Macroeconomic Effects of COVID-19 Across the World

Income Distribution ∗

Titan Alon Minki Kim David Lagakos Mitchell VanVuren

October 29, 2021

Abstract

The macroeconomic effects of the COVID-19 pandemic were most severe for emerging
market economies, representing the middle of the world income distribution. This paper
provides a quantitative economic theory for why emerging markets fared worse, on av-
erage, relative to advanced economies and low-income countries. To do so we adapt a
workhorse incomplete-markets macro model to include epidemiological dynamics along-
side key economic and demographic characteristics that distinguish countries of different
income levels. We focus in particular on differences in lockdown stringency, public in-
surance programs, age distributions, healthcare capacity, and the sectoral composition of
employment. The calibrated model predicts greater output declines in emerging markets,
as in the data, and greater excess mortality, albeit to a smaller extent than what is ob-
served in the data. Quantitatively, stricter lockdowns and a higher share of jobs requiring
social interaction explain a large fraction of the especially severe outcomes in emerging
markets. Low-income countries fared relatively better mainly due to their younger pop-
ulations, which are less susceptible to the disease, and larger agricultural sectors, which
require fewer social interactions.

∗
First draft - comments welcome. Titan Alon, Minki Kim, Mitchell VanVuren: Department of Eco-
nomics, University of California San Diego, 9500 Gilman Drive, La Jolla, CA 92093 (emails: talon@ucsd.edu,
minkikim@ucsd.edu, mvanvure@ucsd.edu). David Lagakos: NBER and Department of Economics, Boston Uni-
versity, 270 Bay State Road, Boston, MA 02215 (email: lagakos@bu.edu).
1. Introduction

While every country has been badly affected by the coronavirus pandemic, the damage it has
wrought varied widely around the world. In this paper, we investigate how and why the pan-
demic’s macroeconomic consequences have differed (so far) across the world income distri-
bution. We focus in particular on variation in output and excess mortality across three broad
groups of countries: low-income economies, emerging markets, and advanced economies, as
classified by the International Monetary Fund’s (IMF). As we detail below, data from a variety of
sources reveal that the pandemic’s cost in terms of lives and livelihoods was roughly U-shaped
in national income, with emerging markets experiencing the worst public health and macroe-
conomic consequences. For instance, GDP per capita in emerging markets declined by 6.7
percent on average from 2019 to 2020, compared to 2.4 percent in advanced economies and
3.6 percent in low-income countries. Excess mortality has exhibited a similar pattern. Accord-
ing to estimates by The Economist, excess mortality was 75 percent higher in emerging markets
than in advanced economies. While credible excess mortality data for low-income countries
are still largely unavailable, the few existing estimates similarly point to lower mortality rates
than in emerging markets.

We assess the extent to which policy responses and certain preexisting differences in economic
and demographic conditions can explain the cross-country variation we observe in the data. In
part, these outcomes could stem from differences in government policy responses to combat
the coronavirus pandemic. While most countries enacted similar “lockdown style” policies and
expanded social insurance programs, the scope of such efforts varied substantially. According
to the Oxford Coronavirus Government Response Tracker, the stringency of lockdown policies
aiming to restrict individual behavior (such as school and workplace closures) were most strict
in emerging markets. The generosity of social insurance programs, in contrast, appears to
increase linearly with a country’s GDP per capita. Accounting for these differences in policy is
important because they can directly affect both fatalities and growth during the pandemic.

The cross-country variation may also arise from the stark underlying differences in economic
and demographic characteristics that predate the pandemic. Low-income economies may have
faced very different public health risks than wealthier ones due to their substantially younger
populations but also their less developed healthcare systems. Furthermore, systematic varia-
tion in the sectoral composition of employment across the world income distribution creates
differences in the ability of workers to preserve income while mitigating health risks or coping
with extended lockdowns. Lower-income countries may benefit from large rural agricultural
sectors, which provide a resilient source of income that can be sustained while limiting con-
tacts. On the other hand, Gottlieb, Grobovsek, Poschke, and Saltiel (2021b) show that in urban

1
areas, the ability to work from home is far more limited in lower income countries. Combining
their estimates with data on urbanization rates, we can measure the share of labor in social and
non-social employment across countries, as in Kaplan, Moll, and Violante (2020), to capture
differences in the ability to work without social interactions. Our composite measure shows
that emerging markets have the highest share of workers in social employment, with their
largely urban workforces concentrated in high-contact sectors such as manufacturing and re-
tail trade. In contrast, low-income countries have the smallest social employment shares, due
to the predominance of rural agricultural work.

To investigate the extent to which these factors can explain the differential mortality and
output losses in the data, this paper follows the newly emerged literature on the macroe-
conomics of pandemics by combining a variant of the SICR model standard in epidemiology
with a workhorse macro model. In particularly, our model builds on the heterogeneous-agent
incomplete-markets model of Aiyagari (1994), Bewley (1977) and Huggett (1996). This set-
ting allows us to capture the individual trade-off between maintaining consumption levels and
preserving health that has been the focus of economic analysis of behavior during the pan-
demic. The model distinguishes between social and non-social jobs, so that individuals differ in
the ability to work effectively from home. We incorporate age heterogeneity following Glover,
Heathcote, Krueger, and Ríos-Rull (2020) and allow death rates to depend on the infected per-
son’s age, consistent with a vast medical literature. Our model also allows for a time-varying
infection rate that captures, in a reduced-form way, the various other non-modeled determi-
nants of disease progression, such as seasonal conditions, improved treatment, or virus mu-
tation. Finally, we include constraints on peak healthcare capacity, which capture differential
ability for healthcare systems to treat many patients at once, stemming from the availability of
hospital beds or supplemental oxygen.

In the model, the propagation of disease depends in large part on individual household choices
on whether or not to work from home. The model thus features a public health externality,
creating space for welfare improving government interventions. We model lockdown policy in
a simple way that is consistent with policy variation observed during the pandemic. Specif-
ically, we feed in time-varying lockdown measures that replicate the changing stringency of
government policies over the course of the pandemic, as measured by the Oxford Coronavirus
Government Response Tracker. In the model, lockdown policies confine individuals to their
home, where they are less likely to become infected but incur income losses depending on
their job type. More stringent lockdowns confine a larger share of the population to their
home. While we do not allow individuals to disobey lockdowns, households can voluntarily
elect to work from home at any point in time.

2
To evaluate the quantitative importance of each of these channels, and their interactions, in
explaining the facts at hand, we parameterize the model to match key pre-pandemic economic
and demographic characteristics of the United States. Parameters governing the epidemiologi-
cal process are set using estimates from the relevant medical literature. We compute the models
equilibrium response to the COVID-19 pandemic as a surprise “MIT shock,” where a small ex-
ogenous fraction of the population becomes infected with the virus, and then allow the disease
to spread endogenously through the populous. We feed in the time-series of vaccination rates,
as reported by OxCGRT, allowing a random fraction of the population to be vaccinated in each
period, consistent with rates we observe in the data. We set the non-parametric component
of the infection probability so that the model’s endogenous disease path (nearly) exactly repli-
cates the time-path of fatalities from COVID-19 in the United States during the pandemic. We
calibrate the productivity penalty incurred during lockdowns to match the cumulative 2019-
2020 year-on-year employment loss in the United States. We also allow for a one-off shock to
aggregate total factor productivity (TFP), which is calibrated to match the cumulative 2019-
2020 year-on-year decline in U.S. real GDP per capita.

We use the calibrated model to simulate how the United States would have fared during the
pandemic if it had counterfactually had the characteristics of emerging economies. Comparing
these counterfactual predictions to the actual outcomes allows us to assess the importance of
each characteristic in explaining the higher GDP declines and mortality rates in emerging mar-
kets. Including all emerging-market characteristics, the model predicts a substantially larger
decline in GDP during the pandemic, consistent with larger decline in emerging markets in the
data. The model also predicts a larger mortality rate with the emerging markets’ characteris-
tics, but quantitatively the gap is significantly smaller than in the data. The latter result implies
that the higher excess mortality in emerging markets was likely driven by factors other than
those modeled here, in particular the greater prevalence of social employment and lower ICU
capacity. Possible missing factors include other existing co-morbidities, less prevalent mask
use, or other other deficiencies in the medical system.

The final set of counterfactuals we run simulate the effects of the pandemic in the United
States assuming it had the features of low-income countries. We find that with the younger
demographics and sectoral composition of employment of low-income countries, the pandemic
would have been much less pronounced in terms of GDP declines and fatalities. The less intense
lockdowns and weaker ICU capacity both would have raised mortality, though only modestly.
The combined effects of all of these features lead to substantially lower mortality, which is
consistent with the limited available evidence on excess deaths in Africa.

We conclude by reporting multiple correlations between cross-country changes in GDP-per-

3
capita during the pandemic and covariates representing the channels embodied in our model.
The data show that agricultural employment shares are strong positive correlates of GDP
changes during the pandemic, while lockdown stringency is a strong negative correlate. Me-
dian age and an economic support index exhibit weaker correlations. Altogether, the covariates
greatly reduces the observed U-shape pattern in GDP declines across the world income distri-
bution. The result suggests that this parsimonious set of variables is empirically relevant in
explaining cross-country macroeconomic outcomes during the pandemic.

Taken together, our analysis suggest that the comparatively worse outcomes experienced by
emerging markets, and comparably better outcomes of low-income countries, may have been in
large part pre-determined by underlying economic and demographic conditions, rather than
by policy failures or successes during the pandemic. The greater size of the social sector in
emerging markets, which limited the ability of individuals to work from home, was an impor-
tant factor in their greater economic losses, whereas their somewhat younger age structure
had only a modest impact on their mortality rates. In low income countries, the large rural
agriculture sectors and young age structure was a central factor in keeping their GDP losses
and mortalities lower than they otherwise would have been. A valuable goal for future re-
search would be to help refine the quantitative importance of different policy decisions across
countries in determining macroeconomic outcomes during the pandemic.

Our work builds on the first generation of papers addressing the aggregate effects of COVID-
19 in the developing world, which were largely written in the early months of the pandemic
(Loayza and Pennings, 2020; Alon, Kim, Lagakos, and VanVuren, 2020; Alfaro, Becerra, and
Eslava, 2020; von Carnap, Almås, Bold, Ghisolfi, and Sandefur, 2020; Djankov and Panizza,
2020). The current paper differs in its efforts to explain observed macroeconomic outcomes
through the first year and a half of the pandemic, in particular the larger declines in GDP
and employment in emerging markets. Sanchez (2021) also notes the larger decline in GDP
middle-income countries, but does not attempt to explain this finding. We also emphasize
the inability of individuals in emerging market economies to work from home, following Got-
tlieb, Grobovsek, Poschke, and Saltiel (2021a,b), though we argue that low-income developing
countries, on account of their large agriculture sectors, are better able to work without social
interactions.

On the modeling front, our study most closely follows the structural macro work on the pan-
demic using models of heterogeneity in income, age and sector of employment (e.g. Acemoglu,
Chernozhukov, Werning, and Whinston, 2020; Bairoliya and Imrohoroglu, 2020; Kaplan, Moll,
and Violante, 2020; Glover, Heathcote, Krueger, and Ríos-Rull, 2020; Brotherhood, Kircher,
Santos, and Tertilt, 2021). Our model of disease dynamics features endogenous behavioral

4
responses to changes in infection rates, even in the absence of government intervention, as in
Greenwood, Kircher, Santos, and Tertilt (2019); Alvarez, Argente, and Lippi (2020); Krueger,
Uhlig, and Xie (2020) and other studies. To our knowledge ours is the first to evaluate the
quantitative predictions of a model of this sort for how the experience of emerging markets
differed from richer (or poorer) countries.

