Alan Kwan et al., Stress Testing Banks’ Digital Capabilities: Evidence from the COVID-19 Pandemic (2023)
Summary
A journal article by Alan Kwan, Chen Lin, Vesa Pursiainen and Mingzhu Tai, Stress Testing Banks' Digital Capabilities: Evidence from the COVID-19 Pandemic, published in the Journal of Financial and Quantitative Analysis, Vol. 59, No. 6, Sep. 2024, pp. 2618–2646. The authors treat the pandemic as a shock that shifted banking services from in-person to digital. They report that banks with better information technology saw larger reductions in branch visits and larger increases in website traffic. They find that stronger IT banks originated more Paycheck Protection Program loans to small business borrowers, especially in areas with more severe outbreaks, higher internet use and higher bank competition, and that those banks attracted more deposit flows and better mobile customer reviews.
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
“Alan Kwan et al., Stress Testing Banks’ Digital Capabilities: Evidence from the COVID-19 Pandemic (2023).”
Original at www.cambridge.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/alan-kwan-et-al-stress-testing-banks-digital-capabilities-evidence-fro-93bad2272a/
, SHA-256 0c27e844f0ecd0b192bf788a67c4a5d9af75c2365b5d7332fd59a3ba6a812b16.Full text
https://doi.org/10.1017/S0022109023000662
JOURNAL OF FINANCIAL AND QUANTITATIVE ANALYSIS Vol. 59, No. 6, Sep. 2024, pp. 2618–2646
© The Author(s), 2023. Published by Cambridge University Press on behalf of the Michael G. Foster
School of Business, University of Washington. This is an Open Access article, distributed under the
terms of the Creative Commons Attribution licence (https://creativecommons.org/licenses/by/4.0), which
permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
doi:10.1017/S0022109023000662
Stress Testing Banks’ Digital Capabilities:
Evidence from the COVID-19 Pandemic
Alan Kwan
University of Hong Kong Faculty of Business and Economics
apkwan@hku.hk
Chen Lin
University of Hong Kong Faculty of Business and Economics
chenlin1@hku.hk (corresponding author)
Vesa Pursiainen
University of St.Gallen and Swiss Finance Institute
vesa.pursiainen@unisg.ch
Mingzhu Tai
University of Hong Kong Faculty of Business and Economics
taimzh@hku.hk
Abstract
Banks’ information technology (IT) capabilities affect their ability to serve customers during
the COVID-19 pandemic, which generates an unexpected and unprecedented shock that
shifts banking services from in-person to digital. Amid mobility restrictions, banks with
better IT experience larger reductions in physical branch visits and larger increases in website
traffic, implying a larger shift to digital banking. Stronger IT banks are able to originate more
Paycheck Protection Program loans to small business borrowers, especially in areas with
more severe COVID-19 outbreaks, higher internet use, and higher bank competition. Those
banks also attract more deposit flows and receive better mobile customer reviews during the
pandemic.
We thank an anonymous referee, Sumit Agarwal, Christoph Basten, Diana Bonfim, Martin Brown,
Murillo Campello, Sebastian Doerr, Mara Faccio (the editor), Andreas Fuster, Tobin Hanspal, Christoph
Herpfer, Yiping Huang, Andrew Karolyi, Xiang Li, Mrinal Mishra, Deniz Okat, Steven Ongena, Mikael
Paaso, Yu Shan, David Solomon, and Yannick Timmer, as well as conference participants at the 2020
Shanghai-Edinburgh Fintech Conference, 2021 Columbia SIPA/BPI Bank Regulation Research Con-
ference, 2021 CEBRA Annual Meeting, 2021 ADBI-ADB-CCAF Conference on “Fintech and COVID-
19,” 2021 Biennial IADI Research Conference, 2021 IWH-FIN-FIRE Workshop on “Challenges to
Financial Stability,” 2021 Swiss Winter Conference on Financial Intermediation, 2021 GdRE Interna-
tional Symposium on Money, Banking and Finance, and 2021 UZH Workshop on Sustainable Banking,
as well as seminar participants at ACPR – Banque de France, Concordia University, WU Vienna
University of Economics and Business, and Kyiv School of Economics for helpful comments. Kwan
acknowledges support from the Research Grant Council of the Hong Kong SAR, China (Project
#C7105-20G). Lin acknowledges the financial support from the National Natural Science Foundation
of China (Project No. 72192841) and the Research Grant Council of the Hong Kong SAR, China (Project
No. T35-710/20-R).
2618
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2619
I. Introduction
A long-running trend has seen banks gradually decrease their reliance on
physical branches and move services online, reducing the importance of geographic
proximity to customers.1 As banks have expanded geographically, proximity to
bank headquarters has also become less important (Berger and DeYoung (2006)).
While information technology (IT) plays an important role in facilitating these
changes, the evidence on the relationship between bank performance and IT invest-
ment is mixed (e.g., Berger (2003), Beccalli (2007), and Koetter and Noth (2013)).
While the recent rise of fintech lenders has intensified the debate about the com-
petitive advantage that technology can provide in credit analysis (e.g., Berg, Burg,
Gombović, and Puri (2020)), wealth management (e.g., D’Acunto, Prabhala, and
Rossi (2019)), and the ability to address underserved clienteles (e.g., Tang (2019)),
recent evidence suggests that brick-and-mortar banking still matters a great deal.2
One challenge to studying the role of IT in the traditional banking sector is
that both bank IT investments and how they serve customers evolve over long
periods of time, possibly endogenously to one another. In this article, we study
how banks’ IT capabilities affect their ability to serve retail customers in response
to an exogenous shock that leads to a shift from physical branch banking to digital
banking services. The COVID-19 pandemic represents a unique natural experi-
ment for our study, as large-scale mobility restrictions imposed by state and local
governments in response to public health risks have led to substantial barriers
to physical banking. Anecdotal evidence suggests that the use of digital banking
has substantially increased during the pandemic.3 For example, Wells Fargo, one
of the largest U.S. banks, reported an 81% increase in the amount of money
deposited using a mobile device in Apr. 2020 relative to Apr. 2019 and a 23%
surge in customers signing on to digital banking since mid-Mar. 2020 (Wells
Fargo Stories: Digital Banking Soars in the COVID-19 Pandemic, available
at https://stories.wf.com/digital-banking-soars-in-the-covid-19-pandemic/). Sur-
vey data suggest that banks have generally seen a double-digit rise in first-time
online accounts, mobile deposits and payments, and overall usage of web and
mobile banking.4 At the same time, the pandemic has forced operational changes
such as moving employees into remote work and branches into temporary or even
permanent closure (Kreiss (2021)).
This unexpected and unprecedented shock to branch banking services pro-
vides us with an opportunity to study the role of banks’ digital capabilities in
affecting how they serve retail customers. There are numerous factors at play.
1
See, for example, Petersen and Rajan (2002) and Keil and Ongena (2020).
2
Branch closings can lead to persistent declines in small business lending (Nguyen (2019)) and less
competitive loan markets (Bonfim, Nogueira, and Ongena (2021)). Herpfer, Mjøs, and Schmidt (2022)
find evidence that reductions in travel time lower banks’ transaction costs and facilitate banking
relationships.
3
See, for example, The Economist (2020).
4
Based on polling research by J.D. Power (https://www.jdpower.com/business/press-releases/2020-
us-retail-banking-satisfaction-study) and Ondot Systems (summary by Payments Journal at: https://
www.paymentsjournal.com/fis-need-updated-digital-roadmaps-to-reflect-the-rapid-shift-to-digital-
banking-during-covid-19/).
https://doi.org/10.1017/S0022109023000662
2620 Journal of Financial and Quantitative Analysis
Customers facing the prospect of COVID-19 may be less willing or able to visit
physical branch services, either because they want to avoid personal risk or because
mobility restrictions prevent them from leaving their home or visiting bank
branches. Likewise, banks may face restrictions keeping the branches open, and
bank employees may prefer to work from home. Both of these demand-side and
supply side effects would result in a reduction of physical branch visits and a greater
shift toward digital banking since the onset of the pandemic.
We argue that banks with strong IT are better able to make this transition from
in-person to digital services. Better IT allows banks to support remote work, to
provide impacted customers with alternative services, or to scale up and improve
existing digital infrastructure. To examine the relationship between IT and bank
service provision during the pandemic, we examine a number of outcome variables
assembled from various data sets, including physical bank branch visits, website
traffic, SBA Paycheck Protection Program (PPP) lending to small businesses,
customers’ propensity to switch banks, deposit flows, as well as mobile customer
reviews. Across these various outcomes, we find that better IT is associated with a
more pronounced shift away from in-person banking toward online banking and
better ability to provide service to retail customers. We also investigate various
mechanisms for this set of findings.
Our study uses an IT index measuring the presence of technologies that we
deem likely to be useful for virtual, automated, or remote work. To construct this
IT index, we use data from Aberdeen Computer Intelligence database, which
tracks technology usage at the establishment level. Of the 63 key technology areas
tracked by Aberdeen, we identify 14 technologies that, ex ante, seem relevant
for facilitating a shift from physical to digital operations. This shift might entail
staff working from home, bank personnel serving customers via website, phone,
or video chat, or increasing automation of previously manual processes. Thus, our
IT_INDEX includes technologies such as VPN and remote access, document
management, collaborative software, customer relationship management soft-
ware, video conference and internet phone software, database software, and
server virtualization. We define IT_INDEX as the sum of 14 dummies, indicating
the presence of each of the relevant technologies, and aggregate it to the bank
level.5 While this is our preferred measure, our findings are robust to other
measures of IT as well.6
Armed with this measure, we begin by studying bank IT and the shift of bank
customers from physical branches to online. First, we study physical visits to bank
branches. Our measure of visits to physical bank branches comes from SafeGraph,
whose anonymized and aggregated data cover approximately 10% of all mobile
devices in the United States. We combine these with data on local mobility restric-
tions imposed amid the COVID-19 pandemic from Keystone Strategy and Spiegel
5
The reasons for using a bank-level index are to reduce noise in the data and to reflect the fact that
much of the digital infrastructure is not likely to be office specific.
6
In robustness checks reported in Section IA.V of the Supplementary Material, we complement this
index with three additional measures: IT_INDEX_OTHER, reflecting all the other technologies that are
not included in the main index, IT_STAFF, measuring the number of IT staff relative to total employees,
and IT_BUDGET, measuring the IT budget scaled by total employees. These additional analyses show
that our findings are not sensitive to the definition of the IT measure.
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2621
and Tookes (2021). We construct an indicator of severe mobility restrictions and
find that banks with stronger IT capabilities experience significantly larger reduc-
tions in customer visits to physical branches after—but not before—the imposition
of severe mobility restrictions, suggesting that better IT enables banks to better
serve their customers without the need for face-to-face interaction.7 A
1-standard-deviation increase in IT_INDEX is associated with a doubling of the
effect of mobility restrictions on branch visits.
We then perform a complementary analysis of traffic to banks’ websites. We
use two data sources to measure bank web traffic. Our main data come from
AlexaRank, which dynamically ranks global websites by their estimated traffic.
This allows us to measure the relative rankings of bank websites’ visitor volumes
over time. While we do not see the actual volume that the ranking is based on, these
rank data allow cross-bank comparisons of changes in website volumes on a weekly
basis. We complement this data with a direct monthly measure of web traffic
provided by SimilarWeb. Our analyses based on both data sets provide robust
evidence that better IT is associated with a significantly larger increase in bank
website traffic in response to mobility restrictions. The economic magnitude is
large: a 1-standard-deviation increase in bank IT increases the effect of mobility
restrictions on bank web traffic by 10% of the within-bank standard deviation.
