Fintech in the time of COVID−19: Technological adoption during crises

Abstract

We document the effects of the COVID−19 pandemic on digital finance and fintech adoption. Drawing on mobile application data from a globally representative sample, we find that the spread of COVID− 19 and related government lockdowns led to a sizeable increase in the rate of finance app downloads. We then analyze factors that may have driven this effect on the demand−side and better understand the “winners” from this digital acceleration on the supply−side. Our overall results suggest that traditional incumbents saw the largest growth in their digital offerings during the initial period, but that “BigTech” companies and newer fintech providers ultimately outperformed them over time. Finally, we drill−down further on the adoption of fintech apps pertaining to both the asset and liability side of the traditional bank balance sheet, to explore the implications that the accelerated trends in digitization may have for the future landscape of financial intermediation.

1. Introduction

Shocks of various kinds can drive technological adoption in unanticipated ways. Moreover, these shocks can also result in longer term changes to societies and economies. The COVID−19 pandemic started out as a shock to public health and healthcare systems. However, the natures of the pandemic coupled with the speed of transmission have led societies to adopt self−imposed behavioral changes (e.g., “social− distancing”) or government−imposed lockdown measures. Despite the high human and economic costs, this has resulted in some beneficiaries and possible silver linings. There is increasing anecdotal evidence that the technology sector and, in particular, companies enabling communication and exchange of goods and services over distance have seen notable increases in adoption and usage.1 Indeed, digital finance adoption is posited to have helped many households and firms mitigate some of the health risks and adverse socioeconomic effects of the pandemic. This accelerated and en masse adoption is likely to have important implications for the market equilibrium between traditional incumbents and newer tech−based players in financial intermediation. However, it is not immediately clear ex−ante, which intermediaries may have gained most traction during the pandemic period or the reasons why.

In this paper, we provide evidence on the acceleration of digital transformation in the financial sector by documenting the impact of COVID−19 on fintech adoption worldwide. We estimate that the pandemic's spread has led to between a 21% and 26% increase in the relative rate of daily downloads of finance−related mobile applications. This uptick is over and above existing growth trends, has largely persisted throughout our study period, and would roughly equate to an aggregate increase of over 900 million app downloads over the protracted COVID period, that would have likely not occurred in the pandemic's absence. Having documented a large main effect of COVID on fintech adoption, we turn to analyzing the provider and product types that saw greatest increases to their uptake. This allows us to examine how the pandemic affected the market structure and evolving landscape of financial intermediation, as well as provide some insights on possible drivers of adoption. In particular, the latter half of our analysis drills down on changes in lending and payment markets (simultaneously on the asset and liability side of the traditional bank balance sheet) that have been accelerated due to COVID, as this has bearing on changing market risks and appropriate policy and regulatory responses.

To study the aforementioned propositions, we draw on historical and real−time data on mobile application downloads, which provides a high dimensional measure that allows us to detect changes on the extensive margin within a small time frame. We extract aggregated country−level download estimates on finance category apps, as well as disaggregated application−level download estimates for a subset of top finance apps for the countries in our sample.2 We have these data for 71 countries for the Android app market and 56 countries for the iOS markets, covering the period from January 1st, 2019 to December 9th, 2020. We merge this with information on the applications’ product types and provider characteristics and use these data to explore differential trends. Our empirical methodology employs panel data regression models to estimate the change in fintech app adoption pre− and post−COVID−19. We control for country− or app−level characteristics and seasonality to account for spatial or temporal trends in the data. The sample covers all global regions, accounts for approximately 80 percent of the global population and over 90 percent of the world economy in terms of nominal GDP.

We use these compiled granular data on providers and products to examine differential app adoption trends and associated implications with respect to market structure. In particular, we posit that at the beginning of the pandemic and government lockdowns, customers may have had a preference for relying on traditional incumbent providers and general banking (i.e.,“bundled” products) apps owing to greater familiarity with them or because they were already on the intensive margin, but previously not using the app−based services. However, as time goes on, newer and more innovative fintech entrants may have a competitive advantage at developing and mobilizing new services and providing more niche “modularized” products at targeted market segments. Similarly, the BigTech companies may be able to utilize their network effects to lower switching costs significantly.

Our second set of findings largely supports these hypotheses. In terms of provider types, we find that traditional incumbents constitute the majority of top downloaded apps (over 50%) and that their apps exhibit a relative increase in adoption of about 7% on aggregate. However, both “BigTech” and fintech startups see a further relative increase of 9% over−and−above the growth rate of traditional incumbents, and this differential grows more pronounced over time. We further analyze characteristics of the top app providers to yield more insights on the supply−side factors associated with increased adoption. We consider a number of company−level characteristics that may be considered proxies for (perceived) trustworthiness versus innovation and analyze their relationship with adoption. For example, we find evidence that app adoption is positively correlated with providers that have greater longevity, are headquartered domestically, and that have more brand recognition (as proxied by number of trademarks)— i.e., indicators signaling reliability, stability, and familiarity. However, as time from COVID progresses, our main proxy for innovation—providers’ number of patents—becomes significantly more positively related to app adoption; conversely, the time elapsed since the providers’ founding date and geographical proximity become more negatively related. Finally, it is worth noting that we also observe that app adoption is independent of greater employment or revenue of companies—i.e., we rule out that larger company size alone is predictive of app adoption.

In terms of product type, we find that general banking apps (i.e., those with bundled services) comprise the largest share of top downloaded finance apps across our country sample (roughly 50%) and do exhibit notable growth during the COVID period. The remaining top finance apps are more thinly spread across pure−play payment, investment, lending, insurance, and government apps, but these unbundled products see even higher relative rates of COVID−induced growth. Pure−play lending apps see particularly pronounced growth in emerging markets and developing economies (EMDE countries) throughout the pandemic period but also in the advanced economies (AE countries) as time from the onset of COVID prolongs. This may speak to the utility of fintech lending as a tool for improving household and small businesses resilience to shocks. However, it may raise longer−term concerns over potential over−indebtedness issues, among other risks related to financial stability and consumer protection.

We interpret the combined evidence from the provider and product type analyses as suggesting that during prolonged shocks, innovative capabilities of providers—which are likely to be associated with more targeted products, niche market segments, and lower costs—can take precedence over their perceived trustworthiness in driving fintech adoption.

Finally, and in a similar vein, the rapid digitization stemming from COVID also has many interesting implications for the future landscape of financial intermediation. To explore this further, we take a holistic viewpoint by analyzing the adoption of fintech apps on both the asset and liability side of the traditional bank balance sheet and the subsequent implications COVID might have for the same. On the asset side, we drill−down on fintech lending apps, which we further classify into legitimate and suspect apps.3 We observe that legitimate apps see greater traction in advanced economies where they compete with traditional incumbents in securing new borrowers. Meanwhile, suspect apps see a significant uptick in developing economies, possibly driven by consumers facing extreme credit constraints and being locked out of the traditional borrowing market. While the former scenario could result in excess competition for borrowers and threaten stability (Chava et al. (2018); Cornaggia et al. (2018); Di Maggio & Yao (2021)), the latter scenario signals a need for more active regulatory intervention to deal with growing consumer protection risks in the app markets.

On the liability side, we see an increase over time in the adoption of (deposit−only) neo−bank apps, which may threaten the deposit volume of traditional incumbents and successively reduce the size of their balance sheet by forcing them to lower lending volumes. However, we note that in advanced economies, neo−banks affiliated with incumbents see relatively greater adoption than de novo unaffiliated neo−banks. This suggests that incumbents were somewhat successful at stemming the loss of their depositor base and retaining access to low cost deposits as a source of loan funding. We do see an increase in payment and e−wallet adoption in these economies. For developing economies, we observe a significant increase in standalone neo−banks. Many traditional banks in these economies have consumers who still prefer to transact physically. The increased adoption of neo−banks amidst this could be borne out of the desire to use online options for transaction absent similar options at traditional financial institutions.

While the scope of our paper does not permit us to go further into the end outcomes for consumer or providers in detail, we provide evidence regarding the possible direction this may take based on the empirical evidence of app adoption. Our results have important takeaways for policy−makers across different dimensions, namely with respect to the evolving market equilibrium in financial intermediation and associated risks, resilience of new fintech lenders firms over a full economic and credit cycle (given their perceived lower borrower quality) and reducing access to deposits for traditional incumbents.

Related Literature: Our paper contributes to two distinct streams of literature. First, it relates to the growing literature on fintech adoption.4 Fintech adoption studies thus far can be broadly grouped into those focusing on i) network effects, ii) individual−level determinants, and iii) country−level predictors. The first group include studies highlighting how network effects couple with idiosyncratic shocks on either the demand or supply side to drive immediate and longer−term effects on fintech adoption (Higgins (2019); Crouzet et al. (2019)). The second group of studies on individual−level determinants tend to emphasize the role of digital literacy and familiarity with new technologies, which tends to be highly related with demographics. For example, Carlin et al. (2017) exploit the introduction of a financial management app to analyze how access to financial information via new technology changes digital consumer credit usage and financial health. Their findings highlight a primary role of demographics, where gender and age drive differences in both likelihood of adoption of the various services, as well as economic outcomes. Cong et al. (2021), meanwhile, use survey data on SME owners to show how digital adoption in e−commerce and digital payment tools increases firm resilience during the pandemic. They similarly find that entrepreneurs’ digital literacy and time invested in learning new technologies play a key role in allowing or inhibiting adoption. Finally, the third group includes a number of academic and policy studies using country−level data (typically cross−sectional) to analyze predictors of fintech adoption (see for example Claessens et al. (2018); Frost et al. (2019); Frost (2020)).5 These studies generally highlight positive associations with digital infrastructure, enabling regulatory frameworks, higher financial literacy levels, and familiarity with new technology (the latter two related with demographics). The relationship between economic development level and adoption appears mixed, with studies tending to show a positive relationship for fintech credit but inverse relationship for other products.