Our study abstracts from many important features of reality that may also be relevant for the
effects of the pandemic outside of the world’s advanced economies, such as negative impacts
through shocks to global supply chains (Cakmakli, Demiralp, and Ozcan, 2020; Bonadio, Huo,
Levchenko, and Pandalai-Nayar, 2021), the ability to issue sovereign debt (Arellano, Bai, and
Mihalache, 2020), or the ability to test and trace infections (Berger, Herkenhoff, and Mongey,
2020). We also abstract from differences in the prevalence of co-morbidities, such as diabetes
and cardiovascular disease, and differential ability or willingness or ability to mask or get
vaccinated. These issues would be valuable to consider in future studies trying to explain
cross-country differences in the macroeconomic effects of the pandemic.

2. Macroeconomic Effects of the Coronavirus Pandemic by Income Level

This section presents the main facts regarding excess mortality and output losses across the
world income distribution resulting from the coronavirus pandemic. Following the IMF classifi-
cation, we focus in particular on three major income groups: low-income economies, emerging
markets, and advanced economies. In 2019, the median GDP per capita of these three country
groups was $1,124, $6,700, and $43,144, respectively, in constant 2010 USD. While there is
interesting variation even with these group, we focus the main part of our analysis on just the
three aggregate groups. Section 5 of the paper looks at empirical patterns in the full set of
countries for which data are available. Here, drawing on various data sources, we show that
both output losses and excess mortality exhibit hump-shaped outcomes with middle income
countries experiencing the worst. We then present in a systematic way the important differ-
ences in policy and underlying economic and demographic conditions. For each, we briefly
discuss their relevance for the pandemic’s impact in order to help motivate the model and
quantitative analysis which follows.

2.1. The Impact of COVID-19

The first fact we highlight is the differential impact of the pandemic on output losses and em-
ployment declines across the world income distribution. Figure 1 displays the data by plotting
changes in output and employment for low-income, emerging, and advanced economies. While
there is considerable variance even within groups, a clear U-shaped patterns emerges in which

5
 Figure 1: GDP and Employment Growth from 2019 to 2020 by National Income

Employment per capita GDP per capita

0
Percent change in 2020 relative to 2019
−2

−2.4

−3.1
−3.6
−4

−4.6

−5.4
−6

−6.7
−8

Low income economies Emerging economies Advanced economies

Note: Employment data comess from he ILO Statistical Database and data on GDP-per-capita is taken
from the World Bank World Development Indicators.

output losses were greatest in emerging economies. GDP-per-capita fell by 6.7 percent and
employment by 5.4 percent in emerging economies, considerably worse than both wealthier
countries where output and employment losses were 4.6 percent and 2.4 percent, respectively,
and lower income countries where those losses stood at 3.6 and 3.1 percent. Figures A.2 and
A.1 illustrate that the relationship also hold sin the un-binned data and Figure A.3 displays
similar trends in cross-country consumption data. Interestingly, these data also suggest that
declines in output and consumption may have been greater in advanced economies than in
low-income ones. Such outcomes are surprising given the tremendous resources and technol-
ogy that wealthy countries brought to bare in combating COVID-19, resources that low-income
countries had no ability to marshall or match in any comparable way.

The second important fact pertains to the fatalities caused by COVID-19. These deaths are
commonly measured using excess mortality, the difference between total deaths in a given
month of the pandemic and those that would be normally expected, measured as expected
deaths during the same month over the previous (typically five) years. Figure 2 displays the
data by comparing mortality outcomes in advanced and emerging economies. As with output
losses, we find that the emerging economies experienced the worst outcomes. According to

6
 Figure 2: Excess Deaths from 2019 to 2020

180 Emerging markets

164.5
Advanced economies

148.1
150
Excess deaths per 100k people

120 112.9

99.5

90

64.1 63.0
60

30

0
Economist Karlinsky & Kobak New York Times

Note: Data sourced from the New York Times and Economist excess mortality trackers, and Karlinsky
and Kobak’s (2021) World Mortality Database.

estimates The Economist, excess deaths in emerging economies stands at 112.9 per hundred
thousand people, which is around 75 percent higher than the average estimate for advanced
economies, which experienced 64.1 excess deaths per hundred thousand. Estimates from the
World Mortality Database of Karlinsky and Kobak (2021) show 164.5 excess deaths per hun-
dred thousand people, or 65 percent larger than the 99.5 deaths per hundred thousand of
advanced ones. The gap is even wider in the New York Times mortality tracker which records
148.1 deaths per hundred thousand in emerging economies, compared to 63 in advanced ones.

Internationally comparably data on excess mortality in low-income countries are more difficult
to find. The most comparable statistics of which we are aware contain very few observations
from low-income countries (see Figure A.4 and Figure A.5). These data, from The Economist
and Karlinsky and Kobak (2021), have two and five observations from the low-income group
respectively. Deaths for this small set of countries average around 100 excess deaths per hun-
dred thousand people, putting them well below the level of the emerging markets. Official
data on deaths from COVID-19 in low-income show remarkably low levels of fatalities (see
e.g. Figure A.6), though there is widespread belief that official statistics undercount deaths

7
 Figure 3: Oxford Lockdown Stringency Index

Low income economies Emerging economies Advanced economies

60
55.9

49.5 49.7
50

48.2
46.0
40

36.5
Index
30
20
10
0

Lockdown stringency Workplace closure

Note: The Government Stringency Index is taken from the Oxford Government Response Tracker (Ox-
CGRT). GDP per capita is expressed at PPP and taken from Penn World Table 9.1 (Feenstra et al., 2015).

there. Our read of the literature is that there is still no clear consensus on what the true death
rates have been in low-income countries, though it seems unlikely that they are worse than the
high rates estimated in emerging markets such India (Deshmukh et al., 2021; Ramachandran
and Malani, 2021), Mexico (Dahal et al., 2021) and Brazil (Yamall Orellana et al., 2021).

Taken together, the data reveal that the impact of the COVID-19 pandemic across the world
income distribution has been highly non-linear. Emerging economies have been hit the hardest
most in terms of output losses and likely in terms of excess mortality as well. Equally surprising
is that the data suggest that low-income countries have fared better than advanced economies
in terms of output losses, and possibly also in terms of mortality rates, despite the far greater
economic and technological resources mustered by the latter to combat the crisis.

2.2. Differences in Policy Response

A natural candidate explanation for the cross-country variation is that they reflect differences
in policy responses to the COVID-19 pandemic. While nearly all countries implemented some
sort of lockdown and transfer programs, they varied widely both in the stringency of restrictions

8
and in the generosity of transfers. The policy distinction matters for how well countries manage
the endogenous path of infections through the public health externality and for the ability of
households to protect themselves by staying home for prolonged periods without income.

By lockdown policies, we refer to those whose primary aim is to restrict individual behavior
and social interactions to stem the spread of disease. These include school closures; workplace
closures; public event cancellations; restrictions on public gatherings; closure of public trans-
port; stay-at-home requirements; public information campaigns; and domestic and interna-
tional travel restrictions. The Oxford Coronavirus Government Response Tracker’s (OxCGRT)
stringency index provides a parsimonious quantifiable measure of how strict these policies were
across countries. Figure 3 plots the index of each country group, and shows that the most strin-
gent lockdown policies were implemented by emerging economies (the un-binned data are
displayed in Figure A.7). When we simulate lockdown policies, we implement them using the
time-series of workplace closures reported by OxCGRT to be consistent with how such policies
are represented in the model. As the data show, cross-country variation in these programs is
similar to the overall stringency of policies. The time-series dynamics of their implementation
within countries also appears similar (see Figure A.14)

Another important dimension of the policy response in nearly all countries was the expansion
of social insurance payments, such as unemployment benefits. These payments are viewed as
critical to offsetting lost income and make isolating at home economically feasible for those
with low savings or little income. However, as the crisis unfolded it quickly became clear that
governments in many developing countries lacked the fiscal capacity to sustain substantial
transfers to major segments of their population for very long. Consequently, we observe sub-
stantially more cross-country variation in the size and scope of social insurance programs than
in lockdown policies.

Figure 4 provides two measures capturing the scope and generosity of transfer programs im-
plemented in response to COVID-19 across the world income distribution. The left side his-
togram plots national pandemic spending as a share of GDP, which includes comprehensive
measures of budgetary fiscal support to individuals and firms estimated by the IMF. While pan-
demic spending appears similar in low-income and emerging economies, they are only about
one-third the spending undertaken by advanced economies which reached nearly 10 percent
of GDP. The right side histogram displays the Oxford’s Government Economic Support Index
which records financial assistance programs such as income replacement and debt relief for
individual citizens. The index should be interpreted as an ordinal measure of economic assis-
tance for individual citizens in that it does not include support to firms or business and does
not take into account the total fiscal value of economic support programs. Nevertheless, the

9
 Table 1: Oxford Covid-19 Government Response Indices in 2020
.

Country Income Group

Index Low-Income Emerging Markets Advanced Economies

Panel A: Included in both Stringency and Health & Containment Indices
School closures 53.8 64.8 50.1
Workplace closures 34.6 47.0 45.1
Cancellation of public events 57.0 69.4 63.7
Restrictions on public gatherings 50.9 59.5 61.3
Closure of public transport 22.5 32.0 17.8
Stay at home requirements 25.0 35.7 24.9
Restrictions on internal movements 32.9 47.7 31.8
International travel controls 57.6 63.6 63.4
Public information campaigns 79.7 83.8 87.0
Panel B: Included only in Health & Containment Index
Contact tracing policy 54.4 61.5 67.6
Facial coverings 43.8 46.4 37.3
Testing policy 37.9 52.2 58.8
Vaccination policy 22.8 31.3 35.3
Protection of the elderly 19.4 40.8 57.3
Panel C: Included only in Economic Support Index
Income support 17.3 29.3 57.8
Contract/Debt relief 31.0 49.6 58.9
Observations 52 67 33
Note: Countries are grouped into low income, emerging markets, and advanced economies using the IMF’s
economic classification of countries. Data in the table is the average level of the Oxford Covid-19 government
response tracker by country income group.

10
 Figure 4: Pandemic Spending and Economic Support

Low income economies Emerging economies Advanced economies

60
50 58.4

39.3
40
30

24.2
20

9.6%
10

3.0% 3.4%
0

Pandemic spending to GDP ratio Economic support index

Note: The left side histogram plots the ratio of pandemic spending to GDP, taken from the IMF. The right side
histogram displays the Oxford Economic Support Index available through the Oxford Coronavirus Government
Response Tracker’s (OxCGRT).

data reveal a similar pattern with spending on economic support rising monotonically with na-
tional income. The greater cross-country variation in economic support policies, as compared
to lockdown policies, is most apparent in thes underlying data which is displayed in appendix
Figures A.8 and A.9.

These cross-country differences in lockdown policies and public insurance programs are even
more apparent when one examines the underlying components of the OxCGRT’s indices which
are displayed in Table 1. The first noticeable feature is that low-income countries have the
least stringent policies in every lockdown category, and in all other categories except "Facial
Coverings." The near opposite is true for emerging economies which have the most stringent
policies across all sub-categories of lockdown measures (Panel A) except "Public Information
Campaigns." The largest deviations in emerging economy lockdowns pertain to the closure of
public transport, stay at home orders, and restrictions on internal movements. This is notable
since these measures likely imposed the largest restrictions on commercial activity, especially in
emerging economies where the ability to work from home is not widespread (see section 2.5)

11
and substituting to e-commerce and delivery services is limited by infrastructure. Finally, it is
interesting to note that the stringency of emerging economy policies does not extend beyond
lockdowns; as Panels B and C show, direct public health interventions and economic support
policies were generally less encompassing in emerging economies. Taken altogether, the scope
of differences in the stringency and aim of policies across the world income distribution offer
ample scope for them to drive the differences in outcomes we observe in the data.

2.3. Differences in Population Structure

It has been well known since the beginning of the pandemic that COVID-19 poses dramatically
greater health risks to older individuals, in particular those over the age of 65 (Ferguson et al.,
2020; Glynn, 2020). Early centers of infection in the west, such as Italy, experienced health
impacts concentrated on those in this older age range, with particularly severe fatality rates
for those in their 80s and 90s. At the same time, the number of deaths linked to COVID-19 for
those under 20 has been negligible, though certainly not zero.