Overall, our findings on bank branch visits and web traffic suggest that banks with
better IT more successfully shift from in-person to digital banking services during
the COVID-19 pandemic.
A related finding is that a higher IT_INDEX is associated with observable
differences in banks’ websites. Using data from BuiltWith, we calculate the number
of “website technologies” present on banks’ websites.8 While not all technologies
are equally important, having more technologies likely implies greater technolog-
ical investment and sophistication. Our IT_INDEX is significantly positively
correlated with website technologies. In addition, in response to the pandemic,
BuiltWith data began identifying the first time that the bank website mentions the
COVID-19 pandemic. We find that a 1-standard-deviation increase in IT_INDEX is
associated with approximately 1–2 days reduction in reaction time, depending on
model specification. This suggests that banks with better IT have more sophisti-
cated web services and adapt faster to the pandemic.
That better IT facilitates banking customers toward digital services is important
in and of itself for public health reasons, but it can also have real economic conse-
quences. Next, we focus on banks’ ability to serve small and medium enterprise
(SME) borrowers by investigating loans originated under the Small Business Admin-
istration (SBA) PPP.9 We consider PPP lending a good setting to test the role of IT in
banks’ retail business, since PPP loans are an unexpected new loan product that banks
7
As we discuss in Section IA.I of the Supplementary Material, we construct our mobility restriction
indicator based on a set of relevant restriction categories that we identify as having the most important
effects on branch visits (based on a variety of statistical models).
8
A website technology can comprise many types, including graphical technologies such as jQuery,
web traffic services such as content-delivery networks or the type of web server, security certificates, and
much more.
9
The PPP program is the U.S. government’s SME support program, structured as government-
guaranteed loans distributed through the banking system.
https://doi.org/10.1017/S0022109023000662
2622 Journal of Financial and Quantitative Analysis
had to set up under substantial time pressure, and during a period when both staff and
customers were potentially unable to physically be in the office. These significant
operational frictions may have impacted banking relationships: Prior evidence shows
that bank relationships remained important for obtaining PPP loans (e.g., Li and
Strahan (2021)), as the pool of PPP funds was limited and as banks prioritized larger
borrowers, possibly at the expense of SMEs (e.g., Balyuk, Prabhala, and Puri (2020)).
Correspondingly, media articles suggest that many SME customers experienced poor
service from their traditional banks and, at least in some cases, switched banks as
a result. For example, The Wall Street Journal writes, “Of businesses that secured
PPP funding, about 28% received their loan from a lender with whom they had no
prior relationship or a bank that was not their primary one (“When Their PPP Loans
Didn’t Come Through, These Businesses Broke Up with Their Banks,” The Wall
Street Journal, July 31, 2020. Available at https://www.wsj.com/articles/when-their-
ppp-loans-didnt-come-through-these-businesses-broke-up-with-their-banks-
11596205736).”
We perform two analyses to test the hypothesis that higher bank IT leads to
better performance in distributing PPP loans. On the extensive margin, we perform
a bank-county-level analysis to examine how the amount of PPP loan origination
varies by the bank’s IT capability. On the intensive margin, we study how IT affects
which bank is chosen by a small business borrower. Turning to our first analysis, we
find that, within the same county, banks with better IT originate significantly larger
amounts of PPP loans during the pandemic, controlling for bank characteristics
as well as the bank’s prior activities in the county. This effect of bank IT is both
statistically significant and economically sizeable. A 1-standard-deviation increase
in IT_INDEX is associated with a 27% increase in PPP loan volumes generated. It is
worth noting that these results are found when including county fixed effects,
controlling for any unobserved factors that might affect local small business credit
demand and quality.
We also investigate the differential relation between IT and PPP performance
across counties with different characteristics. First, if our results are driven by the
ability of banks with good IT to better serve customers during the COVID-19
outbreak, we would expect the estimated effect of IT on PPP loan volumes to be
larger in areas that experienced worse outbreaks of the virus. This is due to the likely
higher need for digital service and the possibly better ability of high-IT banks to
supply digital services despite the difficulties brought about by health risks and
restrictions on banks’ physical operations. The evidence confirms this prediction.
Second, we find that the effect of bank IT is stronger in areas with greater high-
speed internet penetration. This is intuitive, as customers in these areas are more
likely to shift to digital banking than those in areas with worse internet infrastruc-
ture. Third, we study the effect of local banking competition. We find that the effect
of IT is larger in areas with greater bank competition, as measured by a lower
Herfindahl–Hirschman Index (HHI) of SME loans. This suggests that a stronger IT
capability can be particularly valuable to banks operating in a competitive market.
We now turn to our second analysis on PPP lending. If IT is a determinant of a
bank’s ability to help small businesses obtain PPP loans, we might expect firms with
existing relationships with low-IT banks to be more likely to switch to higher-IT
banks for PPP loans when they are exposed to COVID-19. To test these predictions,
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2623
we form a sample of PPP borrowers who were also regular SBA-loan borrowers
prior to the pandemic.10 We find that firms are more likely to switch to better-IT
banks in areas harder hit by COVID-19, and in cases where their existing lenders
have worse IT. This suggests that customers prefer higher-IT banks over their
previous banking relationship when COVID-19 severity—a proxy for the extent
of the shock on digital banking—is greater.
A natural extension of the idea that high-IT banks gain customers during the
pandemic is that such banks may also attract more deposit flows. Hence, we study
the relationship between bank IT and deposit growth during the onset of the
COVID-19 pandemic. We find that banks with stronger IT capabilities experience
significantly higher increases in deposits during the first 2 quarters of 2020. This
suggests that banks that are better able to serve customers during the pandemic also
attract more deposits during the COVID-19 shock, consistent with the idea that
customers switched their banking relationships.
Finally, to study the underlying mechanisms more directly, we collect data
on customer reviews for mobile banking apps from both the Android and Apple
mobile app stores. We find that the average bank receives more reviews during the
pandemic period, and that these reviews have lower scores. However, relative to
banks with a lower IT_INDEX, higher-IT banks receive even more reviews (imply-
ing more app downloads and usage) and relatively more favorable ratings. We
also use a novel textual analysis algorithm from Facebook (Lewis, Liu, Goyal,
Ghazvininejad, Mohamed, Levy, Stoyanov, and Zettlemoyer (2019) to construct
additional measures from the reviews. In particular, we classify positive textual
comments (e.g., easy to use, reliable, effective, or aesthetically pleasing) versus
negative ones (e.g., complaints about an app not working, being slow, lacking
services, or being unintuitive), and compute the fraction of negative and positive
comments each bank receives. Consistent with what we find from numeric scores,
the fraction of positive (negative) comments is relatively higher (lower) for higher-
IT banks during the pandemic.
Overall, our results suggest that banks with stronger IT capabilities can better
serve their customers during the onset of COVID-19 through digital channels.
We also examine two potential interpretations of these findings. While our results
might suggest that banks with better IT can better serve their customers during the
pandemic, it is also possible that banks with better IT could have a more IT-savvy
clientele even before the pandemic, who might be more inclined to shift toward
digital services during the pandemic. Both interpretations are equally interesting but
have different policy implications. While we cannot rule out either interpretation
conclusively, three pieces of evidence are inconsistent with clientele sorting driving
our main results. First, we examine the ex ante relationship between bank and SME
customer IT before the pandemic and find no significant relationship between the
two. In addition, those customer firms’ own remote work decisions respond to the
pandemic similarly, regardless of their banks’ IT. Finally, as noted before, we find
customer firms switch from lower-IT banks to higher-IT banks. Thus, our results are
10
We caution that the PPP data only provide names for borrowers above a threshold size, while the
SBA sample includes many very small businesses. Still, we obtain a sample of 97,758 borrowers present
in both data sets.
https://doi.org/10.1017/S0022109023000662
2624 Journal of Financial and Quantitative Analysis
less likely to be explained by sorting between tech-savvy clienteles and high-IT
banks that existed prior to the pandemic. Rather, it is more likely that IT served as a
source of banks’ competitive advantage in serving and acquiring customers during
the pandemic.
We make several contributions to the literature. First, we provide evidence of
the impact of technology on bank outcomes. Our setting with an unexpected shock
to digital banking services allows us to partly overcome the identification challenge
(i.e., the possibility that both consumer preferences and bank performance may
affect investment in IT), instead of the other way around. A few other papers also
study the role of bank IT during the pandemic. Closely related to our article, Core
and De Marco (2023) study the impacts of bank IT capability on small business
credit provision in Italy during COVID-19 and how bank IT affects the importance
of local branches. The institutional features of the Italian guarantee program and the
granular data enable them to examine not only the probability of lending, but also
the pricing and processing time as well as the relationship between local branches
and lending. In their study, Core and De Marco (2023) measure bank IT using the
rating of bank mobile apps. Our article augments these discussions by utilizing
the rich data sources available in the U.S., which permits us to investigate a set of
other bank outcomes. Our analyses enable us to show how banks’ IT affects their
ability to substitute from in-person to digital banking, and how they serve customers
in the retail lending and deposits markets, as well as their service quality. We also
introduce a new measure of bank IT that pinpoints the specific technologies used by
each bank. Overall, relying on different measures of IT and different institutional
settings, these papers complement and reinforce each other in showing the role of IT
in banking during a pandemic.
Second, our study is related to the literature on the interplay between tradi-
tional banking and fintech services. Rather than estimating the levels of substitution
between banks and fintech lenders, we examine the technological readiness of
banks. Given that banks are heavily regulated and play roles that alternative finance
lenders may not be allowed to replace, this may represent a fruitful direction for
future research as well.
Third, we add to the literature on the economic effects of the COVID-19
pandemic, as greater fungibility between banks’ online and brick-and-mortar ser-
vices can reduce the economic impact of mobility restrictions and increase flexi-
bility in unexpected situations. Despite these benefits, the variation across banks in
IT and its impact imply heterogeneous economic costs of the pandemic. Finally,
beyond banking, it is also likely that a relationship between IT and performance
exists in other industries as well.
II. Relevant Literature
A. Banks, Technology, and Branches
A vast literature studies the implications of technological change on banking
and the role of branch banking. For example, Petersen and Rajan (2002) find
evidence of technology facilitating small business lending to increasingly distant
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2625
customers.11 However, extant studies on the role of IT in bank productivity are
somewhat mixed. Berger (2003) argues that technology in banking has led to
improvements in cost efficiency and lending capacity, while Beccalli (2007) finds
only weak evidence of a link between banks’ IT investment and bank performance.
Koetter and Noth (2013) find incorporating bank IT helps explain bank productiv-
ity. He, Jiang, Xu, and Yin (2021) documents the relationship between bank IT
investments and their usage of soft versus hard information. Pierri and Timmer
(2022) find that higher pre-crisis PC adoption by banks is associated with lower
levels of nonperforming loans during the financial crisis of 2008–2009. Mixed
evidence of IT on bank outcomes may reflect the endogeneity between bank
performance and IT investment. Thus, our study contributes evidence based on
an exogenous shock to the relative attractiveness of digital services in the U.S. As
mentioned previously, other work has explored the Italian loan program using
different IT measures and richer loan-level data (Core and De Marco (2023))
outside the U.S. Also, Branzoli, Rainone, and Supino (2023) study the Italian loan
program using a different measure of IT.
