used: PPPs are extrapolated from the revised 2011 PPP currently used for global poverty monitoring, together with
domestic and US inflation between 2011 and 2017. See Section 3.B of the main paper for details.

39
 Table B.4: Rural/Urban PPPs used in global poverty monitoring
Country, variable Original 2011 PPP Revised 2011 PPP 2017 PPP
A) China
Ratio of urban to rural poverty line (𝝎) 1.29 1.29 1.24
ICP urban share of outlets (𝝀) 0.76 0.76 0.79
Rural PPP 3.038 3.039 3.495
Urban PPP 3.904 3.905 4.318
National PPP 3.696 3.698 4.147
B) India
Ratio of urban to rural poverty line (𝝎) 1.22 1.22 1.22
ICP urban share of outlets (𝝀) 0.74 0.74 0.64
Rural PPP 12.908 13.173 17.092
Urban PPP 15.695 16.018 20.787
National PPP 14.975 15.283 19.469
C) Indonesia
Ratio of urban to rural poverty line (𝝎) 1.18 1.18 1.07
ICP urban share of outlets (𝝀) 0.61 0.61 0.98
Rural PPP 3678.414 3498.876 4746.852
Urban PPP 4352.751 4140.299 5098.196
National PPP 4091.939 3892.218 5089.686
Source: Adapted from Atamanov et al. (2020), Table A.2.
𝑁𝑎𝑡𝑖𝑜𝑛𝑎𝑙 𝑃𝑃𝑃
Note: See main text for details. The urban and rural PPPs are computed as follows: 𝑅𝑢𝑟𝑎𝑙 𝑃𝑃𝑃 =
𝜔×𝜆+(1−𝜆)

𝑈𝑟𝑏𝑎𝑛 𝑃𝑃𝑃 = 𝜔 ∗ 𝑅𝑢𝑟𝑎𝑙 𝑃𝑃𝑃. See online appendix of Ferreira et al. (2016) for details.

40
 Table B.5: Cumulative mean and median of poverty lines of the poorest countries
Harmonized national poverty lines, 2017 PPP
No. Country Year Cumulative Cumulative
Country-specific
mean median
1 Burundi 2013 2.14 2.14 2.14
2 Congo, Dem. Rep. 2012 1.90 2.02 2.02
3 Malawi 2016 1.68 1.91 1.90
4 Central African Republic 2008 2.16 1.97 2.02
5 Niger 2014 1.87 1.95 1.90
6 Mozambique 2014 1.49 1.88 1.89
7 Togo 2015 2.17 1.92 1.90
8 Liberia 2016 3.13 2.07 2.02
9 Madagascar 2012 1.64 2.02 1.90
10 Sierra Leone 2018 3.24 2.14 2.02
11 Chad 2011 2.66 2.19 2.14
12 Guinea-Bissau 2010 2.28 2.20 2.15
13 Ethiopia 2015 2.04 2.19 2.14
14 Burkina Faso 2014 2.16 2.18 2.15
15 Rwanda 2016 1.73 2.15 2.14
16 Guinea 2012 3.40 2.23 2.15
17 Mali 2009 1.95 2.22 2.14
18 Gambia, The 2015 3.87 2.31 2.15
19 Uganda 2016 1.49 2.26 2.14
20 Kiribati 2006 2.73 2.29 2.15
21 Nepal 2010 2.93 2.32 2.16
22 Tanzania 2017 1.62 2.29 2.15
23 Solomon Islands 2012 1.61 2.26 2.14
24 Senegal 2011 2.36 2.26 2.15
25 Lesotho 2017 3.22 2.30 2.16
26 Comoros 2004 4.18 2.37 2.16
27 Tajikistan 2015 4.01 2.43 2.16
28 Haiti 2012 3.67 2.48 2.17
29 Benin 2015 1.77 2.45 2.16
30 Zimbabwe 2017 2.01 2.44 2.16
31 Uzbekistan 2003 0.91 2.39 2.16
32 Papua New Guinea 2009 2.16 2.38 2.16
33 Vanuatu 2010 2.04 2.37 2.16
34 Timor-Leste 2014 3.59 2.41 2.16
35 Tuvalu 2010 4.16 2.46 2.16
36 Sudan 2009 3.58 2.49 2.16
37 Cameroon 2014 2.91 2.50 2.16
38 Zambia 2015 1.78 2.48 2.16
39 Micronesia, Federated States of 2013 3.66 2.51 2.16
40 Kenya 2015 2.41 2.51 2.17
Notes: Countries are shown in ascending order of GDP per capita (2017 PPP), starting with the poorest country. The
full sample consists of 157 countries, excluding South Sudan, the Syrian Arab Republic, the Republic of Yemen
(whose GDP per capita data are missing). Only the first 40 countries are listed here due to space constraints.

41
Table B.6: Updating global poverty lines using harmonized poverty lines: different options
(C) Own harmonized
(A) Harmonized lines (B) Own harmonized
Income lines
(Jolliffe and Prydz 2016) lines (pooled)
classification (pooled, since 2007)
Median Mean N Median Mean N Median Mean N
Low 2.15 2.53 33 2.16 2.43 177 2.14 2.35 59
Lower-middle 3.68 4.49 32 3.80 4.20 394 3.59 4.04 214
Upper-middle 6.32 6.18 32 6.76 7.06 343 6.76 7.04 257
High income 23.65 23.34 29 24.15 23.19 460 23.97 23.08 359
Observations 126 1,374 889
Notes: Panel (A) uses the same implicit national poverty lines originally derived by Jolliffe and Prydz (2016). Panel
(B) and (C) pool together the implicit national poverty lines that we have derived. Our total sample of implicit poverty
lines is 1,381, but 7 lines from the 1980s drop out because they could not be matched into the World Bank’s historical
income classification, which begins from 1988. Panels (B) and (C) provide weighted medians and means for country-
year observations based on the income classification in the year in which the surveys were conducted. We weight the
observations to give greater importance to those close to 2017 and to give equal weight to each country: In the first
step, we assign a survey conducted in 2017 a weight of 1, a survey conducted in 2016 or 2018 a weight of 1/2, a survey
conducted in 2015 or 2019 a weight of 1/3, and so on. Secondly, we rescale the weights such that each country has a
total weight of 1. For example, if a country has three harmonized poverty lines for 2016, 2017, and 2018, the
corresponding weights will be 0.25, 0.5, and 0.25. The weights are based on the relevant data set (N = 1,374) or
subsample in question (N = 889). The sample in panel (A) has one harmonized national poverty for each country, so
each country has a weight of 1 (there are 126 countries in panel (A), 157 in panel (B), and 150 in panel (C)). The
oldest lines (n = 2) in panel (A) are from 2001 (i.e., 10 years before the 2011 ICP reference year). Panel (C) includes
only lines since 2007 (i.e., 10 years before the 2017 ICP). When one line is selected for each country from the most
recent lines (since 2007), the proposed international poverty line would be $2.14 (see Table B.7 below).

Table B.7: Updating global poverty lines using harmonized poverty lines:
full sample vs. subsample
Income (A) Full sample (B) Subsample (since 2007)
classification Median Mean N Median Mean N
Low 2.15 2.42 28 2.14 2.31 25
Lower-middle 3.63 3.95 54 3.59 3.93 51
Upper-middle 6.85 7.05 37 6.95 7.08 36
High income 24.36 23.36 38 24.36 23.36 38
Observations 157 150
Notes: Panel (A) replicates Panel (C) from Table 1 in the main text. The line that is closest to 2017 is selected, one for
each country. If harmonized national poverty lines are available for 2016 and 2018 but not 2017, 2018 is selected.
Panel (B) excludes 7 lines derived from surveys conducted more than 10 years before the 2017 ICP reference year:
AZE2001, COM2004, JAM2004, KIR2006, SYR2003, UZB2003, VEN2006.

42
 Table B.8: Equivalent international poverty line, 2017 PPP: different options
Headcount, % Equivalent Share of Weighted
Year (Revised 2011 IPL (2017 global Equivalent
Region
PPP) PPP) poor IPL
(1) (2) (3) (4) (5)
World 2010 16.02 2.119 1.00 2.16
East Asia & Pacific 2010 10.79 1.973 0.19 0.38
Europe & Central Asia 2010 2.37 1.801 0.01 0.02
Latin America & Caribbean 2010 6.03 2.058 0.03 0.07
Middle East & North Africa 2010 2.04 2.215 0.01 0.01
Rest of the World 2010 0.50 2.076 0.00 0.01
South Asia 2010 25.95 2.146 0.38 0.82
Sub-Saharan Africa 2010 47.47 2.274 0.37 0.85
World 2012 12.89 2.128 1.00 2.19
East Asia & Pacific 2012 6.88 1.979 0.15 0.30
Europe & Central Asia 2012 1.88 1.801 0.01 0.03
Latin America & Caribbean 2012 4.63 2.072 0.03 0.06
Middle East & North Africa 2012 2.23 2.185 0.01 0.02
Rest of the World 2012 0.57 2.114 0.01 0.01
South Asia 2012 19.15 2.147 0.35 0.76
Sub-Saharan Africa 2012 43.90 2.284 0.44 1.01
World 2015 10.14 2.161 1.00 2.22
East Asia & Pacific 2015 2.06 2.012 0.06 0.11
Europe & Central Asia 2015 1.51 1.801 0.01 0.02
Latin America & Caribbean 2015 3.69 2.067 0.03 0.06
Middle East & North Africa 2015 4.28 2.143 0.02 0.05
Rest of the World 2015 0.73 2.132 0.01 0.02
South Asia 2015 13.18 2.148 0.31 0.67
Sub-Saharan Africa 2015 41.95 2.294 0.56 1.29
World 2017 9.27 2.173 1.00 2.23
East Asia & Pacific 2017 1.41 2.023 0.04 0.08
Europe & Central Asia 2017 1.30 1.801 0.01 0.02
Latin America & Caribbean 2017 3.77 2.055 0.03 0.07
Middle East & North Africa 2017 6.34 2.147 0.03 0.07
Rest of the World 2017 0.68 2.143 0.01 0.02
South Asia 2017 9.65 2.147 0.25 0.53
Sub-Saharan Africa 2017 41.18 2.297 0.62 1.43
Notes: This table shows equivalent poverty lines in column (3) that keep the global and regional poverty headcount
ratios in column (2) constant in a pre-specified reference year (year given in column (1)). The results for the world in
column (3) are presented in the main text in Table 2. Column (5) provides an alternative approach, where the
equivalent international poverty line for the world is calculated as the weighted average of the regional equivalent
poverty lines (in column (3)), with the weights being the regions’ respective shares of total millions of poor in column
(4). The weights are based on the revised 2011 PPPs (with a $1.90 poverty line).