A basic demographic difference between advanced and developing economies is that popula-
tions are far younger in the developing world. Since fatality rates from COVID-19 are very low
for young individuals but rise sharply with age, these demographic differences suggest much
smaller populations of vulnerable individuals in the developing world. One can see these de-
mographic differences starkly when looking at cross-country data on the median age. Figure
5 plots the median age against GDP per capita in a set of 158 countries using data from UN
Population Division and Penn World Tables. Data from the UN Population Division show that
countries in the bottom quartile of the world income distribution have a median age of 19.1
years. Nigeria, Africa’s most populous country, has a median age of 17.9, while countries like
Angola and the Democratic Republic of the Congo have median ages of just 16.4 and 16.8 years
old. By contrast richer countries like Italy, the United Kingdom and France have median ages
of 45.9, 40.2 and 41.2, respectively.

Another statistic indicative of the much smaller vulnerable population in the developing world
is the cross-country data on the population above 65. Figure A.10 plots the fraction of the
population that is above 65 against GDP per capita in a set of 162 countries using data from
the World Bank and the Penn World Tables. In the world’s poorest countries the fraction of the
population that is above age 65 is negligible, with an average of around 3 percent for countries
in the bottom quartile of the world income distribution. The older population is much larger
as a fraction of the total in richer economies, and reaches around one quarter of the population
in Japan. Among countries in the topic quartile of the world, the average is about 15 percent
of the population being above age 65.

12
 Figure 5: Median Age of the Population

50

JPN
ITA DEU
PRT
BGR GRC ESP AUT
SVN
LVALTU
HRV FINNLD CHE
HUN
ROU EST
CZE FRA DNK
BEL
BIH KOR MLT
SWE
CAN
40 UKR SRB ABWGBR SGP
BLR POL
SVK TWN NOR LUX
BRB
GEO THA MNE NZL AUSUSA
CHN IRL
Median Age, 2015

ALB ISL
MDA MUS
URY SYCCYP
ARM CHL KWTTTO ARE
LCA
LKA BHS
TUN BRA CRI BHR QAT
VNM AZEARGATG
30 COL TUR ISR
SAU
JAM IRN KAZOMN
SURLBN PAN
MMR MAR FJI IDN
PER
DZA
GRD
MNG
MDVMEX MYS
IND BTN ECU
UZB
BGD SLV ZAFDOM TKM
KGZ NIC PRYEGY
KHM BOL PHL BWA
NPL DJI BLZ
HTI HND
PAK GAB
TJK JOR GNQ
LSO GTM NAM
GHA
20 RWA
YEM COM
MRT IRQ
LBR TGO
MDG
ETH ZWE
GNB KEN SDN COG
SLE GIN
BEN CMR
SENCIVNGA
STP
CAF
BDI MWIMOZ
BFA GMB ZMB
TCD UGA MLI AGO
NER

10
1 2 4 8 16 32 64
Real GDP per capita in 2017 x $1,000

Note: Median age data corresponds to 2015 and is from the UN Population Division. GDP per capita is
expressed at PPP and taken from Penn World Table 9.1 (Feenstra et al., 2015).

It is hard to look at statistics like these and not see how different the impacts of COVID-19 will
be in less developed countries. Concretely, while almost everything about COVID-19 suggests
a more severe impact in less-developed countries, the far younger demographic is clearly in
their favor.

2.4. Differences in Healthcare Capacity

Developing countries typically have substantially less ability to control disease than do richer
countries. Sanitation and hygiene are more of an issue given the lack of widespread piped
water and functioning sewage systems. Health infrastructure, especially hospital and health
clinic capacity, is also less developed. For mild cases of COVID-19 infections, this may make
little differences, as bed rest is likely to suffice in these mild cases. However, for critical cases,
the lack of intensive-care capacity is a clear disadvantage for developing countries in their

13
attempts to save lives during the pandemic.

Figure A.11 plots the number of hospital beds per 10,000 people, as reported by the World
Health Organization (WHO), against GDP per capita. The number of hospital beds is an im-
perfect measure of hospital capacity for many reasons, most importantly because it is not a
bed per se that helps critical patients recover from COVID-19 but trained doctors, equipment
like ventilators, and appropriate pharmaceuticals. Still, for lack of more comprehensive cross-
country data, we take hospital beds as a proxy for medical care capacity.

By this metric there are stark differences in healthcare capacity across countries. Richer coun-
tries, which have quite some range amongst themselves, average around 49 hospital beds per
10,000 people. Countries like Japan and Korea have even more beds per capita, having 134 and
115 beds per 10,000 people, respectively. This is still far higher than the capacity in developing
countries, which is a paltry 12 beds per 10,000 people on average in the bottom quartile of the
income distribution. In Appendix Table B.1, we report the availability of intensive care unit
(ICU) beds and per capita healthcare costs across a limited set of countries. Consistent with
the patterns observed from the number of hospital beds, it appears that low income countries
possess significantly fewer ICU beds than high income countries.

2.5. Differences in Sectoral Composition of Employment

It is widely known that the sectoral composition of employment varies systematically with eco-
nomic development. These differences are important because commercial disruptions brought
on by COVID-19 and the resulting lockdowns differed substantially by occupation. Non-essential
jobs that could not be performed remotely or while socially distancing experienced the largest
and most sustained drops in employment throughout the recession; in contrast, occupations
that were amenable to working from home experienced minimal disruption and some even
flourished during the pandemic. In our model, we highlight two systematic differences in the
composition of employment between advanced and developing economies which are relevant
to the pandemic’s macroeconomic outcomes across countries: the share of rural employment
and the extent to which the urban workforce can work from home.

It is well known that the share of agricultural employment varies widely with economic devel-
opmnt (see Figure A.12). In the poorest countries, up to 70% of the population can be engaged
in agricultural work, often subsistence farming on family plots; in advanced economies, that
share is in the low single digits. The high agricultural share, while often considered a drag on
economic modernization, offers a resilient source of income during pandemics. A good deal of
agriculture in the developing world takes place on household-run farms, allowing it to continue
during “stay-at-home” orders. Even in the absence of lockdowns, farming can often continue

14
 Figure 6: Non-Social Sector Employment Share

80
73.0
70

60.0
60 50
Percent

43.0
40 30
20
10
0

Low Income economies Emerging economies Advanced economies

Note: The non-social sector includes rural employment and urban jobs that can be done from home, as esti-
mated by Gottlieb et al. (2021b). See text for details. GDP per capita is expressed at PPP and is taken from
the Penn World Table 9.1 (Feenstra et al., 2015).

while socially distanced or with contact restricted to household members. Agricultural work-
ers therefore do not face the same stark trade-offs in choosing between protecting their health
or incomes since farming can often continue without substantially increasing the risk of infec-
tion. Consequently, while agricultural workers may be vulnerable because of low wages, their
employment is more resilient to large losses from lockdowns or voluntary self-isolation.

Outside of the agriculture sector, labor markets in lower income countries are characterized by
widespread informality and employment concentrated in high-contact sectors.1 Large informal
sectors will generally make economies more vulnerable to COVID-19 since, like agriculture,
these jobs generally pay low wages while, unlike agriculture, most informal jobs cannot be
performed from home or while socially distancing. To summarize these effects at the country
level, we follow (Kaplan et al., 2020) and aggregate employment into social and non-social
sectors. Social sector workers have limited ability to work from home and suffer large income
1
According to the International Labor Organization (ILO), informality rates in the non-agricultural sector can
be as high as 80% of employment in the lowest income countries, but falls drastically with GDP-per-capita to less
than half that level.

15
losses during lockdowns, while non-social sector workers can substitute more easily to remote
work. We calculate the non-social sector share to include rural employment and all urban jobs
that can be worked from home. For the latter, we use the cross-country estimates of (Gottlieb
et al., 2021a) which are constructed using worker level data on the task-content of jobs in
urban labor markets. Figure 6 displays the resulting estimates of non-social employment and
illustrates that it varies substantially across countries. Emerging market economies have the
lowest ability to work from home, with only 43% employed in non-social, low-contact jobs.
In advanced economies, the non-social share is 60%, due to the greater number of high skill,
professional jobs. However, the non-social share is largest in low-income countries, at 73% of
aggregate employment, driven by the large agricultural labor force.

As a consequence of theses differences in the sectoral composition of employment, emerging

market economies are more exposed to economic losses during the pandemic. Having less jobs
that can be done from home or while socially distanced leads to greater economic losses during
lockdowns and workplace closures. Moreover, in the absence of robust transfers, many social
sector workers can become desperate and so voluntarily elect to continue working, rather than
shelter at home, during times of peak infection. Such decisions will generally provide only
marginal income gains, while amplifying the infection risk for the whole population through
the public health externality. Large social sector employment can therefore be a liability for
emerging market countries fighting COVID-19, as these workers are particularly vulnerable
with limited options to avoid increasing their risk of becoming infected, or infecting others.

3. Model

Our analysis draws on a quantitative heterogenous-agent macroeconomic model with epidemi-

ology as in the SICR model to analyze how policy responses to the COVID-19 pandemic should
differ in developing countries. The model is equipped with several features that vary between
advanced and developing economies that are relevant for the pandemic response, as motivated
by the data presented in the previous section. These include uninsurable idiosyncratic health
and income risks, age heterogeneity, fiscal capacity constraints, healthcare capacity, and avail-
ability to work from home across sectors. This section now presents these features in detail.

3.1. Households and Preferences

The economy is populated by a unit mass of heterogenous individuals who make consumption
and savings decisions subject to idiosyncratic income and health risks. Individuals differ in their
age j ∈ {young adult , old adult} and permanent labor productivity z ∼ G. Time is discrete and

16
each period represents two weeks. Preferences are given by:
–∞ § ª™
X
U =E β jt log(c t ) + ū , (1)
t=0

where the discount factors β jt capture age heterogeneity in the population, and βyoung < βold .
This specification follows the tractable formulation of Glover et al. (2020) that abstracts from
explicitly modeling age, appealing to the logic that pandemics are sufficiently short-lived rel-
ative to entire lifetimes. It thus suffices to model only the expected number of years left to
live, which is captured by the heterogeneity in discount factors. The term ū represents the
flow utility value of being alive, following the specification of Jones and Klenow (2016), and
represents the reason that model households try to avoid fatality risk. Once an individual dies,
they receive a fixed utility level that potentially depends on their individual characteristics, as
we describe below.

There are two sectors, which we denote as social (s = S) and non-social (s = N ). We assume
that households are born with the sector they supply labor and cannot switch sectors. The
social sector represents the workers with little availability of remote work. Examples of the
occupations in the social sector includes waitresses, hair dressers, to name a few. The non-
social sector represent the occupations that can be done with low level of social contacts. Such
occupations include farmers in agricultural sector who can work while distancing from others,
or college professors who can easily work remotely. Households in sector s supply their labor
to a representative firm where they can earn wage ws per effective hour worked.

At the beginning of life, workers draw their permanent productivity, z ∼ G. Incomes in both sec-
tors are also subject to idiosyncratic productivity shocks as in Bewley (1977), Huggett (1993)
and Aiyagari (1994). Specifically, we assume that individual labor productivity in each sector
is composed of the sector-specific permanent component z and an idiosyncratic component v
following the stochastic process:

log vt+1 = ρ v log vt + ε t+1 , ε t+1 ∼ F (0,σ v ). (2)

We include idiosyncratic income risk because developing countries are far from having full
insurance, and so accounting for how people insure themselves in response to policies which
may keep them away from work for prolonged periods of time is a first order consideration.

After observing their income realization, households make consumption and savings decisions
given the interest rate, r, and subject to a no-borrowing condition, a ≥ 0. Formally, the budget

17
constraint of a household in sector s before the pandemic is given by:

c + a0 ≤ (1 − τ)ws zvn + (1 + r)a + T (3)

where τ is the income tax rate and T is government transfers.