B. Fintech and Sources of Technological Advantage
The rapid rise of technology-driven financial institutions has brought attention
to the advantage that technology can provide. For example, Berg et al. (2020) show
that technology and the use of easily accessible digital data can substantially
improve credit analysis, while Iyer, Khwaja, Luttmer, and Shue (2016) find that
peer screening facilitated by peer-to-peer (P2P) platforms does better than credit
scores in predicting defaults.12 In addition, whether fintech firms reach underserved
clienteles and serve as complements or substitutes for traditional banking remains a
debate. For example, Tang (2019) finds evidence that p2p lending serves infra-
marginal bank borrowers but also complements bank lending with respect to small
loans. Danisewicz and Elard (2022) find that access to fintech credit reduces
personal bankruptcies. Another contested issue is whether removing face-to-face
interaction and moving to algorithm-based decision-making increases or decreases
bias and discrimination in credit decisions. For example, Bartlett, Morse, Stanton,
and Wallace (2022) and Fuster, Goldsmith-Pinkham, Ramadorai, and Walther
11
Keil and Ongena (2020) find evidence of technological change contributing to the decrease in
the number of branches. Basten and Ongena (2020) find evidence of banks using online platforms to
increase geographic diversification. D’Andrea and Limodio (2023) study the introduction of high-speed
internet on private-sector lending by banks. Degryse and Ongena (2005) find evidence of spatial price
discrimination in bank lending, with loan rates decreasing with the distance between the firm and the
lending bank. Berger and DeYoung (2006) find that technological progress has facilitated banks’
geographic expansion further away from parent banks. Historically, Lin, Ma, Sun, and Xu (2021) find
the telegraph enabled Chinese banks to expand their branch networks from 1881 to 1936.
12
In addition, D’Acunto et al. (2019) find that robo-advisers can help some clients make better
investment decisions. Fuster, Plosser, Schnabl, and Vickery (2019) show that fintech lenders process
mortgage applications faster and adjust supply more elastically than other lenders without incurring
higher default costs. In addition, Buchak, Matvos, Piskorski, and Seru (2018) show a large factor in the
rise of fintech lenders is regulatory arbitrage. They estimate that regulation accounts for 60% of shadow
bank growth in mortgage lending, while technology accounts for the rest.
https://doi.org/10.1017/S0022109023000662
2626 Journal of Financial and Quantitative Analysis
(2022) present contrasting studies of whether minority borrowers are likely to
benefit from fintech algorithms.
Our study is unique from the fintech literature in that we focus on the tech-
nological capabilities of banks rather than the fintech firms that aim to displace
them. As our study is brought about by the pandemic, we deliver insights on the
ability of banks to substitute online and offline activity and to operate digitally. Our
results suggest that substantial variation in IT readiness remains among banks.
C. COVID-19, the PPP, and the Banking Sector
The COVID-19 pandemic has taken hundreds of thousands of lives, strained
healthcare systems, and forced shutdowns of large parts of the global economy. In
the United States, the pandemic sparked large-scale mobility restrictions across
many states in early Mar. 2020. The literature has studied the effects of these
restrictions on social distancing (e.g., Adalja, Toner, and Inglesby (2020) and Ru,
Yang, and Zou (2021)), as well as the economic impacts on the labor market (e.g.,
Bartik, Bertrand, Cullen, Glaeser, Luca, and Stanton (2020a) and Humphries,
Neilson, and Ulyessea (2020), household consumption (e.g., Andersen, Hansen,
Johannesen, and Sheridan (2022), Baker, Farrokhnia, Meyer, Pagel, and Yannelis
(2020)), and various other parts of the economy.
Most relevant to us, some of these studies focus on the PPP. On Mar. 27, 2020,
the U.S. Congress passed the Coronavirus Aid, Relief, and Economic Security
(CARES) Act, which included $350 billion to fund the PPP. The PPP supported
small businesses by extending forgivable loans. Similar schemes of government-
guaranteed loans have been adopted in a number of other countries (see, e.g.,
Alstadsæter, Bjørkheim, Kopczuk, and Økland (2020), Bennedsen, Larsen,
Schmutte, and Scur (2020), Paaso, Pursiainen, and Torstila (2020), Branzoli et al.
(2023), and Core and De Marco (2023)). Some of these studies focus on alternative
lenders.13 Finally, Fu and Mishra (2022) find evidence of a worldwide increase in
bank mobile application downloads during the pandemic.
III. Data and Methodology
A. Measuring IT Capability
To measure banks’ IT capabilities, we use data from Aberdeen’s Computer
Intelligence Technology database. In 2019, the database covered roughly 3.2
million establishments in the United States, including about 76.4% of all parent
companies with employment headcounts over 25 in the United States.14 We use the
13
Erel and Liebersohn (2020) find evidence of fintech players originating more PPP loans in areas
with fewer bank branches, lower incomes, and a larger minority share of the population, as well as in
industries with little ex ante small-business lending. Bao and Huang (2021) show that fintech lenders
expand lending to constrained borrowers during the pandemic, but also experience a higher default rate
subsequently. Howell, Kuchler, Snitkof, Stroebel, and Wong (2021) examine racial disparities in access
to PPP loans from banks and the role of fintech lenders.
14
To make this assessment, we compare the National Establishment Time Series database 2019 file
from Dun and Bradstreet, which has been used widely in the literature. From discussions with Aberdeen,
its universe of firms is based on the Dun and Bradstreet database.
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2627
Competitive Installs file which identifies 63 major lines of technology installed at
each establishment. Out of these technologies, we manually identify 14 that we
consider likely to be useful for virtual, automated, or remote work, falling into the
following categories: VPN and remote access, document management, collabora-
tive software, customer relationship management software, video conference
and internet phone software, database software, and server virtualization. Our
IT_INDEX is calculated as the sum of 14 dummies indicating the presence of each
of the technologies. We interpret the presence of these technologies as the bank
having that technology capability and the absence thereof as not having that
capability. We aggregate the index to a bank-level measure by employee-weighting
our office-level IT measure.
While we think our choice of technologies relevant to digital operations is
reasonable, one might be concerned about the possibility of measurement error. To
mitigate such concerns, we show in the Supplementary Material that our analysis
is robust to using alternative IT measures. First, we construct IT_INDEX_OTHER,
calculated analogously to our main index, but including only the technologies
excluded from the main measure.15 We show that remote IT technologies (our
main IT_INDEX) are generally more strongly linked to economic outcomes than
IT_INDEX_OTHER. Second, we measure IT_STAFF, calculated as the number of
IT staff divided by total employees for each bank. We note that our measure of IT
staff is not fully precise, as Aberdeen reports ranges of employees rather than exact
counts (e.g., 1–4 employees, 250–499), which we impute at the midpoint. Third, we
measure IT_BUDGET, calculated as the inverse hyperbolic sine of total IT budget
divided by total employees for each bank. In various analyses throughout our
Supplementary Material, we show that our main results remain robust when mea-
suring bank IT using IT_STAFF or IT_BUDGET.
While alternative measures are valuable, our count-based measures are pref-
erable because they allow us to study the composition of technology and identify
technologies that may be particularly relevant during this period. Moreover, we
posit that the breadth of capabilities matters. For example, a very expensive VPN
alone may be less useful than a suite of software services such as VPN, internet
phones, and video conferencing software.
B. Measuring Physical Branch Visits
To measure customer visits to bank branches, we use aggregated mobile phone
data from SafeGraph, a company producing anonymized mobile phone location
statistics. SafeGraph observes 18.75 million devices, which are approximately
5.6% of the U.S. population and about 10% of mobile devices. According to its
analysis of user characteristics, SafeGraph posits that its sample is representative of
the U.S. population based on income characteristics, age, and demographics of its
users. The data are widely used in studies of social distancing during the COVID-19
pandemic (e.g., Charoenwong, Kwan, and Pursiainen (2020), Weill, Stigler,
Deschenes, and Springborn (2020)) and are found to be highly correlated with
15
Examples of the remainder of the 63 categories include technologies such as workstations, phone
provider, internet providers, and generic business categories.
https://doi.org/10.1017/S0022109023000662
2628 Journal of Financial and Quantitative Analysis
other commonly used measures of social mobility (e.g., Jay, Bor, Nsoesie, Lipson,
Jones, Galea, and Raifman (2020)).
The data include the monthly number of visits at points-of-interest identified
by SafeGraph, including bank branches. We perform a name-based matching of
SafeGraph data to the FDIC Summary of Deposits (SOD) data set to obtain branch
details. We are able to match 56,242 branches out of the total 86,367 branches in the
SOD data. Matching this further to Aberdeen data yields a sample of 38,217
branches for which we can study IT and footfall. We then construct a weekly panel
data set of physical branch visits at the branch level.
C. Measuring Banks’ Digital Activities
For our analysis of website traffic, we collect data from AlexaRank via the
AlexaRank API. AlexaRank estimates the number of unique visits from users of
their global panel who visited a site. The underlying data are sourced from website
browser extensions as well as directly from websites which participate in Alexa’s
data collection program. AlexaRank sources its data from millions of internet users
in total. The visits are converted to a ranking score, known popularly as the
“AlexaRank.” AlexaRank then normalizes the data using a proprietary methodol-
ogy that combines average unique visits and page views per visitor over the past
3 months. In Section IA.IVof the Supplementary Material, we include a robustness
analysis using alternative data from SimilarWeb, another provider of web
traffic data.
Next, we obtain data on the historical composition of firms’ websites through
BuiltWith, which indexes website technologies over time. We obtain data from their
Domains API, which provides technologies for every subdomain observed on a site.
For example, they can measure the existence of jQuery, which is a technology used
to implement modern, advanced functionality and user layouts for web pages. IIS,
Nginx, and Apache are competing types of web servers. Importantly for our
purposes, BuiltWith provides a database of the mentions of COVID-19 throughout
different websites over time. This allows us to measure different banks’ reaction
times to the COVID-19 pandemic on their websites. However small or large, we
interpret an earlier website mention of COVID-19 as a response of the organization
through its website.
Finally, we collect customer review data on bank mobile apps. We collect data
from the Android app store and Apple iOS app store. These data allow us to find
the version of an app of a bank, if it exists. We extract the following features: the
number of reviews, the average score (1–5), the fraction of reviews that are
complaints based on textual analysis, and the fraction of reviews that consider
the app reliable and effective based on textual analysis. To construct the last two
variables, we employ a natural language algorithm open-sourced by Facebook and
described by Lewis et al. (2019). We describe the data as well as this algorithm in
more detail in Section IA.VII of the Supplementary Material.
D. Bank Data
Across different analyses, we measure bank financial characteristics using
FDIC call report data, including bank deposits, size (measured by the inverse
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2629
hyperbolic sine of total assets), capitalization (measured by both equity/asset ratio
and tier-1 capital ratio), profitability (measured by return on equity and cost/income
ratio), funding cost, personnel costs, and the number of states in which the bank
operates.
To measure banks’ small business lending during the COVID-19 pandemic,
we obtain loan-level data from the SBA PPP. This program, established by the
CARES Act, is implemented by the SBA with support from the Department of the
Treasury. It provides small businesses with funds to pay up to 8 weeks of payroll
costs, including benefits. Funds can also be used to pay interest on mortgages, rent,
and utilities. The program was launched on Apr. 3, 2020.
Our data set covers the period from the beginning of the program through the
end of June, which is the original deadline of the program and the time by which the
program was effectively over. By June, banks had conducted over 99% of total PPP
lending. We construct a bank-county-level data set of PPP loans during this period.
To assess the volume of PPP loans relative to the SME lending done by the bank in
normal times, we also obtain pre-pandemic SME lending volume, measured as the
total small business lending by the same bank to the same county in 2019 as
reported under the Community Reinvestment Act (CRA).
E. Mobility Restrictions Due to COVID-19
We obtain county-level data on mobility restrictions from Keystone Strategy,
available via GitHub as well as the Spiegel and Tookes (2021) restrictions database.
These data sets describe the start and end dates of various types of restrictions.