By contrast, our paper differs in two important ways, leveraging the unique event and the global scale, frequency, and scope of our data. First, to date, there are no studies to our knowledge that look at how cross−border shocks—including global−level ones—occurring across countries may play a role in leading to convergence in fintech adoption rates or exacerbate existing differences. This is an important concern considering an existing “digital gap” between EMDE countries relative to AE countries (GSMA (2020)). Our evidence generally suggests that a “digital gap” in fintech was not exacerbated by the COVID pandemic, as we see overall signs that the uptick was more pronounced in the EMDE versus AE countries, in relative terms. Second, there are also a lack of fintech adoption studies, to our knowledge, that explore the influence or role of company−level supply side factors. By further compiling disaggregated information on the underlying individual apps and their providers, our data allows us to provide new insights on the company−level characteristics that help drive consumer and SME demand and assess the types of providers who were relative “winners” or “losers” during the crisis. On the one hand, a number of studies and commentators suggest that the decision of individuals and firms to adopt new products and services from a given financial institution (whether traditional or digital) are likely to center on their perceived trustworthiness (Guiso et al. (2004); Gennaioli et al. (2015); A. V. Thakor (2020)). On the other hand, many scholars in this area emphasize that providers leveraging more innovative and novel financial technologies should have competitive advantages in terms of lower costs, creating more targeted products and services, and reducing time to develop and scale new products (Philippon (2016); Das (2019); Frost et al. (2019)). Our findings contribute to this debate by suggesting that the primary drivers of adoption can evolve or change inter−temporally in the face of prolonged shocks. While we understandably do not have direct measures of trust, evidence from our proxy measures are suggestive that factors related to the “trustworthiness” of companies may have played a early role in influencing initial uptake. However, consumers then appear to increasingly put weight on the innovativeness of providers or their products to address their particular needs as the pandemic lingers.

Our findings also inform a second literature stream and ongoing policy debate concerning how the market equilibrium in financial intermediation is evolving between traditional incumbents, “BigTech”, and newer fintech entrants (BIS (2018); Das (2019); Frost et al. (2019); Goldstein et al. (2019); Feyen et al. (2021)). Understanding the respective growth trajectories of these provider types is important as it helps shed light on the changing market structure caused by technology, allows initial assessment of the economic effects of changes, and informs the balance of risks and benefits that financial regulators need to bear in mind when formulating policy. We relate our analysis and evidence to discussion raised in BIS (2018), which describes the interplay between these provider types and suggests several theoretical scenarios, each with respective risks and opportunities. We interpret the evidence from the COVID period as falling somewhere between a movement towards “better−bank” and “distributed bank” scenarios. On the one hand, we observe that some traditional incumbents have adequately digitised themselves or acquired partners (e.g., affiliated neobanks) to retain the customer relationship and core banking services during this period. On the other hand, we also see evidence that financial services became increasingly modularised, with “Bigtech” and a range of newer fintech entrants gaining notably in carving out more specific niche markets during the COVID period. As the “Bigtech” firms gained traction across a wider range of regions, this creates impetus for coordination among bank supervisors for dealing with cross−border fintech, apart from gearing regulation to meet the challenges of the altered financial intermediation landscape. Moreover, given the proliferation and rapid expansion shown by some niche fintech entrants on both the asset and liability side, supervisors will need to more closely and frequently vet and monitor new entrants to mitigate gaps in traditional supervisory and regulatory frameworks.

In Appendix Table A.2, we provide an overview of these two primary literature streams as well as studies from several secondary streams.

Table A.2.

Summary of related literature.

2. Data

In this section, we briefly describe the data sources used for the paper's analyses. We then highlight key trends in terms of fintech adoption, COVID spread and policy responses, and characteristics of the finance apps and companies in our data sample.

2.1. Data sources

Our analysis draws on mobile app download data from the AppTweak platform.6 The platform contains historical and real−time data on mobile app downloads at the aggregate, country, and individual app levels for all major app categories for 71 countries in the Android market and 56 countries in the iOS market. We extract daily information on all finance category mobile apps downloads from January 1st, 2019 to December 9th, 2020 for all countries available through the platform. The categorization of finance apps at the aggregate level are taken directly from the platform as listed by the mobile applications’ developers in the app stores. We further extract disaggregated data on top individual apps to understand which product and provider types drive results. (See Appendix Table A.1 for a list of our country sample and additional details.) These download levels are transformed into logarithmic form and per capita estimates to serve as the main dependent variables for our empirical analysis.

Table A.1.

Overview of countries in data sampleThis table lists the 71 countries for which we have mobile app download data along with their region, development level (advanced economy (AE) or emerging and developing economy (EMDE), date of first confirmed case of COVID−19, and dates of lockdowns (if applicable). Data sources: AppTweak, OxGCRT, and aura vision.

Country	Country code		Region	Development level	1st COVID−19 case	1st Lockdown start	1st Lockdown end	2nd Lockdown start	2nd Lockdown end
Algeria	DZA		Africa	EMDE	02−Mar−20	23−Mar−20	26−Apr−20
Argentina	ARG		South America	EMDE	06−Mar−20	19−Mar−20	07−Jun−20
Australia	AUS		Oceania	AE	26−Jan−20	23−Mar−20	15−May−20	08−Jul−20	22−Nov−20
Austria	AUT		Europe	AE	25−Feb−20	16−Mar−20	14−Apr−20	03−Nov−20	30−Nov−20
Belarus	BLR		Europe	EMDE	04−Mar−20	07−Apr−20	16−Dec−20
Belgium	BEL		Europe	AE	01−Mar−20	18−Mar−20	11−May−20	02−Nov−20	14−Dec−20
Brazil	BRA		South America	EMDE	29−Feb−20	17−Mar−20	10−Jun−20
Bulgaria	BGR		Europe	EMDE	08−Mar−20	13−Mar−20	13−May−20	28−Nov−20	21−Dec−20
Canada	CAN		North America	AE	28−Jan−20	13−Mar−20	04−May−20
Chile	CHL		South America	AE	23−Feb−20	19−Mar−20	17−Jun−20
China	CHN		Asia	EMDE	22−Jan−20	23−Jan−20	08−Apr−20
Colombia	COL		South America	EMDE	10−Mar−20	24−Mar−20	25−May−20
Croatia	HRV		Europe	AE	26−Feb−20	18−Mar−20	11−May−20
Czech Republic	CZE		Europe	AE	01−Mar−20	16−Mar−20	17−Apr−20	22−Oct−20	03−Nov−20
Denmark	DNK		Europe	AE	29−Feb−20	13−Mar−20	10−May−20
Ecuador	ECU		South America	EMDE	01−Mar−20	24−Mar−20	03−Jun−20
Egypt	EGY		Middle East	EMDE	01−Mar−20
Estonia	EST		Europe	AE	03−Mar−20	12−Mar−20	18−May−20
Finland	FIN		Europe	AE	26−Feb−20	27−Mar−20	13−Jul−20
France	FRA		Europe	AE	24−Jan−20	17−Mar−20	11−May−20	30−Oct−20	01−Dec−20
Germany	DEU		Europe	AE	28−Jan−20	17−Mar−20	30−Apr−20	02−Nov−20	01−Oct−20
Greece	GRC		Europe	AE	27−Feb−20	23−Mar−20	01−Jun−20	07−Nov−20	07−Jan−21
Hong Kong	HKG		Asia	AE	23−Jan−20
Hungary	HUN		Europe	AE	04−Mar−20	28−Mar−20	04−May−20
India	IND		Asia	EMDE	02−Feb−20	25−Mar−20	08−Jun−20
Indonesia	IDN		Oceania	EMDE	02−Mar−20	26−Mar−20	31−Jul−20
Ireland	IRL		Europe	AE	03−Mar−20	27−Mar−20	18−May−20	21−Oct−20	01−Dec−20
Israel	ISR		Middle East	AE	26−Feb−20	02−Apr−20	24−Apr−20	18−Sep−20	18−Oct−20
Italy		ITA	Europe	AE	31−Jan−20	09−Mar−20	04−May−20	06−Nov−20	03−Dec−20
Japan		JPN	Asia	AE	22−Jan−20	07−Apr−20	26−May−20
Jordan		JOR	Middle East	EMDE	15−Mar−20	18−Mar−20	21−Apr−20	10−Nov−20	15−Nov−20
Kazakhstan		KAZ	Asia	EMDE	13−Mar−20
Kuwait		KWT	Middle East	AE	25−Feb−20
Latvia		LAT	Europe	AE
Lebanon		LBN	Middle East	EMDE	26−Feb−20	22−Mar−20	18−May−20	14−Nov−20	28−Nov−20
Lithuania		LTU	Europe	AE	11−Mar−20	16−Mar−20	27−May−20	07−Nov−20	28−Nov−20
Malaysia		MYS	Asia	EMDE	25−Jan−20	18−Mar−20	10−Jun−20
Mexico		MEX	North America	EMDE	29−Feb−20	21−Mar−20	06−Jul−20
Netherlands		NLD	Europe	AE	29−Feb−20	16−Mar−20	01−Jun−20
New Zealand		NZL	Oceania	AE	04−Mar−20	26−Mar−20	21−May−20
Nigeria		NGA	Africa	EMDE	09−Mar−20	30−Mar−20	11−May−20
Norway		NOR	Europe	AE	28−Feb−20	12−Mar−20	27−Apr−20
Oman		OMN	Middle East	AE	24−Feb−20	10−Apr−20	10−Jun−20
Pakistan		PAK	Asia	EMDE	25−Feb−20	24−Mar−20	09−May−20
Peru		PER	South America	EMDE	08−Mar−20	16−Mar−20	29−Jun−20
Philippines		PHL	Asia	EMDE	02−Feb−20	15−Mar−20	31−May−20	04−Aug−20	18−Aug−20
Poland		POL	Europe	AE	06−Mar−20	13−Mar−20	04−May−20
Portugal		PRT	Europe	AE	02−Mar−20	19−Mar−20	04−May−20
Qatar		QAT	Middle East	AE	01−Mar−20	11−Mar−20	01−Jul−20
Romania		ROU	Europe	AE	28−Feb−20	25−Mar−20	12−May−20
Russia		RUS	Europe	EMDE	31−Jan−20	30−Mar−20	01−Jun−20
Saudi Arabia		SAU	Middle East	AE	05−Mar−20	03−Mar−20	05−Jun−20
Singapore		SGP	Asia	AE	24−Jan−20	07−Apr−20	02−Jun−20
Slovak Republic		SVK	Europe	AE	08−Mar−20	16−Mar−20	06−May−20
Slovenia		SVN	Europe	AE	05−Mar−20	14−Mar−20	04−May−20
South Africa		ZAF	Africa	EMDE	08−Mar−20	26−Mar−20	01−May−20
South Korea		KOR	Asia	AE	24−Jan−20
Spain		ESP	Europe	AE	09−Feb−20	14−Mar−20	11−May−20
Sweden		SWE	Europe	AE	26−Feb−20
Switzerland	CHE		Europe	AE	27−Feb−20	17−Mar−20	11−May−20
Taiwan	TWN		Asia	AE	24−Jan−20
Thailand	THA		Asia	EMDE	22−Jan−20	25−Mar−20	03−May−20
Tunisia	TUN		Middle East	EMDE	08−Mar−20	22−Mar−20	04−May−20
Turkey	TUR		Middle East	EMDE	13−Mar−20	11−Apr−20	09−May−20
Ukraine	UKR		Europe	EMDE	13−Mar−20	17−Mar−20	11−May−20
Utd. Arab Emirates	ARE		Middle East	AE	29−Jan−20	22−Mar−20	24−Apr−20
United Kingdom	GBR		Europe	AE	31−Jan−20	24−Mar−20	15−Jun−20	05−Nov−20	11−Dec−20
United States	USA		North America	AE	24−Jan−20	17−Mar−20	10−Jun−20	07−Dec−20	28−Dec−20
Uruguay	URY		South America	EMDE	13−Mar−20
Venezuela	VEN		South America	EMDE	14−Mar−20	17−Mar−20	12−Jul−20
Vietnam	VNM		Asia	EMDE	23−Jan−20