3.2. Aggregate Production Technology

The economy produces a single final good by combining capital with labor services supplied
by the three sectors. The aggregate production technology is given by:

Y = AL α K 1−α ,

where A is the total factor productivity and 0 < α ≤ 1 is labor’s share of value-added. We
abstract from the domestic capital market. The aggregate capital stock is composed entirely
of foreign sources, K = K F , which can be rented at an exogenously given international rental
rate r F and which depreciates at rate δ. Aggregate labor depends on the total supply of labor
services from the social and non-social sector,

L = LS + L N

3.3. Credit and Capital Markets

Credit market incompleteness prevents households from borrowing against future earnings.
As a result, individuals must maintain non-negative assets in formulating their consumption
plans subject to (3), giving rise to hand-to-mouth consumers as well as a precautionary savings
motive in response to idiosyncratic health and income risks. The precautionary motive is im-
portant for getting aggregate welfare measurements correct since it creates another feedback
between the epidemiological and economic dynamics, as individuals withhold some consump-
tion to increase precautionary savings in response to the pandemic’s onset.

3.4. Public Health and Hospital Capacity

Households face idiosyncratic health risk which can reduce their labor productivity and in-
crease the probability of dying. Susceptibility to infection is determined in part by economic
decisions taken by households. Once infected, progression of the disease depends on an indi-
vidual’s age and the availability of public health infrastructure offering treatments.

Health risks are modeled using an SICR epidemiological model with five health states: suscep-

18
tible (S), infected (I), critical (C), recovered (R), and deceased (D). We denote by Ntx the mass
of individuals in each health state x ∈ {S,I,C,R,D} at time t and use Nt = NtS + NtI + NtC + NtR
to measure the non-deceased population. Figure 7 illustrates how these states evolve:

C
πy
o u ng: 1 − C
y
1− π o R
old: C )
D N t ,Θ
(
g: 1− π y t C Θ)
youn D N ,
S I 1− π t( t
o R
transmission rate: π I old:
youn
g: π C
old: Cy
π
C
o youn
g: π D
( C
old: Dy t Nt ,Θ) D
π ( C
ot N
t ,Θ )

Figure 7: Dynamics of Health States and Transition Probabilities

The probability a susceptible person becomes infected is given as:

NtI
πIt = β tI ×
Nt

where β tI is the ”behaviorally-adjusted infection rate,” which accounts for both the diseases
biological transmission rate as well as population wide behavioral responses to avoid being
infected. The explicit dependence of β tI on time reflects the time-varying and population-wide
behavioral responses to avoid being infected such as improved hygiene, social distancing, and
learning about the best-pratice behavior during a pandemic.

Individuals who contract the virus experience a proportional drop in productivity of 1 − η for
one model period (two weeks), at which point they either recover or enter a critical health
state. The probability of becoming critically ill depends on an individual’s age and is given by
πCj . Those in critical health are unable to work and require hospitalization. The likelihood
of recovery in the hospital depends again on their age in addition to the availability of public
health infrastructure, such as ICU beds and ventilators. In particular, the fatality rate of a
critically ill patient of age j is given by:

πD if assigned ICU bed
j
πDj t (NtC ,Θ) =
κ × πD if not assigned
j

19
where π Dj is a baseline fatality rate for age j individuals in critical health and κ governs the
impact on fatality rates of strained hospital resources. Whether or not a critically ill patient
receives an ICU bed depends on overall hospital capacity and the number of other patients.
Specifically, letting Θ denote hospital ICU capacity, the probability a new patient receives an
ICU bed is given by min{Θ/NtC ,1}. In other words, all critically-ill patients receive an ICU bed
if hospital capacity constraints are not binding, and beds are rationed amongst the critically-ill
with probability Θ/NtC when constraints bind.

3.5. Voluntary Substitution Away From Workplace and Lockdowns

Voluntary Subtitution While the diseases progression is exogenous, the probability a sus-
ceptible person becomes infected depends on endogenous economic decisions and the preva-
lence of infections in the population. To incorporate the feedback from economic behavior to
infections, we allow individuals to lower the degree of exposure to the virus by voluntarily
substitution away their labor supply to remote work. Specifically, we allow workers to choose
between going to workplace and working remotely in each period. Remote work involves less
social contacts, providing protection from being infected. Specifically, remote work lowers the
probability of infection by ξ.

While it provides protection from being infected, working remotely is also less productive than
going to the workplace. The productivity penalty of working remotely is parameterized by
φs , where s ∈ {S, N }, by assuming that the effective labor supply of an worker in sector s can
provide is given as φs n, where 0 ≤ φs < 1. We assume that φS < φN < 1, implying that the jobs
in the non-social sector are more suited to be done remotely. Consequently, the probability a
susceptible person becomes infected is given by:

β I × N I /N if go to workplace
t t t
πIt =
β I × N I /Nt × ξ if work remotely
t t

Lockdowns Infection rates can be further mitigated by containment policies, such as lock-
downs. As in Kaplan et al. (2020), we model lockdowns contrain a certain fraction of workforce
to work remotely through stay-at-home orders. Under a lockdown, households who would oth-
erwise go to workplace hours are forced to substitute switch to remote work. The stringency of
lockdown varied across time and countries. Following Bick et al. (2020), we assume that 70%
of the workers are forced to work at home under a full lockdown. Because remote work lowers
the number of new infections, lockdowns mitigate the pandemic by exogenously decreasing
the aggregate supply of workplace labor.

20
3.6. Vaccinations

Susceptible individuals can obtain immunity through vaccination as well. In each period, a sus-
ceptible individual draw a nonnegative probability of receiving vaccination. Once vaccinated,
the individual obtains immunity and joins the recovered population. The exact probability of
vaccination in each time period is taken from the actual path of vaccination in the US. We will
explain it in more details in the calibration section.

3.7. Government and Taxation

The government has power to tax, transfer, and impose economic lockdowns subject to the
constraints imposed by limited fiscal capacity and labor market informality. We further require
that the government run a balanced flow budget which satisfies,
Z
Bt + τ y(a, x, v)dQ = T

where y(a, x, v) is pretax income for individual (a, x, v) ∼ Q, τ is the prevailing tax rate, and
T is aggregate transfers to households. In addition to tax revenue, we allow developing coun-
tries access to emergency bonds, B t , which can be used to finance additional welfare transfers
during government imposed lockdowns. The source of these funds is international donors and
multinational institutions such as the IMF, World Bank, and World Health Organization. Funds
borrowed for emergency transfers accrue interest at rate 1 + r F until the pandemic ends, at
which they are repaid through annual annuities. Formally, emergency transfers are given by:


 B̄ during the lockdown
−t e

 t lP
rF

t
B t = − 1+r F × 1 + r F B̄ after pandemic ends

 t l −t s

0 otherwise


where B̄ is the size of per-period emergency transfers during lockdown, which we take para-
metrically, and t s , t e , and t l index the lockdown’s start, the lockdown’s end, and the pandemic’s
end, respectively.

4. Quantitative Analysis

In this section, we discuss the calibration strategy, validate the model’s fit, and present our
counter-factual results. To evaluate the quantitative importance of each channel in explaining

21
the cross-country variation in outcomes, we calibrate the model to match the U.S. economy
and then vary key economic and demographic characteristics of the U.S. to match those of
low-income and emerging economies. For each variation, we display the dynamic path of
output and fatalities predicted by the model. To identify the most salient channels, we report
the cumulative effects of each counterfactual on the U.S. economy compared to the calibrated
benchmark.

4.1. Data Sources and Calibration

For expositional clarity, we divide the calibrated targets into three broad categories correspond-
ing to those governing economic mechanisms, those controlling epidemiological dynamics, and
those delineating differences between the advanced, emerging, and low-income countries.

Table 2: Calibration of Economic Parameters

Var Description Value Source / Target

rF Exogenous interest rate 0.0006 Pre-COVID T-Bills rate 1.5%
ρv Persistence of idiosyncratic income shock 0.91 Floden and Lindé (2001)
σv St.Dev of idiosyncratic income shock 0.04 Floden and Lindé (2001)
α Labor share 0.6 Gollin (2002)
βy Discount factor for the young 0.9984 Glover et al. (2020)
βo Discount factor for the old 0.9960 Glover et al. (2020))
σg Variance of remote / non-remote work taste shock 0.0101 Pre-COVID Remote Workers 8.2%
φn Productivity remote work, non-social sector 1 Barrero et al. (2021)
φs Productivity remote work, social sector 0.62 COVID-19 Employment Declines - 6.4%
A(P) Pandemic Total Factor Productivity 1.042 COVID-19 Output Declines -4.1%

Table 2 reports the parameters that govern the core economic dynamics of the model. Pop-
ulation demographics are modeled using age dependent discount factors accounting for dif-
ferences in the remaining years of life for young and old workers. The age specific discount
factors are taken from Glover et al. (2020), and the stochastic income processes are taken
from Floden and Lindé (2001), who estimate similar income processes in the United States
and Sweden. The taste-shock for remote work σ g is chosen so that 8.2% of the pre-pandemic
laborforce works remotely, consistent with the estimates in Bick et al. (2020). Finally, labor’s
share of income comes from Gollin (2002), and the rental rate of capital is set to the two-week
return on pre-COVID Treasury Bills. We set the productivity penalty for remote work in the
nonsocial sector, φn , to unity, consistent with evidence of small productivity losses for these
workers in most cases, and potentially even productivity gains in some cases (Barrero et al.,
2021). Finally, the penalty for remote work in the social sector, φs , and the TFP shock accom-
panying the pandemic A(P), are jointly calibrated to match aggregate 2019-2020 year-on-year

22
 Table 3: Calibration of Epidemiological Parameters

Var Description Value Source or Target

η Effect of infection on productivity 0.3 Alene et al. (2021)
ξ Reduction of infection probability by working from home 0.6 Mossong et al. (2008)
κ Impact of hospital overuse on fatality 2 Glover et al. (2020)
πCy Rate of young entering C from I 6.7% Ferguson et al. (2020)
πC
o Rate of old entering C from I 38.0% Ferguson et al. (2020)
πDy Rate of young entering D from C 2.7% Glynn (2020)
πD
o Rate of old entering D from C 9.0% Glynn (2020)

employment and output declines in the United States.2

Table 3 reports parameters controlling the epidemiological transmission of disease and their
interactions with public health infrastructure and lockdown policies. We take parameters gov-
erning the fatality infection rates from Glynn (2020) and the rates of infected cases becoming
critical from Ferguson et al. (2020). The effect of hospital congestion on disease fatality rates,
κ, is taken from Glover et al. (2020). The productivity penalty of becoming infected, η, is set
to match a 30 percent share of asymptomatic infection cases, as estimate in the meta-analysis
of Alene et al. (2021). Such a choice is motivated by the observation that those known to be in-
fected cannot work, and so have productivity of zero, while those who are infected but asymp-
tomatic may continue to work unhindered. Finally, we choose the time-varying behavioral-
adjusted infection probability, β tI , so that the model’s endogenous path of fatalities precisely
matches the experience of the United States. The simulated endogenous path of the virus also
account the time path of vaccinations and lockdowns in the U.S.. Vaccination data is taken
from the COVID-19 Data Repository by CSSE at John Hopkins University, and we assume vac-
cination rates continue to grow at 1% per period after the last available data point, until period
60. The time path of lockdown policies comes from the Oxford Coronavirus Government Re-
sponse Tracker (see Figure A.14). We assume lockdown policies are gradually lifted starting in
the last period of available data until they are completely discontinued by period 60. Figure 8
plots the fitted results and validates the model’s ability to replicate these dynamics exactly.

Table 4 summarizes parameters which vary across advanced and developing countries. The
tax rates for the advanced and developing countries are taken from Besley and Persson (2013).
Age demographics ω y come from the World Bank and measure the share of the population
under 65. The youth share in advanced economies corresponds to the U.S. economy, as it
2
Appendix Table B.2 summarizes the internally calibrated parameters and the model’s fit to the data. Note
that TFP in normal times, A(N ) is set to one, so that A(P) should be interpreted as a relative TFP shock in effect
during the Pandemic.