However, many of the restrictions take place simultaneously, and not all of them are
equally relevant to banking activities. To construct a measure of relevant mobility
restriction measures by location, we use a variety of statistical models to identify
those specific restrictions that significantly explain changes in branch visits. We
describe our method of identifying these restrictions at length in Section IA.I of
the Supplementary Material, as well as our validation tests for these restrictions. In
brief, out of 14 different mobility restrictions, our methodology identifies three
restriction categories as being most relevant in explaining branch visits, namely:
restrictions on retail establishments, shelter-in-place, and social-distancing guide-
lines. We define a dummy variable SEVERE_RESTRICTIONS as having at least
two of these three restrictions in place. This covers about 55% of the observations
since March in our branch visit sample (Table 1).
IV. Main Results
A. Bank IT and In-Person Branch Visits
The first aspect of the pandemic that we study is a shift away from in-person
banking. We first present time-series evidence. In Figure 1, we plot the average
weekly bank branch visit volume for banks with above and below sample-median
IT_INDEX values. Our sample period is from Jan. 2020 to Apr. 2020, focusing on
https://doi.org/10.1017/S0022109023000662
2630 Journal of Financial and Quantitative Analysis
TABLE 1
Summary Statistics
Table 1 reports the summary statistics of the main variables in each of our analyses. The variable definitions are discussed in
Section III. The residual standard deviations are estimated for the residual of each variable conditional on the fixed effects
used in the corresponding analysis.
Residual Std.
No. of Obs. Mean Std. Dev. Dev. P25 P50 P75
Branch Visits
IT_INDEX 523,844 5.953 3.028 0.000 2.934 6.795 8.633
SEVERE_RESTRICTIONS 523,844 0.284 0.451 0.159 0 0 1
ihs(BRANCH_VISITS) 523,844 2.751 1.177 0.446 1.818 2.776 3.638
Web Traffic
IT_INDEX 29,946 2.384 1.581 0.000 1.191 2.000 3.000
SEVERE_RESTRICTIONS 29,946 0.278 0.448 0.168 0 0 1
RANK ≤ 100k 29,946 0.056 0.229 0.077 0 0 0
Response Time
IT_INDEX 3,751 2.228 1.921 1.871 1.000 1.925 2.685
ihs(WEBSITE_TECHS) 3,751 5.152 0.589 0.560 4.754 5.171 5.545
RESPONSE_TIME 3,751 23.5 21.3 19.9 14 16 28
PPP
IT_INDEX 30,407 5.124 2.677 2.369 2.876 4.577 7.662
ihs(COVID) 30,407 2.168 0.936 0.000 1.544 2.181 2.827
ihs(PPP) 30,407 14.288 2.489 2.134 12.407 14.326 16.088
Customer Switching
IT_INDEX_PRE_COVID_BANK 97,645 5.095 2.913 2.794 2.251 4.818 8.032
ihs(COVID) 97,645 2.420 0.912 0.891 1.849 2.520 3.032
SWITCH_TO_BETTER_IT 97,645 0.180 0.384 0.380 0 0 0
Deposits
IT_INDEX 29,072 1.850 1.336 0.000 1.000 1.667 2.138
ihs(COVID) 29,072 0.391 0.879 0.340 0.000 0.000 0.123
ihs(DEPOSITS) 29,072 13.115 1.417 0.053 12.185 12.937 13.809
Customer Reviews
IT_INDEX 2,704 2.411 1.659 0.000 1.273 2.000 3.000
ihs(COVID) 2,704 2.597 2.432 2.422 0.000 4.327 4.931
ihs(#REVIEWS) 2,704 0.635 0.265 0.114 0.500 0.713 0.829
SCORE ≥ 4 2,704 0.091 0.063 0.027 0.042 0.076 0.125
COMPLAINTS 2,704 0.297 0.260 0.117 0.441 0.229 0.112
RELIABLE 2,704 3.751 1.480 0.383 2.492 3.333 4.543
FIGURE 1
Branch Visits Versus Bank IT
Figure 1 plots the average weekly number of physical bank branch visits for banks with above- and below-median IT_INDEX.
Both series are scaled relative to the median of the first 5 weeks of the year.
0.10
0.00
–0.10
–0.20
–0.30
–0.40
–0.50
–0.60
20 20 3/ 1/
/1 0
2/
10 020
/2
17 020
20
/2
2
/2
2
02 0
20 /2 0
/2
24 020
/2
2/ 20
3/
9/ 20
3/
16 020
/2
/2
20
2 0
6/1/ /1 /3
23 020
/3 /2
/2
30 020
/3 /2
13 20 /
6/ 20
13 4/
27
/4 0
/2
20 020
/4
27 020
/4
0
20
/2
/2
2
02 0
Low IT High IT
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2631
the window before and after the onset of the pandemic in the United States.16 The
graph is normalized so that low- and high-IT banks have the same pre-pandemic
median during the first 5 weeks of the year.17 In January and February, there is no
noticeable difference between high- and low-IT banks. However, a clear divergence
emerges in mid-Mar. 2020, when COVID-19 cases started to increase and mobility
restrictions were put in place in many communities.
We next study how in-person branch visits change in response to the intro-
duction of mobility restrictions, and how the responses vary by banks’ IT_INDEX.
This analysis exploits staggered local variation in mobility restrictions, which have
been shown by many prior studies to be associated with significant increases in
social distancing and reduced mobility.18 We expect mobility restrictions to result
in fewer branch visits because they significantly increase customers’ costs of
visiting bank branches in person, as well as banks’ costs of operating branch
banking services. In addition, we expect mobility restrictions to have larger impact
on banks with better IT, as their stronger digital capabilities may enable them to
more easily substitute online services for branch services.
To test for the effect of mobility restrictions on branch visits, conditional on
bank IT, we perform a regression analysis of the following form:
(1) ihsðBRANCH_VISITSÞi,t ¼ αi þ γt þ βSEVERE_RESTRICTIONSc,t
×IT_INDEXj þ γSEVERE_RESTRICTIONSc,t
þϕXi,t þ εi,t ,
where ihs(BRANCH_VISITS)i,t is the inverse hyperbolic sine of the weekly num-
ber of visits at branch i in week t. SEVERE_RESTRICTIONSc,t is a dummy
variable indicating our measure of severe restrictions are currently in place in
county c where the branch is located, at week t. IT_INDEXj measures the ex ante
IT capabilities of bank j that owns the branch, and Xi,t is a vector of controls. We
include branch fixed effects (αi ), week fixed effects (γt ), and, depending on the
specification, also county linear trends or county-week fixed effects.
In Section IA.II of the Supplementary Material, we present supporting evi-
dence that this county-level restriction indicator affords a valid difference-in-
difference design. In Figure IA.1 in the Supplementary Material, we present an
event-time plot using the recently proposed estimator by Borusyak, Jaravel, and
Spiess (2022). We first validate the staggered effects of mobility restrictions on
a significant decline of physical branch visits. We then proceed to look at the
differential responses across high- and low-IT banks and show that there is a
divergence in branch visits between high- and low-IT in response to local mobility
restrictions.
16
This horizon is comparable to many other papers that study the COVID-19 setting, such as Bian,
Li, Xu, and Foutz (2022).
17
We use the first 5 weeks to follow the same methodology as Google Community Mobility Reports.
The patterns observed would look very similar if we instead chose the first 4 or 8 weeks.
18
Such studies use restrictions to study the effect of moderating variables such as social connections,
political affiliation, and other factors (e.g., Allcott, Boxell, Conway, Gentzkow, Thaler, and Yang (2020),
Charoenwong et al. (2020), and Ding, Levine, Lin, and Xie (2020).
https://doi.org/10.1017/S0022109023000662
2632 Journal of Financial and Quantitative Analysis
TABLE 2
Branch Visits During Mobility Restrictions
Table 2 tests the impact of mobility restrictions on branch visits for banks during the COVID-19 pandemic. The unit of
observation is a branch-week. The dependent variable, ihs(BRANCH_VISITS) is the inverse hyperbolic sine of the number
of visits recorded in SafeGraph’s Places of Interest file. SEVERE_RESTRICTIONS is a dummy variable indicating severe
restrictions as defined in Section IA.I of the Supplementary Material. The sample period is from Jan. to Apr. 2020.
Heteroscedasticity-robust standard errors clustered by county and bank are reported in parentheses. *, **, and *** denote
statistical significance at the 10%, 5%, and 1% levels.
ihs(BRANCH_VISITS)
1 2 3 4 5
SEVERE_RESTRICTIONS 0.0798*** 0.0348*** 0.0760*** 0.0351***
(0.0093) (0.0066) (0.0117) (0.0052)
IT_INDEX × SEVERE_RESTRICTIONS 0.0224*** 0.0108*** 0.0064**
(0.0033) (0.0026) (0.0029)
Branch FE Yes Yes Yes Yes Yes
Week FE Yes Yes Yes Yes No
County linear trends No Yes No Yes No
County-week FE No No No No Yes
No. of obs. 522,370 522,370 522,370 522,370 515,256
R2 0.8561 0.8606 0.8568 0.8607 0.8685
In Table 2, we present our regression analysis. The results confirm that
following the introduction of severe mobility restrictions, branch visits decrease
substantially more for the high-IT banks relative to low-IT ones. This is consistent
with the notion that banks with better technology are less reliant on physical
branches to serve their customers during the start of the COVID-19 pandemic. In
terms of economic magnitudes, column 2 suggests that branch visits on average
decline by 3.5% in response to severe mobility restrictions. Additionally, column
4 shows that a 1-standard-deviation-higher bank IT_INDEX almost doubles the
average effect of restrictions. This effect remains robust when controlling for
county-by-week fixed effects.
That branch visits drop in response to mobility restrictions could reflect both
customers proactively choosing to avoid bank branches or banks simply curtailing
their branch banking services. Studies like Kreiss (2021) find that branches closed
at an accelerated rate during the pandemic. While we do not know precisely whether
branch service hours are shortened or whether a branch is closed, we conduct two
additional analyses using remote work data from Kwan and Matthies (2022) which
we further describe in Section IA.III of the Supplementary Material. The basic idea
is that bank employees’ intensity of remote work likely reflects banks’ operational
changes in their brick-and-mortar branch services. First, we ask if bank employees’
remote work intensity is affected by the mobility restrictions. In Panel A of
Table IA.3 in the Supplementary Material, we find that employees of strong-IT
banks are more able to work remotely in response to the mobility restrictions. This
finding provides supporting evidence that the pandemic indeed leads to operational
changes by banks. Second, we control for the remote work measure in our branch
visit analysis (Panel B of Table IA.3 in the Supplementary Material). We find
similar results to Table 2: controlling for our proxy of bank operational behavior
(remote work), the relation between IT and mobility restrictions on bank visits
remains significant. These findings provide suggestive evidence that the drop in
branch visits is likely driven by a combination of both customer and bank behavior
as opposed to one channel exclusively.
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2633
B. Bank IT and Website Traffic
We next perform a complementary analysis comprised of studying customers’
online visits to banks’ websites. Our main measure of bank web traffic is based on
the AlexaRank, which provides an ordinal measure of the traffic volume of web-
sites. As mentioned before, AlexaRank tracks internet traffic using a panel of
internet users who have browser extensions installed which Alexa owns. This is
only a fraction of the internet, and according to the data provider, websites with low
levels of traffic are ranked less accurately. This makes the AlexaRank only reliable
for the top 100,000 websites.19 In light of this caveat, we define three outcome
variables based on whether the AlexaRank crosses the 100,000 threshold. The three
outcome variables we use are: i) a dummy indicating whether the bank’s median
rank within a week is in the top 100,000, ii) the percentage of days in the week the
rank is in the top 100,000, and iii) a dummy for whether more than half of the days in
a week are in the top 100,000. As a bank may have multiple establishments in
multiple counties, we further construct a bank-level restriction measure based on
banks’ differential geographic exposures to the pandemic and mobility restrictions.
