Alcedo 2022 Back To Trend Covid Effects On E Commerce In 47 Countries

Summary

NBER Working Paper No. 29729, Back to Trend: COVID Effects on E-commerce in 47 Countries, by Joel Alcedo, Alberto Cavallo, Bricklin Dwyer, Prachi Mishra and Antonio Spilimbergo, dated February 2022 and revised October 2023; it first appeared as IMF Working Paper Number WP/22/19. Using Mastercard online transaction data for 47 economies and 26 industries from January 2018 to April 2023, it studies e-commerce during the pandemic. It reports that the online share of credit card transactions surged, especially as governments made transfers to households in lockdown, and returned to pre-pandemic trends in almost all countries as fiscal support and mobility restrictions waned. It finds little evidence of long-lasting structural change in e-commerce spending.

Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used

Cite this document

“Alcedo 2022 Back To Trend Covid Effects On E Commerce In 47 Countries.” Original at doi.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/alcedo-2022-back-to-trend-covid-effects-on-e-commerce-in-47-countries-e0212cc946/, SHA-256 a9959e5d25e83c49ccc8bf0585e6afcd6f231951ef0c90c575fb0fba707e0a74.

Full text

NBER WORKING PAPER SERIES

BACK TO TREND: COVID EFFECTS ON E-COMMERCE IN 47 COUNTRIES

Joel Alcedo
Alberto Cavallo
Bricklin Dwyer
Prachi Mishra
Antonio Spilimbergo

Working Paper 29729
http://www.nber.org/papers/w29729

NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
February 2022, Revised October 2023

We thank Gita Gopinath for several discussions during initial stages of the project. Sergio Garcia
Monroy and Swapnil Agarwal provided excellent research support. All errors remain our own.
The views expressed here are those of the authors and do not necessarily represent those of the
National Bureau of Economic Research, the IMF, its Management and Executive Board, or IMF
policy. This paper first appeared as IMF Working Paper Number WP/22/19.

NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.

© 2022 by Joel Alcedo, Alberto Cavallo, Bricklin Dwyer, Prachi Mishra, and Antonio
Spilimbergo. All rights reserved. Short sections of text, not to exceed two paragraphs, may be
quoted without explicit permission provided that full credit, including © notice, is given to the
source.
Back to Trend: COVID Effects on E-commerce in 47 Countries
Joel Alcedo, Alberto Cavallo, Bricklin Dwyer, Prachi Mishra, and Antonio Spilimbergo
NBER Working Paper No. 29729
February 2022, Revised October 2023
JEL No. E2,F00,L81,O3

ABSTRACT

We study E-commerce across 47 economies and 26 industries during the COVID-19 pandemic
using online transaction data from Mastercard. The online share of total credit card transactions
surged during the pandemic, especially as governments made transfers to households in
lockdown. As the fiscal support and mobility restrictions waned, online shares went back to pre-
pandemic trends in almost all countries. We find little evidence of long-lasting structural changes
in E-commerce spending patterns.

Joel Alcedo Prachi Mishra
Mastercard Economics Institute IMF Research Department
2000 Purchase St 700 19th Street NW
Purchase, NY 10577 Washington, DC 20431
Joel.Alcedo@mastercard.com pmishra@imf.org

Alberto Cavallo Antonio Spilimbergo
Harvard Business School International Monetary Fund
Morgan Hall 287 700 19th Street, N.W.
Soldiers Field Washington, DC 20431
Boston, MA 02163 aspilimbergo@imf.org
and NBER
acavallo@hbs.edu

Bricklin Dwyer
Mastercard Economics Institute
2000 Purchase St
Purchase, NY 10577
Bricklin.Dwyer@mastercard.com
1. Introduction

E-commerce surged when the COVID-19 pandemic hit in early 2020. 1 In a matter of a few
weeks, as lockdowns were introduced, and in-person purchases were severely limited, online
transactions increased to unprecedented levels in most economies. The sudden surge in online
retail transactions attracted a lot of attention in the media, but there is little academic research
documenting the facts. How big and widespread was the digitalization of consumption during
this period? Were there significant differences across countries and sectors? Perhaps more
importantly, were the effects persistent? Did the COVID-19 pandemic fundamentally alter the
trajectory of E-commerce, or did it merely represent a transient deviation in the ongoing
evolution toward a more digitized retail landscape?

This paper investigates these questions by leveraging a unique database provided by Mastercard
©. The data cover all transactions within the Mastercard network. Our sample comprises of 47
economies and 26 sectors, where Mastercard is more representative of total card transactions, for
the period between January 2018 and April 2023. 2 The time span allows the analysis of both
online and in-person patterns of spending for three years before and three years after the
COVID-19 pandemic started.

We find that while online spending shares surged during the pandemic, after three years, they
dissipated in almost all countries, and across most sectors, with the exception of a few industries
within the retail and healthcare sectors. The findings are consistent with higher government
transfers supporting online spending in the presence of pandemic restrictions, with the effects
dissipating as both fiscal support and mobility restrictions have waned.

1
For the pre-pandemic period, the U.S. Census Bureau estimates that e-commerce rose from 5 percent in 2007 to 11
percent in 2019 (https://www.census.gov/retail/index.html).
2
The list of 47 economies is in Table 1.

1
The rest of the paper is organized as follows. Section 2 reviews the related literature; Section 3
describes the data; Section 4 documents stylized facts; Section 5 explores explanations for the
observed patterns. Section 6 presents additional findings. Section 7 concludes.

2. Related literature

This paper is related to and makes contributions to three strands of literature.

First, a growing body of literature seeks to quantify the benefits to consumers from the internet,
including Goolsbee and Klenow, 2006; Brynjolfsson and Oh, 2012; and Varian, 2013. Related
to our paper in the use of credit card information, Dolfen et. al. (2020) use transaction-level
data from the United States obtained from Visa, Inc. between 2007 and 2017, and estimate that
E-commerce spending reached 8% of consumption by 2017 in the US, yielding consumers the
equivalent of a 1% permanent boost to their consumption. These papers consider consumer
benefits during normal times; but the internet can also bring benefits in terms of resilience to
exceptional shocks such as the COVID-19 pandemic. For instance, Alipour et. al. (2022) uses
data on credit and debit card transactions from Mastercard for German cities to evaluate
geographical relocations of offline consumption after the COVID-19 shock. Auer, Cornelli,
and Frost (2023) assemble a cross-country database on broad retail payment behavior using a
variety of public and proprietary sources; their core dataset spans 18 countries over the period
from December 2019 to December 2020. 3 They document that cash in circulation, use of card-
not-present transactions, and downloads of payment apps all spiked during the early phase of
the pandemic, with less pronounced changes for countries with greater pre-pandemic mobile
adoption; they also note that recent data show that some of the effects have not lasted beyond
the acute phase of the pandemic. For example, “card not present” transactions, or payments
executed remotely without the presence of a physical card, fell again as lockdowns ended, even
below pre-pandemic levels in some jurisdictions. Our paper contributes to this literature by
extending the coverage of the credit and debit card data to 47 economies to look at the more
recent COVID-19 period. We do not consider welfare questions and focus exclusively on the
dynamics of online spending, using the COVID-19 pandemic as a natural experiment. Our

3
Auer, Cornelli, and Frost (2023) note that the “proprietary series were obtained by the authors from the networks,
subject to data agreements”.