Our main explanatory variables are proxies for the spread of COVID−19 and related government policy. For the former, we draw on data from the Oxford COVID−19 Government Response Tracker (OxCGRT) (Hale et al. (2020)). OxCGRT provides daily updates on the number of confirmed cases and confirmed deaths, as well as various policies being implemented at the country−daily level—e.g., school closures, workplace closures, travel restrictions, etc. (These policies in combination are also used by OxCGRT to construct a policy “stringency index”, which we also use in some of our extended analyses.) For the latter, we draw on data from Aura Vision, which provides the start and end date of each country's lockdowns implemented due to COVID−19, if applicable.7

We also draw on a variety of additional data sources to analyze product−, provider−, and country−level determinants of differential effects. Namely, we extract and use app descriptions from Google Play and Apple's app stores to categorize a subset of the top 10 Android and top 10 iOS finance apps by country during the study period into product types, as later detailed in Section 3.4. In practice, this subset of top apps combine to account for roughly half of all finance app downloads in the sample countries and period in absolute terms. We further merge company−level data for the providers of these top apps using data from the Crunchbase platform.8 Finally, we also merge country−level data from the World Bank's World Development Indicator database and the 2017 Global Findex database for some of our exploratory analyses on predictors of differential post−COVID−19 trends in adoption. These latter data are used to group countries into advanced economies (AEs) and emerging and developing economies (EMDEs).

2.2. Descriptive statistics and trends

Fig. 1 presents the average number of daily downloads for finance−related apps across our full country sample. We note signs of uptrends starting from around March, which coincide with the period when the bulk of countries had experienced confirmed cases and the start of lockdowns (occurring outside of China). The increase seems primarily driven by the Android market with a muted effect in the iOS market. The figure suggests a heightened adoption of fintech apps and a greater push towards digitalization during the pandemic. This affirms the general notion that exogenous shocks to the economy or society end up accelerating trends which would have otherwise played out in a more protracted fashion.

Fig. 1.

The impact of COVID−19 on the adoption of fintech mobile apps. This figure depicts the daily number of downloads for finance category Android and iOS mobile applications across a globally representative sample. We use a 14−day moving average for downloads to smooth some day−to−day fluctuations. The sample data covers the period from January 1st, 2019 to December 9th, 2020.

Fig. 2 depicts a scatter plot of per capita daily downloads and COVID−19 deaths across countries. We remove the time dimension from our data by aggregating all the data from 2020 and taking country− level averages of daily downloads and deaths, in per capita terms. Subsequently, we also plot the best fit line and observe a general upward trend. The figure gives us a high−level understanding of how the spread of COVID−19 is associated with fintech adoption. Initial evidence, as depicted by Fig. 2 suggests there is a strong positive correlation between both.9 A cursory interpretation may be that countries above the best fit line show increased fintech adoption whereas those below the line show lower adoption owing to the effects of the pandemic.10 A myriad of reasons such as country−level demographics, product or company−level characteristics, and broader demand−side differences in trust in providers and digital technology could be instrumental in driving these effects. Furthermore, it is plausible that key drivers of adoption may have evolved over time as the pandemic progressed. We thus turn to exploring some of the demand− and supply−side factors that may explain these differences later in the paper.

Fig. 2.

Scatterplot of Finance Category Mobile App Downloads per capita vs. Confirmed COVID−19 Deaths per capita. This figure depicts the relationship between average daily number of downloads per capita for finance category mobile applications and average daily confirmed new deaths of COVID−19 per capita. We roughly restrict to the period after the first signs of the COVID−19 outbreak (i.e., since January 1st, 2020). The underlying data includes a globally representative sample of mobile apps from both the Android and iOS platforms. Country codes are listed in appendix Table A.1. Data sources: AppTweak and OxCGRT.

Further descriptive statistics at the aggregate−, country−, product−, and provider−level are provided in Table 1 . Across our 71 country sample and including both the iOS and Android platforms, the mean daily finance app downloads during our study period is around 12.9 million. The most downloaded product types are general banking apps (i.e., platforms offering access to multiple products, typically including savings or current accounts, along with other services), which comprise just under half of the top downloaded apps. They are followed by payment, investment, and loan apps, which make up roughly 30%, 7%, and 6%, respectively. Finally, the remaining share are a mix of insurance, government, and other miscellaneous apps. At the provider−level, traditional incumbents, “Bigtech” companies, fintech incumbents, and fintech startups comprise around 50%, 10%, 22% and 17% of the top downloaded finance apps, respectively. In practice the majority of the providers are traditional commercial banks with mobile app offerings, but we observe a growing set of de novo neobanks, as well as some spin−off neobanks that appear to be affiliated with the traditional commercial banks. (See Sections 3.3 and 3.5 for details on the categorization of provider types used in this paper.) It is worth noting that while the “Bigtech” players thus appear to comprise a smaller share in terms of number of apps, their apps often have greater scale in terms of absolute numbers of users and in terms of being a top app across multiple country or regional markets. This is in contrast with apps for traditional incumbents, which are more often limited to a single or a regional market. Additional summary statistics on the characteristics of providers of the top apps are in line with this broad categorization. For example, we see that the average years since founding is quite high, along with a correspondingly high number of trademarks, patents, active products, media articles, etc. These latter averages sometimes hide notable variation or skewed distributions for some of these company characteristics, however. For example, the traditional incumbents intuitively tend to have longer lifespans, more trademarks, and tend to be from the same region. Bigtech companies meanwhile are of moderate age and tend to have a particularly high number of patents, media articles, and are more likely to be foreign. Finally, fintech startups and neobanks tend to be quite young but also have a disporportionate number of patents and apps. Thus, we conduct log transformation where relevant for use in our subsequent regression models.

Table 1.

Summary statistics. This table provides summary statistics on aggregated and daily finance mobile app downloads, as well as COVID−19 cases, deaths, and government lockdowns. We also provide summary statistics on the sample countries, products, and provider characteristics. The data sample covers the android and iOS mobile finance app markets for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. Figures denoted by a † are based on statistics as of December 9th, 2020. Panels C, D, E, and F are calculated based on a subset of the top 10 finance apps by sample country and app store, for which we extract and merge relevant product and provider information the study period. Data sources: AppTweak, Crunchbase, OxCGRT, Kaggle, WDI, global findex.

Panel A. “Global” finance mobile app market and COVID−19 statistics at a glance			#
Total # of finance category mobile app downloads across sample countries (in 1,000′s)			8072,469
Mean # of daily app downloads (in 1,000′s)			12,897
Total # of COVID−19 cases across sample countries (in 1,000′s)†			61,900
Total # of COVID−19 deaths across sample countries (in 1,000′s)†			1411
Total # of sample countries with confirmed cases†			70
Total # of sample countries with lockdowns†			61
Total population across sample countries (in million's), 2019 estimates			6061
Panel B. Country−level statistics	Obs.	Mean	Std. Dev.	Min	Max
# of daily downloads (in 1,000′s)	50,268	181.64	472.6	0	5297.8
# of daily downloads per capita	50,268	0.00220	0.00202	0	0.01804
# of daily downloads per capita (2019)	26,280	0.00195	0.00222	0	0.01804
# of COVID−19 cases (in 1,000′s)†	71	871.6	2239.2	0	15,200.0
# of COVID−19 deaths (in 1,000′s)†	71	19.9	44.9	0	286.2
OxCGRT government stringency index (0 to 100)	50,268	24.2	32.1	0	100
# of days since 1st COVID−19 case†	70	290.6	17.6	268	321
# of days in lockdown †	61	72.9	36.8	28	246
Panel C. Product type, top 10 finance apps by country & device	Obs.	%
General banking app	602	48.86
Payment app	374	30.36
Investment app	91	7.39
Lending app	72	5.84
− Lending, real	61	4.95
− Lending, suspect	11	0.89
Government app	17	1.38
Insurance app	9	0.74
Miscellaneous app	67	5.44
Total	1232	100.00
Panel D.1 Provider type, top 10 finance apps by country & device	Obs.	%
Traditional incumbents	617	49.72
“Bigtech”	128	10.31
Fintech incumbents	279	22.48
Fintech startups	217	17.49
Total	1241	100.00
Panel D.2 Provider type, top 10 finance apps by country & device	Obs.	%
Commercial banks	612	60.24
Neobanks (affiliated w/ incumbents)	15	1.48
Neobanks	79	7.78
Payments / e−wallets	310	30.51
Total	1016	100.00
Panel E. Provider characteristics, top finance apps	Obs.	Mean	Std. Dev.	Min	Max
Years since founding	1106	55.71	59.10	.94	330.28
# of registered trademarks	554	125.61	290.74	0	1312
Same country	1241	.58	.49	0	1
Same region	1241	.71	.45	0	1
# of registered patents	554	1642.52	5723.01	0	27,097
# of apps	703	15.72	16.91	1	109
# of active products	722	29.20	37.91	0	207
IPO status, public	1118	.38	.49	0	1
# of media articles	978	2362.54	6467.00	1	35,479