Here, our bank-level restriction indicator equals 1 if at least half of the bank
branches are located in counties where our county-level restriction indicator
equals 1.20
Based on these measures, we test for the effect of mobility restrictions on bank
website traffic, conditional on bank IT, using regressions of the following form:
(2) MEDIAN_RANK ≤ 100ki,t ¼ αi þ γt þ βSEVERE_RESTRICTIONSi,t
× IT_INDEXi þ θSEVERE_RESTRICTIONSi,t
þϕXi,t þ εi,t ,
where MEDIAN_RANK ≤ 100ki,t is a dummy taking the value of 1 if the median
rank of the bank i in AlexaRank is in the top 100,000 websites in week t. We can also
replace this outcome variable with the other two measures of bank web traffic we
introduced above. SEVERE_RESTRICTIONSi,t is a dummy variable taking the
value of 1 if our county-level restriction measure is in place in at least half of the
bank’s branch locations, at week t. IT_INDEXi measures the pre-pandemic IT
capabilities of bank i, and Xi,t is a vector of controls. We include bank fixed effects
(αi ) and week fixed effects (γt ).
The results on web traffic mirror those on branch visits. After mobility restric-
tions come into place, banks with better IT exhibit significantly larger increases
in website traffic relative to banks with weaker IT. We report regression results in
Table 3. The odd-numbered columns show the effect of restrictions alone. Accord-
ing to estimates in column 2, a bank with a 1-standard-deviation-higher IT_INDEX
has a 77-BPS-higher probability of receiving higher web traffic such that the
AlexaRank is above 100,000. This is about 10% of the within-bank standard
19
For a discussion of this issue, please see https://attentioninsight.com/what-is-alexa-rank-and-its-
value/.
20
Our results are robust to alternative measures of bank-level restrictions. For example, the results
remain robust if we define the bank-level restriction indicator by considering at least 25% or any
branches are exposed.
https://doi.org/10.1017/S0022109023000662
2634 Journal of Financial and Quantitative Analysis
TABLE 3
Website Traffic During Mobility Restrictions
Table 3 tests the impact of mobility restrictions on bank website traffic during the COVID-19 pandemic. The unit of observation
is a bank-week. We construct three outcome variables based on whether or not the AlexaRank crosses the 100,000 threshold.
This threshold is chosen because documentation suggests that AlexaRank is more reliable for the top 100,000 websites due to
the sparsity of data for lower-ranked websites. MEDIAN_RANK ≤ 100_k indicates whether the bank’s median rank in the week
is in the AlexaRank top 100,000. %(RANK ≤ 100k) is the percentage of days in the week where the bank’s rank is in the top
100,000. %(RANK ≤ 100k) ≥ 50% is a dummy indicating whether the rank is in the top 100,000 for more than half of the days
during the week. SEVERE_RESTRICTIONS is a bank-level dummy indicator that equals 1 if more than half of the bank
branches are located in counties with severe restrictions as defined in Section IA.I of the Supplementary Material. The
sample period is from Jan. to Apr. 2020. Heteroscedasticity-robust standard errors clustered by bank are reported in
parentheses. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels.
%
MEDIAN_RANK ≤ 100k %(RANK ≤ 100k) (RANK ≤ 100k) ≥ 50%
1 2 3 4 5 6
SEVERE_RESTRICTIONS 0.0045* 0.0039* 0.0040*** 0.0037** 0.0067*** 0.0063***
(0.0023) (0.0022) (0.0016) (0.0015) (0.0021) (0.0020)
IT_INDEX × SEVERE_RESTRICTIONS 0.0049*** 0.0023*** 0.0040**
(0.0015) (0.0009) (0.0016)
Bank FE Yes Yes Yes Yes Yes Yes
Week FE Yes Yes Yes Yes Yes Yes
No. of obs. 29,946 29,946 29,946 29,946 29,946 29,946
R2 0.8880 0.8882 0.9589 0.9590 0.8920 0.8921
deviation. The results are similar when we measure relative web traffic using the
other two variables mentioned above. We also present an event-study analysis in
Figure IA.2 in the Supplementary Material. This figure suggests a larger jump in
web traffic by high-IT banks at the onset of mobility restrictions.
To address the concern that the web traffic analysis is based on a rank measure
that does not capture the absolute changes in web traffic volume, we conduct a
robustness analysis using an alternative source of web traffic data provided by
SimilarWeb, which uses similar data collection methodologies as AlexaRank, but
produces visitor counts instead of a rank variable although only at the monthly
level. We report these analyses in Table IA.4 in the Supplementary Material. This
test is relatively less powerful, as analysis at the monthly level obscures variation
across banks in terms of how quickly they were impacted by mobility restrictions
during the middle of March. Still, like with our AlexaRank analysis, we find that
high-IT banks receive more web traffic in response to mobility restrictions.
To provide further support for the above findings, we next check whether our
IT_INDEX is correlated with characteristics of bank websites, which likely play
a large role in digital provision of financial services. Using data from BuiltWith,
we calculate the number of “web technologies” present on banks’ websites. A web
technology can comprise many types, including graphical technologies such as
jQuery, web traffic services such as content-delivery networks or the type of web
server, security certificates, and much more. While not all technologies are equally
important or useful, a greater number of such technologies implies greater techno-
logical investment and sophistication, and thus should be characteristic of banks
with better IT.
In columns 1 and 2 in Table 4, we perform a cross-sectional regression analysis
of the number of website technologies conditional on the bank’s IT_INDEX. Our
IT_INDEX is significantly positively correlated with website technologies, even
though we do not directly measure these technologies in the index.
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2635
TABLE 4
Website Characteristics and COVID-19 Response Time
Table 4 characterizes the relation between IT_INDEX and website characteristics. The unit of observation is a bank.
ihs(WEBSITE_TECHS) is the inverse hyperbolic sine of the number of observed BuiltWith technologies by Feb. 2020.
RESPONSE_TIME is measured as the number of days from Mar. 13, 2020 (the declaration of national emergency) that it
took for the bank to mention COVID-19 on its website as reported by BuiltWith. Negative coefficients imply earlier responses.
Bank controls include size (measured by the inverse hyperbolic sine of total assets), capitalization (measured by both equity/
asset ratio and tier-1 capital ratio), profitability (measured by return on equity and cost/income ratio), funding cost, and
personnel costs. In columns 3 and 4, we also include fixed effects for decile bins of average COVID cases per capita across
each bank’s branch network, which controls for the cross-bank variation of COVID severity that might affect banks’ speed of
response. Heteroscedasticity-robust standard errors clustered by bank are reported in parentheses. *, **, and *** denote
statistical significance at the 10%, 5%, and 1% levels.
ihs(WEBSITE_TECHS) RESPONSE_TIME
1 2 3 4
IT_INDEX 0.1148*** 0.0589*** 0.9771*** 0.3911**
(0.0072) (0.0054) (0.1870) (0.1935)
Bank controls No Yes No Yes
HQ state FE Yes Yes No No
HQ state × COVID decile FE No No Yes Yes
No. of obs. 3,750 3,741 3,651 3,642
R2 0.2305 0.4574 0.1373 0.1735
One particularly interesting aspect that BuiltWith tracks is the time when
websites mention the COVID-19 pandemic. We define a response time variable
as the number of days between Mar. 13— when a national emergency was declared
—and the first time the bank website mentions the pandemic. Columns 3 and 4 in
Table 4 show the results of a regression analysis of the response time to COVID-19
as implied by bank websites. We find that banks with better IT react significantly
faster to the pandemic on their websites. A 1-standard-deviation increase in
IT_INDEX is associated with approximately 1–2 days reduction in reaction time,
depending on model specification. This suggests that firms which have better IT are
able to respond more quickly to the pandemic.
C. Bank IT and PPP Lending
1. PPP Loan Volumes
Next, we focus on the amount of small business loans originated under the
SBA PPP. This can be viewed as the extensive margin of PPP lending – we later
examine the intensive margin in 2. We calculate the total volume of PPP loans
originated in each county by each bank. To test for the effect of bank IT on PPP loan
origination volume, we first perform the following regression analysis:
(3) ihsðPPPi,c Þ ¼ αc þ γhq þ βIT_INDEXi þ ϕXi,c þ δXi þ εi,c ,
where ihs(PPPi,c Þ is the inverse hyperbolic sine of the volume of PPP loans
originated by bank i in county c. Xi,c is a vector of pre-pandemic bank-county level
controls, including the inverse hyperbolic sine of the amount of CRA small business
lending by the same bank in the same county and the log level of deposits to
branches of the same banks in the same county in the year before the pandemic.
Xi is a vector of bank characteristics, including bank size (measured by the inverse
hyperbolic sine of total assets), capitalization (measured by both equity/asset ratio
https://doi.org/10.1017/S0022109023000662
2636 Journal of Financial and Quantitative Analysis
and tier-1 capital ratio), profitability (measured by return on equity and cost/income
ratio), funding cost, personnel costs, and the number of states the bank operates
in. αc refers to the county fixed effects, and γhq is a dummy variable indicating the
state of the bank headquarter. It should be noted that by controlling for county fixed
effects, we are comparing PPP lending by strong- and weak-IT banks within the
same location. This empirical strategy effectively controls for the potential demand
effects driven by local economic or pandemic conditions, allowing us to focus on
the supply side.
The results are shown in columns 1 and 2 in Table 5. Across all the specifi-
cations, stronger bank IT, as measured by a higher IT_INDEX, is related to a
significantly larger amount of PPP lending during the pandemic. This result is
robust when controlling for the ex ante SME lending as well as for deposits of the
same bank in the same county, which control for the ex ante market share of each
bank in each market. The effect is economically significant: According to the point
estimate in column 2 (e.g., a 1-standard-deviation increase in bank IT leads to a 27%
increase in PPP lending to small businesses), which accounts for about 12.5% of the
residual standard deviation of PPP lending.
If our results are driven by the ability of banks with good IT to better serve
customers digitally during the COVID-19 outbreak, we might expect the estimated
effect of IT on PPP loan volumes to be larger in areas with worse outbreaks of
COVID-19. There are two main reasons for this. First, although PPP applications
were filed mostly online, loan officers and other branch personnel were often
involved in screening loans or assisting customers in the application process.21
In counties with larger outbreaks of COVID-19, in-person supports to customers
and offline work by loan officers was more difficult, and thus stronger-IT banks had
greater advantages processing and screening PPP applications through their digital
infrastructure. Second, areas with high COVID-19 numbers saw greater loan
demand (e.g., Bartik, Cullen, Glaeser, Luca, Stanton, and Sunderam (2020b)),
and banks likely suffered longer backlogs especially if they had poorer IT.22
Customers, either experiencing or anticipating slow response times from worse-
IT banks, would be likelier to switch to banks with better IT, as slow processing
means longer waiting time and even a lower likelihood of obtaining a loan, given
that the program has a limited amount of funds.
To test this prediction, we study the impact of bank IT conditional on the
severity of the outbreak. We perform a regression analysis including an interaction
between the IT_INDEX and the inverse hyperbolic sine of the number of confirmed
COVID-19 cases per thousand people in the county. The result is shown in column
3 in Table 5. According to the estimates, if the number of confirmed COVID-19
cases increases by 1-standard-deviation, the estimated effect of bank IT is about
17% larger compared to the average effect.
21
For example, when Pinnacle Bank started to accept PPP loan applications, it explicitly advised its
customers that “in order to apply, contact your local branch or loan officer.”