2
findings on the lack of persistence in online shares are consistent with Auer, Cornelli, and Frost
(2023), though the latter use completely different data sources.

Second, our paper is related to the large literature on the economic effects of COVID-19 (e.g.,
Bloom, Fletcher and Ye, 2021, Chetty et. al., 2020). Soon after COVID-19 hit, some scholars
predicted significant consequences for the retail industry (Roggeveen and Sethuraman, 2020).
Wang et al. (2022) provide an extensive literature review on the effect of COVID-19 on a
variety of consumer behaviors, ranging from purchases of (un)healthy food, panic buying,
impulsive buying and stockpiling, among others. Historically, consumer behavior has
frequently evolved, transitioning from itinerant merchants to mail-order catalogs, and
subsequently from convenience stores to department stores (Moon et al. 2021). Various
channels have frequently coexisted, with multiple theories attempting to elucidate preferences
for different retail avenues. Protection Motivation Theory, for example, posits that perceived
risk levels influence shopping preferences, whereas the Theory of Planned Behavior
underscores the role of planning and intentions. Moon et al. (2021) contends that, in the
context of COVID-19 in Korea, the evidence favors the relevance of Protection Motivation
Theory. Sheth (2020) argues that changes in consumer habits would persist after the pandemic.
Keane and Neal (2021) document consumer panic or hoarding during the pandemic; and
although this does not necessarily lead to more online expenditure but, in period of lockdown,
few alternatives were available. In sum, there are several reasons why the surge in E-commerce
could have been contingent to COVID-19 pandemic with a transient effect (consistent with, for
example, theories of panic or impulsive buying), or alternative theories suggestive of more
permanent changes in consumer behavior. Whether the surge was a permanent shift or simply a
large (but temporary) spike is ultimately an empirical question.

Finally, our paper is part of the growing literature that uses private data to improve and
complement survey-based methods of economic measurement. Examples include the use of
online price data for inflation measurement in Cavallo (2013) and the recent work by Chetty et
al. (2020) and Carvalho et al (2020) to track economic activity in real-time during COVID-19. 4
Closer to our paper, Alandangady et al. (2019) uses credit and debit card transactions to create

4
Other important examples are Choi and Varian (2012), Einav and Levin (2014), Glaeser at al (2017), and Abraham
et al (2020).

3
daily estimates of retail spending that can approximate the official Census retail surveys in the
United States. Our work contributes to this literature by showing how real-time credit card data
can be used to measure E-commerce sales and improve the understanding of consumption
patterns during times of crisis in many economies.

3. Data & Methodology

The digitalization of retail transactions (e-commerce) can be evaluated through various metrics.
While the total volume of online spending saw an uptick during the COVID-19 pandemic, the
overall consumer expenditure in the U.S. recovered too by early 2021 (Chetty et al., 2023). Our
analysis, therefore, focuses on the share of online spending, defined as the ratio of online
transactions to the total number of observable transactions. This share is indicative of both the
preferences and constraints of consumers and sellers with respect to online transactions.

Credit and Debit Card Data

Our primary dataset is the universe of all credit and debit card transactions that were cleared
through the Mastercard network between January 2018 and April 2023 in more than 200
economies and territories. 5 We limit our sample to 47 economies where Mastercard has a
significant market share of total card transactions, accounting for over 20 percent of the card
market. 6 Online Appendix A provides a more detailed description of the data.

The data contain the total dollar amount of the transaction, and a sectoral classification for the
merchant associated with the transaction, spanning 26 sectors. Importantly, although the credit
card data capture the aggregate transaction amounts, they do not include information on the
specific items bought, along with their prices and quantities. We define "online" spending as any
transaction wherein the card—be it physical or virtual—was not present at the point of sale. This
broad categorization encompasses a variety of payment methods, including internet-based
transactions via web browsers or mobile devices, as well as telephone or mail-order purchases.

5
Nilson Report issue 1199 (June 2021) noted the global share of purchase transactions for Visa at 40%, UnionPay at
32%, Mastercard at 24%, and Rest at 4%.
6
Appendix B shows that the online share patterns are similar when we use data from economies that are below that
market share threshold.

4
All other transactions are classified as "offline." The transaction data are aggregated on a
monthly basis for each country.

For each economy in the sample, the “online spending share” of total card consumption in
economy c at time t, given by

(1)

𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑐𝑐,𝑡𝑡
𝑠𝑠𝑐𝑐,𝑡𝑡 =
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑐𝑐,𝑡𝑡

𝑠𝑠𝑐𝑐,𝑡𝑡 is calculated using monthly transaction level data from Mastercard.

We normalize all series to 100 before the pandemic and report the normalized shares. 7
Importantly, we can compare the behavior of these shares over time across countries, and in
particular the degree of reversion to pre-pandemic trends. We also report the deviations from
pre-pandemic trends.

4. E-commerce during COVID: Heterogeneity and transience

In this section we, present stylized facts on online shares across economies and sectors. Although
the levels and dynamics differ, we find surprising similarity in the transitory nature of the e-
commerce surge across countries.

Online and total expenditures

Before looking at the share of transactions that take place online, it is helpful to understand the
behavior of online and total expenditure separately. During the first few months of the pandemic,
two forces were at work. On one hand, consumers did not spend, so total expenditure fell. On the

7
For confidentiality reasons, we are not able to publish the online shares at the country-level. We can, however,
report the online shares (without normalization) at the sector level (Tables 4 and 5), and also use these at
country-time levels in the empirical analysis (Tables 2 and 3).

5
other hand, the lockdowns were forcing those who wanted to spend to go online. In some
countries this led to an immediate increase in online spending. In others, online expenditure fell,
but by less than total expenditure.

These two cases are illustrated in Figure 1, which shows the amount of total and online spending
for the US and Brazil. Consumers’ spending is seasonal, so we present seasonally adjusted
indices using X13-ARIMA. In both countries, online spending was much higher three years after
the pandemic started. But the short-term dynamics were different. The US illustrates the case
where online expenditure rose despite the contraction of total spending. Online expenditure in
Brazil, on the other hand, fell but by less than total expenditure. As a result, the share of online
expenditure rose in both countries.