3. Results

3.1. What is the estimated impact of the spread of COVID−19 on fintech adoption?

To provide a concise way of estimating the effect of COVID−19 on fintech adoption during our study period, we start with a general empirical specification as depicted in Eq. 1:

$$y_{it}={\beta }_0+{\beta }_{1}COVID{19}_{it}+Pos{t}_t+{\theta }_i+{\gamma }_m+u_{it}$$ (1)

where the dependent variable y is either 1) the relative (logarithmic) change in daily downloads or 2) the absolute number of daily downloads in per capita terms for country i and at time (day) t. COV ID19 denotes a dummy that is set to one at time t, either after country i had its first confirmed cases (Post−Confirmed case) or during its lockdown period(s) (In Lockdown), respectively. Post is a general dummy for the period after the first COVID case was detected in our entire sample and is used to capture any general upward trend that may have been occurring in absence of COVID. Specifically, it is set to one for the period after Jan 23, 2020 and zero from January 1st 2019 to January 22nd, 2020. We further include location−level dummies (θ_i) to control for unit fixed effects, and month−of−the−year dummies (denoted by γ_m) to control for aggregate seasonal trends in the finance app market. We alternatively interact the location and month dummies in our preferred specification to control for more disaggregated seasonality in downloads (e.g., country−level seasonality), which would substitute (θ_i × γ_m) in lieu of the location and month−of−the−year dummies.

Table 2 presents results for this model estimating the impact of COVID−19 on fintech adoption at the country−level using the Post−Confirmed case variable in Panel A and In Lockdown variable in Panel B.11 The results across both table panels are fairly consistent and provide further supporting evidence that the COVID−19 outbreak led to an increase in finance app adoption, even after taking into account any general time trend and controlling for country fixed characteristics and aggregate− or country−level seasonality in app downloads. In relative terms, we estimate an increase in downloads of around 20.6 percent beyond what would have been an expected trend since countries’ first confirmed COVID−19 case. While not reported in the table, it is worth noting that the general Post dummy depicts an overall upward trend of between 4% and 8% in relative terms, depending on the specification. Thus, the relative increase brought on by the spread of COVID represents a fairly dramatic increase over and above existing trends. During the more narrow period of lockdowns, there was a further 6.7% increase. These figures are based on our preferred specification in column 3, which controls for country−level seasonality. It is worth noting that the effect captured by the In Lockdown variable is likely reduced relative to the Post−confirmed case variable as the former is a more precise and narrowly defined time dummy that turns on and then off again when a country enters and exits its lockdown(s). As such, the general “Post” dummy is seen to absorb more of the effect (including the effect that was being attributed to the country−level impact of COVID cases). We interpret the In Lockdown variable as capturing any remaining impact of COVID on fintech adoption over and above the ongoing uptick driven by COVID cases.

Table 2.

Effect of COVID−19 on finance mobile app adoption. This table presents coefficient estimates for panel regression models estimating the country−level relationship between the spread of COVID−19 on the relative (logarithmic) and per capita changes in daily downloads for finance category mobile apps. Post−confirmed case denotes a dummy indicator that turns on after a given country saw its first confirmed COVID−19 case. In lockdown denotes a dummy indicator that turns on and off during each countries’ lockdown period(s). All models include a general post dummy indicator that turns on after the first confirmed COVID−19 case worldwide. Depending on the model specification, we also include country fixed effects, month−of−the−year (MoY) dummies to control for aggregate seasonality in finance app downloads, and country MoY interaction dummies to control for country−level seasonality. The data sample covers the android and iOS mobile finance app markets for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. Standard errors are clustered at the country level and in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, OxCGRT, and aura vision.

We alternatively calculate and estimate our results in terms of the daily rate of downloads in per capita terms. These specifications also confirm a significant absolute increase, after taking into consideration the countries’ market size. As depicted in Table 1, there was a daily average of roughly 1.95 finance app downloads for every 1000 individuals across the 71 countries in our dataset in the period roughly prior to the advance of COVID−19 (i.e., in 2019). For the roughly 6.06 billion individuals that would comprise our sample countries, this translates into an average of just over 11.8 million finance category app downloads daily from the Android and iOS platforms in the pre−COVID−19 period. In Table 2, we observe that following the onset of confirmed COVID−19 cases or during the period of related lockdowns, respectively, the daily rate of downloads per capita (10^{− 3}) increased by between 0.53 and 0.18, respectively—i.e., from the baseline of 1.95 to between 2.48 and 2.13 apps for every 1000 individuals. (See Column 6, which again reflects our preferred and most rigorous specification.) This reflects per capita download increases of around 27 percent for the Post−Confirmed case models and 9 percent for the In Lockdown models, which fall roughly in line with our prior relative estimates.

If we aggregate this effect across the 71 countries in our full sample, of which 70 have had a confirmed COVID−19 case and 61 have had a lockdown period (see Table 1 or Appendix Table A.1), this translates to an increase of approximately 3.2 million finance app downloads per day “globally” driven by COVID−19 cases and a further 1.1 million daily within the lockdown periods, respectively. Furthermore, if we take into account the average number of days since confirmed COVID−19 cases or average days in lockdown in our sample (see Table 1), this equates to an aggregate increase of around 935 million additional finance app downloads since the onset of COVID cases and a further 81 million additional finance app downloads during the lockdown periods, relative to prior trends.

3.2. Does the effect of COVID−19 on app adoption persist across time?

We next explore time varying dimensions of the effect of COVID−19 on finance app adoption. We draw on the distinct waves of policy stringency across countries to examine how the end of the lockdowns or additional lockdown waves may have affected app adoption. This allows us to shed light on whether the additional increase in digitization of financial intermediation driven by government lockdowns was a temporary phenomenon.

First, we examine whether the end of the first lockdown periods coincided with any break in finance app adoption trends. To do so, we restrict our time series to the period after the lockdown start dates of all countries in our sample. We create an alternate dummy indicator, End lockdowns, that turns on after the end date of the respective countries’ first lockdowns and substitute this as our main explanatory variable in our baseline specification (Eq. 1). Any countries with a second lockdown wave are excluded from this particular analysis to avoid confounding effects. Table 3 Panel A summarizes the results and show that the end of the first wave of lockdowns does not appear to lead to a significant break in the trends in app adoption that were ongoing during the lockdown. There are some minor signs of decreases in our preferred specifications (in columns 3 and 6), however, they are insignificant in both statistical and economical terms—e.g., amounting to barely a l% relative decrease.

Table 3.

Change in finance mobile app adoption across lockdown waves. This table presents coefficient estimates for panel regression models estimating the country−level relationship between the end of government lockdowns or sequential lockdown waves and the relative (logarithmic) and per capita changes in daily downloads for finance category mobile apps. In panel A, we restrict our analysis to the period after the start date of our sample countries’ first lockdowns. Countries which have had a second lockdown period as of December 9th, 2020 are excluded from this analysis to avoid confounding effects. End of first lockdown denotes a dummy indicator that turns on after the end date of each country's first lockdown period, where applicable. In Panel B, we run the analysis using the full time period from January 1st, 2019 to December 9th, 2020. First wave of lockdowns and Second wave of lockdowns denote dummy indicators that turn on and off during each countries’ first and second lockdown periods, respectively, and where applicable. Depending on the model specification, we include country fixed effects, month−of−the−year (MoY) dummies to control for aggregate level seasonality in finance app downloads, and country MoY interactions to control for country−level seasonality. Details on country lockdowns can be found in appendix Table A.1. Standard errors are clustered at the country−level and in parentheses *** p<0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, OxCGRT, and aura vision.

Second, as countries have differed in terms of their lockdown requirements, we examine some of the dynamics by creating separate dummy indicators for the First wave of lockdowns and Second wave of lockdowns. Our a priori expectation is that the first wave of lockdowns produced a greater shock and drove a greater increase in adoption than the second wave of lockdowns. The results, as depicted in Table 3 Panel B, seem to support this general intuition. We can observe that the economic magnitude of effects for the first wave of lockdowns generally are substantially larger than in the second wave of lockdowns. (This is particularly the case in our preferred specifications in columns 3 and 6, which include the most rigorous controls.) Nevertheless, we can still observe a significant increase in adoption during the second lockdown periods that is of significance both statistically and economically. In other words, the adoption does not appear to be restricted to a one−off event, but shows some responsiveness to successive shocks. Along with the descriptive evidence from Fig. 1, this provides preliminary evidence that finance app adoption levels may remain higher than pre−COVID levels.12

3.3. What provider and product types saw greatest changes in adoption?

Having documented a large main effect of COVID on fintech adoption, we turn to analyzing the provider and product types that saw greatest increases to their uptake. This allows us to examine how the pandemic affected the market structure and evolving landscape of financial intermediation, and provide some insights on possible drivers of adoption. It is not immediately clear ex−ante, which may have gained most traction during the pandemic period. Moreover, given the protracted nature of the pandemic, it is plausible that the drivers of adoption and providers and products seeing greatest growth evolved over time. In particular, one may posit that at the beginning of the pandemic and government lockdowns, customers may have had a preference for relying on traditional incumbent providers and general banking (i.e.,“bundled” products) apps owing to greater familiarity with them or because they were already on the intensive margin (but previously not using the app−based services). However, as time goes on, newer and more innovative fintech entrants may be faster at developing and mobilizing new services, providing more niche “modularized” products at targeted market segments. Similarly, the BigTech companies may have been able to utilize their network effects to lower switching costs significantly.