22
Anecdotally, the PPP loan processing time could reach as high as several weeks if the applicants are
processed manually, but only a few days or even shorter when the process is automated. For example, see
https://markets.businessinsider.com/news/stocks/automation-anywhere-launches-ai-powered-banking-
bot-to-expedite-sba-loan-processing-from-3-weeks-to-3-days-1029098646.
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2637
TABLE 5
PPP Lending During the Pandemic
Table 5 estimates the effect of bank IT on PPP lending to small businesses. The unit of observation is a bank-county. The
dependent variable is ihs(PPP), the inverse hyperbolic sine of the amount of PPP loans originated by each bank in each
county. ihs(COVID) is the inverse hyperbolic sine of the number of total confirmed COVID-19 cases per 1,000 people in the
county until June 30, 2020. INTERNET_USE is the percentage of households having access to high-speed internet. HHI is
the Herfindahl–Hirshman Index of ex ante SME lending across banks in the county. Bank controls include size (measured by
the inverse hyperbolic sine of total assets), capitalization (measured by both equity/asset ratio and tier-1 capital ratio),
profitability (measured by return on equity and cost/income ratio), funding cost, personnel costs, and the number of states
where the bank operates. Bank-county controls include ihs(CRA) and ihs(DEPOSITS). ihs(CRA) is the inverse hyperbolic sine
of the pre-pandemic amount of CRA small business lending by the same bank to the same county. ihs(DEPOSITS) is the
inverse hyperbolic sine of the amount of pre-pandemic deposits of the bank in the same county. Heteroscedasticity-robust
standard errors clustered by county and bank are reported in parentheses. *, **, and *** denote statistical significance at the
10%, 5%, and 1% levels.
ihs(PPP)
1 2 3 4 5
IT_INDEX 0.1530*** 0.0993***
(0.0298) (0.0308)
IT_INDEX × ihs(COVID) 0.0184***
(0.0061)
IT_INDEX × INTERNET_USE 0.0098***
(0.0025)
IT_INDEX × HHI 0.2950***
(0.0672)
Bank-county controls No Yes Yes Yes Yes
Bank controls No Yes No No No
HQ state dummy Yes Yes Yes Yes Yes
County FE Yes Yes Yes Yes Yes
Bank FE No No Yes Yes Yes
No. of obs. 30,166 30,166 30,163 30,130 30,163
R2 0.2852 0.6521 0.7077 0.7082 0.7080
We then explore the importance of local market characteristics in the role of
bank IT. We first study the level of internet use, measured by the share of households
with access to broadband internet, as a proxy for customers’ propensity to use and
differentiate between banks’ online services. The result, shown in column 4 in
Table 5, suggests that bank IT matters more when the customer base has better
internet infrastructure. This is intuitive, as customers in these areas are more likely
to shift to digital banking than those in areas with low internet use.
Finally, we study how the role of bank IT varies across locations with different
levels of banking competition. As a measure of market concentration, we calculate
the Herfindahl–Hirschman Index (HHI) of SME loans in the county before the
pandemic. Column 5 shows that the effect of IT_INDEX on PPP loan volumes is
higher when the market concentration is lower (as measured by a lower HHI index),
suggesting that a strong IT capability is more valuable to a bank when it faces
greater competition in a local market.
It is worth noting that these specifications control for both the county and bank
fixed effects. Thus, a stronger IT capability can have a stronger effect on PPP
lending in counties with a greater COVID-19 shock, better internet, and greater
banking competition. These within-bank comparisons also help us further rule out
the concern that our results might be driven by unobserved bank characteristics that
are correlated with bank IT.
https://doi.org/10.1017/S0022109023000662
2638 Journal of Financial and Quantitative Analysis
TABLE 6
Likelihood of Customer Switching Banks for PPP Loan
Table 6 estimates small businesses’ likelihood of switching to better IT banks during the pandemic. The sample of small
businesses are those taking SBA 7a loans between 2011 and 2019 which we can match with the Aberdeen Computer
Intelligence database. The dependent variable is a dummy taking the value of 1 if the borrower’s PPP lender has a higher
IT_INDEX than this same borrower’s pre-COVID relationship bank (via SBA loans). Heteroscedasticity-robust standard errors
clustered by borrower county are reported in parentheses. *, **, and *** denote statistical significance at the 10%, 5%, and 1%
levels.
SWITCH_TO_BETTER_IT
1 2 3
ihs(COVID) 0.0128*** 0.0215*** 0.0219***
(0.0041) (0.0049) (0.0045)
IT_INDEX_PRE_COVID_BANK 0.0262*** 0.0262***
(0.0017) (0.0016)
ihs(COVID) × IT_INDEX_PRE_COVID_BANK 0.0058***
(0.0013)
Industry (NAICS) FE Yes Yes Yes
No. of obs. 97,548 97,548 97,548
R2 0.0209 0.0568 0.0584
2. Likelihood of Switching Banks
If weak IT results in poor service to customers during the pandemic, we might
expect customers to switch to banks with better ITwhen choosing lenders that provide
PPP loans. This can be viewed as the intensive margin of PPP lending, whereas our
previous analysis in Table 5 might be viewed as the extensive margin. Furthermore,
we might expect such bank switching to be related to both the IT capability of the
existing bank as well as to the severity of the COVID-19 impact at the location of the
customer. We can empirically test these predictions for SME customers that are
present in both SBA 7a programs ex ante and the PPP program ex post. We match
borrowers across these two samples and form a panel of 97,758 firms that are PPP
borrowers in 2020 and SBA borrowers between 2011 and 2019.23 To identify
borrowers that switch banks, we look at firms that take PPP loans who also have
an SBA loan, and we define a binary outcome variable that takes the value of 1 if the
firm changes to a bank with a higher IT_INDEX than its pre-COVID-19 lender had.
Our hypothesis is that a firm facing more COVID-19 exposure should be more likely
to switch to a lender that has a higher IT score than their pre-COVID-19 lender.
Our results are shown in Table 6. Consistent with our prediction, we find that
a borrower is more likely to switch to a bank with better IT if it is in a county
more affected by the pandemic. Controlling for the IT strength of the pre-COVID
bank, the estimates in column 2 suggest that a 1-standard-deviation increase in
the severity of COVID-19 would result in an approximately 2-percentage-point-
higher likelihood that a borrower switches to a stronger-IT bank after the onset
of the pandemic. This increase is roughly 11% against the mean probability of
switch (2/18). The interaction term in column 3 further suggests that, given a
1-standard-deviation increase in COVID severity, if a borrower is already related
to a bank with a 1-standard-deviation-higher IT_INDEX, then the probability that
23
The public data of PPP lending only name borrowers with borrowing amounts above $150,000,
while normal SBA loans target relatively smaller borrowers. Thus, the intersection of the two sets of
firms only represents a fraction of the universe.
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2639
TABLE 7
Deposits During the Pandemic
Table 7 estimates the relation between bank IT and deposits during the pandemic. The dependent variable is ihs(DEPOSITS),
the inverse hyperbolic sine of the quarterly bank-level deposits. The sample period is from 2019 Q1 to 2020 Q2.
Q1_2020_ONWARD is a dummy variable that equals 1 for the first 2 quarters of 2020. Bank controls include size
(measured by the inverse hyperbolic sine of total assets), capitalization (measured by both equity/asset ratio and tier-1
capital ratio), profitability (measured by return on equity and cost/income ratio), funding cost, and personnel costs. Industry
(NAICS) refers to the industry classifications from the North American Industrial Classification system. Heteroscedasticity-
robust standard errors clustered by bank are reported in parentheses. *, **, and *** denote statistical significance at the 10%,
5%, and 1% levels.
ihs(DEPOSITS)
1 2 3 4
IT_INDEX × Q1_2020_ONWARD 0.0066*** 0.0033***
(0.0010) (0.0006)
IT_INDEX × ihs(COVID) 0.0028*** 0.0012***
(0.0005) (0.0003)
ihs(COVID) 0.0094*** 0.0037***
(0.0014) (0.0012)
Bank controls No Yes No Yes
Bank FE Yes Yes Yes Yes
Quarter FE Yes Yes Yes Yes
No. of obs. 29,062 27,472 29,062 27,472
R2 0.9986 0.9993 0.9986 0.9993
this borrower switches to another bank with even higher IT would be 1.5-
percentage-points lower relative to a borrower related with an average-IT bank.
Taken together, the results suggest that a stronger COVID-19 shock leads to a
greater probability of switching to a bank with better IT.
D. Bank IT and Deposits
Next, we examine the dynamic of bank deposits over the first 2 quarters of
2020 as another measure of a bank’s ability to serve its customers during the crisis.
Using quarterly bank-level data from call reports, we run the following panel
analysis to estimate the role of bank IT in their ability to attract deposits:
(4) ihsðDEPOSITSÞi,t ¼ αi þ γt þ βIT_INDEXi × Q1_2020_ONWARD
þϕXi,t1 þ εi,t ,
where ihs(DEPOSITS)i,t is the inverse hyperbolic sine of the deposits of bank i in
quarter t. αi and γt are bank and quarter fixed effects. Q1_2020_ONWARD is a
dummy variable that equals 1 for the first 2 quarters in 2020 (the first wave of
COVID-19), and 0 otherwise. Xi,t1 is a vector of lagged controls, including bank
size (measured by the inverse hyperbolic sine of total assets), capitalization (mea-
sured by both equity/asset ratio and tier-1 capital ratio), profitability (measured by
return on equity and cost/income ratio), funding cost, and personnel costs.
The results are shown in Table 7. Columns 1 and 2 compare deposits of banks
with different IT levels in the first 2 quarters of 2020 (the first wave of COVID-19)
versus those in 2019, with column 2 including bank controls. Columns 3 and 4 focus
on the interaction between IT_INDEX and the inverse hyperbolic sine of COVID-
19 cases. Both analyses show that better bank IT is associated with a significantly
https://doi.org/10.1017/S0022109023000662
2640 Journal of Financial and Quantitative Analysis
greater increase in deposits during the onset of the pandemic. During the first 2
quarters of 2020, a 1-standard-deviation-higher bank IT explains higher deposit
flows equivalent to 8.3% of the residual standard error.24 Moreover, banks with
stronger IT also see a higher deposit sensitivity to COVID-19. We present a
graphical version of this result in Figure IA.3 in the Supplementary Material. This
figure reveals that a gap in deposit levels between banks with high versus low IT is
insignificant before the pandemic but becomes statistically and economically sig-
nificant starting in 2020.
E. Alternative IT Measures and Restrictions Measures
We conduct a series of robustness analyses to mitigate measurement concerns
regarding how we construct the bank IT_INDEX and mobility restrictions. First, in
Section IA.Vof the Supplementary Material, we repeat our main analyses employ-
ing three alternative measures of bank IT. First, our results remain robust when we
measure bank IT using all the other technologies (IT_INDEX_OTHER) that are not
included in the main IT measure. However, when we include both the main IT
measure and IT_INDEX_OTHER simultaneously in our regression, our main
measure is stronger in nearly all analyses, which further corroborates that our main
IT_INDEX reflects the technologies that are most relevant for our analysis. This test
also helps mitigate concerns of omitted variables. If IT were correlated with other
unobserved characteristics, it is less clear why such characteristics would be spe-
cifically correlated with our specific 14 technologies which are highly relevant to
remote operations. Second, our results remain similar when we measure bank IT
using IT_STAFF, the number of IT staff relative to the total number of employees,
which reflects a bank’s stock of technology labor force or IT_BUDGET, the log
amount of IT budget per employee, which reflects a bank’s financial investment in
technology.