We further note that the increase in online spending was not caused by a contraction in the
availability of cash (which can only be used offline). Indeed, Figure 2 shows some indicators for
cash in both the US and Brazil. The top charts (Figure 2a) show a measure of stock of cash in
circulation from the IMF’s World Economic Outlook database. In both countries, the availability
of cash increased at the onset of the pandemic, reflecting the effects of expansionary monetary
policies. It continued to grow in the US while partially retracted in Brazil. The bottom charts
(Figure 2b) show ATM withdrawals from Mastercard, a flow measure which can reflect the
preference for holding cash on the part of consumers. In the US, cash withdrawals increased in
mid-2020 and early 2021, peaking between 40% and 55% higher than before the pandemic
started. In Brazil, ATM withdrawals fell by about 40% in 2020 in line with the movements in
total expenditure, but increased thereafter, and reached pre-pandemic levels by end of 2022.

Online shares in 47 countries

Next, we document stylized facts on the evolution of online shares in 47 countries. We report the
shares at two points in time (peak relative to 2019 average, and latest available, i.e. April 2023),
and focus on the deviations during the COVID-19 pandemic with respect to economy-specific
trends. In all cases, the pre-COVID trend is estimated in each economy using a regression of
online shares on a monthly time trend between 2018 and 2019.

6
Figure 3 provides two examples of the dynamic behavior of the online share 𝑠𝑠𝑐𝑐,𝑡𝑡 in the retail
sector (solid line) compared to the pre-COVID-19 trend in each economy (dashed line). In the
United States, the online share of spending in retail peaked shortly after the pandemic started,
rising nearly by 30%. This share rose again when the economy faced waves of COVID-19 cases,
and eventually reverted to the level predicted by the pre-COVID trend by early 2023. In Brazil
the online share initially had a similar pattern but has fallen in 2022 and is now at a level below
the one predicted by the pre-pandemic trends.

Table 1 summarizes the dynamic behavior of the online share for our sample of 47 economies.
Results are ordered by the latest deviation relative to the pre-COVID trends. The shares reported
in Column 1 are normalized to the average level for 2019, so these numbers need to be
interpreted as indices relative to the pre-pandemic levels. Column 2 reports the online shares in
April 2023 (latest available), while 3 Column reports differences relative to pre-pandemic trends.
As shown at the bottom of the table, on average the online share rose by 57 percent at the peak,
and then fell to 16 percent in 2023. Although current levels are 16 percent higher than before the
pandemic started, they are, in fact, only 1 percentage point above the average level predicted by
pre-pandemic trends. Median values tell a similar story of a surge and then a return back to trend
over time.

The transitory nature of surge can be further illustrated by computing the global (weighted)
average difference between the online shares and the predicted pre-crisis trends, both at the peak
and with the latest data. 8 We find that the share of total transactions was 3.1 percentage points
higher at the peak, but this number fell to just 0.6 percentage points by April 2023. At a global
level, therefore, on average, only one-fifth of the deviation during the peak persists in the latest
data.

Although on average countries are back to the levels predicted by pre-pandemic trends, there are
still significant differences across countries in the numbers shown in the last column. In 30

8
The average global deviation is calculated using the following 4 steps: i) calculate the deviations of the online
shares from trend in each economy (as reported in Table 1), then ii) calculate the weight of each "economy-wide
deviation" as the share of Mastercard spending of that economy over the total spending in the world, and iii)
multiply each economy-wide deviation by the weight, using the Q3 2021 weight and (iv) compute the weighted
sum which is defined as the average global deviation.

7
percent of the economies, these deviations are still numerically positive, though small and close
to zero in all of these. For some smaller economies such as Bahrain, Croatia, and Slovenia, the
shares are significantly higher. 9 But in all the other economies, including the United States and
many developed economies, the online shares are now almost close to or significantly below the
predicted pre-COVID trend levels.

5. What explains transience and heterogeneity in online shares?

The previous section has shown that E-commerce jumped at the onset of the pandemic, but the
share of E-commerce quickly receded to pre-COVID-19 trend, with some heterogeneity across
countries.

What can explain the heterogeneity across economies in the level and persistence of online
shares? Figure 4 illustrates the strong correlation between global online shares and the strictness
of the COVID-19 pandemic movement restrictions, as measured by Google’s index of residential
mobility, especially at the beginning of the crisis in the second quarter of 2020, when the
lockdowns severely restricted mobility in most economies. However, the correlation declined as
the pandemic continued. This is consistent with the impact of COVID-19 lockdowns and other
restrictions on economic activity declining over time as economic agents learned how to cope
with restrictions (see e.g. ECB, 2021).

To explore the issue in a more systematic way, we present a panel regression of the deviation in
online shares from the predicted pre-COVID-19 trends on mobility, fiscal support during
COVID-19, and other covariates (Table 2). 10 Given that levels may be picking up cross-sectional
differences in pre-existing capabilities, we estimate the regression in changes. The dependent
variable is measured by the change in online shares over the previous month; time-varying
explanatory variables are also measured in differences. Time fixed effects are included in all
regressions, but we do not include country fixed effects as several controls are time-invariant. 11

9
Note that some cross-sectional variation was normal also before the pandemic. The cross-sectional variation after
COVID-19 is smaller than before the pandemic.
10
Summary statistics for all the variables used in the regressions are reported in Table A1.
11
The main findings are robust to estimation in levels (Table 3). The only time varying variables are Google
mobility and number of new cases. The results reported in Column (1) of Table 2 remain robust when we include
country fixed effects.

8
Column [1] shows that residential mobility and intensity of the pandemic, as measured by the
number of new cases, correlate positively with the deviation from trend in the latest numbers. In
Column [2], we include a number of additional controls. Residential mobility and intensity of the
pandemic continue to be important drivers of online shares. Richer economies also returned
faster to trend once the pandemic receded, as suggested by the negative coefficient on per capita
income, though the estimates are statistically indistinguishable from zero.

In Column [3], we include the interaction of fiscal support during COVID-19 with the severity of
the restrictions. The interaction term is positive and statistically distinguishable from zero at
conventional levels. The estimates indicate that greater fiscal support was associated with higher
online shares when restrictions were more severe. Therefore, government transfers supported
spending by increasing consumption, which, in the presence of pandemic restrictions, could
mostly be done online.

Notably, the average effect of fiscal spending on online shares (when evaluated at the average
change in mobility) is also positive in Column [3]. The average change in mobility in the sample
is 36.6 percentage points, therefore, the average effect of fiscal spending on E-commerce shares
can be computed based on the coefficients in Column (3) as -0.0012+36.6*0.0063=0.23.
Notably, a one percentage point (pp) higher change in residential mobility is associated with a
0.4 pp higher increase in online share gap, almost double the average effect of a one pp higher
fiscal support.

Firm and Consumer Online Capabilities

One potential explanation for the rise in online shares observed during the pandemic is the fact
that consumers and firms learned to transact online. In fact, some commentators anticipated a
‘collateral benefit’ of the lockdown period; according to this, many buyers and sellers would
have been ‘obliged’ to learn online shopping. However, the lack of persistence in the surge in
online shares suggests that this online capability was likely not a binding constraint.