43
 Table B.9: Updating the $1.90 IPL with 2017 PPPs
Poverty Line Update PPP & Update PPP & CPI Take $1.90 as
Original CPI (update CPI II (do not update given, update
Country
2011 PPP in $1.90) CPI in $1.90) PPP & CPI
(1) (2) (3) (4)
Ethiopia 2.03 2.37 2.38 2.23
Ghana 3.07 2.28 2.33 1.44
Gambia, The 1.82 1.88 1.88 1.97
Guinea-Bissau 2.16 2.45 2.45 2.15
Mali 2.15 2.49 2.49 2.20
Mozambique 1.26 1.44 1.37 2.06
Malawi 1.34 1.55 1.61 2.28
Niger 1.49 1.47 1.47 1.88
Nepal 1.47 1.83 1.95 2.52
Rwanda 1.5 1.81 1.79 2.27
Sierra Leone 2.73 2.88 3.73 2.60
Chad 1.28 1.60 1.51 2.24
Tajikistan 3.18 3.27 3.27 1.95
Tanzania 0.88 1.06 1.06 2.29
Uganda 1.77 1.90 1.96 2.11
Mean 1.88 2.02 2.08 2.15
Notes: Column (1) reproduces national poverty lines from Ferreira et al. (2016). Column (2) updates both PPPs and
CPIs to the 2017 ICP reference year, while incorporating the revisions to the CPI data used in deriving the $1.90 line.
Column (3) updates both PPPs and CPIs to the 2017 ICP reference year, without incorporating the revisions to CPI
data used in deriving the $1.90 line. Column (4) takes the $1.90 line as given for all countries and updates the $1.90
line with both PPPs and updated CPIs to the 2017 reference year. See Appendix E, section 2 for the derivations. If we
updated the IPL with US inflation between 2011 and 2017 (i.e., approximately 9%), the new IPL would be $2.07 .

44
 Table B.10: Global and regional changes in poverty in 2017 at $3.20/day
Poverty Poverty Change in
Change in Change in
Region rate, % rate, % millions of
poverty, pp poverty, %
(2011 PPP) (2017 PPP) poor
World 24.25 24.82 0.57 2.35 42.79
East Asia & Pacific 8.43 10.52 2.09 24.85 43.30
Europe & Central Asia 4.63 6.70 2.06 44.51 10.13
Latin America & Caribbean 9.31 10.14 0.83 8.87 5.21
Middle East & North Africa 18.30 16.32 -1.99 -10.85 -7.57
Rest of the World 0.89 0.91 0.02 2.31 0.23
South Asia 43.43 44.29 0.86 1.98 15.43
Sub-Saharan Africa 67.34 65.06 -2.28 -3.39 -23.94
Notes: This table compares poverty in 2017 estimated with the revised 2011 and 2017 PPPs at the global and regional
levels. The poverty lines used are $3.20/day (2011 PPP) and $3.65/day (2017 PPP).

Table B.11: Countries with the largest changes in millions of poor in 2017 at $3.20/day
Poverty rate, Poverty rate, Change in Change in Change in
Country
% (2011 PPP) % (2017 PPP) poverty, pp poverty, % millions of poor
China 3.80 5.72 1.92 50.41 26.56
Nigeria 71.74 64.67 -7.08 -9.86 -13.50
Indonesia 24.64 29.59 4.95 20.10 13.11
Pakistan 33.95 38.86 4.91 14.47 10.21
Uzbekistan 43.66 71.46 27.80 63.68 9.00
India 45.17 45.81 0.64 1.41 8.53
Angola 68.03 47.81 -20.23 -29.73 -6.03
Egypt, Arab
24.79 18.58 -6.21 -25.05 -5.99
Rep.
Ghana 28.72 48.13 19.41 67.56 5.65
Ethiopia 61.33 56.50 -4.83 -7.88 -5.14
Notes: This table compares poverty in 2017 estimated with the revised 2011 and 2017 PPPs at the country level. The
poverty lines used are $3.20/day (2011 PPP) and $3.65/day (2017 PPP). The top 10 countries with the largest absolute
changes in millions of poor are shown here. Countries are ranked in descending order of absolute changes in millions
of poor.

45
 Table B.12: Global and regional changes in poverty in 2017 at $5.50/day
Poverty Poverty Change in Change in
Change in
Region rate, % rate, % poverty, millions of
poverty, %
(2011 PPP) (2017 PPP) pp poor
World 43.53 47.81 4.28 9.83 321.22
East Asia & Pacific 27.62 36.21 8.60 31.13 177.79
Europe & Central Asia 12.63 16.43 3.81 30.15 18.70
Latin America & Caribbean 22.97 27.59 4.63 20.15 29.16
Middle East & North Africa 43.08 44.09 1.00 2.33 3.83
Rest of the World 1.29 1.49 0.20 15.56 2.20
South Asia 78.48 82.66 4.18 5.33 74.96
Sub-Saharan Africa 86.23 87.61 1.39 1.61 14.58
Notes: This table compares poverty in 2017 estimated with the revised 2011 and 2017 PPPs at the global and regional
levels. The poverty lines used are $5.50/day (2011 PPP) and $6.85/day (2017 PPP).

Table B.13: Countries with the largest changes in millions of poor in 2017 at $5.50/day
Poverty rate, Poverty rate, Change in Change in Change in millions
Country
% (2011 PPP) % (2017 PPP) poverty, pp poverty, % of poor
China 19.88 29.18 9.30 46.77 128.91
India 79.52 83.37 3.85 4.85 51.59
Indonesia 55.79 65.17 9.38 16.81 24.82
Pakistan 74.52 82.05 7.53 10.10 15.65
Brazil 20.23 25.93 5.70 28.15 11.84
Iraq 47.59 21.04 -26.55 -55.79 -9.97
Myanmar 54.3 68.2 13.91 25.61 7.42
Ghana 54.34 77.94 23.60 43.44 6.87
Philippines 60.02 66.25 6.23 10.38 6.55
Iran,
Islamic 13.24 20.94 7.70 58.13 6.21
Rep.
Notes: This table compares poverty in 2017 estimated with the revised 2011 and 2017 PPPs at the country level. The
global poverty lines used are $5.50/day (2011 PPP) and $6.85/day (2017 PPP). The top 10 countries with the largest
absolute changes in millions of poor are shown here. Countries are ranked in descending order of absolute changes in
millions of poor.

46
 Table B.14: Global and regional changes in societal poverty in 2017
Poverty Poverty Change in
Change in Change in
Region rate, % rate, % millions of
poverty, pp poverty, %
(2011 PPP) (2017 PPP) poor
World 27.31 27.55 0.23 0.86 17.59
East Asia & Pacific 22.58 23.55 0.97 4.29 20.04
Europe & Central Asia 17.10 18.18 1.09 6.35 5.33
Latin America & Caribbean 26.36 26.67 0.31 1.18 1.96
Middle East & North Africa 26.01 25.47 -0.54 -2.09 -2.07
Rest of the World 15.19 15.25 0.06 0.39 0.66
South Asia 31.32 31.84 0.52 1.65 9.29
Sub-Saharan Africa 48.28 46.60 -1.68 -3.48 -17.62
Notes: This table compares societal poverty in 2017 estimated with the revised 2011 and 2017 PPPs at the global and
regional levels. With the revised 2011 PPP, the societal poverty line (SPL) used is given as max($1.90, $1.00 + 0.5 *
median consumption/income). With the 2017 PPP, the societal poverty line (SPL) used is given as max($2.15, $1.15
+ 0.5 * median consumption/income).

Table B.15: Countries with the largest changes in millions living in societal poverty in 2017
Poverty rate, Poverty rate, Change in Change in Change in millions
Country
% (2011 PPP) % (2017 PPP) poverty, pp poverty, % of poor
China 21.04 22.07 1.03 4.88 14.25
Nigeria 41.36 37.57 -3.79 -9.16 -7.23
Congo,
72.65 64.42 -8.24 -11.34 -6.7
Dem. Rep.
India 31.43 31.87 0.45 1.42 5.98
Uzbekistan 32.18 44.93 12.75 39.62 4.13
Indonesia 29.36 30.86 1.49 5.09 3.95
Pakistan 30.88 32.69 1.81 5.87 3.77
Angola 51.81 43.58 -8.24 -15.9 -2.46
Iraq 21.31 15.82 -5.49 -25.78 -2.06
Egypt,
30.53 28.54 -1.99 -6.52 -1.92
Arab Rep.
Notes: This table compares societal poverty in 2017 estimated with the revised 2011 and 2017 PPPs at the country
level. With the revised 2011 PPP, the societal poverty line (SPL) used is given as max($1.90, $1.00 + 0.5 * median
consumption/income). With the 2017 PPP, the societal poverty line (SPL) used is given as max($2.15, $1.15 + 0.5 *
median consumption/income). The top 10 countries with the largest absolute changes in millions of poor are shown
here. Countries are ranked in descending order of absolute changes in millions of poor.

47
 Figure B.1: Share of global poor by income group in 2017

Notes: This chart shows the share of global poor in 2017 in lower-income countries (LIC), lower-middle-income
countries (LMIC), upper-middle-income countries (UMIC) and high-income countries (HIC) based on the LIC, LMIC,
and UMIC global poverty lines.

Figure B.2: Delta ratio by income group

48
Notes: This chart shows the mean delta ratio for each income group based on the World Bank’s income classification
for 2017 (see more details on the delta ratio in Section 3 of the paper). The delta ratios are either weighted by countries’
population sizes in 2017 or weighted equally. The weights are population sizes of countries in 2017. The relative
changes in global poverty lines derived with data from low-income countries (LIC), lower-middle-income countries
(LMIC) and upper-middle-income countries (UMIC) when moving to the 2017 PPPs are marked in the chart.

Figure B.3: Comparison of residuals in price level indices (PLIs) across regions

Notes: This graph shows the mean log residuals and absolute mean log residuals in price level indices (PLIs) for the
2017 PPP for household consumption expenditure across regions. The residual is defined as the published PLI (log)
minus predicted PLI (log). Error bars represent 95% confidence intervals.

49
 Figure B.4: Changes in extreme poverty in Sub-Saharan Africa, 2017

Notes: This chart shows estimates of extreme poverty in 2017 for countries in Sub-Saharan Africa. For countries,
without a survey in 2017, the estimates are based on extrapolations or interpolations from recent surveys. Extreme
poverty is measured using the IPL of $1.9 (revised 2011 PPP) or $2.15 (2017 PPP). Marker size is proportional to
absolute change in millions of poor. The dotted line is a 45-degree line.

Table B.16: Decomposition of changes to global poverty lines

Jolliffe and Update Update Update Obs Update Obs Update Obs
Prydz with with national income country
(original revised 2017 poverty groups availa-
2011 PPP) 2011 PPP PPP lines bility
(1) (2) (3) (4) (5) (6)
Low 1.91 1.85 2.15 2.17 33 2.16 25 2.15 28
Lower-middle 3.21 3.21 3.68 3.76 32 3.58 37 3.63 54
Upper-middle 5.47 5.65 6.32 7.15 32 6.95 30 6.85 37
High-income 21.70 21.70 23.65 27.39 29 24.31 34 24.36 38
Notes: The table shows what is driving the changes to the global poverty lines, starting from the global lines replicating
Jolliffe and Prydz (2016) in column (1), to the final global poverty lines derived in this paper in column (6). The
intermediate columns all make one change to the derivation of the harmonized national poverty lines, allowing for a
(path dependent) decomposition of what is driving the final derivation. Columns (2) and (3) update the harmonized
poverty lines from Jolliffe and Prydz (2016) with the revised 2011 PPPs and 2017 PPPs, respectively. Column (4)
updates with new harmonized poverty lines derived from new survey data for the exact same countries that Jolliffe
and Prydz (2016) used. Column (5) updates the income classification for the countries that with new data have changed
income groups. Column (6) adds the countries for which national poverty lines have been made available after Jolliffe
and Prydz (2016) made their analysis and removes the countries for which the national poverty lines or welfare
aggregates have been removed due to quality concerns.