To provide evidence on this, we build on the baseline model by further disaggregating the data on individual apps (denoted by p) and interacting the COV ID19 dummies with variables capturing provider and product categories or characteristics to formally test whether they predict differential trends. Specifically, we extract daily download data for the top 10 finance applications in the Android and iOS markets for each country in our sample during the period from January 1st, 2019 to December 9th, 2020. We identify the apps’ product type and company and merge data from the Crunchbase database, which contains dates of company founding, details on acquisitions and funding, in addition to a wide range of other company characteristics. A generalized model is depicted in Eq. 2:

$$y_{pit}={\beta }_0+\sum _{n-1}^b{\beta }_{1p}COVID{19}_{pit}\times Moderator+Pos{t}_t+{\sigma }_{ip}\times {\gamma }_m+u_{pit}$$ (2)

Where the level of analysis is thus moved to the application−country−day level (pit), Moderator denotes further provider− or product−level categorization or characteristics used to test for differential results, and σ_ip denotes country−application fixed effects. We test for intertemporal changes by 1) substituting in a main explanatory variable which captures the log of days since the first COVID case within a country or 2) by adding in separate dummies for the 1st and 2nd wave of lockdowns. All other variables are same as previously described in Eq. 1. Note that we ultimately decide to use country−application × month−of−the−year interactions as we find they most rigorously control for other disaggregated seasonality trends. We run the analysis for the full country sample and also subset to advanced economies (AEs) and emerging and developing economies (EMDEs), separately, to see if different patterns emerge by the countries’ level of economic development.

3.3.1. What provider types have been “winners” from COVID−19?

We start by examining a high−level question raised by scholars and policymakers—both prior to and since COVID—concerning if and in what ways the market structure has evolved between traditional incumbents and newer fintech and BigTech companies. To do so, we combine provider lists and typologies from Gomber et al. (2017); BIS (2018); Frost et al. (2019) to roughly categorize the app providers in our sample into four categories:

• 1Traditional incumbents: founding date prior to 2000;
• 2“BigTech” companies: Alibaba/Ant Financial Group, Baidu, Facebook, Google, Mercado Libre, Rakuten, PayPal, Samsung, Square, Tencent;13
• 3Fintech incumbents: non−BigTech and founding date between 2000 and 2015;
• 4Fintech startups: non−BigTech and founding date from after 2015 to present−day.

Table 4 provides results from specifications that adapt Eq. 2 to include interaction terms for these provider categories. We observe that there is a significant main effect following the onset of COVID cases for traditional incumbents (the baseline level), who exhibit a 7% relative increase in app adoption across the full sample. The traditional incumbents in the EMDE sample are seen to exhibit a larger relative increase than the AE sample (12% as opposed to 4%). However, “BigTech” companies and fintech startups both saw relative increases in their finance app adoption that was a further 8% and 9% greater than the traditional incumbents (i.e., combined increases of around 14% and 16%), respectively. The growth in adoption for these latter two provider types was again larger in the EMDE compared to AE countries. A possible explanation for why growth is generally higher in the EMDE countries may be linked to lower ex−ante levels of financial inclusion and financial health. In such settings, there is more space for all provider types to increase clientele without having to compete and potentially more impetus for households and firms from the pandemic shock to adopt novel products given their greater vulnerability. Finally, fintech incumbents do not show significant signs of differential trends–that is, they also exhibited growth due to COVID but at a rate that is comparable to that of the traditional incumbents. For the models using the In Lockdown dummy, we do not see any statistically significant changes in the results for the traditional incumbents or fintechs, suggesting that the growth in their demand actually starts to occur earlier in the pandemic before lockdowns became commonplace. Meanwhile, we do see suggestions that finance apps from “Bigtech” apps exhibit a further uptick during the lockdown period on aggregate.

Table 4.

Effect of COVID−19 on finance app adoption by provider type. This table presents coefficient estimates for a panel regression model estimating the relationship between the spread of COVID− 19 on relative (logarithmic) change in daily app downloads for finance category mobile apps. Post−confirmed case is a dummy indicator that turns on after a given location saw its first confirmed COVID−19 case. In lockdown denotes a dummy indicator that turns on and off during each countries’ lockdown period(s). All models also include a general post dummy indicator that turns on after the first confirmed COVID−19 case worldwide. We include interaction terms by provider category to test for differential effects. The base level is traditional incumbents. We run separate specifications including all countries in the sample (All), subset to advanced economies (AE) and subset to emerging and developing economies (EMDE). Appendix Table A.1 provides the country categorization. The data sample covers top android and iOS mobile finance apps for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. Standard errors are clustered at the country−application level and in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, Crunchbase, OxCGRT, and aura vision.

	DV = Relative (ln) daily app downloads			DV = Relative (ln) daily app downloads
Base level = Traditional incumbents	(All)	(EMDE)	(AE)	(All)	(EMDE)	(AE)
Post−Confirmed case = 1	0.068∗∗∗ (0.014)	0.119∗∗∗ (0.027)	0.037∗∗∗ (0.013)
Post−Confirmed case = 1 × Fintech incumbents	0.024 (0.025)	0.024 (0.048)	0.037 (0.028)
Post−Confirmed case = 1 × Bigtech−	0.076∗∗ (0.036)	0.192∗ (0.107)	0.066∗∗ (0.027)
Post−Confirmed case = 1 × Fintech startups	0.093∗∗∗ (0.033)	0.134∗∗ (0.054)	0.024 (0.036)
In Lockdown = 1				−0.016	−0.033	−0.010
				(0.015)	(0.028)	(0.016)
In Lockdown = 1 × Fintech incumbents				−0.001 (0.027)	−0.060 (0.054)	0.043 (0.030)
In Lockdown = 1 × Bigtech				0.092∗∗ (0.039)	0.217∗ (0.112)	0.067∗∗ (0.027)
In Lockdown = 1 × Fintech startups				0.049 (0.038)	0.030 (0.062)	0.029 (0.043)
Observations	881,942	343,660	538,282	881,942	343,660	538,282
R2 Additional controls:	0.907	0.903	0.908	0.906	0.902	0.908
Country−application × MoY interactions	Yes	Yes	Yes	Yes	Yes	Yes

These findings contrast, however, with early results from the same analysis capturing the start of the pandemic, which indicated different trends for certain provider types. Appendix Table A.4 reports on results from similar specifications based on earlier data (covering up until August 9th, 2020). We observe that the main effect is instead largest for traditional incumbents (the baseline level), who exhibit between a 13% to 23% relative increase in app adoption across the aggregated sample. “BigTech” and the fintech providers are also shown to exhibit COVID−induced increases in adoption, but their economic magnitude is considerably smaller. In other words, in the opening stages of the pandemic, it appeared that top performing traditional incumbents, rather than “Bigtech” or smaller fintechs, disproportionately benefited from COVID−19, at least in terms of gaining increased consumer uptake of their digital offerings.

Consequently, we combine the disaggregated provider−level analysis with exploration on time−varying dimensions of COVID−19 to formally test whether and to what degree adoption may have evolved over time across the different provider types. We first substitute in a continuous time variable which captures the log of days since the first COVID case within a country as the main independent variable and then further interact this with the provider categories. This captures whether the rates of adoptions for any of these provider categories changes as a function of time from COVID exposure within a country and, if so, whether this rate change is positive or negative. Secondly, we again construct and use independent dummies to separate the effects of 1st and 2nd lockdown waves, for countries where this is applicable.

Appendix Table A.5 depicts results and generally support our earlier stated hypotheses. In the first set of specifications, which draw on time from 1st COVID cases within country, we observe that all provider types saw a relative increase in adoption as time from COVID progressed. However, this time−varying increase was notably larger for the Bigtech companies–particularly in the AE countries–and the fintech startups—particularly in the EMDE countries, over and above the base level. Moreover, in the second set of specifications, we observe that the traditional incumbents generally see a drop off in adoption during the second wave of lockdowns. This is in line with intensive margin clients, most likely, fulfilling their immediate needs for general banking services or transferring over to the digital offerings of incumbents in the early pandemic period. Meanwhile, BigTech finance apps saw some uptick in the first lockdown but it grew even larger by the second wave. By contrast, the fintech startups seem to have had little initial traction in the first wave but then particularly high growth in the second wave of lockdowns. This may suggest that the traditional incumbents and BigTech may have benefited relative to fintech startups from greater familiarity and brand recognition earlier on but that the protracted economic shocks may have then led more agile and targeted fintech startups to meet consumer's needs with more innovative and targeted products.

3.3.2. What provider characteristics drive adoption?

To build on the preceding analysis, we further examine more granular provider characteristics to better understand what may have contributed to the success of particular providers during the pandemic period. Based on the literature, we identify variables from the Crunchbase data that can be seen as general proxies related to trustworthiness vs. innovativeness (and by association, expected lower cost). For proxies that are related to trust, we use the 1) Number of years since a company was established as a measure of stability, whether the 2) Provider is headquartered in the same country (as opposed to foreign entrants) as measures of geographic/cultural proximity, and 3) Number of registered trademarks and 4) Number of media articles as measures of reliability and familiarity, respectively. For proxies related to innovation and efficiency, we use their 5) Number of patents, 6) Number of mobile applications, and 7) Number of active products. We again adapt Eq. 2 including interaction terms with company characteristics to empirically test for their influence on differential trends.

As mentioned before, existing literature points out that the decision of individuals and firms to adopt the digital products and services of a given financial institution are likely to ultimately center on the perceived trustworthiness versus relative cost of providers (A. V. Thakor (2020)). On the issue of trust, Goldstein et al. (2019) argue that part of the consumer movement that enabled the earlier growth in fintech was a reaction to Global Financial Crisis and loss of trust in the mainstream financial sector and traditional incumbents. On the issue of relative cost, there is general consensus that traditional providers have higher operating costs than fintechs (Philippon (2016); Das (2019); A. V. Thakor (2020)). Whereas the former still heavily rely on brick and mortar networks, have higher staffing, and greater regulatory burden, the latter by definition are expected to leverage more innovative technologies that should reduce both the suppliers’ operational costs and costs for consumers (Das (2019)).