9
To explore this further, we develop two new monthly measures of online learning at the country
level. The first is “Firm capability”, defined as the proportion of active firms that ever sold
online since the start of the sample in 2019. The idea underlying this measure is that sellers who
learned how to do transactions online and set up the necessary infrastructure have paid the fixed
cost and will be able to do online transactions also in the future. The second is “Consumer
capability”, defined as the proportion of active consumers that ever bought online, building on
the same idea. Figure 5 plots the median consumer and firm capability measures over time. On
average, the measures of online capability kept increasing over time, suggesting that the online
capabilities are more widespread than ever before. The fact that the share of online transactions
has fallen from its peak, may not, therefore, be consistent with the story that technological
constraints were not binding on average.

At the same time, such constraints could still play some role in explaining some of the
differences we observe across countries. We included these measures of online capabilities in the
regressions in Table 2, column 4. 12 We find the estimated coefficient on firm capability to be
positive and statistically significant, while consumer capability is statistically indistinguishable
from zero.

Robustness checks

Table 3 reports additional checks to confirm the results presented in Table 2. Column [1] repeats
the final column in Table 1 as the baseline. Columns [2]-[3] includes alternative measures of
mobility (retail, and grocery and pharmacy, which are negatively correlated with pandemic
restrictions) and Column [4] uses the Oxford stringency measure. Column [5] drops the internet
penetration variable. Columns [6]-[7] employ alternative specifications – with changes measured
over 12-month period, and estimation in levels rather than changes respectively. The findings
remain broadly similar, and in particular, the result that government transfers increased online
consumption – especially during times of high restrictions on mobility – continues to be strong
and robust.

12
The number of observations reduces in Column [4] due to insufficient information to compute consumer and firm
capabilities. Appendix Table A2 repeats Table 2 with all specifications repeated on the smaller sample. The
results remain robust.

10
Notably, the estimated coefficients on online capabilities are not robust to alternative
specifications. Overall, we interpret the evidence as providing little evidence of learning effects.

6. Additional Findings

In this section we explore the role of specific sectors that were particularly affected. We also
discuss if the Mastercard data comprise a representative sample.

Restaurants, bars, and some retail categories experienced the largest increases in online
spending shares, though divergence across sectors does not appear to persist.

Table 4 presents a snapshot of the share of online spending by sector, which is perhaps most
revealing. 13 It shows the online share by sector pre-COVID, at the peak, at the latest time, and
differences between the latest observation and the pre-COVID trend. Among these broad
categories, restaurants-bars had the largest increase in online spending shares, rising more than
25 percentage points from 7.3 percent in 2019 to 32.6 percent at the peak. At peak values,
services and retail also had large percentage point increases, though lower than that for
restaurants-bars. By April 2023, however, the deviation turned negative for all the restaurants-
bars, retail, and services sub-sectors.

These broad categories mask substantial heterogeneity across subsectors. Table 5 shows the
difference between the latest online shares and the pre-crisis trend for 26 disaggregated sectors in
our dataset. In particular, auto-rental, clothing stores, drug stores, and electrical appliances still
report high online shares. Some of these sectors, in particular, drug stores, also had relatively low

13
Sectoral averages are created by computing a weighted average of the deviation in online share from the trend at
the economy-sector level, where the weights are the shares of Mastercard spending in the economy-sector as a ratio
of global spending in that sector. Notably, there are few sectors (e.g. airlines and travel agencies), with typically
very high online penetration; for these we introduce an upper limit on the pre-pandemic time trend, constraining the
value below 100% (as eCommerce shares cannot exceed 100%). The industries whose online penetration has
achieved the upper limit are included in Table 5. These include, for example, “mail order”, “airline”, “travel
agencies”, and “utilities”. As our methodology for estimating online share applies a scaling factor of (total card
spending / total consumption), the scaled online shares for these industries will converge to an upper limit of card
spend/total consumption when the 100% threshold is met.

11
pre-COVID-19 e-commerce shares. On the other hand, the share of online spending in services,
specifically hotel-motel, recreation, mail order, and other transport have all fallen to significantly
below their pre-pandemic trend.

Similar to economy-wide results, there is divergence across sectors in pre-Covid trends. On
average, sectors that were leaders in e-commerce before COVID-19 reported mildly higher
increases in online shares (Figure 6). Most of the increase during the peak occurred in sectors
with a middle-range in online. Sectors such as health care, electrical appliances, and sporting
goods-toys reported the largest increases compared to the pre-crisis trend.

We also find these effects to be (largely) transitory, similar to the evidence across economies
(Figure 7), but with the exception of certain sectors such as those in retail, including drug stores,
clothing stores, and electrical appliances. The temporary nature looks to be particularly
pronounced in industries with higher pre-pandemic shares of e-commerce, as well as in
industries such as travel and entertainment.

Overall, the sectoral evidence supports the main findings in this paper: i) there is a divergence
across sectors in E-commerce shares, with ii) early E-commerce adopters reporting mildly higher
peak increases during COVID-19; and (iii) despite the sharp acceleration, the effects appear to be
transitory with some exceptions in sectors such as in retail and healthcare.

Are Mastercard share representative?

We now compare our Mastercard shares with broader estimates for a subset of economies with
alternative results obtained from survey data that incorporates other cards and payment methods.
We do this in Figure 8 for the United States and the United Kingdom, where official survey
estimates of “online retail spending” are published on a monthly basis by the US Census Bureau
and the UK Office of National Statistics.

Although there are differences in levels, the overall dynamic patterns are similar. In particular,
the official data also suggests that the online spending share in the United States has returned to

12
the pre-COVID trend levels, while the United Kingdom remains slightly above the trend-
predicted levels.

7. Conclusions

During the COVID-19 pandemic, online purchases surged in almost all economies. This paper
documents the extent of this surge in a systematic and comparable way, using a unique, large
dataset covering 47 economies and 26 sectors, for close to three years before and after the
pandemic. There is heterogeneity across countries and sectors in pre-COVID trends of share of
online spending. Online spending shares surged during the pandemic across all economies, but
the increase has reversed as the pandemic receded, with the exception of few sectors in retail,
and healthcare.

The lack of persistence is surprising and contrary to expectations. A common story often seen in
the press is that the pandemic accelerated the trend towards digitalization, forcing people to learn
new digital skills; and this learning was going to stay. Our results support the quick uptake of E-
commerce, but there is little support for any lasting learning by locking effects.

What can explain these patterns? During the first phase (surge to the spike), demand for E-
commerce relative to in-person commerce surged. More importantly, the interaction between
mobility and government transfers played a crucial role in explaining the temporary nature of
these dynamics; as these effects waned, online shares came back to pre-pandemic trends.