23
 Figure 8: Predicted and Actual COVID-19 Mortality in the United States

0.200 Cumulative deaths (Data)

0.175
Cumulative deaths (Model)
Percentage of population

0.150

0.125

0.100

0.075

0.050

0.025

0.000
22 Mar 14 June 6 Sept. 29 Nov. 21 Feb. 16 May 8 Aug 31 Oct. 23 Jan. 1 May 7 Aug 13 Nov.
2020 2020 2020 2020 2021 2021 2021 2021 2022 2022 2022 2022

Note: Time path of U.S. COVID-19 mortality taken from the COVID-19 Data Repository by the Center for
Systems Science and Engineering (CSSE) at John Hopkins University.

is our benchmark calibration, and we set the shares for emerging and low-income countries
to their group averages. The share of workers in the social sector, ωs , is constructed using
estimates from Gottlieb et al. (2021b) on the share of urban labor that can work from home
and adjusting the ratio to account for the rural population. Specifically, we take the shares of
urban and rural labor from the UN Population Division and assuming the entire rural sector is
non-social, calculate the ωs as the weighted average of the urban and rural populations.

The flow value of life, ū, is calibrated using the value of statistical life (VSL) approach. Follow-
ing Glover et al. (2020), we set the per-period statistical value of life to $515,000 for advanced
economies, equal to 11.4 times average US consumption. The value for ū is then computed so
that the behavioral response to a marginal increase in the risk of death is consistent with the
VSL. Specifically, we get ū by solving,

dc
VSL = | E(u)=k,ρ=0 = ln(c̄) − ū
dρ

where ρ is the risk of death and c̄ is average consumption. Absent better evidence, we assume
the VSL has unitary income elasticity and adjust ū for developing countries accordingly.

24
 Table 4: Calibration of Parameters Varying Across Advanced and Developing Economies

Advanced Emerging Low-Income Source or

Var Description Economies Economies Economies Target
US MID DEV
ū Flow value of being alive 11.4c̄ 11.4c̄ 11.4c̄ Glover et al. (2020)
τ Marginal tax rate 0.25 0.20 0.15 Besley and Persson (2013)
ωy Share of young in population 83% 84% 92% UN Population Division
ωs Share of social sector workforce 40% 57% 27% Gottlieb et al. (2021b)/IPUMS
Θ Hospital capacity per capita 0.00042 0.00025 0.00011 Glover et al. (2020) / WHO

The final cross-country parameter to be set govern the ICU hospital capacity in developing and
developed countries. One challenge is that while many countries report hospital bed capacity,
few developing countries distinguish explicitly between general hospital capacity and ICU ca-
pacity in the data. To address this, we assume the ratio of hospital beds to ICU beds is constant
across countries, and calibrate Θ by adjusting WHO data on the availability of hospital beds
in the top and bottom quartiles of country income levels (as in Figure A.11) by the ratio of
hospital beds to ICU beds taken from Glover et al. (2020).

4.2. Economic and Demographic Sources of Cross-Country Differences

Figure 9 and 10 plot the dynamic path of GDP-per-capita and fatalities as a percentage of
population during the COVID-19 pandemic in the United States in each of our counterfactual
simulations. The top panels display results for cumulative fatalities, the bottom panels display
results for output. Each figure provides five simulated paths: the benchmark U.S. calibration
and the four counterfactual exercises which vary demographics, the sectoral composition of
employment, public healthcare capacity, and the stringency of lockdowns in the United States.
Figure 9 reports counterfactuals that endow the U.S. economy with the characteristics of low-
income countries; Figure 10 reports the results of endowing the U.S. with emerging market
economy characteristics.

Looking across the panels, one can see that all four mechanisms play an important role to some
degree, but differences in age demographics and the sectoral composition of employment are
the most quantitatively prominent. In determining the trajectory of fatalities, age demograph-
ics are the most important for understanding differences between low-income and advanced
economies, while the sectoral composition of employment is most relevant for differences be-
tween emerging markets and advanced economies. The high agricultural employment share
in low-income countries also greatly reduces fatalities there. In emerging market economies,
lockdown policies also played an important role, on par with age-demographics, suggesting
the especially stringent policies enacted there were tied to the more serious public health

25
Figure 9: Time Path of Cumulative Deaths and GDP: Low Income Economies

US Calibration
0.20
Lockdown Intensity
Percentage of population

Sectoral Composition
0.15

0.10 Age Structure

ICU Capacity

0.05

0.00
22 Mar 14 June 6 Sept 29 Nov 21 Feb 16 May 8 Aug 31 Oct 23 Jan 1 May 7 Aug 13 Nov
2020 2020 2020 2020 2021 2021 2021 2021 2022 2022 2022 2022

(a) Cumulative Death, US with Low Income Economies’ Features

105
Sectoral Composition
Percentage of Pre-Pandemic Steady State

100 Lockdown Intensity US Calibration

Age Structure ICU Capacity

95

90
22 Mar 14 June 6 Sept 29 Nov 21 Feb 16 May 8 Aug 31 Oct 23 Jan 1 May 7 Aug 13 Nov
2020 2020 2020 2020 2021 2021 2021 2021 2022 2022 2022 2022

(b) GDP per capita, US with Low Income Economies’ Features

26
Figure 10: Time Path of Cumulative Deaths and GDP: Emerging Economies

0.25
Sectoral Composition
ICU Capacity

0.20 US Calibration
Percentage of population

0.15 Lockdown Intensity

Age Structure

0.10

0.05

0.00
22 Mar 14 June 6 Sept 29 Nov 21 Feb 16 May 8 Aug 31 Oct 23 Jan 1 May 7 Aug 13 Nov
2020 2020 2020 2020 2021 2021 2021 2021 2022 2022 2022 2022

(a) Cumulative Death, US with Emerging Economies’ Features

105
Age Structure
Percentage of Pre-Pandemic Steady State

100

US Calibration ICU Capacity

95
Lockdown Intensity

90 Sectoral Composition

85
22 Mar 14 June 6 Sept 29 Nov 21 Feb 16 May 8 Aug 31 Oct 23 Jan 1 May 7 Aug 13 Nov
2020 2020 2020 2020 2021 2021 2021 2021 2022 2022 2022 2022

(b) GDP per capita, US with Emerging Economies’ Features

27
 Table 5: Cumulative Counterfactual Effects of the COVID-19 Pandemic
Panel (a): GDP Changes from 2019 to 2020
Data Model

All Features Age/Sector/ICU

Advanced Economies -4.60 -4.01 -4.01

Emerging Economies -6.70 -7.36 -6.40
Ratio 1.46 1.84 1.60
Panel (b): Excess Mortality
Data Model

All Features Age/Sector/ICU

Advanced Economies 64.10 197.39 197.39

Emerging Economies 112.90 208.03 236.55
Ratio 1.76 1.05 1.20

emergency; Our simulations suggests deaths would have been considerably higher without
them.

The output counterfactuals exhibit less variation than what we see in fatalities, suggesting
the mechanisms we study contribute more equally to observed economic declines. Among
the channels, only the sectoral composition of employment stands out as having an especially
important quantitative role. In low-income countries, economic losses were moderated by
a large agricultural sector that was minimally disrupted by lockdowns and social distancing
requirements. In emerging markets, high levels of urban employment in jobs that cannot be
done from home explains a substantial part of their larger economic losses.

To assess what may be driving the especially bad outcomes observed in emerging markets,
Table 5 reports the cumulative effect of our counterfactuals on 2019-2020 year-on-year changes
in GDP and fatalities. For comparison, the first data column displays the data for advanced
and emerging economies discussed in the introductory sections (see Figures 1 and 2). The
second data column reports the simulation outcomes when all features are allowed to vary
(i.e. demographics, sectoral employment, ICU capacity, and lockdown policies). The entry for
advanced economies corresponds to our benchmark calibration to the United States data; the
entry for emerging economies corresponds to the simulation which endows the United States
with all the features of emerging economies. The third column reports results when we endow
the United States with only the age demographics, sectoral employment, and ICU capacity of

28
emerging economies. We distinguish these features since we view them as largely immutable
throughout the pandemic’s duration. To facilitate comparisons, the final row of each column
reports the ratio of outcomes in emerging markets relative to advanced economies.

In panel (a) we see that the model does relatively well at replicating variation in GDP. In the
data, GDP in advanced economies contracted by -4.6% while emerging economies shrank by
-6.7%. The benchmark model generates a -4.01% decline in advanced economies–matching
the U.S. data target – and predicts a decline of -7.3% for emerging economies. The model
therefore over-accounts for GDP declines, predicting that emerging markets would experience
contractions in GDP that are 86% greater than those advanced economies, while the data show
declines that are roughly 46% larger. One reason the model may over-predict GDP declines is
that official lockdowns could overstate de facto lockdowns in emerging markets, where gov-
ernments have more limited enforcement capability.

Panel (b) reports excess mortality per hundred thousand people in advanced economies and
emerging markets, both in the data and full counterfactual. The model substantially over-
predicts the total fatality rate since the benchmark advanced economy calibration is set to
match the United States, which has been a outlier in terms of reported COVID-19 mortality
amongst advanced economies. Endowing the United States with all the features of an emerging
market economy leads to a 5% rise in excess mortality. Since the data show mortality was 76%
higher in emerging markets, the counterfactual simulation can only explain about 6.5% of
the overall difference. These results suggest that there may exist other important public health
differences between countries that are missing from our model. Examples include lower overall
healthcare capacity in developing economies and a greater prevalence of co-morbidities.

Finally, in light of the large differences in emerging economies, it is natural to ask if there is
anything emerging market economies could have done differently to improve their outcomes.
While we do not model the optimality of different policies, our framework allows us to study
the extent to which outcome differences depend on features that are outside the control of
policymakers throughout the pandemic’s duration. In particular, we view a country’s age de-
mographics, sectoral composition of employment, and healthcare capacity to be largely fixed
throughout the pandemic. That governments cannot choose the age of their population is ob-
vious. Similarly, it’s generally widely held that the industrial composition of the economy is
rigid in the short-run. While public healthcare capacity can in principle be expanded (and
was, rather rapidly in a few places like China), we believe that emerging market economies by
and large only had limited ability to do so during the pandemic, especially given the concur-
rent global competition for medical equipment, oxygen, and protective gear. The final column
of Table 5 reports the cumulative counterfactual impact on output and fatalities if only these

29
immutable characteristics varied between emerging markets and advanced economies. For
output, these characteristics alone lead to a -6.4% decline in GDP, compared with -6.7% in the
data. For mortality, these fixed features lead to a 20% rise in fatalities, accounting for over 25%
of the 76% mortality gap observed in the data. Taken together, the simulations suggest that the
unusually bad outcomes in emerging markets were largely outside the control of policymakers,
depending instead on prevailing demographic and structural differences that cannot be easily
changed. In fact, the more stringent policy response of policymakers in emerging markets ap-
pears to have drastically reduced the fatalities they’ve experienced during the pandemic while
leading to an additional 1 percentage point decline in GDP.

5. Empirical Correlates of GDP Declines During the Pandemic

In this section we explore the empirical correlates of changes in GDP per capita from 2019 to
2020, focusing on the same variables emphasized in the model. We make no claim at uncover-
ing causal patterns in this section. Instead, we assess the extent to which correlations between
aggregate income changes during the pandemic and a country’s demographic, economic, and
policy characteristics are consistent with the model’s predictions and quantitative exercises.

We begin with the basic relationship between declines in GDP per capita and pre-pandemic
level of GDP per-capita. The first column of Table 6 shows that this relationship is U-shaped,
as we argued earlier. Both the level and quadratic coefficients on GDP per capita in 2019 are
statistically significant at the five-percent level, with the former negative and latter positive.
The second column includes controls for the agricultural employment share. The variable
exhibits a significant positive correlation with changes in GDP, holding constant differences in
national income, means that countries with larger percentages of their workforce in agriculture
also experienced smaller declines in national income, all else equal. Interesting, the coefficients
on GDP-per-capita and its square are now statistically indistinguishable from zero, with the
former switching signs. The third column includes median age as a control which exhibits no
significant correlation, somewhat puzzlingly. The fourth column controls for the stringency
of lockdowns, which is positive and statistically significant. The fifth column adds controls
for the generosity of economic support programs during the pandemic, which turns out ot be
statistically insignificant.