Table 5 summarizes initial results from this analysis. We focus on results in relative terms and based on the Post−Confirmed case dummy for brevity. We observe that our proxy for stability (Number of years since a company was established) has a negative differential trend on app adoption during the time of COVID. In a similar vein, we also see signs of decreased relative adoption rates for more local rather than foreign providers. This provides a counter−example to the general reasoning from Degryse & Ongena (2005) and Fisman et al. (2017) that geographic and cultural proximity eases frictions in establishing customer−provider relationships, and shows that this can be overcome in the context of digital providers. We posit that this latter result is particularly driven by the “Bigtech” providers, whose network size and brand recognition allow them to overcome barriers typically faced by foreign entrants. By contrast, we see a strong positive relationship between number of patents, our main proxy for provider innovation, and app adoption. Meanwhile, we do see some suggestion that companies with a larger Number of registered trademarks, our proxy for visibility and name recognition, had a positive uptake, but this relationship does not appear particularly significant. Greater familiarity alone does not automatically lead to adoption, however, as we see signs of a slight negative relationship between Number of media articles on providers and adoption of their apps. A possible explanation for this may have to do with a recent bias towards negative coverage of the financial sector in line with D'Acunto (2017), who shows how greater exposure to anti−market rhetoric can lower individual's use of financial institutions. Finally, it is worth briefly noting that additional results suggest that companies did not see greater adoption just by sake of having larger ex−ante size alone. For example, results from additional models in Appendix Table A.7 show that app adoption does not seem to be particularly related to companies with greater employment and even appears somewhat inversely related to companies’ total annual estimated revenue size.

Table 5.

Effect of COVID−19 on finance mobile app adoption by company characteristics. This table presents coefficient estimates for a panel regression model estimating the relationship between the spread of COVID−19 on the relative (logarithmic) change in daily app downloads for finance category mobile apps. Post−confirmed case is a dummy indicator that turns on after a given location saw its first confirmed COVID−19 case. The base level varies based on model specification and is labelled in the top row. All models include a general post dummy indicator that turns on after the first confirmed COVID−19 case worldwide. The data sample covers top android and iOS mobile finance apps for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. Standard errors are clustered at the country−application level and in parentheses. *** p<0.01, ** p<0.05, * p<0.10. Data sources: AppTweak, Crunchbase, OxCGRT, and aura vision.

DV = Relative (ln) # of daily app downloads
Base levels (for categorical interaction variables)=	(1)	(2) Outside region	(3) Outside country	(4)	(5)	(6)	(7)	(8)
Post−Confirmed case  = 1	0.157∗∗∗	0.142∗∗∗	0.106∗∗∗	0.070∗∗	0.112∗∗∗	0.067∗∗∗	0.062∗∗	0.096
Post−Confirmed case = 1 × Log (Years since founded)	(0.036) −0.020∗∗ (0.010)	(0.020)	(0.015)	(0.033)	(0.020)	(0.018)	(0.029)	(0.063)
Post−Confirmed case = 1 × Provider from same region		−0.062∗∗ (0.025)
Post−Confirmed case = 1 × Provider from same country			−0.014 (0.021)
Post−Confirmed case = 1 × Log (Number of trademarks)				0.007 (0.009)
Post−Confirmed case = 1 × Log (Number of articles)					−0.006 (0.004)
Post−Confirmed case = 1 × Log (Number of patents)						0.011∗∗ (0.005)
Post−Confirmed case = 1 × Log (Number of apps)							0.019 (0.013)
Post−Confirmed case = 1 × Log (Number of products active)								−0.007 (0.021)
Observations	782,430	881,942	881,942	392,912	690,423	392,912	501,098	511,664
R2	0.916	0.907	0.907	0.921	0.918	0.921	0.916	0.917
Additional controls: Country−application × Month−of−the−Year interactions	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes

We gain further insights on how drivers of adoption evolved by combining this disaggregated analysis of provider characteristics with exploration on time−varying dimensions of COVID−19, as depicted in Appendix Table A.6. We again draw on a continuous time variable capturing the log days since the first COVID case within a country as the main dependent variable and interact this with the same provider characteristic variables. In these specifications, we substitute in country and month−of−the−year interactions (in lieu of country−application and month−of−the−year) to allow us to observe both the main and interaction effects of these provider characteristics. We find that, in general, the years since a company was founded, being a more local provider (e.g., providers from the same country see a larger relative increase in adoption than those from the same region), and having more company trademarks and patents have strongly positive relationships on app adoption. However, as time from COVID increases, the relationship between company longevity and being a domestic provider on adoption becomes more negative and there is no further positive differential for companies with more trademarks. Contrarily, we observe that having more patents (our proxy for innovation) is not only positively related to adoption, but also that this effect becomes larger as time from COVID increases.

In summary, the combined results from our analyses on provider types and characteristics seem well aligned with the original hypotheses. They suggest that the onset of the pandemic might have seen more demand effects at play, with factors related to perceived “trustworthiness” (e.g., longevity, reliability, familiarity) of providers playing an initial role for drawing in new users or keeping existing (but non−digital users) from switching. In this early scenario, traditional incumbents were well placed to benefit as they were able to capitalize on their familiarity and thus increase adoption by converting existing physical channel clients to online or app based clients. However, it is indeed plausible that as the pandemic lingered, the needs of both households and firms evolved. As consumers and providers adjusted to the “new normal” of higher levels of digitization, supply effects from the providers’ side come to the fore. Consequently, institutions with greater innovation, robust online offerings and digital infrastructure seem to have become increasingly competitive, with newer fintechs and BigTechs particularly being able to capitalize on these core strengths.

3.4. What types of products have seen greatest adoption?

In this section, we study the product types that saw the most change in adoption following the onset of COVID−19. Such comparisons can provide further insight into what drove consumer demand during the crisis period and additional evidence on changes to the market structure. For example, we can directly test whether bundled or unbundled (“moduralized”) services gained more traction. We also explore how this may have differed across countries of different development levels.

To examine this, we again draw on the download data for the top 10 finance applications by country for the Android and iOS platforms, respectively, during the period from Jan 1st 2019 to December 9th, 2020. We use app store descriptions to categorize the apps into the following product types: 1) general banking, 2) payment, 3) lending, 4) insurance, 5) investment, 6) government, and 7) miscellaneous. The general banking apps are multi−product platforms often linked to major banks or neobanks. They generally include access to some form of deposit account, but also are frequently bundled with other payment, loans, investment, and insurance products, among other add−on services. Meanwhile, the other categories are pure−play unbundled apps which appear specialized in an individual product category. We then rerun our model in Eq. 2 using these product categories to test for differential effects.

Table 6 summarizes results in terms of relative change in app downloads. The baseline level constitutes the general banking apps and is used to capture the main effect for bundled products. For this base level, we observe a modest 5% percent relative increase in downloads across the full country sample when estimated using the Post−Confirmed case dummy. This effect appears primarily driven by the EMDE countries rather than AE countries, with a 13% relative increase in the former compared to a neglible change in the latter. Meanwhile, examining the comparable specifications using the In Lockdown dummy, we see some signs of a weakly negative relationship. Like our previous provider−level results, we interpret this to suggest that the uptick in general banking app adoption had already began occurring as the pandemic confirmed cases spread − i.e., prior to the lockdowns.

Table 6.

Effect of COVID−19 on finance mobile app adoption by product type. This table presents coefficient estimates for a panel regression model estimating the country application−level relationship between the spread of COVID−19 on relative (logarithmic) change in daily downloads for finance category mobile apps. Post−Confirmed case is a country−level dummy indicator that turns on after a given location saw its first confirmed COVID−19 case. In lockdown denotes a dummy indicator that turns on and off during each countries’ lockdown period(s). We include interaction terms by product category to test for differential effects. The base level is general banking apps. All models include a general Post dummy indicator that turns on after the first confirmed COVID−19 case worldwide. We run separate specifications including all countries in the sample (All), subset to advanced economies (AE) and subset to emerging and developing economies (EMDE). Appendix Table A.1 provides the country categorization. The data sample covers top android and iOS mobile finance apps for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. Standard errors are clustered at the country−application level and in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, OxCGRT, and aura vision.

	DV = Relative (ln) daily app downloads			DV = Relative (ln) daily app downloads
Base level = General banking apps	(All)	(EMDE)	(AE)	(All)	(EMDE)	(AE)
Post−Confirmed case = 1	0.049∗∗∗	0.128∗∗∗	−0.005
	(0.013)	(0.022)	(0.014)
Post−Confirmed case = 1 × Payment	0.070∗∗∗ (0.024)	0.046 (0.047)	0.099∗∗∗ (0.023)
Post−Confirmed case = 1 × Lending	0.209∗∗∗ (0.066)	0.260∗∗ (0.119)	0.128∗∗∗ (0.046)
Post−Confirmed case = 1 × Insurance	0.042 (0.050)		0.126∗∗ (0.051)
Post−Confirmed case = 1 × Investment	0.151∗∗∗ (0.035)	0.042 (0.067)	0.229∗∗∗ (0.038)
Post−Confirmed case = 1 × Government	0.250∗ (0.137)	0.209 (0.188)	0.144∗∗ (0.068)
Post−Confirmed case = 1 × Miscellaneous	−0.025 (0.051)	0.015 (0.084)	0.021 (0.059)
In Lockdown = 1				−0.029∗	−0.026	−0.040∗∗
				(0.015)	(0.025)	(0.017)
In Lockdown = 1 × Payment				0.071∗∗ (0.028)	0.017 (0.060)	0.117∗∗∗ (0.027)
In Lockdown = 1 × Lending				0.034 (0.060)	−0.010 (0.124)	0.061 (0.040)
In Lockdown = 1 × Insurance				−0.042 (0.068)		0.027 (0.068)
In Lockdown = 1 × Investment				0.113∗∗∗ (0.043)	−0.016 (0.070)	0.206∗∗∗ (0.053)
In Lockdown = 1 × Government				0.227∗ (0.133)	0.270 (0.189)	0.054 (0.070)
In Lockdown = 1 × Miscellaneous				−0.054 (0.042)	−0.082 (0.072)	−0.001 (0.048)
Observations	865,031	338,032	526,999	865,031	338,032	526,999
R2	0.909	0.904	0.911	0.907	0.902	0.910
Additional controls: Country−application × MoY interactions	Yes	Yes	Yes	Yes	Yes	Yes

Turning attention to unbundled product types, across the overall sample, we observe that the onset of COVID cases led to more considerable relative increases in the adoption of pure−play apps of payment (7%), lending (21%), investment (15%), and government apps (25%), although the latter is of somewhat weaker significance. Notably, the orientation of demand across the product types does differ between the EMDE and AE country sample. For example, the EMDE countries exhibit a particularly larger relative increase in uptake of lending apps, amounting to a combined increase of 39% (compared to a combined increase of 12% in the AE countries). Meanwhile, the EMDE countries saw insignificant changes in investment apps and no pure−play insurance apps made their top performing finance category app lists. By contrast, the AE countries saw stronger relative increases in a greater variety of other product types. For example, we observe that they primarily account for the overall effect in terms of increased uptake in payment, investment, and insurance apps. Moreover, we see signs in the AE countries that the more narrow onset of lockdown periods led to further upticks in the payment and investment apps. It is worth noting that, in practice, we observe that many traditional incumbents have launched standalone payment apps as well, in addition to their general banking platforms. This may presumably be to meet the increasing competition from Bigtech and fintech players. Thus, in combination with the prior provider−level results in Section 3.3, this suggests that most of the increase in uptake for the traditional incumbents in AE countries during the COVID period was attributed to increased demand for their payment products rather than their general banking apps.