There is still some heterogeneity: for 30 percent the economies, online spending shares remain
higher than pre-pandemic trends. Furthermore, the scale up in e-commerce appears more longer-
lasting in few sectors, including categories of retail, and healthcare. These results could indicate
some extensive margin effects (new and younger customers continue to shop online). With travel
and gathering restrictions and government transfers now lifted around the world, the long-term
effects are important questions for future research.

Overall, the results show consumer spending patterns display a great deal of inertia. COVID-19
brought about a surge in E-commerce, but as the crisis subsided, spending patterns went back to
trend. In that sense, there is little evidence of a “long COVID” effect on e-commerce.

13
14
References

Abraham, Katharine G., Ron S. Jarmin, Brian Moyer, and Matthew D. Shapiro. “Big Data for
Twenty-First Century Economic Statistics: The Future Is Now.” National Bureau of
Economic Research, 2020.
Aladangady, Aditya, Shifrah Aron-Dine, Wendy Dunn, Laura Feiveson, Paul Lengermann, and
Claudia Sahm. “From Transactions Data to Economic Statistics: Constructing Real-Time,
High-Frequency, Geographic Measures of Consumer Spending.” National Bureau of
Economic Research, 2019.
Alcedo, Joel, Alberto Cavallo, Bricklin Dwyer, Prachi Mishra, and Antonio Spilimbergo. "E-
commerce During COVID: Stylized Facts from 47 Economies." NBER Working Paper
Series, No. 29729, February 2022.

Alipoura, Jean-Victor, Oliver Falckb,∗, Simon Krausec, Carla Krolaged, Sebastian Wichert
(2022), “The Future of Work and Consumption in Cities After the Pandemic: Evidence
from Germany”, CESifo Working Paper No. 10000
Auer, Raphael, Giulio Cornelli and Jon Frost (2023), “The pandemic, cash and retail payment
behaviour: insights from the future of payments database”, BIS Working Paper.
Bloom, Nicholas and Fletcher, Robert and Yeh, Ethan, The Impact of Covid-19 on Us Firms
(January 2021). NBER Working Paper No. w28314, Available at SSRN:
https://ssrn.com/abstract=3772594.
Brynjolfsson, Erik and JooHee Oh, “The Attention Economy: Measuring the Value of Free
Digital Services on the Internet,” 2012.
Carvalho, Vasco M., Stephen Hansen, Álvaro Ortiz, Juan Ramon Garcia, Tomasa Rodrigo, Sevi
Rodriguez Mora, and Pep Ruiz de Aguirre. “Tracking the Covid-19 Crisis with High-
Resolution Transaction Data” (2020).
Cavallo, Alberto. “Online and Official Price Indexes: Measuring Argentina’s Inflation.” Journal
of Monetary Economics 60, no. 2 (2013): 152–65.
Cavallo, Alberto. “More Amazon Effects: Online Competition and Pricing Behaviors.” Jackson
Hole Economic Symposium Conference Proceedings - Federal Reserve Bank of Kansas
City, 2018.
Chetty, Raj and Friedman, John and Hendren, Nathaniel and Stepner, Michael and Team, The
Opportunity Insights, The Economic Impacts of Covid-19: Evidence from a New Public
Database Built Using Private Sector Data (June 2020). NBER Working Paper No. w27431,
Available at SSRN: https://ssrn.com/abstract=3637732. Quarterly Journal of Economics,
forthcoming.
Choi, Hyunyoung, and Hal Varian. “Predicting the Present with Google Trends.” Economic
Record 88 (2012): 2–9.
Couture, Victor, Benjamin Faber, Yizhen Gu, and Lizhi Liu, “Connecting the Countryside via E-
Commerce: Evidence from China,” American Economic Review: Insights, 2020.

15
Dolfen, Paul, Liran Einav, Peter J. Klenow, Benjamin Klopack, Jonathan D. Levin, and Larry
Levin, 2020, “Assessing the Gains from E-Commerce”, mimeo, Stanford University, May
2020.
ECB (2021), “The impact of containment measures across sectors and countries during the
COVID-19 pandemic”, ECB Economic Bulletin, Issue 2/2021.
Einav, Liran, and Jonathan Levin. “The Data Revolution and Economic Analysis.” Innovation
Policy and the Economy 14, no. 1 (2014): 1–24.
Glaeser, Edward L., Hyunjin Kim, and Michael Luca. “Nowcasting the Local Economy: Using
Yelp Data to Measure Economic Activity.” National Bureau of Economic Research, 2017.
Goolsbee, Austan and Peter J. Klenow, “Valuing Consumer Products by the Time Spent Using
Them: An Application to the Internet,” American Economic Review, 2006, 96 (2), 108–
113.
Jo, Yoon J., Misaki Matsumura, and David E. Weinstein, “The Impact of ECommerce on
Relative Prices and Consumer Welfare,” 2019. National Bureau of Economic Research
Working Paper 26506.
Keane, Michael, Timothy Neal, 2021, “Consumer panic in the COVID-19 pandemic” Journal of
Econometrics, Volume 220, Issue 1, 2021, pp. 86-105.
Kinda, 2019, “E-commerce as a Potential New Engine for Growth in Asia”, IMF Working Paper
Working Paper No. 19/135, July 1, 2019.
Mian, Atif, Kamalesh Rao, and Amir Sufi. “Household Balance Sheets, Consumption, and the
Economic Slump.” The Quarterly Journal of Economics 128, no. 4 (2013): 1687–1726.
Moon, JiHyo, Yunseon Choe, and HakJun Song. 2021. "Determinants of Consumers’
Online/Offline Shopping Behaviours during the COVID-19 Pandemic" International
Journal of Environmental Research and Public Health 18, no. 4: 1593
Reinsdorf, Marshall and Paul Schreyer (2020), “Measuring consumer inflation in a digital
economy”, in Barbara M. Fraumeni eds Measuring Economic Growth and Productivity,
Academic Press, 2020, Pages 339-362.
Roggeveen, Anne L., Raj Sethuraman, 2020. “How the COVID-19 Pandemic May Change the
World of Retailing. ”Journal of Retailing Volume 96, Issue 2, June 2020, Pages 169-71.
Sheth, Jagdish, 2020 “Impact of Covid-19 on consumer behavior: Will the old habits return or
die?” Journal of Business Research Volume 117, September 2020, Pages 280-283.
Syverson, Chad, “Challenges to Mismeasurement Explanations for the US Productivity
Slowdown,” Journal of Economic Perspectives, 2017, 31 (2), 165–86.
Turrell, Arthur, Bradley J. Speigner, Jyldyz Djumalieva, David Copple, and James Thurgood.
“Transforming Naturally Occurring Text Data into Economic Statistics: The Case of
Online Job Vacancy Postings.” National Bureau of Economic Research, 2019.
Varian, Hal, “The Value of the Internet Now and in the Future,” The Economist, 2013.