Column six of Table 6 adds all the covariates at once. This specification shows that agricul-
ture’s employment share remains a strong positive correlate of GDP changes, while lockdown
stringency remains a strong negative correlate. Median age and the economic support index
continue to be insignificant. This results do not change significantly under alternative speci-
fications of the regression model (see Table B.3). Collectively, the inclusion of these controls

30
 Table 6: Correlates of GDP per Capita Change from 2019 to 2020

Dependent variable: GDP per capita change from 2019 to 2020

Independent variables (1) (2) (3) (4) (5) (6)

GDP per capita in 2019 -0.10** 0.037 -0.17* -0.076* -0.11 -0.052
(0.046) (0.068) (0.094) (0.044) (0.068) (0.11)
GDP per capita in 20192 0.0014** 0.00021 0.0020* 0.0011* 0.0014* 0.00084
(0.00066) (0.00071) (0.0010) (0.00063) (0.00080) (0.0011)
Agriculture emp. share 0.076*** 0.062**
(0.027) (0.030)
Median age 0.083 0.074
(0.079) (0.082)
Lockdown stringency -0.13*** -0.13**
(0.043) (0.053)
Economic support 0.0042 0.024
(0.036) (0.038)
Constant -4.21*** -8.03*** -5.67*** 2.38 -4.29*** -2.97
(0.60) (1.66) (1.48) (2.07) (1.09) (3.34)

Observations 144 144 144 140 140 140

R2 0.031 0.071 0.037 0.129 0.030 0.163
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

eliminates the statistical significance of the original U-shape pattern in GDP-per-capita, and
substantially reduce the magnitude of the original correlations. We take this as suggestive
evidence that these variables are important empirical determinants of macroeconomic perfor-
mance across the world income distribution, at least thus far, during the pandemic.

6. Conclusion

The macroeconomic impact of the COVID-19 pandemic was most severe in emerging market
economies, which represent the middle of the world income distribution. This paper provides a
quantitative economic theory to explain why these economies fared so poorly compared to both
poorer and wealthier nations. Our model is motivated by key economic and demographic dif-

31
ferences across the world income distribution, including variation in lockdown policies, public
insurance, demographics, healthcare capacity, and the sectoral composition of employment.

Our quantitative model predicts greater declines in employment and output in emerging mar-
ket economies, as in the data. It also predicts the higher excess mortality in middle income
countries, albeit to a substantially smaller extent than in the data. The modest excess mortality
predictions of the model suggest that the higher COVID-19 fatalities in middle-income coun-
tries is likely driven by factors other than ICU capacity and the ability to work from home (e.g.
co-morbidities, hospital quality, etc.). Among the channels we study, age demographics and
the sectoral composition of employment are the most quantitatively important. Low-income
countries fare well because of their younger demographic and large agricultural population,
which provide a resilient source of income during lockdowns and while socially distancing. A
large share of jobs which require social interaction and stringent government lockdowns ex-
plains a large fraction of the worse outcomes in emerging market economies. Quantitatively,
the results suggest that cross-country differences are mostly driven by factors outside the short-
term control of government officials, and so there is likely little policy makers in middle-income
countries could have done differently to avert the especially severe outcomes they experienced.

Overall, our findings suggest that much of the variation in aggregate outcomes across country
income groups during the pandemic can be attributed to a small set of economic characteristics
and broad policy choices. Though substantial gaps are still left unexplained by these factors,
suggesting that other forces must be playing important roles. Absent from this study are pol-
icy decisions regarding school closings (e.g. Fuchs-Schündeln, Krueger, Ludwig, and Popova,
2020), mask use (e.g. Abaluck et al., 2021; Karaivanov et al., 2021), testing and tracing poli-
cies (e.g. Berger et al., 2020), and vaccine provision (e.g. Arellano, Bai, and Mihalache, 2021).
Future research could also fruitfully assess the quantitative importance of other policy choices
for cross-country macroeconomic performance during the pandemic.

Another key limitation of our analysis is that it relies on a large exogenous time-varying com-
ponent of the infection rate in order to match the observed path of excess deaths in the United
States. In reality, however, much of the time variation in infection probabilities is likely due
public policy choices that are not modeled here. These include policies that increase the preva-
lence of mask wearing, the development of better treatments for the infected, the rate of vacci-
nation, or general knowledge about how COVID-19 can and cannot be transmitted. Future re-
search should more explicitly consider the role these factors play in determining cross-country
differences in aggregate outcomes during the pandemic.

32
References
ABALUCK, J., L. H. KWONG, A. STYCZYNSKI, A. HAQUE, M. A. KABIR, E. BATES-JEFFRIES, E. CRAW-
FORD , J. B ENJAMIN -C HUNG , S. R AIHAN , S. R AHMAN , S. B ENHACHMI , N. Z AMAN , P. J. W INCH ,
M. M. HOSSAIN, H. M. REZA, A. A. JABER, S. G. M. F. L. B. A. RAHMAN, T. S. HUQ, S. P. LUBY,
AND A. M. M OBARAK (2021): “The Impact of Community Masking on COVID-19: A Cluster
Randomized Trial in Bangladesh,” Unpublished Working Paper, Yale University.

ACEMOGLU, D., V. CHERNOZHUKOV, I. WERNING, AND M. D. WHINSTON (2020): “A Multi-Risk

SIR Model with Optimally Targeted Lockdown,” NBER Working Paper No. 27102.

AIYAGARI, S. R. (1994): “Uninsured Idiosyncratic Risk and Aggregate Saving*,” The Quarterly
Journal of Economics, 109, 659–684.

AKTAY, A., S. BAVADEKAR, G. COSSOUL, J. DAVIS, D. DESFONTAINES, A. FABRIKANT,

E. GABRILOVICH, K. GADEPALLI, B. GIPSON, M. GUEVARA, ET AL. (2020): “Google COVID-
19 Community Mobility Reports: Anonymization Process Description (version 1.0),” arXiv
preprint arXiv:2004.04145.

ALENE, M., L. YISMAW, M. A. ASSEMIE, D. B. KETEMA, B. MENGIST, B. KASSIE, AND T. Y. BIRHAN

(2021): “Magnitude of Asymptomatic COVID-19 Cases Throughout the Course of Infection:
A Systematic Review and Meta-Analysis,” PloS One, 16, e0249090.

ALFARO, L., O. BECERRA, AND M. ESLAVA (2020): “EMEs and COVID-19: Shutting Down in a
World of Informal and Tiny Firms,” NBER Working Paper No. 27360.

ALON, T. W., M. KIM, D. LAGAKOS, AND M. VANVUREN (2020): “How Should Policy Responses to
the COVID-19 Pandemic Differ in the Developing World?” NBER Working Paper No. 27273.

ALVAREZ, F. E., D. ARGENTE, AND F. LIPPI (2020): “A Simple Planning Problem for COVID-19
Lockdown,” NBER Working Paper No. 26981.

ARELLANO, C., Y. BAI, AND G. P. MIHALACHE (2020): “Deadly Debt Crises: COVID-19 in Emerg-
ing Markets,” NBER Working Paper No. 27275.

——— (2021): “COVID-19 Vaccination in Emerging Markets,” Unpublished Working Paper,

University of Rochester.

BAIROLIYA, N. AND A. IMROHOROGLU (2020): “Macroeconomic Consequences of Stay-at-Home

Policies During the COVID-19 Pandemic,” COVID Economics Vetted and Real-Time Papers, 13,
71–90.

BARRERO, J. M., N. BLOOM, AND S. J. DAVIS (2021): “Why Working from Home Will Stick,”
Working Paper 28731, National Bureau of Economic Research.

BERGER, D. W., K. F. HERKENHOFF, AND S. MONGEY (2020): “An SEIR Infectious Disease Model
with Testing and Conditional Quarantine,” NBER Working Paper No. 2690.

BESLEY, T. AND T. PERSSON (2013): “Taxation and Development,” in Handbook of Public Eco-
nomics, Elsevier, vol. 5, 51–110.

33
BEWLEY, T. (1977): “The Permanent Income Hypothesis: A Theoretical Formulation,” Journal
of Economic Theory, 16, 252 – 292.

BICK, A., A. BLANDIN, AND K. MERTENS (2020): “Work from Home After the COVID-19 Out-
break,” Federal Reserve Bank of Dallas Working Paper 2017.

B ONADIO, B., Z. HUO, A. A. LEVCHENKO, AND N. PANDALAI-NAYAR (2021): “Global Supply

Chains in the Pandemic,” Journal of International Economics, 133.

BROTHERHOOD, L., P. KIRCHER, C. SANTOS, AND M. TERTILT (2021): “An Economic Model of the
Covid-19 Pandemic with Young and Old Agents: Behavior, Testing and Policies,” Unpublished
Working Paper, University of Mannheim.

CAKMAKLI, C., S. DEMIRALP, AND S. K. OZCAN (2020): “COVID-19 and Emerging Markets: An
Epidemiological Model with International Production Networks and Capital Flows,” IMF
Working Paper No. 20/133.

DAHAL, S., J. M. BANDA, A. I. BENTO, K. MIZUMOTO, AND G. CHOWELL (2021): “Characterizing

All-Cause Excess Mortality Patterns During COVID-19 Pandemic in Mexico,” BMC Infectious
Diseases, 21.

DESHMUKH, Y., W. SURAWEERA, C. TUMBE, A. BHOWMICK, S. SHARMA, P. NOVOSAD, S. H. FU,

L. NEWCOMBE, H. GELBAND, P. BROWN, AND P. JHA (2021): “Excess Mortality in India from
June 2020 to June 2021 During the COVID Pandemic: Death Registration, Health Facility
Deaths, and Survey Data,” Unpublished Working Paper, University of Toronto.

DJANKOV, S. AND U. PANIZZA, eds. (2020): COVID-19 in Developing Economies, London, UK:
Centre for Economic Policy Research Press.

FEENSTRA, R. C., R. INKLAAR, AND M. P. TIMMER (2015): “The Next Generation of the Penn
World Table,” American Economic Review, 105, 3150–82.

FERGUSON, N., D. LAYDON, G. NEDJATI GILANI, N. IMAI, K. AINSLIE, M. BAGUELIN, S. BHA-

TIA , A. B OONYASIRI , Z. C UCUNUBA P EREZ , G. C UOMO -DANNENBURG , ET AL . (2020): “Report
9: Impact of Non-Pharmaceutical Interventions (NPIs) to Reduce COVID-19 Mortality and
Healthcare Demand,” Unpublished Working Paper, Imperial College London.

FLODEN, M. AND J. LINDÉ (2001): “Idiosyncratic Risk in the United States and Sweden: Is
There a Role for Government Insurance?” Review of Economic Dynamics, 4, 406–437.

FUCHS-SCHÜNDELN, N., D. KRUEGER, A. LUDWIG, AND I. POPOVA (2020): “The Long-Term Dis-
tributional and Welfare Effects of Covid-19 School Closures,” Unpublished Working Paper,
University of Pennsylvania.

GLOVER, A., J. HEATHCOTE, D. KRUEGER, AND J.-V. RÍOS-RULL (2020): “Health Versus Wealth:
On the Distributional Effects of Controlling a Pandemic,” NBER Working Paper No. 27046.

GLYNN, J. R. (2020): “Protecting Workers Aged 60–69 Years from COVID-19,” The Lancet In-
fectious Diseases; (published online April 16).

34
GOLLIN, D. (2002): “Getting Income Shares Right,” Journal of Political Economy, 110, 458–474.

GOTTLIEB, C., J. GROBOVSEK, M. POSCHKE, AND F. SALTIEL (2021a): “Lockdown Accounting,”

The B.E. Journal of Macroeconomics.

——— (2021b): “Working from Home in Developing Countries,” European Economic Review,
133.