This difference in the orientation of demand may be linked to generally lower starting levels of financial inclusion and financial health in EMDEs compared with AE countries, prior to the onset of COVID. This appears to have driven the former cohort to exhibit increased demand for basic banking services and alternative lending options. As others have documented, increased financial inclusion and alternative fintech products can provide useful short−term tools for allowing households and firms to deal with exogenous shocks and increase financial resilience (Demirguc−Kunt et al. (2018); Cong et al. (2021); Suri et al. (2021)). Indeed, our findings do seem to suggest that there were consumers that increasingly drew on novel new lending sources to deal with the shock, in both the EMDE and AE countries. For example, we conduct additional analyses that test how sensitive the adoption of different products was to the duration of the COVID shock and the separate lockdown waves within countries at the aggregate level and then broken out by EMDE and AE countries (see Appendix Table A.8). We observe that the demand for lending apps in EMDE countries notably grew in strength from the outset of the pandemic relative to other product categories and wasn't particularly more prominent as lockdown waves came in. Meanwhile, in the AE countries, demand for similar lending apps also grew as a function of time, but became particularly prominent in the longer run during the second wave of lockdowns. This is consistent with AE consumers generally having higher ex−ante levels of financial health to buffer from the initial shocks, but that a subset may have become more vulnerable as the economic shock grew increasingly prolonged. However, in the AE countries, we also see signs of a subset of consumers that had greater capacity to branch out and leverage a wider set of financial tools, including risk mitigation products (e.g., pure−play insurance apps) and asset building tools (e.g., pure−play investment apps).

Taken as a whole, our analysis suggests that most product types saw a general increase in adoption due to COVID, but that unbundled (“moduralized”) apps saw particularly notable gains relative to bundled products. On the one hand, such “unbundling” has benefits, particularly with respect to increasing competition and improved choice and prices for customers. For example, it can help move beyond the “broad, but shallow” nature of financial inclusion today and create a deeper range of targeted services. On the other hand, this trend will also create new risks such as making regulatory oversight more challenging (due to the proliferation of providers in increasingly niche market segments) and potentially affecting solvency and financial stability of traditional incumbents, should trends shift towards customer disintermediation. To this end, we turn in the remaining sections to examining implications on financial intermediation more directly.

3.5. Implications for financial intermediation

The increased adoption of fintech due to the pandemic could have possible consequences for financial intermediaries and the structure of the financial services industry via changes in lending and payment markets (simultaneously on the asset and liability side of the traditional bank balance sheet). In this section, we further delve into disaggregated trends for the top downloaded finance apps to understand how they could potentially impact the finance industry and the overall stability of the financial system in the near future.

3.5.1. Impact on digital lending

We first study in more detail how COVID−19 affected the (already growing) demand for fintech lending channels and use this to draw a few salient implications for traditional lenders. Our analysis in Section 3.4 and Table 6 indicated a sharp increase in the demand for digital lending apps driven by COVID, and that this uptick was stronger and driven more by the emerging and developing economy sample. Given this initial finding, we dig deeper into the nature of these apps and their providers, to provide some insights on whether and in what ways traditional financial intermediaries are likely to have been affected.

Upon closer inspection of the underlying apps, we separate the lending apps into two different categories: authentic apps and suspect apps, depending on whether a lending app appears genuine in offering services to borrowers or it shows signs of being either fraudulent or predatory. To do so in practice, we manually review the lending apps making it into the top apps during the study period by going over their websites (where applicable) and Google app store reviews for each app. For suspect apps, we manually study the reviews for each app, along with other indicators drawn from the app's meta data. The reviews get sorted by importance automatically which allows us to infer which apps are more likely to be suspect. We find many authentic personal lending apps that existed from prior to the pandemic period, some of which have licensing with local regulators. However, we also observe numerous examples of top downloaded lending apps that exhibit signals of being either predatory or potentially fraudulent. For example, there are a group of personal loan apps that are newly−launched at the time of lockdowns, initially buttressed by what appear to be falsified reviews and ratings before receiving a subsequent large uptick in uptake followed by a wave of negative feedback. We further observe that many of them lack legitimate contact emails or provider websites and are eventually removed from the app store platforms (albeit subsequently replaced by similar ones in their place in subsequent waves). Another subset of suspect apps does not exhibit falsified reviews, but have particularly low ratings and disproportionate reviews from real users indicating predatory practices (e.g., high interest rates, harassment during debt collection, scraping of private information, etc.).

Table 7 depicts the relative increases in adoption for the authentic and suspect lending apps across emerging markets and advanced economies, relative to the base level of general banking apps. We use a continuous time variable capturing the log of days since the first COVID case within a country, as it provides richer insights into the temporal nature of the adoption of these lending apps. That is, early into the pandemic, individuals and firms might have been able to use their cash reserves to tide through the initial phase. However, as the pandemic prolonged, it is likely that demand for alternative credit would increase as cash reserves get exhausted. Panel A depicts the results where the increase in uptake of authentic lending apps is 3% above the base level of app downloads whereas the increase in uptake for suspect apps is 15% above the base level, and grows stronger the further time progresses from the 1st detected COVID case within a country. On further disaggregation, we observe that the increase in authentic lending apps comes primarily from advanced economies whereas the associated increase in suspect lending apps comes from developing economies. As before, the effects for these more narrowly subset models become stronger the further we are from the outbreak of COVID. Panel B further establishes the results we observe in Panel A. Here we use dummies to study the effect of successive lockdowns instead of using a continuous variable. Our dummies switch on for each lockdown wave. We observe that the increase in uptake of the authentic lending apps is concentrated in the AE countries after the second wave of lockdowns. In fact, while general banking apps there see a 12% reduction in their downloads during the second wave, the authentic lending apps see an aggregate increase of around 10%. Meanwhile, for the EMDE countries the uptake of suspect lending apps rises sharply in the second wave with a commensurate decrease in authentic lending apps in these geographies.

Table 7.

Effect of COVID−19 on app adoption across lending−related products. This table presents coefficient estimates for a panel regression model estimating the relationship between the spread of COVID−19 on relative (logarithmic) change in daily app downloads for finance category mobile apps. Days since first COVID case is a continuous variable capturing time elapsed (transformed into logarithmic form) since each countries’ first confirmed COVID case. First wave of lockdowns and second wave of lockdowns denote dummy indicators that turn on and off during each countries’ first and second lockdown periods. Post denotes a general dummy indicator that turns on after first confirmed COVID−19 case worldwide. The base level consists of general banking apps (that include some form of lending product amongst its menu of options). We further label pure−play lending apps, sub−setting between those that appear from legitimate providers versus those that show signs of being predatory or fraudulent. We include interaction terms for these categories to test for differential effects. We omit all other product categories for purposes of this analysis. All models include a general post dummy indicator that turns on after the first confirmed COVID−19 case worldwide. We run separate specifications including all countries in the sample (All), subset to advanced economies (AE) and subset to emerging and developing economies (EMDE). Appendix Table A.1 provides the country categorization. The data sample covers top android and iOS mobile finance apps for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. Standard errors are clustered at the country−application level and in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, Crunchbase, OxCGRT, and aura vision.

	DV=Relative (ln) daily app downloads			DV=Relative (ln) daily app downloads
Base level = General banking apps	(All)	(EMDE)	(AE)	(All)	(EMDE)	(AE)
Panel A. Time from COVID cases within country ln(Days since 1st COVID case in country)	0.013∗∗∗	0.034∗∗∗	0.000
	(0.003)	(0.006)	(0.004)
ln(Days since 1st COVID case in country) × Pure−play lending app (real)	0.030∗∗ (0.014)	0.036 (0.031)	0.027∗∗ (0.010)
ln(Days since 1st COVID case in country) × Pure−play lending app (suspect)	0.146∗∗∗ (0.045)	0.121∗∗∗ (0.045)
Panel B. Patterns across successive lockdown waves
First wave of lockdowns = 1				0.011	−0.019	0.018
				(0.014)	(0.024)	(0.016)
First wave of lockdowns = 1 × Pure−play lending app (real)				0.012 (0.079)	0.015 (0.162)	0.011 (0.044)
First wave of lockdowns = 1 × Pure−play lending app (suspect)				0.011 (0.178)	−0.068 (0.178)
Second wave of lockdowns = 1 Second wave of lockdowns = 1 × Pure−play lending app (real)				−0.180∗∗∗ (0.021) 0.191∗∗ (0.075)	−0.025 (0.051) −0.475∗∗∗ (0.049)	−0.121∗∗∗ (0.019) 0.218∗∗∗ (0.076)
Second wave of lockdowns = 1 × Pure−play lending app (suspect)				0.835∗∗∗ (0.019)	0.598∗∗∗ (0.049)
Observations	473,910	200,019	273,891	473,910	200,019	273,891
R2	0.913	0.915	0.905	0.911	0.911	0.905
Additional controls: Country−application × Month−of−the−Year interactions	Yes	Yes	Yes	Yes	Yes	Yes

We interpret these combined result to suggest that there was demand for personal or consumer loans during the COVID period that was unmet by traditional financial intermediaries or incumbent credit card companies, and instead being fulfilled through non−traditional channels. In some cases, the demand seems so strong that borrowers are willing to experiment with non−certified providers to meet their requirements, particularly in the EMDE countries. The increase in fintech lending over and above the traditional banking apps thus presents interesting challenges for these incumbents as well as regulators. The literature has emphasized that fintech lenders tend to bottom fish borrowers traditionally excluded from the lending market rather than cream skim (De Roure et al. (2016); Jagtiani & Lemieux (2018)). However, it is plausible that because of the pandemic many borrowers who were marginally able to access credit were shut out of the borrowing market due to credit rationing by traditional banks and thus they resorted to lending apps to satisfy their credit requirements. As a result, the rise of fintech lending due to COVID posits an empirical understanding of whether the uptake was due to apps that are complementary to the existing financial system or whether they are substitutes. If the uptake is more towards apps that are of the latter type, it could imply consequences for financial stability owing to the increased competition between fintech lenders and traditional banks (Chava et al. (2018); Cornaggia et al. (2018); Di Maggio & Yao (2021)).