16
Wang, Xueqin, Wenming Shi, Kum Fai Yuen (2022) “A synthesised review of pandemic-driven
consumer behaviours and applied theories: Towards a unified framework” International
Journal of Disaster Risk Reduction Volume 77, July 2022.

17
Figure 1. Retail spending during the pandemic

Notes. Figure 1 reports retail spending, seasonally adjusted using X13-ARIMA SEATS, adjusted with data starting
in 2014.

18
Figure 2. Cash during the pandemic

Figure 2a Cash in circulation in the US and Brazil

Figure 2b ATM Withdrawals in the US and Brazil

Notes. Figure 2a reports currency in circulation from the IMF’s World Economic Outlook database; the Figure 2b
reports ATM withdrawals from Mastercard.

19
Figure 3. Examples of Retail Online Spending Share During COVID

Notes: The “online spending share” of total card consumption in the retail sector in a given year is defined by
𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑐𝑐,𝑡𝑡
Equation (1) in the text, as 𝑠𝑠𝑐𝑐,𝑡𝑡 = . The online share indexed to January 2020=1.
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑐𝑐,𝑡𝑡

20
Figure 4. Residential mobility (time spent at home) vs. e-commerce deviation from trend
Figure 4a

Figure 4b

21
Notes: E-commerce gaps are seasonally adjusted to account for spikes around the holidays. Both
the global residential mobility and E-commerce gaps are computed by taking an equal weight
average for the 47 economies in our sample. Google COVID-19 Mobility Reports Discontinued
in October 2022.

22
Figure 5. Firm and Consumer Capability Over Time

Notes. “Firm capability” is defined at the country-time level, as the proportion of active firms that ever sold
online since the start of the sample in 2018. “Consumer capability” is defined as the proportion of active
consumers that ever bought online. Figure 5 plots the median consumer and firm capability measures over
time.

23
Figure 6. Peak versus pre-COVID trend in online spending shares across sectors

Figure 7. Latest versus pre-COVID trend in online spending shares across sectors

Notes: “Latest” refers to online spending share in April 2023. The pre-COVID trend is estimated
in each sector using a regression of monthly online shares on a time trend between 2018-2019.

24
Figure 8: Online shares in retail sales in the United States and the United Kingdom
Comparison between modeled estimates and national statistics

25
Table 1. Online Spending Share During COVID-19 in all Economies in the Sample

Crisis peak Latest minus pre-
Economy Latest
(relative to 2019) COVID trend
Bahrain 1.77 (2022/02) 1.60 0.60
Croatia 2.47 (2022/01) 1.79 0.50
Slovenia 2.2 (2022/01) 1.42 0.49
Slovakia 1.89 (2022/02) 1.35 0.31
Ecuador 1.78 (2022/01) 1.18 0.30
Norway 1.33 (2022/02) 0.94 0.29
Serbia 1.68 (2022/01) 1.40 0.24
Italy 1.58 (2022/01) 1.16 0.24
Argentina 1.55 (2022/01) 1.16 0.17
Austria 2.01 (2022/03) 1.13 0.13
Jamaica 1.18 (2022/01) 1.10 0.10
United Kingdom 1.34 (2022/01) 0.97 0.06
United States 1.25 (2022/01) 1.11 0.06
Cambodia 1.54 (2021/07) 1.31 0.05
Sweden 1.31 (2022/01) 0.88 0.04
Hungary 1.82 (2022/01) 1.23 0.03
Australia 1.16 (2022/01) 1.04 0.01
Singapore 1.45 (2022/02) 1.19 0.00
Canada 1.38 (2022/01) 1.01 -0.01
Poland 1.71 (2022/02) 1.31 -0.01
Germany 1.52 (2022/03) 0.93 -0.02
Thailand 1.65 (2022/03) 1.26 -0.03
New Zealand 1.39 (2022/03) 0.96 -0.04
Philippines 1.66 (2022/01) 1.10 -0.04
Dominican Republic 1.36 (2022/01) 1.14 -0.05
Czech Republic 2.01 (2022/02) 1.27 -0.06
United Arab Emirates 1.22 (2021/02) 1.21 -0.07
Netherlands 1.39 (2022/02) 0.93 -0.07
Nicaragua 1.18 (2021/09) 0.90 -0.07
Bulgaria 1.31 (2022/01) 1.06 -0.09
Indonesia 1.55 (2022/02) 1.45 -0.09
Denmark 1.32 (2022/02) 0.90 -0.09
Brazil 1.37 (2022/02) 1.19 -0.11
Greece 1.58 (2022/01) 1.37 -0.11
Luxembourg 1.51 (2022/01) 0.93 -0.14
Lithuania 1.5 (2022/02) 1.01 -0.15
Barbados 1.75 (2022/01) 0.81 -0.16
Romania 1.59 (2022/02) 1.10 -0.17
Montenegro 1.9 (2022/01) 1.43 -0.18
Egypt 1.61 (2022/02) 1.47 -0.20
Malaysia 1.68 (2022/03) 1.19 -0.23
Zimbabwe 1.43 (2021/12) 1.10 -0.24
Costa Rica 1.33 (2022/02) 1.09 -0.28
Somalia 1.67 (2021/03) 0.82 -0.46

Mean 1.57 1.16 0.01
Median 1.55 1.14 -0.03
Standard Deviation 0.28 0.21 0.21

26
Notes. For each economy in the sample, the “online spending share” of total card consumption in any economy for a
𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑐𝑐,𝑡𝑡
given year is defined by Equation (1) in the text, as 𝑠𝑠𝑐𝑐,𝑡𝑡 = . “Latest” is the online share
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑐𝑐,𝑡𝑡
indexed to January 2020=1. Country-wise daily new covid cases data from ourworldindata.org. Monthly average of
daily new cases are taken; month and year with the highest average is defined as peak. The peak is measured relative
to the average online share in the economy for 2019.

27
Table 2. Correlates of deviation in online shares from the predicted pre-COVID trends
Dependent variable: Month-on-month change in online share gap (in pp)
[1] [2] [3] [4]