GREENWOOD, J., P. KIRCHER, C. SANTOS, AND M. TERTILT (2019): “An Equilibrium Model of the
African HIV/AIDS epidemic,” Econometrica, 87, 1081–1113.

HUGGETT, M. (1993): “The Risk-Free Rate in Heterogeneous-agent Incomplete-Insurance

Economies,” Journal of Economic Dynamics and Control, 17, 953–969.

——— (1996): “Wealth Distribution in Life-Cycle Economies,” Journal of Monetary Economics,

38, 469–94.

JONES, C. I. AND P. J. KLENOW (2016): “Beyond GDP? Welfare across Countries and Time,”
American Economic Review, 106, 2426–57.

KAPLAN, G., B. MOLL, AND G. L. VIOLANTE (2020): “The Great Lockdown: Macroeconomic and
Distributional Effects of Covid-19,” Unpublished Manuscript, University of Chicago.

KARAIVANOV, A., S. E. LU, H. SHIGEOKA, C. CHEN, AND S. PAMPLONA (2021): “Face Masks,
Public Policies and Slowing the Spread of COVID-19: Evidence from Canada,” Journal of
Health Economics, 78.

KARLINSKY, A. AND D. KOBAK (2021): “The World Mortality Dataset: Track-

ing Excess Mortality across Countries During the COVID-19 Pandemic,”
Https://doi.org/10.1101/2021.01.27.21250604.

KRUEGER, D., H. UHLIG, AND T. XIE (2020): “Macroeconomic Dynamics and Reallocation in an
Epidemic,” NBER Working Paper No. 27047.

LOAYZA, N. AND S. M. PENNINGS (2020): “Macroeconomic Policy in the Time of COVID-19: A

Primer for Developing Countries,” World Bank Research and Policy Brief 147291.

MOSSONG, J., N. HENS, M. JIT, P. BEUTELS, K. AURANEN, R. MIKOLAJCZYK, M. MASSARI,

S. SALMASO, G. S. TOMBA, J. WALLINGA, ET AL. (2008): “Social Contacts and Mixing Pat-
terns Relevant to the Spread of Infectious Diseases,” PLoS Med, 5, e74.

PRIN, M. AND H. WUNSCH (2012): “International Comparisons of Intensive Care: Informing

Outcomes and Improving Standards,” Current Opinion in Critical Care, 18, 700.

RAMACHANDRAN, S. AND A. MALANI (2021): “All-Cause Mortality During SARS-CoV-2 Pan-

demic in India: Nationally-Representative Estimates Independent of Official Death Registry,”
Unpublished Working Paper, University of Toronto.

SANCHEZ, J. M. (2021): “COVID-19’s Economic Impact around the World,” Federal Reserve
Bank of St. Louis Regional Economist.

35
VON CARNAP, T., I. ALMÅS, T. B OLD, S. GHISOLFI, AND J. SANDEFUR (2020): “The Macroeco-
nomics of Pandemics in Developing Countries: an Application to Uganda,” COVID Economics
Vetted and Real-Time Papers Issue 27, CEPR, 104–122.

YAMALL ORELLANA, J. D., G. M. DA CUNHA, L. MARRERO, R. ISMERIO MOREIRA,

L. DA COSTA LEITE, AND B. LESSA HORTA (2021): “Excess Deaths During the COVID-19 Pan-
demic: Underreporting and Regional Inequalities in Brazil,” Cadernos de Saúde Pública, 37.

36
 Appendix

A. Appendix Figures

Figure A.1: GDP-per-capita Growth from 2019 to 2020

10%

GIN
ETH
TJK VNM IRL
BEN BGD EGY CHN
UZB IRN TUR LTU NZL
0%

TGO UGA
BFATZA CIV LAO SRB
BLR
CAF
MWI CMR PAK
SEN GHA PRIKORAUS NOR
BDI NER
COD KEN SWZ PRY
JORIDN LVA
POL
GMB NIC GTM ALB RUS
ROUEST FIN DNKCHE
MOZTCD
SLEAFG NPL
MLI HTI
SDN NGA
MRT
KHM
UKRGAB
LKABIH
MKD
BGR
KAZ ISR SWE
NLD
USA
LBR GNB BRA
AZE HUN
SVK DEU
RWA ZMB PNG CRI SVN
CZE
CYP SAU QAT
GEOTHA URY CAN SGP
MDG AGO MDA
MNG CHLMYSPRT BEL
HKG
AUT
DZA
Growth rate

MAR ARM
ZAFCOL DOM
GNQ TTOGRC
HRV
SLV FRA
ITA
−10%

ZWE IND BOL ECU MEX BHR

KGZ NAMTUN BWA
COG MMR
TLS HND JAM
PHL ARG GBR
ESP
LSO
IRQPER

MUS
−20%

PAN
LBN
−30%

LBY

0.5 1 2 4 8 16 32 64 128
GDP per capita in 2019, PPP x ($1,000)

Note: GDP-per-capita data comes from the World Bank World Development Indicators. GDP per capita
is expressed at PPP and is taken from the Penn World Table 9.1 (Feenstra et al., 2015).

37
 Figure A.2: Employment Growth from 2019 to 2020
5%

BDI MNG
TZA ZMB
0%

TLS
NER MLI PNG GEOSRB NZL
JPN
CYP
MWI SLE BEN LAO ALB THA POL KOR
GRC
HRV
ROU
HUN FINDNK
GBR NLD
NOR
BFA CMR CIV
MRT NAM BWA LVA ITA
MYS
RUS SVNFRA BEL CHE
MOZTGO SVK SWE
MDG GIN
LSO KEN GHA VNM IDN CHNBLR CZE
LTU DEU IRL
ZWETJK KHM NIC AGO TUN GABGNQMUS
LKA PRT
EST AUT
AUS
SEN ISRSAU
COD TCD GNB
AFG
ETH
RWA SDN SWZ BIH BHR SGP
GMB PRY
MDALBY ESP ARE QAT
LBR UGA NGA UZB JAM IRQUKRLBN TUR
−5%

COG PAK JOREGY BGR

Growth rate

HTI PHL KAZ

OMN
BGD IRN HKG
DZA AZE CAN USA
NPL MMR MAR ARM PRI
IND KWT
KGZ URY
ZAF MKD TTO
BOL MEX
−10%

ECU BRA
DOM
GTM
ARG
CRI CHL
PAN
COL
SLV
−15%

HND

PER
−20%

0.5 1 2 4 8 16 32 64 128
GDP per capita in 2019, PPP x ($1,000)

Note: Employment data comes from the ILO Statistical Database. GDP per capita is expressed at PPP
and is taken from the Penn World Table 9.1 (Feenstra et al., 2015).

38
 Figure A.3: Consumption-per-capita Growth from 2019 to 2020
10%

EGY GEO
5%

NGA
SLE MNG
JOR BGR
BDI BEN
NPL TUR
BGD UZB UKR
UGA SEN VNM
0%

LTU
SRB
BWA EST
SVKCYP
MLITZA NIC CHNBLR POL DNK
TCD IRN
ALB GNQ HUN AUS
HTI CMR
Growth rate

TGO COG GTM LKABIH

MKD KOR
CZEFIN
IDN GRC
ROU
COD SDNPAK DOM HRV DEU
SWE
GNB KHM CRI MYSPRT SAU NLDCHE
−5%

PRYGAB
DZA NOR
ZAFCOL RUS PRI
MRT MDA
BRA LVA CAN IRL
LBR RWA HND MARPHL URY FRABEL
SVN AUT
PER CHL ISR
ITA
IND SLV ECU ESP
NAM
−10%

MMR MEX
JAM ARM GBR
KGZ AGO
ARG
LBN
−15%

MUS

GIN
−20%

0.5 1 2 4 8 16 32 64 128
GDP per capita in 2019, PPP x ($1,000)

Note: Consumption data comes from the World Bank World Development Indicators. GDP per capita is
expressed at PPP and is taken from the Penn World Table 9.1 (Feenstra et al., 2015).

39
 Figure A.4: Excess Deaths Estimated by The Economist
400

Low income Middle income High income

ARM
Excess deaths per 100,000 people
300

MKD
PER
BGR
MEX RUS
SRB
ECU BIH LTU
ALB
BOL MDA ROU
200

AZE ITA
KAZPOLESP
CZE BEL
HRV SVN USA
GEO HUN GBR
KGZ
NIC ZAF
EGY PRT CHE
UKR
COL
BRA SVK
100

XKX AUT
NLD
SLV CHL SWE
FRA
PAN
UZB PRY BLR GRC
LVA CAN
IDNLBN ISR DEU
OMN EST
TUN CRI FIN QAT
THA CYP
KOR DNK NOR IRL
SGP
PHL
JAM MYS JPNAUS
0

MUS
MNG NZL
URY
-100

0.5 1 2 4 8 16 32 64 128
GDP per capita in 2019, PPP x ($1,000)

Note: Data sourced from the Economist excess mortality tracker.

40
 Figure A.5: Excess Deaths Estimated by Karlinsky & Kobak (2021)

PER
600

Low income
Emerging markets
500

Advanced economies BGR

MKD
Excess deaths per 100k people

SRB
400

ECU MEX
RUS LTU
BOL ALB POLCZE
SVK
300

ARM
ZAFBIH ROU
BRA MNE HUN
UKR
MDA HRV ITA
ESP USA
200

AZE
COL KAZ PRT
SVN
XKX CHLLVA
PAN GBR
PRY
LBN EST BEL
KGZ GEO FRAAUT
NLD
TJK NIC EGY ARG SWECHE
100

SLV IRN GRC MLT

UZB
GTM BLR ISR DEU
TUN OMN CYPCAN HKG IRL
THA
CRI FINISL QATLUX
PHL
JAM KOR
JPNAUS
DNK SGPMAC
MUSMYS
NZL NOR
0

MNG URY
-200 -100

SYC

0.5 1 2 4 8 16 32 64 128
GDP per capita in 2019, PPP x ($1,000)

Note: Data sourced from Karlinsky & Kobak (2021)’s World Mortality Database.

41
 Figure A.6: Official COVID-19 Deaths in the United States, Mexico and Ghana

Note: This figure plots cumulative official deaths from COVID-19, according to Our World in Data, in the
three focus countries: the United States, Mexico and Ghana.

42
 Figure A.7: Oxford Lockdown Stringency Index
100
80

HND LBY ARG

PER CHN
NPL BGD IND BOL
PHL AZE
SLV IRQPRYCOLDOM CHL
GTM KAZPAN
Stringency index

MMR OMN PRI

UGA
RWA ZWE JAM GAB KWT
MAR DZA
MNG GEO
LBN ITA QAT
60

KGZ AGOVNM JORECU TTO

KEN BRA MEX ISR GBR
ETH SWZ MYS ESP SAU USA IRL
HKG
TCD
LBR PAK IDN
EGY ALB CRI TUR BHR
CAN
AUS
GMBGIN
NGA ZAF
UKR
MDA CYPFRA
GRCPRT
DEU
MOZ COG
SDN UZB BWA BEL
LSO IRN BIHSRB ROU SVN
RUSHUN KOR SWE NLDARE SGP
MDG
TGO LKA THA POL AUT
MWI
COD HTI SVK CZE DNK
NAMTUN HRV
CMRPNGGHA LVALTU CHE
NOR
CAF AFG
MLI SEN BGR
URY
40

ZMB MRT FIN

TJK KHMCIV
BEN
ESTNZL
SLE BFA LAO
TLS JPN
MUS
NER BLR
TZA
20

BDI NIC
0

0.5 1 2 4 8 16 32 64 128
GDP per capita in 2019, PPP x ($1,000)

Note: The Government Stringency Index is taken from the Oxford Government Response Tracker (Ox-
CGRT). GDP per capita is expressed at PPP and taken from Penn World Table 9.1 (Feenstra et al., 2015).