50
Appendix C. More details on the stability of the ICP methodology

The System of National Accounts (SNA) is the most widely used framework for measuring
economic activities across countries. The conceptual framework of the ICP methodology is
centered around the SNA’s definition of GDP from the expenditure side; the sum of final
expenditures on consumption, gross capital formation, and net exports (World Bank 2013; 2020).
The most recent ICP rounds were based on different versions of the SNA: the 2005 and original
2011 rounds were based on the 1993 SNA, whereas the revised 2011 and 2017 rounds were based
on the 2008 SNA. Even though both versions share the same fundamental structure, the 2008 SNA
introduced some changes to the 1993 SNA (OECD 2013; World Bank 2020). These changes have
implications for GDP, either by re-allocating components and/or changing the level of GDP. For
example, research and development (R&D) has been capitalized for the first time, which has two
implications: (1) it re-classifies R&D from government final consumption expenditure to
government gross fixed capital formation, and (2) it adds to the level of GDP via the cost of
depreciation that has to be imputed and included (OECD 2013).29

A few changes have been introduced into the ICP expenditure classification in the 2017 round at
different levels of aggregation, mainly in light of the SNA changes. The ICP classifies expenditure
and price data as main aggregates, categories, groups, classes, and basic headings, in descending
order.30 For example, gross capital formation is a new main aggregate of GDP that now combines
two main aggregates in the (original) 2011 round (i.e., gross fixed capital formation and changes
in inventories and acquisitions less disposal of valuables). Even though there are still 155 basic
headings in the 2017 round as in the 2011 round, some basic headings have been merged (e.g.,
opening value of inventories and closing value of inventories in the 2011 round have been merged
as changes in inventories in the 2017 round) and new ones have been added (e.g., actual and
imputed rentals for housing was a basic heading in the 2011 round but is now separated as actual
rentals for housing and imputed rentals for housing under the housing rentals category). A key
point to note is that aggregation is done from the level of basic headings up to the desired aggregate
(e.g., GDP, household final consumption, or other levels) so that all these changes at the level of
basic headings do not have a significant impact on the aggregates. Apart from changes in national
accounts structures caused by SNA revisions, national accounts expenditure data of countries get
routinely revised or rebased to incorporate new information. This primarily explains the revisions
to the original 2011 PPPs.

29
As an example, OECD (2013) shows that the capitalization of R&D increases Australia’s GDP by 1.25-1.5%. Other
SNA changes relate to the classification and/or valuation of weapon systems and ammunitions, computer software
and databases, financial intermediation services indirectly measured (FISIM), output of central banks, and output for
own use, among others. A more detailed description of the SNA changes can be found in OECD (2013).
30
A basic heading consists of a well-defined group of goods or services at the lowest level of ICP expenditure
classification. Rice and bread are examples of basic headings under the class of bread and cereals, which forms a part
of the group called food. Food is within the Food and nonalcoholic beverages category, under the main aggregate of
individual consumption by households. PPPs are first estimated at the level of basic headings and are aggregated along
higher levels of aggregation up to the level of GDP, which sums up the main aggregates of expenditure.

51
The overall methodology for selecting the basket of goods and services the ICP collects prices on
has remained quite stable. The ICP relies on sampling theory to determine the number of products
to be priced in each basic heading, the number of survey outlets, and the number of times surveys
will be conducted in the reference year. For example, only 10-15 items may be priced for the rice
basic heading while 70-100 items may be priced for the more heterogeneous garments basic
heading to obtain a similar level of precision in the estimates for the basic-headings PPPs (World
Bank 2013). The exact items that are priced, as well as where or when surveys are exactly
conducted, are subject to expert advice ex-ante and validation ex-post. The list of items also
requires at least partial updates for each comparison (e.g., IT equipment, cars, home entertainment,
etc.). In the 2005 round, there was a separate price and expenditure survey with a “ring” list of
1,000 items that were priced in 18 selected “ring” countries across all regions, to link prices within
and across regions. The ring list of items has been found to have a bias for traded goods, which
are often unrepresentative, scarce, and expensive in developing countries (Deaton and Aten 2017).
As a result, the 2005 round overstated price levels in most developing countries. The lessons
learned from the 2005 round brought about significant improvements in ICP product specification.
In subsequent rounds, the list of items in the price surveys conducted in all participating economies
has been carefully chosen to ensure that the items are both representative within countries and
comparable across countries, as best as possible (World Bank 2013; 2015; 2020). A global core
list (GCL) of items is priced as part of the price and expenditure surveys, for linking prices within
and across regions. The GCL items are marked as “important” or “less important” and prices are
weighted accordingly, so that price levels in developing countries are not overstated. Each country
decides its own sets of important and less important items, depending on their national
consumption patterns. Also, importance indicators apply only within basic headings, not across.
While the general approach of obtaining the price and expenditure data remains similar, especially
between the 2011 and 2017 rounds, data quality assurance has been improved with the latest
available technology, particularly in the 2017 round (World Bank 2020).

As described in more detail in World Bank (2013; 2015; 2020), the computational steps the ICP
follows to produce the final global PPPs have also remained largely unchanged. Since the major
change involving the linking of regional prices—namely, the 2005 “ring” method vs. 2011 or 2017
“Global Core List” (GCL) method—the common elements of PPP estimation have not changed.
At the global level, the Country Product Dummy (CPD) method is used to estimate PPPs. This
method estimates a regression model that produces in one step PPPs that are both transitive and
base-country invariant. The resulting PPPs are multilateral (i.e., the basic-heading PPPs for one
economy depend on those of all other participating economies). PPPs are calculated for countries
within regions, first at the basic heading level using the Country Product Dummy (CPD) method
and then at the aggregate level using the Gini, Eltetö and Köves, and Szulc (GEKS) method, using
prices in the GCL and a regional list, and expressed in the currency of a reference country in the
region. Finally, the PPPs are linked across regions solely using items in the GCL to produce the

52
global PPPs, expressed in US dollars, in a way that ensures regional fixity (i.e., the ratio of real
expenditures between any pair of participating economies within a region remains the same after
linking prices across regions).

The ICP data coverage varies across rounds, which potentially impacts the global PPPs used for
measuring global poverty. Given that the PPPs are multilateral, PPPs would, in principle, be
affected if the number of economies changes between any two ICP rounds. In practice, however,
the total number of countries participating is less of an issue due to the regional fixity principle. In
particular, the differences in country coverage between the 2011 and 2017 rounds mainly come
from small Pacific Islands (e.g., Cook Islands and Kiribati) that participated in the 2011 round.
However, since these countries were linked via three bridge-countries that participated from other
regions—Australia (Eurostat-OECD), Fiji (Asia and the Pacific), and New Zealand (Eurostat-
OECD)—the inclusion of these islands did not impact the rest of the countries in any manner.
Overall, comparability of different ICP rounds may still be an issue in cases where countries move
regions (e.g., Colombia and Costa Rica have been moved from Latin America and the Caribbean
to OECD in the 2017 round) or where countries join or drop from the comparison altogether.

Within countries, the price data coverage also varies between rounds, although with the currently
available metadata it is not possible to assess this everywhere for past ICP rounds. In the 2005
round, for instance, price survey outlets were predominantly in capital cities and urban areas
(World Bank 2008). In the 2017 round, this is still the case in some countries, including many
high-income countries. Survey frames do not necessarily imply overestimation or underestimation
of national prices, as the ICP requests countries to submit nationally representative annual prices.
However, the ICP acknowledges that national average prices are often hard to achieve in large
economies with settlements and populations that are substantially rural (World Bank 2015), and
ultimately the ICP does not know whether prices have been adjusted appropriately.

53
Appendix D. Deciding when to impute 2017 PPPs for exception countries

While we argue that the ICP methodology has largely stabilized, the change from 2011 to 2017
PPPs can still lead to erratic changes in poverty estimates at the country level. In this part of the
Appendix, we describe the analysis we have done to identify the countries where we have decided
to deviate from the official 2017 PPPs.

Step 1: Select all countries where it is worthwhile to take a deeper look

Countries that fulfill any of the following conditions are selected for more detailed analysis:

a) Is there a large change in the real value of welfare when moving to the 2017 PPPs?
As described in the main text and following Ferreira et al. (2016), we select all countries
with delta ratios more than two standard deviations from the mean using the CPIs and
revised 2011 PPPs used for global poverty monitoring. From Figure 2A (in the main text),
we select Angola, Belize, the Arab Republic of Egypt, Ghana, Guinea, Iraq, Jordan,
Liberia, São Tomé and Príncipe, and Suriname.

b) Is there a big difference in the movements of CPIs and PPPs?

We select all countries with delta ratios more than two standard deviations from the mean
using IFS CPIs and official 2011 PPPs. This is very similar to the previous condition,
except for the slightly different CPIs and PPPs (see the main text for details). From Figure
2B (in the main text), we select Angola, Belize, Bolivia, Central African Republic, Guinea,
Liberia, Nigeria, São Tomé and Príncipe, Suriname, and Trinidad and Tobago.

c) Are the 2017 PPPs very different from what we would expect based on the country’s
characteristics?
We select all countries whose log difference in the PPP residual is more than two standard
deviations from the mean. The PPP residual is defined as the difference in the logs of the
published PPP and predicted PPP. PPPs are predicted using the seemingly unrelated
regressions (SUR) model the ICP uses to predict PPPs for non-benchmark countries.

The ICP model for predicting 2017 PPPs for non-benchmark countries is given as:

𝑃𝐿𝐼𝑖 − 𝑃𝐿𝐼𝑈𝑆𝐴 = 𝑏 ∗ (𝑋𝑖 − 𝑋𝑈𝑆𝐴 ) + 𝑒𝑖

where 𝑃𝐿𝐼𝑖 is the price level index of country 𝑖, calculated as the ratio of the PPP
conversion factor to the market exchange rate, and 𝑋 is a vector of explanatory variables:
GDP per capita in U.S. dollars (based on market exchange rates), imports as a share of
GDP, exports as a share of GDP, age dependency ratio, and dummies for Sub-Saharan

54
 Africa, the EU, island economies, and landlocked developing economies. Interaction terms
between GDP per capita and the country-group-dummy variables are also included.31

From Figure D.1, we select Barbados, Central African Republic, Egypt, Djibouti, Sudan,
and Suriname. Barbados has no surveys in PovcalNet, so we drop it from the set of
countries.

Figure D.1: Outlier countries based on the residual criterion

d) Are revised 2011 PPPs currently imputed for global poverty monitoring?
We also include countries whose revised 2011 PPPs are currently imputed (Ferreira et al.
2016; Atamanov et al. 2018; 2020). This adds Egypt, Iraq, Jordan, the Lao People’s
Democratic Republic, Myanmar, and the Republic of Yemen. The Republic of Yemen is
not included in the rest of the investigation because it has no official 2017 PPP.