Residential Mobility,
monthly avg % 0.5134*** 0.5131*** 0.4395*** 0.3998***
[0.04] [0.04] [0.05] [0.05]
New Covid Cases per
million, logs 0.0836** 0.0853** 0.0704* 0.0718*
[0.04] [0.04] [0.04] [0.04]
Pre-covid trend online
share -0.0014 -0.0019 0.0006
[0.00] [0.00] [0.00]
COVID Fiscal spending as
% 2019 GDP -0.001 -0.0012 0.0027
[0.01] [0.01] [0.01]
Internet penetration, 2019
(%) -0.0012 -0.0008 -0.0013
[0.01] [0.01] [0.01]
Residential mobility *
Fiscal spending, 2019 0.0063*** 0.0073***
[0.00] [0.00]
GDP per capita-2019, in
'000' -0.0002 -0.0001 -0.0023
[0.00] [0.00] [0.01]
Firm capability (%) 1.1818**
[0.56]
Consumer capability (%) 0.2081
[0.22]

r2 0.5578 0.5579 0.5639 0.5471
N 1336 1336 1336 1082
Time FE Yes Yes Yes Yes
# countries 43 43 43 34
# months 33 33 33 33
Notes: Variables in differences are online share, Residential Mobility, Firm and Consumer Capabilities. The
differences are computed by taking first difference month over month. Residential mobility on any given day of the
week is measured as the percentage change from the baseline value, which is defined as the median value, for the
corresponding day of the week, during the 5-week period Jan 3–Feb 6, 2020. GDP per capita is in $, constant prices,
PPP 2017 international dollars, taken from the IMF WEO. “Firm capability”, defined at the (country, time) level, as
the proportion of active firms that ever sold online since the start of the sample in 2018. “Consumer capability” is
defined as the proportion of active consumers that ever bought online. Figure 5 plots the median consumer and firm
capability measures over time. Robust standard errors in parentheses. ***, **, and * denote statistical significance at
1, 5, and 10 percent levels.

28
Table 3. Correlates of deviation in online shares from the predicted pre-COVID trends. Robustness

[1] [2] [3] [4] [5] [6] [7]

Grocery
Baseline Retail mob and phar
Ox str Ex internet Yoy Level
Residential Mobility, monthly
avg % 0.3998*** -0.1547*** -0.1424*** 0.0858*** 0.3999*** 0.3102*** 0.2719***
[0.05] [0.02] [0.02] [0.02] [0.05] [0.04] [0.04]
New Covid Cases per million,
logs 0.0718* 0.0319 0.1127*** 0.1033** 0.0699* 0.5091*** 0.4163***
[0.04] [0.04] [0.04] [0.04] [0.04] [0.09] [0.09]
Pre-covid trend online share 0.0006 0.0003 0.0046 0.0037 0.0007 -0.0590*** 0.0027
[0.00] [0.00] [0.00] [0.01] [0.00] [0.01] [0.01]
COVID Fiscal spending as %
2019 GDP 0.0027 -0.0026 0.0028 0.0031 0.003 -0.0056 0.0222
[0.01] [0.01] [0.01] [0.01] [0.01] [0.02] [0.02]
Internet Penetration, 2019
(%) -0.0013 -0.006 -0.0084 0.0006 0.0206 0.0255*
[0.01] [0.01] [0.01] [0.01] [0.02] [0.02]
Residential mobility * Fiscal
spending, 2019 0.0073*** -0.0017*** -0.0038*** 0.0025*** 0.0073*** 0.0155*** 0.0098***
[0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00]
GDP per capita-2019, in '000' -0.0023 0.0002 -0.006 -0.006 -0.0031 0.0292** -0.0064
[0.01] [0.01] [0.01] [0.01] [0.00] [0.01] [0.02]
Firm capability (%) 1.1818** 0.7023 1.6310*** 1.5475** 1.1815** 0.0008 -0.0707***
[0.56] [0.51] [0.62] [0.60] [0.56] [0.11] [0.02]
Consumer capability (%) 0.2081 0.0253 0.2193 0.0969 0.2094 0.2585*** 0.0657***
[0.22] [0.21] [0.23] [0.23] [0.21] [0.09] [0.01]

r2 0.5471 0.5803 0.4932 0.4243 0.5471 0.5233 0.3198
N 1082 1082 1082 1171 1082 708 1105
Country FE No No No No No No No
Time FE Yes Yes Yes Yes Yes Yes Yes
# countries 34 34 34 34 34 34 34
# months 33 33 33 33 33 33 33
Notes: Column [1] repeats the baseline specification (column [4]) in Table 2. Columns [2]-[3] includes alternative
measures of mobility (retail, and grocery and pharmacy, which are negatively correlated with pandemic restrictions)
and Column [4] uses the Oxford stringency measure. Column [5] drops the internet penetration variable. Columns
[6]-[7] employ alternative specifications – with changes measured over 12-month period, and estimation in levels
rather than changes respectively. In all columns excluding Columns [6—[7], variables in differences are online
share, Residential Mobility, Firm and Consumer Capabilities. The differences are computed by taking first
difference month over month. In Column [6], online shares, and consumer and firm capabilities are in differences,
computed as 12-month changes. In Column [7], all variables are in levels. Residential mobility on any given day of
the week is measured as the percentage change from the baseline value, which is defined as the median value, for
the corresponding day of the week, during the 5-week period Jan 3–Feb 6, 2020. GDP per capita is in $, constant
prices, PPP 2017 international dollars, taken from the IMF WEO. “Firm capability”, defined at the (country, time)
level, as the proportion of active firms that ever sold online since the start of the sample in 2018. “Consumer
capability” is defined as the proportion of active consumers that ever bought online. Figure 5 plots the median
consumer and firm capability measures over time. Robust standard errors in parentheses. ***, **, and * denote
statistical significance at 1, 5, and 10 percent levels.

29
Table 4. Divergence Across Selected Sectors

Latest vs
2019 Peak
Latest pre-covid
average crisis
sector trend
All Categories 35.9 46.4 43.9 0.6
Restaurants-Bars 7.3 32.6 16.3 -0.2
Retail 21.8 33.2 26.7 -1.4
Services 67.5 82.2 73.0 -0.5

30
Table 5. Heterogeneity across sectors in persistence of online shares

Latest vs
2019 Peak pre-covid
sector average crisis Latest trend Rank
Auto Rental 30.19 42.31 40.72 6.86 1
Clothing Stores 22.65 77.95 35.13 5.69 2
Drug Stores 10.31 19.43 16.75 5.43 3
Electric-Appliance 45.70 78.46 59.16 4.19 4
Travel Agencies 86.13 94.29 94.15 4.17 5
Education 73.45 88.41 79.29 3.49 6
Interior Furnishings 21.90 43.74 30.89 2.41 7
Utilities 84.77 90.82 88.23 2.21 8
Professional
Services 85.48 91.58 90.94 1.29 9
Airline 93.02 97.64 97.64 1.28 10
Food Stores-
Warehouse 4.92 9.69 8.32 1.28 11
Health Care 34.35 57.95 40.55 1.21 12
Gas Stations 0.48 1.44 1.42 0.50 13
Restaurants-Bars 7.30 32.63 16.34 -0.18 14
Vehicles 19.66 28.39 24.17 -0.57 15
Other Transport 66.74 79.13 70.66 -0.72 16
Discount Stores 15.83 33.23 22.15 -1.21 17
Department Stores 28.68 63.92 50.27 -1.44 18
Other Services 64.95 79.22 71.14 -2.42 19
Mail Order 97.95 99.16 96.88 -3.12 20
Sporting-Toy Stores 27.31 59.38 32.04 -3.16 21
Recreation 60.14 79.44 65.80 -3.38 22
Hardware 27.47 34.39 31.37 -3.58 23
Other Retail 46.15 66.50 52.05 -4.86 24
Repair Shops 32.22 43.10 34.58 -6.56 25
Hotel-Motel 38.65 58.14 52.66 -9.73 26
All Categories 35.89 46.35 43.89 0.64