43
 Figure A.8: Pandemic Spending as Share of GDP

USA
25
Pandemic spending to GDP ratio (%)
20

NZL

GBR HKG
TLS JPN AUS SGP
15

CAN
GRC

AUT
DEU
LSO
MUS ISR
10

HUN IRL
BRA LVA ITA
MNG THA CHL BEL
POLESPFRA
PER SVN
CYP CHE
LTU
TGO RWA
TCD
KGZ ZAFGEOSRB BHR
GNB BOL HRVPRTCZE DNK
BDI LBR MRT IRN CHN
5

UZB IDN LBY BGRMYS

RUS NLD
SVK KOR SWE
COD KHM COL
MKD ARG NOR
BFA ZWE
SLE MLI SEN GTM
IND SWZ UKR PANEST
GMB TJK GHA PHLTUNPRYBIH
ETH BEN KEN HND MAR
COG SLV TTO
KAZ FIN ARE
AFG AZE ROU SAU
MOZ MDG UGAGIN ZMB PAK ALB
GABDOM TUR
CAF HTI CIV
BGD
NIC
MMR JAMEGY
VNMNAM MDA BWACRI KWT
NER
MWI SDN
CMR
PNG JOR
ECU ARM URY
NGA DZALKA GNQMEX
BLR QAT
LAO IRQ
0

0.5 1 2 4 8 16 32 64 128
GDP per capita in 2019, PPP x ($1,000)

Note: Data on pandemic spending come from the IMF Fiscal Monitor Database. GDP per capita is ex-
pressed at PPP and taken from Penn World Table 9.1 (Feenstra et al., 2015).

44
 Figure A.9: Economic Support Index
100

CYP HKG
80

GBR
IRL
ISR AUT
Economic support index

THA BHR
JPN SGP
SVK ESP
ROU DNK
GAB BEL
NLD
HND
URY TUR PRTNZL
60

MWI ECU COL ARGCHL FRA

AUS
PER GRCPRICZE
MYS
HRV
SLVTUN
EGY
PRY ITAFIN
SVN
LVALTU
PAK MAR BGR
IND MNG CAN
RWA SEN PNG GTM ZAF MUS PAN EST
HUN USA
NPL ALB OMN SAU
CIV LAO SWE
TLS BGD POL KOR QAT
40

TGO KGZ UZB JAM IRNBRA SRB CRI RUS ARE

BEN
HTI PHL DZA LBN DEU
TCD AZE
LKA CHE
IRQ BWA
MDG MLIGIN TJK VNM JOR UKR CHN NOR
CAF UGA KEN MDA
COD BOL GEO KWT
LSO KHM SWZ IDN KAZ
ETH SDN MRT AGO TTO
20

ZWEZMB DOM
MMR MEX
NER COG
CMR NAM
NGA
GHA BIH
SLE
BDI
GMB
AFG
MOZ
LBR BFATZA NIC LBY BLR
0

0.5 1 2 4 8 16 32 64 128
GDP per capita in 2019, PPP x ($1,000)

Note: Oxford Coronavirus Government Response Tracker’s Economic Support Index.

45
 Figure A.10: Fraction of the Population Older than Age 65

30

JPN
Population over 65 (% of Population)

ITA
PRT
GRC FINDEU
BGR
20 HRV
LVA FRASWE
MLT DNK
EST
LTU ESP NLD
SVN
CZE AUT
HUN GBRBEL
SRB ROU
POL CAN
UKR
BIHBRB
SVK NZL AUSUSA
GEO BLRURY
MNE RUS ISL
KOR
ALBMKD CYP
ABW

THA ISR
MDA ARM ARG MUS CHL
CHN
LKA TTO
10 VCT GRD CRI
JAM BRA ATG TUR
SLV TUN
PERCOL PAN
BOL MEX SYC
KAZ BHS
VNM VEN
MAR ECU SUR DOM
LBN
PRY DZA MYS
NPL IND BTN IDN IRN
AZE
MMR FJI EGYZAF
HTI BGD
LSO NIC PHL
HND
DJI SYR
KHM GTM
CPV
PAK LAO BLZ BRN
SWZ UZB MNG BWA TKM
ETH BEN SDN JOR NAM
GAB MDV
LBR
COD SLE
MDG GIN MRT
COMSEN
TJK
STP NGA GHA IRQ SAU
MWIMOZ
CAF NER TGO
TCD BFA
ZWE
YEM
RWA
GNB GMB CMR
MLI KEN COG GNQ KWT
OMN BHR
BDI UGA ZMB AGO

0
1 2 4 8 16 32 64
Real GDP per capita in 2017 x $1,000

Note: This figure plots the proportion of population ages over 65 and above as a percentage of total population
across 162 countries. GDP per capita is from Penn World Table 9.1 (Feenstra et al., 2015). Population data
is World Bank staff estimates using the World Bank’s total population and age/sex distributions of the United
Nations Population Division’s World Population Prospects: 2019 Revision.

46
 Figure A.11: Hospital Beds per 10,000 People

150

JPN
Hospital Beds per 10,000 Population

KOR
BLR

100
RUS
UKR LVA
TKM
DEU
AZE LTU
KAZ AUT
HUN
MNG
BRB POL FRA
MDA ARM BGR ROU
BEL
TJK SVK
SRB EST
UZB HRV MLT
50 ARG
GRC SVN
KGZ GEO CYPITA NLD
FINDNK
CHN MDV
MNE
ATG SYC AUS
ESP ISL
LKA
BIH MUSPRT
SUR BHS ISRSAUSWE
ZWE STP ALB
NAMZAF LBN URY GBR USA
NZL
BHRTTO
CANBRN
VNM GRD
FJI BRA
COM CPVSWZ EGY THA PAN
CHL
TUR
ZMB TUN GNQ MYS KWT
JAM
BTN BWA
JORECUDZADOM MEX OMN
RWA CMR COG PER
LCA
COL
KEN
LSODJI SYR
CAF MWI VEN
PRY IDNGAB CRI IRQ
TGOGNB GMB PAK
GHA
NIC
MMR
BOL
SLVMAR
BLZ
BDILBR MOZ HTIYEM MLI BGD
KHM SDNAGO IND
HND
BEN NGA GTM
PHL
NER TCD
SLEBFA
UGAGINNPL MRT
MDG SEN
ETH IRN
0
1 2 4 8 16 32 64
Real GDP per capita in 2017 x $1,000

Note: This figure plots the number of hospital beds available per 10,000 inhabitants in 153 countries. GDP
per capita is at PPP and taken from the Penn World Table 9.1 (Feenstra et al., 2015). The hospital bed data
are from the World Health Organization’s Global Health Observatory.

47
 Figure A.12: Size of the Agricultural Sector

90

80
MOZ RWA
UGA
Share of Agricultural Employment

KHM
70 MLI
SLE
MWI ZMB
NGA CHN
TZA
CMR
60
THA
LBR VNM
IND
50
MNG TUR
KGZ GHA
IDN
40
ARM
SDN MAR
PHL
BOL
30 COL
PRY ROM
FJI
IRQ PER
ECU LCA
20 IRN
JAM BRA MYS
CRI PAN
MEX
CHL
10 DOM VEN GRC
ITA IRL
JOR HUN SVNESP AUT CHE
PRT DEU
ISR FRA
CAN
GBRNLD
USA
0

1 2 4 8 16 32 64
Real GDP per capita in 2017 x $1,000

Note: Agriculture employment data is taken from the IPUMS database. GDP per capita is expressed at PPP
and is taken from the Penn World Table 9.1 (Feenstra et al., 2015).

48
 Figure A.13: Changes in Mobility Across Countries During Lockdown Periods

60
Average Percentage Change from Baseline

40 ZWE BOL BWA

PAN SGP
ECUPER MYS
ZAF PRT
TUR LUX
RWA BRB
DOM
COL ITA
FRA
NZL
GHAIND PRY ESP
During Lockdown Period

NPL HND JOR ARG BEL ARE

GBR
BGD JAM
SLV GEO CRI
NAM CHL SAU AUT IRL QAT
20 NGAPAK VEN
FJI IRQ
LBN
GRC
OMN KWT
HRV
SVK TTO
BRATHA HUN
ROUPOL
LTU FIN AUSNOR
USA
NLD
DNK
DEU
MNG

-20 MNG
NGA FJI THA
AUS QAT
BRA TTO USA
POL KWT DEU
-40 GHA SLV
CRI
OMN SAU
CHL
ROU LTU FINNLD ARE
HUN
PRY NAM
VEN
PAK INDJAM SVK DNK
NOR
ARG
RWA IRQ HRV
GRC
BGD IRL
LBN
ZWE COL
PER BWA PRT NZL AUT
MYS
-60 HND ZAF
GEO ITABEL
GBR
FRA
ESP
JOR DOM SGP
NPL TUR LUX
BRB PAN
ECU
BOL
-80
1 2 4 8 16 32 64
Real GDP per capita in 2017 x $1,000

Workplace Places of Residence

Note: This figure plots the average percentage changes of the mobility metric in the ’Places of Residence’ and
’Workplace’ categories in the Google Community Mobility Report (Aktay et al., 2020), during the lockdown
periods for the 65 countries which had implemented or are implementing lockdown. GDP per capita is from
Penn World Table 9.1 (Feenstra et al., 2015). The average across all 65 countries is 23.44 percent. The slope of
the fitted line is 1.52, with p-value of 0.354 for the ’Workplace’ category. For the ’Places of Residence’ category,
the slope of the fitted line is -1.52, with p-value of 0.083.

49
Figure A.14: Time-Series of Lockdown Policies and Economic Support in the United States

90 100 Workplace closure

Lockdown stringency
80
70
60
Index

Economic support
50
40
30
20
10
0

Jan. March May July Sept. Nov. Jan. March May July Sept.
2020 2020 2020 2020 2020 2020 2021 2021 2021 2021 2021

Note: This figure displays the time-series of Oxford Lockdown Stringency Index, Economic Support Index,
and Workplace Closures for the United States.

50
B. Appendix Tables

Table B.1: ICU Bed Availability Across Countries

Country ICU beds per 100,000 Per capita healthcare cost

population

United States 20.0-31.7 $7,164

Canada 13.5 $3,867
Denmark 6.7-8.9 $3,814
Australia 8.0-8.9 $3,365
South Africa 8.9 $843
Sweden 5.8-8.7 $3,622
Spain 8.2-9.7 $2,941
Japan 7.9 $2,817
UK 3.5-7.4 $3,222
New Zealand 4.8-5.5 $2,655
China 2.8-4.6 $265
Trinidad and Tobago 2.1 $1,237
Sri Lanka 1.6 $187
Zambia 0 $80

Source: Table 1 in Prin and Wunsch (2012). Healthcare cost includes all public and private expenditures.

51
 Table B.2: Internally Calibrated Parameters and Model Fit

Data Model Parameters Description

U.S. GDP Decline, ’19-’20 -4.10% -4.01% A(P) Pandemic TFP

U.S. Employment Decline, ’19-’20 -6.40% -6.36% φs Productivity of remote work, social sector
Fraction Remote Workers pre COVID 8.20% 8.14% σg Variance of remote work taste shock

52
 Table B.3: Multiple Correlates of GDP per Capita Change from 2019 to 2020

Dependent variable: GDP per capita change from 2019 to 2020

Independent variables (1) (2) (3) (4) (5)

GDP per capita in 2019 -0.10** 0.037 -0.052 -0.027 -0.052

(0.046) (0.068) (0.096) (0.096) (0.11)

GDP per capita in 20192 0.0014** 0.00021 0.00095 0.00066 0.00084

(0.00066) (0.00071) (0.00098) (0.00096) (0.0011)

Agriculture emp. share 0.076*** 0.083*** 0.065** 0.062**

(0.027) (0.029) (0.029) (0.030)

Median age 0.12 0.080 0.074

(0.082) (0.082) (0.082)

Lockdown stringency -0.12*** -0.13**

(0.043) (0.053)

Economic support 0.024

(0.038)

Constant -4.21*** -8.03*** -10.6*** -3.09 -2.97

(0.60) (1.66) (2.53) (3.22) (3.34)

Observations 144 144 144 140 140

R2 0.031 0.071 0.084 0.157 0.163
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

53