In summary, in step 1 of the analysis we select the following countries for a more detailed analysis:
Angola, Belize, Bolivia, Central African Republic, Djibouti, Egypt, Ghana, Guinea, Iraq, Jordan,

31
Note that this model is not targeted towards predicting the price levels of the poor, for which other variables may
be more relevant. This is a general issue with PPPs and CPIs used for global poverty monitoring. We thank Stefan
Dercon for raising this issue. We nonetheless use the model by ICP to stay consistent with the PPPs used for
benchmark countries, which likewise are not tailored towards the prices faced by the poor.

55
Lao PDR, Liberia, Myanmar, Nigeria, São Tomé and Príncipe, Sudan, Suriname, Trinidad and
Tobago.

Step 2: Remove countries where imputing PPPs would lead to even larger changes

For the countries identified in steps 1a and 1b, we predict PPPs using the ICP model (while
excluding all the countries that have been identified in step 1). If the predicted PPP makes the delta
ratio more extreme, the country is removed from the list of countries worthwhile a deeper look.
The reason for this is that our proposed solution to dealing with countries selected as outliers will
be to partially rely on imputed PPPs. If imputed PPPs make the country an even stronger outlier,
our proposed solution will not improve the situation. From Table D.1, using imputed PPPs would
not improve the situation for Angola, Central African Republic, Ghana, Jordan, and Liberia, so
they are excluded from the subsequent analysis.

Table D.1: Delta ratios with official and imputed 2017 PPP

Delta ratio Delta ratio

Country
increases/decreases further?
Official PPP Imputed PPP
1. Angola 1.80 2.14 Yes
2. Belize 0.83 1.05 No
3. Bolivia 1.43 1.33 No
4. Central African Republic 1.24 1.93 Yes
5. Egypt, Arab Rep. 1.77 0.88 No
6. Ghana 0.75 0.73 Yes
7. Guinea 1.46 1.27 No
8. Iraq 1.90 1.70 No
9. Jordan 1.53 1.57 Yes
10. Liberia 1.50 1.81 Yes
11. Nigeria 1.44 1.18 No
12. São Tomé and Príncipe 1.46 1.30 No
13. Suriname 1.45 0.96 No
14. Trinidad and Tobago 1.43 1.17 No
Note: The countries whose delta ratios become more extreme with the imputed PPP are in bold. When the delta ratio
increases further, poverty decreases further. When the delta ratio decreases further, poverty increases further. Also see
Figure D.3, panel (A) below.

3: Criteria to suggest that the 2017 PPPs may be problematic

We use four criteria to evaluate whether the 2017 PPP for a country may be problematic. Given
that no information can cast definite doubt on the 2017 PPPs—if so, the ICP team would have

56
dealt with the issue already—each of these criteria provides a signal of evidence on whether the
2017 PPPs are fit for global poverty monitoring.

a) Are the CPIs fit for purpose?

As described in the main text, an extreme delta ratio could arise from issues with either
PPP or the CPI. We evaluate the quality of the CPI by whether it classifies expenditures
using COICOP, and whether the weights have national coverage and are no older than 10
years from 2017. If none of these issues applies, we do not have evidence that there is a
problem with the CPIs, which may suggest that the problem could be with the PPPs,
creating an argument for imputing PPPs (see Table D.2).

Table D.2: CPI quality indicators

CPI
CPI expenditure CPI weights
Country classification Flag
weights coverage reference year
system
1. Belize COICOP National 2009 Yes
2. Bolivia Other National 2016 No
3. Djibouti Other Capital City 2013 No
4. Egypt, Arab Rep. COICOP National 2010 Yes
5. Guinea Other Capital City 2002 No
6. Iraq COICOP National 2012 Yes
7. Lao PDR Other National 2010 No
8. Myanmar Other National 2012 No
9. Nigeria Other National 2004 No
10. São Tomé and Príncipe COICOP National 2014 Yes
11. Sudan Other National 2007 No
12. Suriname COICOP National 2014 Yes
13. Trinidad and Tobago COICOP National 2009 Yes
14. Angola COICOP National 2009 Yes
15. Central African Republic Other National 2005 No
16. Ghana Survey National 2005.67 No
17. Jordan Other National 2010 No
18. Liberia COICOP National 2016 Yes
Note: Countries with a “Yes” have no obvious problems with their CPIs. Countries dropped in step 2 above are still
added (in italics) for completeness.

b) Is there a large decline in the share of items priced?

The share of items priced is defined as the ratio of the sum of global core list and regional
items priced to the sum of all global and regional items. Figure D.2 plots the share of items

57
 priced in the 2017 round against the 2011 round. For countries in red, the log difference in
the share of items priced in 2011 and 2017 is more than two standard deviations from the
mean (evaluated over all countries, not just the ones plotted). Among the countries
identified above, only Angola shows up as an outlier by this criterion. In Angola, the share
of items priced almost doubles between the two rounds, which is an improvement. None
of the countries under consideration has a large decline in the share of items priced.

Figure D.2: Share of items priced: 2011 vs. 2017 rounds

Note: Countries with a marked change in the share of items priced are shown in red, including Angola (AGO),
Bermuda (BMU), Bonaire (BON), Equatorial Guinea (GNQ), Jamaica (JAM), and Panama (PAN). Outlier
countries under consideration are shown in blue, including those that should be dropped based on step 2. All
other countries that participated in the 2011 and 2017 rounds are suppressed. The dotted line is a 45-degree
line.

c) Does the imputed PPP make the poverty rate more consistent with related indicators that
do not rely on PPPs, such as the multidimensional poverty index (MPI) and the age
dependency ratio (ADR)?

58
 We regress the monetary poverty rate (in 2017 PPPs with a $2.15 IPL) on the related
indicator (MPI or ADR) with regional dummies and interactions. The countries identified
in step 1 are excluded from the regression. Using the estimated parameters, we then predict
the monetary poverty rate for these countries. Finally, we compute the difference between
the predicted poverty rate and (a) the poverty rate with official 2017 PPP and (b) the
poverty rate with imputed 2017 PPP. We create a flag for a country if the former difference
is greater than the latter difference (see Tables D.3 and D.4). Such a case would imply that
the poverty rate with imputed 2017 PPPs is closer to our MPI- or ADR-based priors than
the poverty rate with official 2017 PPPs.

Table D.3: Multidimensional poverty index and monetary poverty

Log difference between Log difference between
predicted poverty from predicted poverty from
Country Year Flag
MPI and poverty with MPI and poverty with
official 2017 PPP imputed 2017 PPP
1. Belize 2015 1.79 1.45 Yes
2. Bolivia 2008 0.12 0.06 Yes
3. Djibouti 2006 1.37 0.56 Yes
4. Egypt, Arab Rep. 2014 2.21 2.43 No
5. Guinea 2018 1.36 0.96 Yes
6. Iraq 2018 3.41 2.17 Yes
7. Lao PDR 2017 0.57 0.77 No
8. Myanmar 2015 1.37 1.54 No
9. Nigeria 2018 0.26 0.08 Yes
10. São Tomé and Príncipe 2014 0.21 0.43 No
11. Sudan 2014 1.42 0.37 Yes
12. Suriname 2018 1.84 2.05 No
13. Trinidad and Tobago 2011 1.82 1.46 Yes
14. Angola 2015 0.55 0.86 No
15. Central African Republic 2010 0.07 0.37 No
16. Ghana 2014 0.18 0.22 No
17. Jordan 2017 0.41 0.13 Yes
18. Liberia 2013 0.46 0.95 No
Notes: We use the multidimensional poverty index (MPI) from the Oxford Poverty and Human Development Initiative
(OPHI) for the latest available year, which is indicated in the table (Alkire et al. 2020). These estimates are matched
with the lined-up, annual estimates of poverty in PovcalNet. For countries with a “Yes”, the poverty rate with imputed
PPP is more in tune with what we would expect from the MPI than with the official PPP. Countries that are dropped
in step 2 above are still added (in italics) for completeness.

59
 Table D.4: Age dependency ratio and monetary poverty
Log difference Log difference
between predicted between predicted
Country Year poverty from ADR poverty from ADR Flag
and poverty with and poverty with
official 2017 PPP imputed 2017 PPP
1. Belize 1999 0.40 0.09 Yes
2. Bolivia 2019 0.76 0.59 Yes
3. Djibouti 2017 3.22 2.38 Yes
4. Egypt, Arab Rep. 2017 1.07 2.28 No
5. Guinea 2012 1.01 0.69 Yes
6. Iraq 2012 3.90 2.98 Yes
7. Lao PDR 2018 0.90 0.66 Yes
8. Myanmar 2017 0.66 0.38 Yes
9. Nigeria 2018 0.47 0.12 Yes
10. São Tomé and Príncipe 2017 0.43 0.18 Yes
11. Sudan 2014 1.23 0.19 Yes
12. Suriname 1999 1.25 1.53 No
13. Trinidad and Tobago 1992 1.20 0.55 Yes
14. Angola 2018 0.72 0.94 No
15. Central African Republic 2008 0.36 0.06 Yes
16. Ghana 2016 0.47 0.51 No
17. Jordan 2010 3.75 3.75 No
18. Liberia 2016 0.09 0.52 No
Notes: We use estimates of poverty for the latest survey year available in PovcalNet, which is indicated in the table.
These estimates are matched with annual estimates of ADR in WDI. For countries with a “Yes”, the poverty rate with
imputed PPP is more in tune with what we would expect from ADR than with the official PPP. We use “Age
dependency ratio, young (% of working-age population)”, the ratio of younger dependents (i.e., people younger than
15) to the working-age population (i.e., those aged 15-64) from WDI for the analysis reported here. The flags are the
same when we use “Age dependency ratio (% of working-age population)”, the ratio of all dependents (i.e., people
younger than 15 or older than 64) to the working-age population (i.e., those aged 15-64). Countries that should be
dropped based on step 2 above are still added (in italics) for completeness.

d) Does the poverty economist find the official 2017 PPP less appropriate?

Following Ferreira et al. (2016), we consulted with country poverty economists (PE)—
World Bank staff specialized in the measurement of poverty in a country—to assess
whether the imputed PPP seems more appropriate for the countries identified in step 1. Our
partial reliance on poverty economists reflects our view that the prior three steps might
have missed relevant country-specific information and nuances. There is, however, a trade-
off between using country-specific information and preserving comparability across
countries. For that reason, we consider the judgment of poverty economists as one input
alongside the other indicators. Poverty economists could not influence the set of countries

60
 we considered outliers nor the PPPs we use for countries not considered outliers. Their
responses are provided in Table D.5.