31
Appendix
Tables
Appendix Table A1. Summary Statistics (based on Table 2 Column 4)

Variable Obs Mean Std. Dev. Min Max
Online share gap (online share-trend) (in pp), NSA 1082 0.0 3.4 -15.7 19.1
Residential Mobility, monthly avg % 1082 36.6 21.9 2.8 97.3
Pre-covid trend online share 1082 0.1 4.1 -15.5 34.1
COVID Fiscal spending as % 2019 GDP 1082 40.8 17.7 12.0 100.0
GDP per capita-2019, in '000' 1082 13.7 10.8 1.1 43.4
Residential mobility * Fiscal spending, 2019 1082 1.0 71.9 -567.3 844.8
Firm capability (%) 1082 7.1 2.4 -1.7 12.1
Consumer capability (%) 1082 78.5 18.0 26.6 99.7
Internet penetration (2019, %) 1082 0.0 3.4 -15.7 19.1
Notes: Variables in differences: online share gap (LHS), Residential Mobility, Firm and Consumer Capabilities
(RHS). First difference month over month.

32
Appendix Table A2 Robustness Constant Sample

[1] [2] [3] [4]

Residential Mobility,
monthly avg % 0.4974*** 0.4969*** 0.3956*** 0.3998***
[0.04] [0.04] [0.05] [0.05]
New Covid Cases per
million, logs 0.0876** 0.0926** 0.0706* 0.0718*
[0.04] [0.04] [0.04] [0.04]
Pre-covid trend online
share -0.0002 -0.0008 0.0006
[0.00] [0.00] [0.00]
COVID Fiscal
spending as % 2019
GDP -0.0024 -0.0026 0.0027
[0.01] [0.01] [0.01]
Internet penetration,
2019 (%) -0.0022 -0.0015 -0.0013
[0.01] [0.01] [0.01]
Residential mobility *
Fiscal spending, 2019 0.0079*** 0.0073***
[0.00] [0.00]
GDP per capita-2019,
in '000' 0.0017 0.0017 -0.0023
[0.01] [0.01] [0.01]
Firm capability (%) 1.1818**
[0.56]
Consumer capability
(%) 0.2081
[0.22]

r2 0.5333 0.5334 0.5446 0.5471
N 1082 1082 1082 1082
Country FE No No No No
Time FE Yes Yes Yes Yes
# countries 43 43 43 34
# months 33 33 33 33

Notes: Variables in differences: online share (LHS), Residential Mobility, Firm and Consumer
Capabilities (RHS). First difference month over month.

33
Appendix A: Detailed overview of database used in our study.
Payment Channels and the Scope of Study
Payments can occur through a variety of channels, including cash, card, check, and deposits.
This study focuses on aggregated & anonymized card transactions within the Mastercard
network. A related database from Mastercard was used by Mian et al (2013).
Although card transactions mainly serve the retail and services sector, specific segments like
vehicle sales are underrepresented as deposits often take precedence over card payments.
Card payments are the predominant method for online transactions, translating to over-
representativeness of e-commerce relative to household expenditures in aggregate. Although this
translates to an inherent payment form bias, e-commerce shares in our database align well in
markets whose official statistics agencies report e-commerce sales.
Our study adds value by providing a more granular look into the transience of e-commerce
spend, across different industries and many more markets than is typically not readily available.
Data Structure & Dimensionality

• Payment Channels: A significant component of our study is defining online spending.
For this purpose, we categorize E-Commerce as transactions where neither the cardholder
nor the card are physically present. In transactions where tap to pay is initiated via a
mobile wallet with a linked card, which represent a small portion of aggregate volumes,
the card itself is not physically present.
• Localization: The data is relative to the card's issuing bank country. For instance, a
transaction made using a Canadian card for an online purchase from a U.S. merchant will
be attributed to Canada, regardless of the merchant's location.
• Sectoral Definitions & Industry Representativeness: Industries within our data set are
identified using Merchant Category Codes (MCC) - a standard in the payments industry.
While research often correlates MCCs with other industry classifications like NAICS or
SIC, the core insights one could extrapolate from card payment transaction data remain
consistent. Specifically, our figures exclude cash transactions and sectors not typically
represented through card payments, such as rent or automotive sales. Hence, industry
categorization should be contextualized within these parameters.
The database offers details on:
• Date & time: the date and time in which the transaction occurred
• Type of card: whether the card is a credit, debit, or pre-paid card
• Merchant location: what is the address on record for the merchant where the transaction
took place?
• Industry classification: what industry is the merchant classified as
• Channel: whether the payment was an online or brick & mortar transaction
• Transaction amount: the value of the transaction made

34
Importantly, Mastercard does not have access to the following dimensions:
• Cardholder-specific details, like location, income, or account balances.
• Breakdown of individual items or SKUs in a purchase. Mastercard only observes the total
payment amount.
Mastercard’s Reach
The Mastercard network has a presence in 210 countries and territories, connected to 20,000
financial institutions and over 80 million merchant locations with more than 2.9 billion cards in
force. In 2022, Mastercard processed approximately 125.7 billion transactions, amounting to a
total value of nearly 8.2 trillion U.S. dollars from purchases and cash disbursements.
For this research, we thoroughly examine aggregated & anonymized transaction data. Our focus
is on capturing industry and sector patterns by market monthly. Our figures represent aggregated
and anonymized Mastercard transactions based on the card’s issuing country.
The use of aggregated and anonymized data underscores our commitment to data privacy and
ethical research practices. This database provides invaluable insights into transaction trends,
allowing for a nuanced understanding of industry shifts and market dynamics. By focusing on
Mastercard transactions, we study a significant portion of global transactional data, enabling us
to deliver robust insights.

35
Appendix B: Mastercard’s Market Share Cut-off
To increase the chances of having a representative sample in the paper we limited our data to
economies where Mastercard has a significant market share. Changing this cutoff value has little
impact on our results. To show this, we computed the online share for economies where
Mastercard has a small share of the card market and compared it to the results we can obtain
from the 47 economies included in our sample. Figure B1 shows that online shares are highly
correlated between these two groups, with a correlation coefficient of 0.97.

Figure B1: Online shares in economies where Mastercard has >=20% and <20% market share.

File and source

File: alcedo-2022-back-to-trend-covid-effects-on-e-commerce-in-47-countries.pdf
Size: 743,661 bytes
SHA-256: a9959e5d25e83c49ccc8bf0585e6afcd6f231951ef0c90c575fb0fba707e0a74
Our copy: alcedo-2022-back-to-trend-covid-effects-on-e-commerce-in-47-countries.pdf
Original: doi.org