Second, we show that our results are not sensitive to how we classify “severe”
mobility restrictions. In Section IA.VI of the Supplementary Material, we measure
restrictions by simply counting the number of restrictions in place instead of
defining a dummy variable indicating whether restrictions are above a high thresh-
old. Our results indicate that higher restriction counts relate to statistically signif-
icant drops in footfall and jumps in web traffic.
V. Additional Analysis
A. Customer Reviews
Our previous results suggest that banks with a higher IT_INDEX shifted
operations to digital channels faster, resulting in more deposit inflow and small
business loan volumes. These results are financial outcomes, which we believe
could stem from high-IT banks being able to provide higher-quality services to
customers during the pandemic. To test this more directly, we collect information
about customer satisfaction with digital banking services as expressed through
24
The standard deviation of bank IT is 1.34 in this sample, while deposit growth has a within-bank
residual standard error of 0.053.
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2641
TABLE 8
Customer Reviews
Table 8 presents an analysis of customer reviews. We obtain mobile app reviews from the Google Playstore and Apple App
Store and process them using NLP algorithms outlined in Section IA.VII of the Supplementary Material. The pre-period is from
Mar. to Dec. 2019 and the post-period is the same period in 2020. The dependent variable is shown above each column. ihs
(#REVIEWS) is the inverse hyperbolic sine of the number of customer reviews. SCORE ≥ 4 is the fraction of ratings scoring at
least 4 out of 5. COMPLAINTS is the fraction of reviews indicating complaints. RELIABLE is the fraction of reviews conveying
the feeling that the application is effective. In Panel A, POST is a dummy indicating the period of 2020. In Panel B, ihs(COVID)
refers to the inverse hyperbolic sine of the average number of cases per capita across bank branches by the end of 2020.
Bank controls include size (measured by the inverse hyperbolic sine of total assets), capitalization (measured by both equity/
asset ratio and tier-1 capital ratio), profitability (measured by return on equity and cost/income ratio), funding cost, and
personnel costs. Heteroscedasticity-robust standard errors clustered by bank are reported in parentheses. *, **, and ***
denote statistical significance at the 10%, 5%, and 1% levels.
ihs(#REVIEWS) SCORE ≥ 4 COMPLAINTS RELIABLE
1 2 3 4
Panel A. Two-Period Analysis
IT_INDEX × POST 0.0337*** 0.0072** 0.0018** 0.0073**
(0.0123) (0.0034) (0.0008) (0.0034)
POST 0.1753*** 0.0568*** 0.0140*** 0.0732***
(0.0473) (0.0153) (0.0038) (0.0156)
Bank controls Yes Yes Yes Yes
Bank FE Yes Yes Yes Yes
No. of obs. 2,194 2,194 2,194 2,194
R2 0.9385 0.8237 0.8084 0.7420
Panel B. COVID-19 Exposure
IT_INDEX × ihs(COVID) 0.0075*** 0.0015** 0.0004** 0.0015**
(0.0026) (0.0007) (0.0002) (0.0007)
ihs(COVID) 0.0318*** 0.0117*** 0.0030*** 0.0156***
(0.0097) (0.0031) (0.0008) (0.0032)
Bank controls Yes Yes Yes Yes
Bank FE Yes Yes Yes Yes
No. of obs. 2,194 2,194 2,194 2,194
R2 0.9386 0.8238 0.8105 0.8022
mobile phone app reviews. Mobile app reviews are a major source of data that we
can use to directly measure the quality of a bank’s online services.
We collect over 2.4 million mobile app reviews from the Apple and Google
mobile application stores. We aggregate these reviews into a two-period panel and
compare banks’ app reviews from Mar. to Dec. 2020 versus the same period in
2019. We conduct a two-period test here because there may be a lag between when a
person downloads an app and when they review said app. A user may only review
an app after using it for a while. For example, a user might feel compelled to review
the app when the app does not work for a specific transaction, which might occur
several months or even longer after installing an app.
We construct four outcome variables: the inverse hyperbolic sine of the count
of reviews, the fraction of reviews with four stars or greater (following Core and
De Marco (2023)), the fraction of textual comments that convey complaints, and the
fraction of comments which describe the app as “reliable.” The last two measures
are computed based on a state-of-the-art natural language processing algorithm
recently open-sourced by Facebook, called the BART Natural Language Inference
model. Some apps are very sparsely rated (or the app store does not return a large
number of reviews), and thus we require apps to have at least three reviews in both
https://doi.org/10.1017/S0022109023000662
2642 Journal of Financial and Quantitative Analysis
periods to qualify for our analysis. We further describe our data collection process
as well as the textual analysis in Section IA.VII of the Supplementary Material.
We present our results in Table 8. We first describe Panel A, which simply
compares banks before and after the pandemic. After the COVID-19 pandemic
begins, customers of banks issue a greater number of reviews relative to 2019, and
the increase in the number of reviews is greater for high-IT banks. These results
likely imply more app downloads and/or usage for these banks, and in that sense
concur with our earlier findings that online traffic increases in response to the
pandemic, and that high-IT banks especially attract more customers to digital
channels. Meanwhile, for the average bank, the fraction of high scores decreases
on average, likely because new mobile customers are more demanding of mobile
apps or more issues are detected when customers start using mobile apps more
intensively during the pandemic period. However, the interaction term suggests that
relatively speaking, despite more negative customer reviews for the average bank,
high-IT banks receive a higher fraction of above-4 scores. The results are robust if
we measure the reviews using the fraction of positive or negative textual comments.
Finally, Panel B of this analysis suggests that these results also obtain if we compare
across banks with different degrees of exposure to COVID-19. Overall, these
results provide direct evidence that high-IT banks are perceived by customers to
provide higher-quality service after the onset of the pandemic.
B. Customer IT Preferences and Banks’ IT Ability
Our article suggests that banks with better IT had a superior ability to operate
during the pandemic. Alternatively, another plausible interpretation is that high-IT
banks might potentially draw in an IT-savvy clientele even prior to the pandemic,
and such IT-savvy customers may shift their banking activity toward digital chan-
nels faster in response to the pandemic, irrespective of the bank’s IT capability. Both
mechanisms are interesting, but provide different interpretations for policymakers.
To assess this demand-centric view, we test whether clients who are more
technologically sophisticated tended to choose to work with banks with better IT
before the pandemic. We test this using Small Business Administration (SBA) 7a
loan data to identify pre-pandemic lending relationships between SME customers
and banks. We match SBA borrowers and lender banks to Aberdeen and compute
the same IT_INDEX for SMEs as before. We are able to obtain the IT information
for both the lenders and borrowers for 94,150 SBA borrowers between 2011 and
2019.25
The results are shown in Panel A of Table 9. We do not find a positive
relationship between customer IT and bank IT in any of our regressions. Hence,
our findings that high-IT banks can better provide credit to SMEs through PPP
loans are not likely explained by their clientele’s relative familiarity with digital
technology. This finding complements our previous finding that customers
“switched” toward high-IT banks. It implies that customers did not seek out banks
with good IT systems before the pandemic, but began to do so during the pandemic.
25
Comparing Aberdeen, which is based on Dun and Bradstreet Data, to the NETS data, we find that
Aberdeen provides broader coverage of larger establishments. Thus, our matched sample is more
concentrated on the relatively larger SMEs.
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2643
TABLE 9
Bank IT Versus Customer IT
In Table 9, Panel A shows the pre-pandemic relationship between bank and customer IT. The dependent variable is the
IT_INDEX of the SME customer who took SBA loans between 2011 and 2019, and the independent variable is the IT_INDEX of
the corresponding SBA lender. Loan controls include the inverse hyperbolic sine of loan size and the number of jobs the
borrower claims to support with the loan, loan term, interest rate, and a dummy for whether the borrower is a corporation. Panel
B examines whether bank IT explains the customers’ response to COVID-19 as evidenced by their propensity of remote work.
The dependent variable is the estimated fraction of firm employees working remotely that week. The sample period is from Jan.
to Apr. 2020. IT_INDEX_BANK and IT_INDEX_CUSTOMER are the IT_INDEX of the lending bank and the SME customer who
had a borrowing relationship with the bank before the pandemic. SEVERE_RESTRICTIONS is our dummy restriction measure
for the customer’s county. Heteroscedasticity-robust standard errors clustered by customer county-bank (Panel A) or
customer county (Panel B) are reported in parentheses. *, **, and *** denote statistical significance at the 10%, 5%, and
1% levels.
Panel A. Customer Versus Bank IT (Prepandemic)
IT_INDEX_CUSTOMER
1 2 3 4
IT_INDEX_BANK 0.0091 0.0021 0.0082 0.0014
(0.0071) (0.0071) (0.0066) (0.0069)
Loan controls No Yes No Yes
County FE No No Yes Yes
Industry FE Yes Yes Yes Yes
No. of obs. 94,038 94,038 93,631 93,631
R2 0.2430 0.2458 0.2708 0.2735
Panel B. Customer Firms’ Working from Home Behavior
CUSTOMER_REMOTE_WORK
1 2 3
SEVERE_RESTRICTIONS 0.0052***
(0.0019)
IT_INDEX_CUSTOMER × SEVERE_RESTRICTIONS 0.0090***
(0.0005)
IT_INDEX_BANK × SEVERE_RESTRICTIONS 0.0004
(0.0003)
Firm FE Yes Yes Yes
Week FE Yes No No
County-week FE No Yes Yes
No. of obs. 1,078,918 1,071,608 1,071,608
R2 0.5919 0.6040 0.6036
In addition, we show in Panel B of Table 9 that customers of ex ante high-IT
banks do not exhibit stronger responses to the pandemic in their internal operations.
Using a firm-level, big-data-based measure of remote work developed by Kwan and
Matthies (2022), we show that although a firm is more likely to adopt remote work
when COVID-19 is more severe in the local area, the firm responds to COVID-19
similarly, regardless of the lending banks’ IT_INDEX. This finding further suggests
the bank IT_INDEX is not picking up a bank’s tech-savvy clientele.
One potential concern is that we might not be capturing the impact of IT but
rather of a characteristic with which IT is highly correlated. Technological sophis-
tication may be only one component of a bank’s response to COVID-19, which
might also include other operational aspects, human resource practices, corporate
governance, or other features. But it is hard to imagine a characteristic that would be
so highly correlated with technological sophistication that it would render techno-
logical sophistication itself wholly irrelevant. In addition, as we showed before, it is
specifically the 14 remote technologies we study that show strong correlations to
bank outcomes during the pandemic. Such a characteristic that is not technology
https://doi.org/10.1017/S0022109023000662
2644 Journal of Financial and Quantitative Analysis
would also have to become important in a way that matches the timing of mobility
restrictions in a county in driving fewer branch visits and more web traffic. It would
also have to explain the bank’s ability to respond to COVID-19 on its website.
Given we are controlling for county and bank fixed effects, such a characteristic
would also have to explain the bank’s ability to lend in high-internet counties, and
be correlated with information and internet technology, but not be IT itself. While
we cannot rule out all omitted variables, it seems highly likely that IT is not merely a
proxy for some omitted variable.
VI. Conclusion
We find that banks’ IT capabilities affect their ability to serve customers during
the COVID-19 pandemic. Better bank IT is associated with both a larger shift from
offline to online as well as better ability to serve retail customers. In the areas harder
hit by the pandemic, customers are more likely to switch to banks with better
IT. While the COVID-19 outbreak represents an unprecedented shock for digital
banking services, our findings have broader implications beyond the pandemic as
they suggest a general negative relationship between technology and reliance on
physical branches.