Table D.5: Summary of responses

Panel (A) Panel (B)
Flags
Proble
Delta PPP (sum of
Country Items m with
CPIs MPI ADR PE ratio residual panel A
priced 2011
outlier outlier Yes's)
PPP
1. Belize Yes No Yes Yes Yes Yes No No 4
2. Bolivia No No Yes Yes No Yes No No 2
3. Djibouti No No Yes Yes No No Yes No 2
4. Egypt, Arab
Yes No No No Yes Yes Yes Yes 2
Rep.
5. Guinea No No Yes Yes Yes Yes No No 3
6. Iraq Yes No Yes Yes Yes Yes No Yes 4
7. Lao PDR No No No Yes Yes Yes No Yes 2
8. Myanmar No No No Yes No Yes No Yes 1
9. Nigeria No No Yes Yes Yes Yes No No 3
10. São Tomé and
Yes No No Yes Yes Yes No No 3
Príncipe
11. Sudan No No Yes Yes Yes No Yes No 3
12. Suriname Yes No No No N/A Yes Yes No 1
13. Trinidad and
Yes No Yes Yes Yes Yes No No 4
Tobago
14. Angola Yes No No No No Yes No No 1
15. Central
No No No Yes No Yes Yes No 1
African Republic
16. Ghana No No No No No Yes No No 0
17. Jordan No No Yes No No Yes No Yes 1
18. Liberia Yes No No No No Yes No No 1
Note: In panel (A), we have the factors identified in Step 3 providing suggestive evidence as to whether there might
be issues with the official 2017. In panel (B), we include the questions raised in Step 1 with which we selected the 18
problem countries. There is no poverty economist for Suriname.

Step 4: Decide on when to use imputed PPPs

a) For the 18 countries identified, we consider two options: either the official PPP, or the
average between the imputed PPP (using the ICP model) and the official PPP. We prefer
the average of official and imputed, instead of the imputed PPP by itself, because we think
both PPPs represent signals that are worthy accounting for. The choice between these
options is guided by the analysis in step 3 (see Table D.5 above). The more flags a country
has, the more likely we are to decide not to use the official PPP.

61
 Figure D.3: Poverty estimates in 2017 with different PPPs

Notes: This chart shows extreme poverty estimates in 2017 using revised 2011 and 2017 PPPs. Panel (A) shows the
poverty estimates for countries dropped in step 2 because using an imputed PPP would make the delta ratio even more
extreme. Panel (B) shows outlier countries where there is little evidence against using the official PPP to estimate
poverty. Panel (C) shows outlier countries where we propose to use the average of official and imputed PPPs. This
figure uses the international poverty lines of $1.90 (2011 PPP) and $2.15 (2017 PPP) for estimating extreme poverty.
See Table D.6 below for the poverty estimates and associated delta ratios.

62
 In the end, we have decided to use the average of official and imputed PPPs for Belize,
Egypt, Guinea, Iraq, Nigeria, São Tomé and Príncipe, Sudan, and Trinidad and Tobago,
while for the remaining outlier countries we use the official 2017 PPPs. This is equivalent
to all countries with at least three flags and Egypt. According to the official PPP, Egypt has
the lowest price level index in the world (World Bank 2020) and Egypt thus represents an
extreme outlier. Egypt is also the only country that is identified as possibly having a
problematic 2017 PPP estimate based on the delta criterion, PPP residual criterion, and
countries with exceptional 2011 PPPs (see Table B.5, panel (B)).

See Figure D.3 and Table D.6 for the poverty estimates for all 18 countries under review for the
year 2017 with different PPPs.

Table D.6: Delta ratios and poverty changes

2011 2017 PPP (official) 2017 PPP (imputed) 2017 PPP (average)
Country PPP
Poverty Change Delta Poverty Change Delta Poverty Change Delta
Poverty
(A) Use official 2017 PPP: Imputed PPP leads to even larger changes
Jordan 0.10 0.04 -0.06 1.53 0.03 -0.07 1.57
Ghana 12.34 24.68 12.33 0.75 25.72 13.38 0.73
Liberia 44.55 27.64 -16.91 1.50 17.97 -26.58 1.81
Angola 45.07 26.50 -18.57 1.80 20.41 -24.66 2.14
Central African Republic 72.37 69.03 -3.35 1.24 49.44 -22.93 1.93
(B) Use official 2017 PPP: Official PPP seems appropriate
Myanmar 1.36 1.99 0.63 1.05 1.51 0.14 1.10
Bolivia 6.55 4.93 -1.61 1.43 5.68 -0.87 1.33
Lao PDR 10.72 7.61 -3.11 1.24 6.26 -4.46 1.31
Djibouti 17.01 19.08 2.07 1.06 8.24 -8.77 1.63
Suriname 17.53 15.53 -2.01 1.45 19.30 1.77 0.96
(C) Use average 2017 PPP
Trinidad and Tobago 0.28 0.17 -0.10 1.43 0.26 -0.02 1.17 0.21 -0.07 1.29
Iraq 1.24 0.02 -1.22 1.90 0.09 -1.15 1.70 0.05 -1.19 1.80
Egypt, Arab Rep. 2.97 0.31 -2.65 1.77 9.78 6.81 0.88 1.93 -1.04 1.24
Sudan 10.22 7.39 -2.83 1.26 21.78 11.56 0.87 12.72 2.50 1.05
Belize 13.66 20.05 6.39 0.83 14.39 0.73 1.05 17.50 3.83 0.93
Guinea 24.33 12.67 -11.66 1.46 18.84 -5.49 1.27 15.90 -8.43 1.36
São Tomé and Príncipe 35.64 22.94 -12.70 1.46 29.20 -6.44 1.30 26.12 -9.52 1.38
Nigeria 41.36 28.09 -13.27 1.44 38.86 -2.50 1.18 33.41 -7.96 1.31
Notes: Poverty is in percentages (%), and poverty change is in percentage points (pp). Poverty estimates with the
average of official and imputed PPPs are not applicable for panels (A) and (B). Countries are sorted in ascending order
of poverty rate, as measured by the revised 2011 PPP.

63
 Figure D.4: Sensitivity of extreme poverty trends to alternative PPPs

Notes: This figure shows the trends in extreme poverty at the global and regional levels using the revised 2011 PPPs
and 2017 PPPs. The solid red line uses 2017 PPPs officially published by the ICP, except for 14 countries where
alternative PPPs are used (see Table B.3), while the dashed red line uses only 2017 PPPs officially published by the
ICP.

Table D.7: Sensitivity of changes in extreme poverty in 2017 to alternative PPPs

A. 2017 PPP B. 2017 PPP (ICP)
Region Change in Change in Change in Change in
poverty, pp millions of poor poverty, pp millions of poor
World -0.20 -15.36 -0.41 -30.96
East Asia & Pacific 0.41 8.45 0.41 8.45
Europe & Central Asia 1.48 7.26 1.48 7.26
Latin America & Caribbean 0.31 1.93 0.31 1.94
Middle East & North Africa 0.02 0.09 -0.72 -2.76
Rest of the World 0.00 0.01 0.00 0.01
South Asia 0.06 1.01 0.06 1.01
Sub-Saharan Africa -3.25 -34.10 -4.46 -46.86
Notes: This table shows the changes to extreme poverty estimates and millions of poor when moving from the revised
2011 PPPs to the 2017 PPPs. Extreme poverty is estimated using $1.90 (2011 PPP) and $2.15 (2017 PPP). Panel (A)
uses 2017 PPPs officially published by the ICP, except for 14 countries where alternative PPPs are used (see Table
B.3), while panel (B) uses only 2017 PPPs officially published by the ICP.

64
Appendix E. Methodological details

Section 1: Update $1.90 with revised 2011 PPPs

Equation 1 expresses how the national poverty lines of the 15 poor countries denominated in the
original 2011 PPPs were originally derived by Ferreira et al. (2016) from the national poverty lines
estimated in Ravallion et al. (2009).

𝐶𝑃𝐼2011 𝑃𝑃𝑃2005
𝑃𝐿2011𝑖 = 𝑃𝐿2005𝑖 * 𝐶𝑃𝐼2005𝑖 * 𝑃𝑃𝑃2011𝑖 (1)
𝑖 𝑖
where:
• 𝑃𝐿2011 is the national poverty line in the original 2011 PPP
• 𝑃𝐿2005 is national poverty line Ravallion et al. (2009) estimated in 2005 PPP
• 𝐶𝑃𝐼2011 is consumer price index (CPI) for 2011 as used by Ferreira et al. (2016)
• 𝐶𝑃𝐼2005 is consumer price index (CPI) for 2005 as used by Ferreira et al. (2016)
• 𝑃𝑃𝑃2011 is the original 2011 PPP
• 𝑃𝑃𝑃2005 is the 2005 PPP

The international poverty line (IPL), denominated in the original 2011 PPP, is given as:
1
𝐼𝑃𝐿2011 = 15 ∗ ∑15
𝑖=1 𝑃𝐿2011𝑖 (2)

We use the following equation to update the IPL in (2) with the revised 2011 PPPs only.
𝐶𝑃𝐼2011 𝑃𝑃𝑃2005
𝑃𝐿2011𝑖 = 𝑃𝐿2005𝑖 * 𝐶𝑃𝐼2005𝑖 * 𝑃𝑃𝑃𝑟2011𝑖 (3)
𝑖 𝑖
where:
• 𝑃𝑃𝑃𝑟2011 is the revised 2011 PPP

We use the following equation to update the IPL in (2) with CPI revisions only.
𝐶𝑃𝐼𝑟2011 𝑃𝑃𝑃2005𝑖
𝑃𝐿2011𝑖 = 𝑃𝐿2005𝑖 * 𝐶𝑃𝐼𝑟2005𝑖 * (4)
𝑖 𝑃𝑃𝑃2011𝑖
where:
• 𝐶𝑃𝐼𝑟2011 is the updated consumer price index (CPI) for 2011
• 𝐶𝑃𝐼𝑟2005 is the updated consumer price index (CPI) for 2005

We use the following equation to update the IPL in (2) with both PPP and CPI revisions.
𝐶𝑃𝐼𝑟2011 𝑃𝑃𝑃2005
𝑃𝐿2011𝑖 = 𝑃𝐿2005𝑖 * 𝐶𝑃𝐼𝑟2005𝑖 * 𝑃𝑃𝑃𝑟2011𝑖 (5)
𝑖 𝑖

65
Section 2: Update $1.90 with 2017 PPPs

Using equation (5) above, the following expression updates the 15 national poverty lines selected
by Ravallion et al. (2009) to 2017, while updating the CPIs used by Ferreira et al. (2016)

𝐶𝑃𝐼𝑟2011 𝐶𝑃𝐼2017 𝑃𝑃𝑃2005 𝑃𝑃𝑃𝑟2011𝑖

𝑃𝐿2017𝑖 = 𝑃𝐿2005𝑖 * 𝐶𝑃𝐼𝑟2005𝑖 *𝐶𝑃𝐼𝑟2011𝑖 * 𝑃𝑃𝑃𝑟2011𝑖 * (6)
𝑖 𝑖 𝑖 𝑃𝑃𝑃2017𝑖

Here 𝑃𝐿2017 is the national poverty line in 2017 PPP. Similarly, the following equation updates
the poverty lines to 2017 while using the CPIs that Ferreira et al. (2016) had originally used,

𝐶𝑃𝐼2011 𝐶𝑃𝐼2017 𝑃𝑃𝑃2005 𝑃𝑃𝑃𝑟2011𝑖

𝑃𝐿2017𝑖 = 𝑃𝐿2005𝑖 * 𝐶𝑃𝐼2005𝑖 ∗ 𝐶𝑃𝐼𝑟2011𝑖 ∗ 𝑃𝑃𝑃𝑟2011𝑖 * (7)
𝑖 𝑖 𝑖 𝑃𝑃𝑃2017𝑖

Lastly, when we take $1.90 as given for all 15 countries, and update both PPP and CPI to 2017 we
use the following expression:

𝑃𝑃𝑃𝑟2011𝑖 𝐶𝑃𝐼2017
𝑃𝐿2017𝑖 = 1.9 * * 𝐶𝑃𝐼𝑟2011𝑖 (8)
𝑃𝑃𝑃2017𝑖 𝑖

In all these three cases, the international poverty line (IPL), denominated in 2017 PPP, is given as:

1
𝐼𝑃𝐿2017 = 15 ∗ ∑15
𝑖=1 𝑃𝐿2017𝑖 (9)

Section 3: Derive harmonized national poverty lines

The harmonized national poverty line is derived from the following equation.
𝑧
𝐹(𝑧) = ∫0 𝑓(𝑦) 𝑑𝑦 = 𝑃 (10)
where:
• 𝐹 is a cumulative distribution function of income, denominated in original or revised 2011
PPP per person per day (as in PovcalNet)
• 𝑓(𝑦) is a probability distribution function of consumption or income, denominated in
original or revised 2011 PPP per person per day (as in PovcalNet)
• 𝑃 is the national poverty headcount ratio reported in WDI
• 𝑧 is the harmonized national poverty line that yields a poverty headcount ratio of 𝑃

Section 4: Update harmonized poverty lines with revised 2011 PPPs

Let 𝑧2011 be 𝑧 in equation (10) if the income distribution is denominated in the original 2011
PPP, and let 𝑧𝑟2011 be 𝑧 in equation (10) above if the income distribution is denominated in the
revised 2011 PPP. Then

66
 𝑃𝑃𝑃2011
𝑧𝑟2011 = 𝑧2011 * 𝑃𝑃𝑃𝑟2011 (11)

Section 5: Update harmonized poverty lines with 2017 PPPs

The harmonized poverty line, denominated in 2017 PPP, is derived as:
𝑃𝑃𝑃𝑟2011 𝐶𝑃𝐼2017
𝑧2017 = 𝑧𝑟2011 * * 𝐶𝑃𝐼2011 (12)
𝑃𝑃𝑃2017

Section 6: Derive equivalent poverty lines

Let the global poverty rate in a reference year (e.g., 2010) be given as:
1.9
𝐹(1.9) = ∫0 𝑓(𝑦(𝑃𝑃𝑃2011)) 𝑑𝑦 = 𝑃∗ (13)
with an international poverty line of $1.90 per person per day in revised 2011 PPP and a global
welfare (probability distribution) function, 𝑓(𝑦(. )) expressing daily income per person in revised
2011 PPP. 𝑃∗ is the global poverty rate.

The equivalent poverty line in 2017 PPP, 𝑧̃ , is derived by solving the following equation for a
given reference year.
𝑧̃
𝐹(𝑧̃ ) = ∫0 𝑓(𝑦(𝑃𝑃𝑃2017)) 𝑑𝑦 = 𝑃∗ (14)
with a global welfare (probability distribution) function, 𝑓(𝑦(. )) expressing daily income per
person in 2017 PPP.

Section 7: Convert PovcalNet distributions from revised 2011 PPPs to original 2011 PPPs
The income distribution, denominated in original 2011 PPP, is derived as:
𝑃𝑃𝑃𝑟2011
𝑦2011 = 𝑦_𝑟2011 * (15)
𝑃𝑃𝑃2011
where:
• 𝑦_𝑟2011 is the income distribution, denominated in revised 2011 PPP

Section 8: Convert PovcalNet distributions from revised 2011 PPPs to 2017 PPPs
The income distribution, denominated in 2017 PPP, is derived as:
𝑃𝑃𝑃𝑟2011 𝐶𝑃𝐼2017
𝑦2017 = 𝑦_𝑟2011 * * 𝐶𝑃𝐼𝑟2011 (16)
𝑃𝑃𝑃2017

Section 9: Explain changes in poverty with adopting the 2017 PPPs

The change in poverty is a function of the relative change in income per capita (given a poverty
line and the distribution). Re-arranging (16),
𝑦2017 𝐶𝑃𝐼2017 𝑃𝑃𝑃2017
= ⁄ = 𝛿 (17)
𝑦_𝑟2011 𝐶𝑃𝐼2011 𝑃𝑃𝑃𝑟2011
Thus, the rate of change in income per capita depends on the rates of change in CPI and PPP. If,
for example, the PPP increases faster than the CPI between the 2011 and 2017 ICP reference years,
income per capita decreases and poverty increases at a given poverty line.

67
Appendix F. Global poverty lines in local currency units

The following table expresses the four global poverty lines typical of low-income countries (LIC),
lower-middle-income countries (LMIC), upper-middle-income countries (UMIC), and high-
income countries (HIC) in local currency units in 2020 prices. Global poverty lines determined
from both revised 2011 and 2017 PPP rounds are provided in the table.

For example, the 2011 PPP exchange rate for Albania is 54.65 Albanian lek per US dollar, and the
inflation in Albania from 2011 to 2020 was 16.97%. This means that the $1.90 line in 2011 USD
lek
equates 1.90$ ∗ 54.65 ∗ 1.1697 = 121.46 lek in 2020 prices, which is the first entry in the
$
table below.

Table F.1: Global poverty lines in local currency units in 2020 prices
Economy PPP LIC LMIC UMIC HIC
Albania 2011 121.46 204.56 351.59 1387.19
Albania 2017 113.84 193.26 362.70 1289.82
Algeria 2011 88.81 149.58 257.09 1014.33
Algeria 2017 89.44 151.84 284.96 1013.37
Angola 2011 627.76 1057.28 1817.20 7169.70
Angola 2017 394.99 670.56 1258.45 4475.30
Argentina 2011 85.10 143.33 246.35 971.96
Argentina 2017 68.04 115.51 216.78 770.91
Armenia 2011 382.20 643.70 1106.36 4365.09
Armenia 2017 378.64 642.81 1206.37 4290.11
Australia 2011 3.39 5.71 9.81 38.72
Australia 2017 3.43 5.83 10.94 38.90
Austria 2011 1.87 3.16 5.43 21.41
Austria 2017 1.88 3.18 5.98 21.25
Azerbaijan 2011 0.92 1.55 2.66 10.49
Azerbaijan 2017 1.13 1.92 3.60 12.79
Bangladesh 2011 76.02 128.04 220.06 868.25
Bangladesh 2017 74.74 126.89 238.13 846.85
Belarus 2011 1.23 2.08 3.57 14.08
Belarus 2017 1.55 2.63 4.94 17.55
Belgium 2011 1.90 3.20 5.50 21.70
Belgium 2017 1.90 3.22 6.04 21.47
Belize 2011 2.33 3.93 6.75 26.64
Belize 2017 2.83 4.81 9.02 32.07
Benin 2011 475.67 801.12 1376.93 5432.61
Benin 2017 485.82 824.77 1547.85 5504.48
Bhutan 2011 53.19 89.58 153.96 607.45
Bhutan 2017 48.86 82.95 155.68 553.63

68
 Table F.1: Global poverty lines in local currency units in 2020 prices (continued)
Economy PPP LIC LMIC UMIC HIC
Bolivia 2011 7.48 12.60 21.66 85.47
Bolivia 2017 5.90 10.02 18.81 66.89
Bosnia and Herzegovina 2011 1.63 2.74 4.71 18.56
Bosnia and Herzegovina 2017 1.70 2.88 5.40 19.22
Botswana 2011 11.39 19.19 32.98 130.11
Botswana 2017 11.64 19.76 37.09 131.90
Brazil 2011 5.08 8.56 14.71 58.05
Brazil 2017 5.55 9.43 17.69 62.92
Bulgaria 2011 1.68 2.84 4.87 19.23
Bulgaria 2017 1.70 2.89 5.43 19.31
Burkina Faso 2011 455.35 766.91 1318.13 5200.63
Burkina Faso 2017 431.67 732.84 1375.32 4890.93
Burundi 2011 1681.06 2831.25 4866.22 19199.44
Burundi 2017 1457.43 2474.25 4643.45 16513.07
Cabo Verde 2011 97.16 163.64 281.26 1109.69
Cabo Verde 2017 105.43 178.99 335.92 1194.60
Cameroon 2011 544.41 916.90 1575.91 6217.70
Cameroon 2017 540.32 917.28 1721.48 6121.92
Canada 2011 2.78 4.69 8.06 31.80
Canada 2017 2.90 4.93 9.26 32.91
Central African Republic 2011 774.54 1304.49 2242.10 8846.10
Central African Republic 2017 704.13 1195.38 2243.38 7977.92
Chad 2011 572.98 965.02 1658.63 6544.07
Chad 2017 551.18 935.72 1756.08 6244.98
Chile 2011 966.73 1628.17 2798.42 11041.05
Chile 2017 1114.85 1892.64 3551.95 12631.46
China 2011 8.53 14.36 24.68 97.38
China 2017 9.59 16.28 30.56 108.66
Colombia 2011 3215.11 5414.92 9306.89 36719.89
Colombia 2017 3343.95 5676.94 10653.98 37887.74
Comoros 2011 493.11 830.50 1427.42 5631.81
Comoros 2017 488.03 828.52 1554.89 5529.52
Congo, Dem. Rep. 2011 2440.70 4110.65 7065.17 27875.32
Congo, Dem. Rep. 2017 2043.47 3469.15 6510.59 23152.99
Congo, Rep. 2011 720.26 1213.07 2084.96 8226.10
Congo, Rep. 2017 647.20 1098.73 2062.00 7332.88
Costa Rica 2011 821.50 1383.58 2378.03 9382.40
Costa Rica 2017 842.26 1429.89 2683.49 9543.05
Côte d'Ivoire 2011 504.03 848.89 1459.02 5756.51
Côte d'Ivoire 2017 540.13 916.97 1720.89 6119.85
Croatia 2011 8.63 14.54 24.99 98.59
Croatia 2017 8.40 14.26 26.77 95.19
Cyprus 2011 1.42 2.39 4.12 16.24
Cyprus 2017 1.45 2.46 4.62 16.45

69
 Table F.1: Global poverty lines in local currency units in 2020 prices (continued)
Economy PPP LIC LMIC UMIC HIC
Czech Republic 2011 32.34 54.47 93.63 369.41
Czech Republic 2017 31.81 54.00 101.35 360.41
Denmark 2011 17.08 28.77 49.45 195.11
Denmark 2017 17.31 29.38 55.14 196.08
Djibouti 2011 222.70 375.08 644.66 2543.48
Djibouti 2017 237.14 402.59 755.55 2686.89
Dominican Republic 2011 51.15 86.15 148.07 584.22
Dominican Republic 2017 57.38 97.42 182.82 650.16
Ecuador 2011 1.23 2.08 3.57 14.08
Ecuador 2017 1.23 2.09 3.92 13.95
Egypt, Arab Rep. 2011 15.03 25.31 43.50 171.62
Egypt, Arab Rep. 2017 13.66 23.19 43.52 154.77
El Salvador 2011 1.08 1.81 3.11 12.29
El Salvador 2017 1.12 1.89 3.56 12.64
Estonia 2011 1.28 2.15 3.70 14.58
Estonia 2017 1.35 2.29 4.30 15.30
Eswatini 2011 12.76 21.49 36.94 145.73
Eswatini 2017 14.69 24.94 46.81 166.47
Ethiopia 2011 30.81 51.88 89.17 351.84
Ethiopia 2017 28.99 49.22 92.36 328.47
Fiji 2011 2.32 3.91 6.71 26.49
Fiji 2017 2.19 3.71 6.97 24.78
Finland 2011 2.01 3.39 5.83 23.00
Finland 2017 2.06 3.49 6.55 23.28
France 2011 1.81 3.05 5.24 20.67
France 2017 1.86 3.15 5.92 21.06
Gabon 2011 836.53 1408.90 2421.54 9554.08
Gabon 2017 745.16 1265.05 2374.13 8442.89
Gambia, The 2011 36.20 60.97 104.79 413.43
Gambia, The 2017 39.28 66.68 125.14 445.02
Georgia 2011 2.06 3.46 5.95 23.47
Georgia 2017 2.25 3.82 7.17 25.50
Germany 2011 1.75 2.94 5.05 19.94
Germany 2017 1.76 2.98 5.59 19.89
Ghana 2011 2.95 4.97 8.55 33.72
Ghana 2017 4.45 7.56 14.19 50.47
Greece 2011 1.42 2.39 4.12 16.24
Greece 2017 1.38 2.35 4.41 15.68

70
 Table F.1: Global poverty lines in local currency units in 2020 prices (continued)
Economy PPP LIC LMIC UMIC HIC
Guatemala 2011 10.27 17.30 29.73 117.31
Guatemala 2017 10.51 17.85 33.50 119.12
Guinea 2011 11838.66 19938.80 34269.82 135210.00
Guinea 2017 9854.55 16729.82 31397.07 111654.39
Guinea-Bissau 2011 534.37 899.99 1546.86 6103.08
Guinea-Bissau 2017 509.72 865.34 1624.00 5775.28
Guyana 2011 255.94 431.06 740.88 2923.11
Guyana 2017 258.90 439.52 824.86 2933.38
Haiti 2011 97.28 163.85 281.61 1111.08
Haiti 2017 109.64 186.13 349.32 1242.25
Honduras 2011 28.41 47.86 82.25 324.53
Honduras 2017 26.26 44.58 83.66 297.51
Hungary 2011 315.21 530.88 912.45 3600.02
Hungary 2017 347.97 590.75 1108.66 3942.64
Iceland 2011 341.30 574.82 987.98 3898.03
Iceland 2017 364.06 618.05 1159.91 4124.87
India 2011 49.17 82.81 142.33 561.56
India 2017 48.12 81.69 153.31 545.20
Indonesia 2011 10815.32 18215.28 31307.52 123522.39
Indonesia 2017 11858.56 20131.97 37781.92 134360.22
Iran, Islamic Rep. 2011 60201.79 101392.49 174268.34 687567.82
Iran, Islamic Rep. 2017 67542.39 114664.99 215193.20 765271.01
Iraq 2011 2019.23 3400.80 5845.13 23061.70
Iraq 2017 1270.34 2156.62 4047.36 14393.24
Ireland 2011 1.88 3.17 5.46 21.53
Ireland 2017 2.11 3.58 6.72 23.91
Israel 2011 8.40 14.15 24.31 95.93
Israel 2017 9.14 15.52 29.12 103.57
Italy 2011 1.66 2.80 4.82 19.00
Italy 2017 1.68 2.85 5.34 18.99
Jamaica 2011 189.92 319.86 549.76 2169.06
Jamaica 2017 171.65 291.41 546.90 1944.88
Japan 2011 231.41 389.75 669.88 2642.98
Japan 2017 246.49 418.46 785.34 2792.81
Jordan 2011 1.01 1.71 2.93 11.58
Jordan 2017 0.75 1.27 2.38 8.47
Kazakhstan 2011 279.37 470.52 808.71 3190.71
Kazakhstan 2017 320.79 544.59 1022.04 3634.57
Kenya 2011 118.85 200.17 344.05 1357.43
Kenya 2017 103.95 176.47 331.19 1177.77

71
 Table F.1: Global poverty lines in local currency units in 2020 prices (continued)
Economy PPP LIC LMIC UMIC HIC
Kiribati 2011 2.04 3.44 5.91 23.30
Kiribati 2017 2.12 3.60 6.75 24.01
Korea, Rep. 2011 1910.14 3217.08 5529.35 21815.79
Korea, Rep. 2017 2145.06 3641.61 6834.25 24304.00
Kosovo 2011 0.77 1.29 2.22 8.77
Kosovo 2017 0.80 1.35 2.54 9.03
Kyrgyz Republic 2011 43.94 74.01 127.20 501.85
Kyrgyz Republic 2017 45.48 77.21 144.90 515.29
Lao PDR 2011 8193.58 13799.72 23718.27 93579.36
Lao PDR 2017 7466.18 12675.14 23787.58 84593.50
Latvia 2011 1.21 2.04 3.50 13.80
Latvia 2017 1.26 2.14 4.01 14.25
Lebanon 2011 3613.81 6086.42 10461.04 41273.54
Lebanon 2017 3308.51 5616.78 10541.07 37486.20
Lesotho 2011 11.62 19.57 33.63 132.70
Lesotho 2017 13.04 22.14 41.55 147.76
Liberia 2011 250.73 422.27 725.78 2863.55
Liberia 2017 189.34 321.44 603.25 2145.28
Lithuania 2011 1.12 1.88 3.23 12.75
Lithuania 2017 1.12 1.90 3.56 12.65
Luxembourg 2011 2.09 3.52 6.05 23.85
Luxembourg 2017 2.16 3.67 6.89 24.50
Madagascar 2011 2333.90 3930.79 6756.04 26655.65
Madagascar 2017 2474.20 4200.39 7882.93 28033.31
Malawi 2011 743.51 1252.22 2152.25 8491.62
Malawi 2017 694.72 1179.40 2213.39 7871.28
Malaysia 2011 3.51 5.91 10.15 40.06
Malaysia 2017 3.73 6.33 11.88 42.24
Maldives 2011 20.26 34.12 58.65 231.40
Maldives 2017 20.79 35.29 66.22 235.51
Mali 2011 452.14 761.50 1308.83 5163.95
Mali 2017 437.23 742.27 1393.02 4953.88
Malta 2011 1.33 2.25 3.86 15.23
Malta 2017 1.40 2.37 4.45 15.83
Mauritania 2011 280.46 472.35 811.85 3203.12
Mauritania 2017 288.80 490.29 920.14 3272.20
Mauritius 2011 44.11 74.28 127.68 503.74
Mauritius 2017 41.16 69.88 131.15 466.39

72
 Table F.1: Global poverty lines in local currency units in 2020 prices (continued)
Economy PPP LIC LMIC UMIC HIC
Mexico 2011 23.90 40.24 69.17 272.91
Mexico 2017 23.83 40.46 75.93 270.03
Micronesia 2011 2.10 3.54 6.09 24.03
Micronesia 2017 2.18 3.70 6.95 24.72
Moldova 2011 16.46 27.72 47.65 188.01
Moldova 2017 15.38 26.11 49.01 174.29
Mongolia 2011 2065.69 3479.06 5979.63 23592.35
Mongolia 2017 2232.10 3789.38 7111.57 25290.20
Montenegro 2011 0.94 1.59 2.73 10.77
Montenegro 2017 0.92 1.57 2.95 10.48
Morocco 2011 8.69 14.63 25.15 99.23
Morocco 2017 9.49 16.11 30.23 107.49
Mozambique 2011 50.86 85.67 147.24 580.92
Mozambique 2017 52.08 88.41 165.92 590.03
Myanmar 2011 952.66 1604.49 2757.71 10880.43
Myanmar 2017 1030.63 1749.67 3283.62 11677.24
Namibia 2011 15.32 25.80 44.35 174.98
Namibia 2017 16.40 27.84 52.24 185.79
Nauru 2011 3.01 5.07 8.72 34.40
Nauru 2017 3.13 5.31 9.96 35.44
Nepal 2011 88.78 149.53 257.01 1014.01
Nepal 2017 75.71 128.53 241.21 857.81
Netherlands 2011 1.91 3.21 5.52 21.77
Netherlands 2017 1.93 3.28 6.15 21.88
Nicaragua 2011 27.44 46.21 79.43 313.38
Nicaragua 2017 28.42 48.25 90.55 322.00
Niger 2011 478.41 805.74 1384.86 5463.90
Niger 2017 544.56 924.49 1735.00 6170.01
Nigeria 2011 434.05 731.04 1256.47 4957.36
Nigeria 2017 376.01 638.34 1197.97 4260.23
North Macedonia 2011 49.54 83.44 143.41 565.80
North Macedonia 2017 49.31 83.71 157.10 558.68
Norway 2011 22.43 37.78 64.93 256.19
Norway 2017 24.56 41.70 78.26 278.29
Pakistan 2011 86.60 145.86 250.69 989.08
Pakistan 2017 91.16 154.75 290.43 1032.82
Panama 2011 1.19 2.01 3.45 13.63
Panama 2017 1.08 1.83 3.44 12.23

73
 Table F.1: Global poverty lines in local currency units in 2020 prices (continued)
Economy PPP LIC LMIC UMIC HIC
Papua New Guinea 2011 6.36 10.71 18.41 72.63
Papua New Guinea 2017 6.60 11.21 21.04 74.82
Paraguay 2011 5759.64 9700.45 16672.65 65781.19
Paraguay 2017 5937.75 10080.37 18917.96 67276.12
Peru 2011 3.84 6.46 11.10 43.81
Peru 2017 4.29 7.28 13.67 48.61
Philippines 2011 45.31 76.31 131.16 517.48
Philippines 2017 46.14 78.33 147.00 522.77
Poland 2011 4.07 6.86 11.79 46.53
Poland 2017 4.26 7.23 13.58 48.28
Portugal 2011 1.40 2.36 4.05 15.99
Portugal 2017 1.43 2.43 4.56 16.23
Romania 2011 4.13 6.96 11.96 47.18
Romania 2017 4.33 7.35 13.78 49.02
Russian Federation 2011 61.08 102.87 176.80 697.57
Russian Federation 2017 60.24 102.27 191.94 682.58
Rwanda 2011 768.69 1294.63 2225.15 8779.23
Rwanda 2017 714.66 1213.26 2276.94 8097.26
Samoa 2011 3.97 6.69 11.49 45.34
Samoa 2017 4.12 7.00 13.13 46.71
São Tomé and Príncipe 2011 38.16 64.27 110.47 435.84
São Tomé and Príncipe 2017 31.35 53.22 99.89 355.22
Senegal 2011 513.16 864.27 1485.47 5860.84
Senegal 2017 537.73 912.88 1713.22 6092.57
Serbia 2011 109.10 183.74 315.81 1246.01
Serbia 2017 108.84 184.77 346.77 1233.17
Seychelles 2011 20.06 33.79 58.07 229.11
Seychelles 2017 20.74 35.20 66.07 234.95
Sierra Leone 2011 8628.24 14531.77 24976.48 98543.55
Sierra Leone 2017 6915.81 11740.79 22034.09 78357.73
Slovak Republic 2011 1.19 2.01 3.46 13.64
Slovak Republic 2017 1.34 2.27 4.26 15.14
Slovenia 2011 1.40 2.36 4.06 16.03
Slovenia 2017 1.42 2.41 4.52 16.08
Solomon Islands 2011 17.29 29.12 50.06 197.50
Solomon Islands 2017 17.96 30.49 57.21 203.46
South Africa 2011 14.93 25.15 43.23 170.55
South Africa 2017 15.82 26.85 50.39 179.20