It is not yet clear to what extent physical branch banking will return once
the threat of COVID-19 decreases. Given there was already a long-running trend
toward reduced reliance on branches and increasingly digital banking services, it
seems likely that investment in improving IT capabilities may help to both better
position banks for the future, as well as reduce their vulnerability to extreme shocks
such as the COVID-19 pandemic. The latter point might also have important policy
implications for the stability of the financial system more broadly. However, the
empirical relationship between IT and banking activities observed today suggests
that variation among banks remains substantial—and that there is less-than-
complete readiness in the U.S. banking sector.
Supplementary Material
To view supplementary material for this article, please visit http://doi.org/
10.1017/S0022109023000662.
References
Adalja, A. A.; E. Toner; and T. V. Inglesby. “Priorities for the US Health Community Responding to
COVID-19.” JAMA - Journal of the American Medical Association, 21202 (2020), 2019–2020.
Allcott, H.; L. Boxell; J. Conway; M. Gentzkow; M. Thaler; and D. Yang. “Polarization and Public
Health: Partisan Differences in Social Distancing During the Coronavirus Pandemic.” Journal of
Public Economics, 191 (2020), 104,254.
Alstadsæter, A.; J. B. Bjørkheim; W. Kopczuk; and A. Økland. “Norwegian and US Policies Alleviate
Business Vulnerability Due to the COVID-19 Shock Equally Well.” National Tax Journal, 73
(2020), 805–828.
Andersen, A. L.; E. T. Hansen; N. Johannesen; and A. Sheridan. “Consumer Responses to the COVID‐
19 Crisis: Evidence from Bank Account Transaction Data.” Scandinavian Journal of Economics,
124 (2022), 905–929.
https://doi.org/10.1017/S0022109023000662
Kwan, Lin, Pursiainen, and Tai 2645
Baker, S. R.; R. A. Farrokhnia; S. Meyer; M. Pagel; and C. Yannelis. “How Does Household Spending
Respond to an Epidemic? Consumption During the 2020 COVID-19 Pandemic.” Review of Asset
Pricing Studies, 10 (2020), 834–862.
Balyuk, T.; N. Prabhala; and M. Puri. “Indirect Costs of Government Aid and Intermediary Supply
Effects: Lessons from the Paycheck Protection Program.” NBER Working Paper No. 28114 (2020).
Bao, Z., and D. Huang. “Shadow Banking in a Crisis: Evidence from FinTech During COVID-19.”
Journal of Financial and Quantitative Analysis, 56 (2021), 2320–2355.
Bartik, A. W.; M. Bertrand; Z. Cullen; E. L. Glaeser; M. Luca; and C. Stanton. “The Impact of COVID-
19 on Small Business Outcomes and Expectations.” Proceedings of the National Academy of
Sciences, 117 (2020a), 17,656–17,666.
Bartik, A. W.; Z. B. Cullen; E. L. Glaeser; M. Luca; C. T. Stanton; and A. Sunderam. “The Targeting and
Impact of Paycheck Protection Program Loans to Small Businesses.” NBER Working Paper
No. 27623 (2020b).
Bartlett, R.; A. Morse; R. Stanton; and N. Wallace. “Consumer-Lending Discrimination in the FinTech
Era.” Journal of Financial Economics, 143 (2022), 30–56.
Basten, C., and S. Ongena. “The Geography of Mortgage Lending in Times of FinTech.” Working Paper,
University of Zurich (2020).
Beccalli, E. “Does IT Investment Improve Bank Performance? Evidence from Europe.” Journal of
Banking & Finance, 31 (2007), 2205–2230.
Bennedsen, M.; B. Larsen; I. Schmutte; and D. Scur. “Preserving Job Matches During the COVID-19
Pandemic: Firm-Level Evidence on the Role of Government Aid.” Covid Economics, 27 (2020),
1–32.
Berg, T.; V. Burg; A. Gombović; and M. Puri. “On the Rise of FinTechs: Credit Scoring Using Digital
Footprints.” Review of Financial Studies, 33 (2020), 2845–2897.
Berger, A. N. “The Economic Effects of Technological Progress: Evidence from the Banking Industry.”
Journal of Money, Credit and Banking, 35 (2003), 141–176.
Berger, A. N., and R. DeYoung. “Technological Progress and the Geographic Expansion of the Banking
Industry.” Journal of Money, Credit and Banking, 38 (2006), 1483–1513.
Bian, B.; J. L. Li; T. Xu; and N. Z. Foutz. “Individualism During Crises.” Review of Economics and
Statistics, 104 (2022), 368–385.
Bonfim, D.; G. Nogueira; and S. Ongena. ““Sorry, We’re Closed” Bank Branch Closures, Loan Pricing,
and Information Asymmetries.” Review of Finance, 25 (2021), 1211–1259.
Borusyak, K.; X. Jaravel; and J. Spiess. “Revisiting Event Study Designs: Robust and Efficient
Estimation.” Working Paper, Stanford University (2022).
Branzoli, N.; E. Rainone; and I. Supino. “The Role of Banks’ Technology Adoption in Credit Markets
During the Pandemic.” Working Paper, Bank of Italy (2023).
Buchak, G.; G. Matvos; T. Piskorski; and A. Seru. “Fintech, Regulatory Arbitrage, and the Rise of
Shadow Banks.” Journal of Financial Economics, 130 (2018), 453–483.
Charoenwong, B.; A. Kwan; and V. Pursiainen. “Social Connections with COVID-19-Affected Areas
Increase Compliance with Mobility Restrictions.” Science Advances, 6 (2020), 1–10.
Core, F., and F. De Marco. “Information Technology and Credit: Evidence from Public Guarantees.”
Management Science, forthcoming (2023).
D’Acunto, F.; N. Prabhala; and A. G. Rossi. “The Promises and Pitfalls of Robo-Advising.” Review of
Financial Studies, 32 (2019), 1983–2020.
D’Andrea, A., and N. Limodio. “High-Speed Internet, Financial Technology and Banking in Africa.”
Management Science, forthcoming (2023).
Danisewicz, P., and I. Elard. “The Real Effects of Financial Technology: Marketplace Lending and
Personal Bankruptcy.” Working Paper, Tilburg University (2022).
Degryse, H., and S. Ongena. “Distance, Lending Relationships, and Competition.” Journal of Finance,
60 (2005), 231–266.
Ding, W.; R. Levine; C. Lin; and W. Xie. “Social Distancing and Social Capital: Why US Counties
Respond Differently to COVID-19.” NBER Working Paper No. 27393 (2020).
Erel, I., and J. Liebersohn. “Does FinTech Substitute for Banks? Evidence from the Paycheck Protection
Program.” NBER Working Paper No. 27659 (2020).
Fu, J., and M. Mishra. “Fintech in the Time of COVID-19: Technological Adoption During Crises.”
Journal of Financial Intermediation, 50 (2022), 100,945.
Fuster, A.; P. Goldsmith-Pinkham; T. Ramadorai; and A. Walther. “Predictably Unequal? The Effects of
Machine Learning on Credit Markets.” Journal of Finance, 77 (2022), 5–47.
Fuster, A.; M. Plosser; P. Schnabl; and J. Vickery. “The Role of Technology in Mortgage Lending.”
Review of Financial Studies, 32 (2019), 1854–1899.
He, Z.; S. Jiang; D. Xu; and X. Yin. “Investing in Lending Technology: IT Spending in Banking.”
Working Paper, University of Chicago (2021).
https://doi.org/10.1017/S0022109023000662
2646 Journal of Financial and Quantitative Analysis
Herpfer, C.; A. Mjøs; and C. Schmidt. “The Causal Impact of Distance on Bank Lending.” Management
Science, 69 (2022), 1–18.
Howell, S. T.; T. Kuchler; D. Snitkof; J. Stroebel; and J. Wong. “Lender Automation and Racial
Disparities in Credit Access.” NBER Working Paper No. 29364 (2021).
Humphries, J. E.; C. Neilson; and G. Ulyessea. “The Evolving Impacts of COVID-19 on Small
Businesses Since the CARES Act.” Working Paper, Yale University (2020).
Iyer, R.; A. I. Khwaja; E. F. P. Luttmer; and K. Shue. “Screening Peers Softly: Inferring the Quality of
Small Borrowers.” Management Science, 62 (2016), 1554–1577.
Jay, J.; J. Bor; E. O. Nsoesie; S. K. Lipson; D. K. Jones; S. Galea; and J. Raifman. “Neighbourhood
Income and Physical Distancing During the COVID-19 Pandemic in the United States.” Nature
Human Behaviour, 4 (2020), 1294–1302.
Keil, J., and S. Ongena. “It’s The End of Bank Branching As We Know It (And We Feel Fine).” Technical
Report, Swiss Finance Institute Research Paper No. 20-83 (2020).
Koetter, M., and F. Noth. “IT Use, Productivity, and Market Power in Banking.” Journal of Financial
Stability, 9 (2013), 695–704.
Kreiss, K. “Bank Branches and COVID-19: Where are Banks Closing Branches During the Pandemic?”
FEDS Notes, Dec. 17, 2021 (2021).
Kwan, A., and B. Matthies. “Measuring Remote Knowledge Work Using Big Data.” Working Paper,
University of Hong Kong (2022).
Lewis, M.; Y. Liu; N. Goyal; M. Ghazvininejad; A. Mohamed; O. Levy; V. Stoyanov; and L.
Zettlemoyer. “Bart: Denoising Sequence-to-Sequence Pre-training for Natural Language Genera-
tion, Translation, and Comprehension.” Technical Report, arXiv preprint arXiv:1910.13461 (2019).
Li, L., and P. E. Strahan. “Who Supplies PPP Loans (and Does It Matter)? Banks, Relationships, and the
COVID Crisis.” Journal of Financial and Quantitative Analysis, 56 (2021), 2411–2438.
Lin, C.; C. Ma; Y. Sun; and Y. Xu. “The Telegraph and Modern Banking Development, 1881–1936.”
Journal of Financial Economics, 141 (2021), 730–749.
Nguyen, H.-L. Q. “Are Credit Markets Still Local? Evidence from Bank Branch Closings.” American
Economic Journal: Applied Economics, 11 (2019), 1–32.
Paaso, M.; V. Pursiainen; and S. Torstila. “Entrepreneur Debt Aversion and Financing Decisions:
Evidence from COVID-19 Support Programs.” Working Paper, University of St. Gallen (2020).
Petersen, M. A., and R. G. Rajan. “Does Distance Still Matter? The Information Revolution in Small
Business Lending.” Journal of Finance, 57 (2002), 2533–2570.
Pierri, N., and Y. Timmer. “The Importance of Technology in Banking During a Crisis.” Journal of
Monetary Economics, 128 (2022), 88–104.
Ru, H.; E. Yang; and K. Zou. “Combating the COVID-19 Pandemic: The Role of the SARS Imprint.”
Management Science, 67 (2021), 5606–5615.
Spiegel, M., and H. Tookes. “Business Restrictions and COVID-19 Fatalities.” Review of Financial
Studies, 34 (2021), 5266–5308.
Tang, H. “Peer-to-Peer Lenders Versus Banks: Substitutes or Complements?” Review of Financial
Studies, 32 (2019), 1900–1938.
The Economist. “How the Digital Surge Will Reshape Finance.” The Economist, Oct. 8, 2020 (2020).
Weill, J. A.; M. Stigler; O. Deschenes; and M. R. Springborn. “Social Distancing Responses to COVID-
19 Emergency Declarations Strongly Differentiated by Income.” Proceedings of the National
Academy of Sciences, 117 (2020), 19,658–19,660.
File and source
- File
- kwan-2023-stress-testing-banks-digital-capabilities-evidence-from-the-covid-19-pandemic.pdf
- Size
- 414,633 bytes
- SHA-256
- 0c27e844f0ecd0b192bf788a67c4a5d9af75c2365b5d7332fd59a3ba6a812b16
- Original
- www.cambridge.org