74
 Table F.1: Global poverty lines in local currency units in 2020 prices (continued)
Economy PPP LIC LMIC UMIC HIC
South Sudan 2011 429.05 722.62 1241.99 4900.23
South Sudan 2017 334.51 567.89 1065.78 3790.12
Spain 2011 1.61 2.70 4.65 18.33
Spain 2017 1.54 2.62 4.91 17.47
Sri Lanka 2011 126.36 212.82 365.78 1443.19
Sri Lanka 2017 133.94 227.39 426.74 1517.57
St. Lucia 2011 4.19 7.06 12.13 47.87
St. Lucia 2017 4.50 7.64 14.34 51.00
Sudan 2011 79.29 133.55 229.54 905.63
Sudan 2017 85.78 145.62 273.29 971.87
Suriname 2011 12.19 20.53 35.29 139.25
Suriname 2017 9.50 16.13 30.27 107.64
Sweden 2011 18.49 31.14 53.52 211.15
Sweden 2017 20.68 35.11 65.89 234.31
Switzerland 2011 2.88 4.86 8.35 32.94
Switzerland 2017 2.94 4.99 9.36 33.28
Syrian Arab Republic 2011 759.41 1279 2198.28 8673.21
Syrian Arab Republic 2017 788.58 1338.75 2512.45 8934.79
Taiwan, China 2011 32.98 55.55 95.48 376.71
Taiwan, China 2017 36.35 61.72 115.82 411.89
Tajikistan 2011 5.85 9.85 16.93 66.82
Tajikistan 2017 6.78 11.51 21.60 76.83
Tanzania 2011 1921.85 3236.81 5563.26 21949.60
Tanzania 2017 1794.46 3046.41 5717.24 20331.69
Thailand 2011 26.23 44.17 75.92 299.53
Thailand 2017 28.83 48.94 91.85 326.65
Timor-Leste 2011 1.23 2.08 3.57 14.10
Timor-Leste 2017 0.94 1.60 3.01 10.70
Togo 2011 500.81 843.47 1449.71 5719.77
Togo 2017 548.62 931.38 1747.93 6215.99
Tonga 2011 3.93 6.62 11.37 44.88
Tonga 2017 4.08 6.93 13.00 46.23
Trinidad and Tobago 2011 11.77 19.82 34.07 134.42
Trinidad and Tobago 2017 10.29 17.46 32.78 116.56
Tunisia 2011 2.10 3.54 6.09 24.03
Tunisia 2017 1.93 3.28 6.15 21.87
Turkey 2011 5.33 8.98 15.43 60.87
Turkey 2017 5.33 9.04 16.97 60.35
Turkmenistan 2011 5.11 8.60 14.78 58.31
Turkmenistan 2017 4.58 7.78 14.59 51.89

75
 Table F.1: Global poverty lines in local currency units in 2020 prices (continued)
Economy PPP LIC LMIC UMIC HIC
Tuvalu 2011 2.78 4.68 8.05 31.77
Tuvalu 2017 2.89 4.90 9.20 32.72
Uganda 2011 2812.38 4736.65 8141.11 32120.40
Uganda 2017 2876.69 4883.68 9165.27 32593.57
Ukraine 2011 16.20 27.29 46.90 185.06
Ukraine 2017 18.52 31.45 59.01 209.87
United Arab Emirates 2011 5.06 8.53 14.65 57.82
United Arab Emirates 2017 6.03 10.24 19.22 68.35
United Kingdom 2011 1.72 2.89 4.97 19.60
United Kingdom 2017 1.76 2.99 5.61 19.96
United States 2011 2.19 3.68 6.33 24.97
United States 2017 2.27 3.85 7.23 25.72
Uruguay 2011 65.02 109.51 188.21 742.59
Uruguay 2017 68.09 115.59 216.93 771.43
Uzbekistan 2011 3337.31 5620.74 9660.64 38115.62
Uzbekistan 2017 4968.84 8435.47 15830.96 56298.13
Vanuatu 2011 270.05 454.81 781.71 3084.20
Vanuatu 2017 280.42 476.06 893.43 3177.22
Venezuela, RB 2011 2103457343.59 3542664999.72 6088955468.27 24023697029.37
Venezuela, RB 2017 2184265599.02 3708171830.90 6959171792.23 24748237205.67
Vietnam 2011 20343.08 34262.03 58887.87 232339.41
Vietnam 2017 18441.94 31308.41 58756.87 208951.45
West Bank and Gaza 2011 5.03 8.47 14.56 57.43
West Bank and Gaza 2017 4.47 7.59 14.24 50.66
Yemen, Rep. 2011 931.33 1568.56 2695.96 10636.77
Yemen, Rep. 2017 967.11 1641.84 3081.26 10957.58
Zambia 2011 11.46 19.29 33.16 130.83
Zambia 2017 12.33 20.93 39.29 139.72
Zimbabwe 2011 12.95 21.82 37.50 147.94
Zimbabwe 2017 13.02 22.11 41.49 147.54
Notes: Poverty increases (decreases) when the global poverty line in local currency units increases (decreases) with
the 2017 PPP, relative to the revised 2011 PPP. These results are mechanically similar to the results from the analysis
of the delta ratio. For example, a relatively high delta ratio of 1.80 for Angola indicates a reduction in extreme poverty
(see Table D.6), as the local currency units of the IPL in 2017 PPP reduce relative to the revised 2011 PPP.

76
Appendix references

Alkire, S., U. Kanagaratnam, and N. Suppa. 2020. “The Global Multidimensional Poverty Index
(MPI).” 49. OPHI MPI Methodological Notes. Oxford: Oxford Poverty and Human
Development Initiative, University of Oxford.
Allen, Robert C. 2017. “Absolute Poverty: When Necessity Displaces Desire.” American
Economic Review 107 (12): 3690–3721. https://doi.org/10.1257/aer.20161080.
Atamanov, Aziz, Dean Jolliffe, Christoph Lakner, and Espen Beer Prydz. 2018. “Purchasing
Power Parities Used in Global Poverty Measurement.” Global Poverty Monitoring Technical
Note 5 (September).
Atamanov, Aziz, Christoph Lakner, Daniel Gerszon Mahler, Samuel Kofi Tetteh-Baah, and Judy
Yang. 2020. “The Effect of New PPP Estimates on Global Poverty: A First Look,” no. May.
Bai, Yan, and William A. Masters. 2020. “Retail Food Prices at Purchasing Power Parity Exchange
Rates: A First Look at Aggregate ICP 2017 Data.” World Bank Data Blog. 2020.
https://blogs.worldbank.org/opendata/retail-food-prices-purchasing-power-parity-exchange-
rates-first-look-aggregate-icp-2017.
Chen, Shaohua, and Martin Ravallion. 2010. “The Developing World Is Poorer than We Thought,
but No Less Successful in the Fight against Poverty.” The Quarterly Journal of Economics,
no. November: 1577–1625.
Deaton, Angus. 2001. “Counting the World’s Poor: Problems and Possible Solutions.” World Bank
Research Observer 16 (2): 125–47. https://doi.org/10.1093/wbro/16.2.125.
Deaton, Angus, and Bettina Aten. 2017. “Trying to Understand the PPPs in ICP 2011: Why Are
the Results so Different.” American Economic Journal: Macroeconomics 9 (1): 243–64.
https://doi.org/10.1257/mac.20150153.
Deaton, Angus, and Paul Schreyer. 2022. “GDP, Wellbeing, and Health: Thoughts on the 2017
Round of the International Comparison Program.” Review of Income and Wealth 68 (1): 1–
15. https://doi.org/10.1111/roiw.12520.
Ferreira, Francisco H.G., Shaohua Chen, Andrew Dabalen, Yuri Dikhanov, Nada Hamadeh, Dean
Jolliffe, Ambar Narayan, et al. 2016. “A Global Count of the Extreme Poor in 2012: Data
Issues, Methodology and Initial Results.” Journal of Economic Inequality 14 (2): 141–72.
https://doi.org/10.1007/s10888-016-9326-6.
Inklaar, Robert, and D S Prasada Rao. 2017. “Cross-Country Income Levels over Time: Did the
Developing World Suddenly Become Much Richer?” American Economic Journal:
Macroeconomics 9 (1): 265–90.
Inklaar, Robert, and Prasada Rao. 2020. “ICP PPP Time Series Implementation.” Washington,
DC. https://thedocs.worldbank.org/en/doc/f32e966db6ec404699b3381b32e2e589-
0050022021/original/2-01-RA-Item-01-ICP-PPP-Time-Series-Implementation-Rao-and-
Inklaar.pdf.
Jolliffe, Dean, and Espen Beer Prydz. 2016. “Estimating International Poverty Lines from
Comparable National Thresholds.” Journal of Economic Inequality 14 (2): 185–98.
https://doi.org/10.1007/s10888-016-9327-5.
OECD. 2013. “The 2008 SNA – Changes from the 1993 SNA.” In National Accounts at a Glance
2013, 90–92. Paris: OECD Publishing. https://doi.org/10.1787/na_glance-2010-28-en.
Ravallion, Martin. 2020. “Book Review of Measuring Poverty around the World.” The Journal of
Economic Inequality 18 (1): 131–36. https://doi.org/10.1007/s10888-019-09430-w.
Ravallion, Martin, Shaohua Chen, and Prem Sangraula. 2009. “Dollar a Day Revisited.” World

77
 Bank Economic Review 23 (2): 163–84. https://doi.org/10.1093/wber/lhp007.
Tetteh-Baah, Samuel Kofi, Aziz Atamanov, Christoph Lakner, Daniel Gerszon Mahler, Marko
Rissanen, William Vigil-Oliver, Mizuki Yamanaka, and Judy Yang. 2020. “Why Have the
2011 PPPs Been Revised and What Does It Mean for Estimates of Poverty?” World Bank
Data Blog, 2020. https://blogs.worldbank.org/opendata/why-have-2011-ppps-been-revised-
and-what-does-it-mean-estimates-poverty.
World Bank. 2008. Global Purchasing Power Parities and Real Expenditures: 2005 International
Comparison Program. Washington, DC: World Bank.
———. 2013. Measuring the Real Size of the World Economy: The Framework, Methodology,
and Results of the International Comparison Program (ICP). Washington, DC: World Bank.
https://doi.org/10.5860/choice.51-0989.
———. 2015. Purchasing Power Parities and the Real Size of World Economies: A
Comprehensive Report of the 2011 International Comparison Program. Washington, DC:
World Bank.
———. 2016. “Monitoring Global Poverty: A Cover Note to the Report of the Commission on
Global Poverty, Chaired by Prof. Sir Anthony B. Atkinson.” Washington, DC: World Bank.
http://pubdocs.worldbank.org/en/733161476724983858/MonitoringGlobalPovertyCoverNot
e.pdf.
———. 2020. Purchasing Power Parities and the Size of World Economies: Results from the 2017
International Comparison Program. Washington, DC: World Bank.
https://doi.org/10.1596/978-1-4648-1530-0.

78