3.5.2. Addressing predatory and fraudulent fintech apps

We delve further into the issue of predatory and fraudulent lending apps by elucidating an example from India in more detail. We extract a larger list of the top 16 downloaded personal lending apps in the Indian market during the study period. We review the individual apps and their underlying providers and categorize them into those that are legitimate (based on registration with the Reserve Bank of India) versus suspect (based on meeting the criteria previously detailed). We then aggregate the downloads for these apps by category and depict their time series trends during the pandemic period.

Fig. 3 depicts results graphically. We observe that in the months leading up to the pandemic, the legitimate lending apps cumulatively saw healthy demand of roughly 200,000 downloads per day. As confirmed cases start to rise in India, the loan apps from suspect providers begin to gain traction. Moreover, as the first lockdowns commence, the suspect loan apps overtake and appear to erode take−up from the legitimate loan apps. In other words, we see evidence of a high degree of substitution from consumers in certain markets between using the personal loan apps coming from legitimate versus illegitimate providers. Moreover, we observe another wave of new suspect loan apps released and similarly gain traction over a longer duration towards the end of the year. At the peak, the suspect loan apps were being downloading by over 500,000 users a day. Finally, it is worth noting that while take−up of the legitimate loan apps does slowly start to recover, it does not appear to reach a level comparable to the pre−pandemic trend.

Fig. 3.

Indian personal loan mobile application downloads during COVID−19. This figure depicts daily downloads of selected lendings apps prior to and following the onset of COVID−19 cases and lockdowns in the country. We extract a larger list of the top 16 performing lending apps in the Indian market during the study period. We review the individual apps and their underlying providers and categorize them into those that are “legitimate” (based on registration with the Reserve Bank of India) versus “suspect” (based on meeting the criteria previously detailed). Data sources: AppTweak and OxCGRT.

Given our data, we cannot directly test the extent to which the uptick in suspected predatory or fraudulent digital lending apps has spill−over effects in the traditional credit market (e.g., reducing borrowing from financial intermediaries such as commercial banks and credit unions). On the one hand, it is possible that traditional intermediaries and the digital app−based lenders serve slightly different target populations. On the other hand, studies have documented in various settings that fintech and traditional lenders do compete over some segments of borrowers, which can lead to increased loan non−performance (Chava et al. (2018); Cornaggia et al. (2018); Di Maggio & Yao (2021)). If this occurs in the Indian case, fraudulent lenders are likely to further exacerbate overindebtedness issues which are likely to also increase systematic risk for traditional formal lenders.

This descriptive analysis highlights that financial regulators and supervisors will need to increasingly address consumer protection issues in these app markets, which have until now been mostly lightly monitored. Left unchecked, such widespread instances could erode trust in legitimate financial providers and further hinder efforts to promote financial inclusion in vulnerable populations, who are likely to be the primary users of such apps. On the supply side, in terms of ex−ante vetting, efforts should be made to work with country regulators and the app stores to change some basic requirements for listing finance−related apps on their platforms, e.g., at a minimum, providers of finance−related apps should have to provide minimal website, valid email address, and further minimum legal information. In terms of ex−post monitoring, various high frequency data on mobile app downloads/usage and mobile app store data could be combined and used in order to create “regtech” systems for flagging and reporting highly suspect apps in real−time to local regulators, app stores, or other relevant stakeholders. On the demand side, part of the issue is likely to stem from lower average levels of financial and digital literacy in markets with many first−time users of formal financial services. In the long term, it is likely that these issues will become less pronounced, as studies have shown that even relatively inexperience and unsophisticated consumers of financial technology quickly learn−by−doing and can adapt behaviors to counter exploitative financial intermediaries (Breza et al., 2020). However, in the short term, demand−side interventions to inform consumers would be an important intermediate step (e.g., just in time announcements which highlight key signals and patterns in apps to avoid).

3.5.3. Effect on neo−banks, wallets and payment banks

In this section, we analyze the effect of COVID on fintech apps of payment banks, e−wallets, and neo− banks – i.e., institutions that operate exclusively on the liability side of the traditional bank balance sheet. We classify apps into four different categories: i) a base level consisting of traditional commercial banks, ii) Neo−banks affiliated with incumbents (i.e., those that were incubated or acquired by traditional banks themselves), iii) De−novo neo−banks (i.e., those that are standalone and not affiliated with any large financial institution), and iv) payment apps and e−wallets (apps which facilitate payments at merchants or vendors by allowing consumers to store cash). The increase in the adoption of these apps is expected to be one−off at the start of the pandemic to facilitate ease of payments and transactions online owing to social distancing norms and associated lockdowns.

Table 8 depicts that affiliated neo−banks have higher adoption rates in the AE countries by about 11% over and above the base level (traditional banks). Meanwhile, de−novo neo−banks see a 19% increase relative to the base level in the EMDE countries. A possible reason for de−novo neo−banks being more successful in EMDE settings may again be linked to the fact that financial inclusion is not complete in these economies and thus there is space for new “banks” to grow their clientele without having to encroach upon that of the traditional banks. Payments and wallets see a uniform increase across both AE and EMDE countries with the economic magnitude (10%) being higher in the former as compared to the latter (6%). However, the result for AE countries is statistically more significant than the result for EMDEs. We see statistically weaker changes in the results using the lockdown dummies possibly hinting that these changes were already starting to occur early into the pandemic before lockdowns became commonplace.

Table 8.

Effect of COVID−19 on app adoption across traditional banks, neobanks, and e−wallet platforms. This table presents coefficient estimates for a panel regression model estimating the relationship between the spread of COVID−19 on relative (logarithmic) change in daily app downloads for finance category mobile apps. Post−Confirmed case is a dummy indicator that turns on after a given location saw its first confirmed COVID−19 case. In lockdown denotes a dummy indicator that turns on and off during each countries’ lockdown period(s). Post denotes a general dummy indicator that turns on after first confirmed COVID−19 case worldwide. The base level is traditional financial intermediaries (including commercial banks, credit unions, and SLAs). We further label providers who are independent (de novo) neobanks, neobanks that are (affiliated) with traditional intermediaries, or payment/e−wallets. We include interaction terms for these provider categories to test for differential effects. All other provider types are omitted for purposes of this analysis. All models include a general Post dummy indicator that turns on after the first confirmed COVID−19 case worldwide. We run separate specifications including all countries in the sample (All), subset to advanced economies (AE) and subset to emerging and developing economies (EMDE). Appendix Table A.1 provides the country categorization. The data sample covers top Android and iOS mobile finance apps for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. Standard errors are clustered at the country−application level and in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, Crunchbase, OxCGRT, and aura vision.

	DV = Relative (ln) daily app downloads			DV = Relative (ln) daily app downloads
Base level = Traditional incumbents	(All)	(EMDE)	(AE)	(All)	(EMDE)	(AE)
Post−Confirmed case = 1	0.049∗∗∗	0.103∗∗∗	0.017
	(0.012)	(0.022)	(0.014)
Post−Confirmed case = 1 × Neobanks (affiliated w / incumbents)	0.080 (0.064)	0.006 (0.102)	0.108∗∗ (0.052)
Post−Confirmed case = 1 × Neobanks (de novo)	0.037 (0.045)	0.185∗∗ (0.087)	−0.003 (0.047)
Post−Confirmed case = 1  × Payments / e−wallets	0.067∗∗∗ (0.025)	0.095∗ (0.051)	0.059∗∗ (0.026)
In Lockdown = 1				−0.015	−0.033	−0.009
				(0.013)	(0.025)	(0.015)
In Lockdown = 1 × Neobanks (affiliated w/ incumbents)				0.042 (0.071)	−0.012 (0.096)	0.027 (0.066)
In Lockdown = 1 × Neobanks (de novo)				0.029 (0.052)	0.095 (0.108)	0.018 (0.055)
In Lockdown = 1 × Payments / e−wallets				0.049∗ (0.029)	0.026 (0.063)	0.068∗∗ (0.027)
Observations	720,668	266,850	453,818	720,668	266,850	453,818
R2	0.915	0.916	0.908	0.914	0.915	0.907
Additional controls: Country−application X Month−of−the−Year interactions	Yes	Yes	Yes	Yes	Yes	Yes

The increase in adoption of apps that focus on the liability side of the traditional bank balance sheet has interesting consequences for financial intermediation in the future. In most advanced economies, interest rates have been zero or close to zero for close to a decade now. This has led to a proliferation of no−frills app−only bank accounts that have been able to attract customers while keeping their fixed costs low (as they do not operate brick and mortar branches). However, the erosion of deposits from traditional banks has resulted in the reduction of their balance sheet size while at the same time stripping them of access to low−cost funding. Consequently, traditional banks in advanced economies have started acquiring or launching neo−banks of their own to compete with pure−play unaffiliated neo−banks to stem the loss of their deposit base. We observe from Table 8 that this trend seems to have picked up after COVID. Future research can focus on the impact of neo−banks on the deposit base of traditional banks, their subsequent loan disbursals and possible credit rationing.

4. Conclusion

We study the effects of COVID−19 on fintech adoption and usage. Our first and main finding is to document how the pandemic's spread and related government lockdowns drove significant increases in adoption of finance−related mobile applications, whether in relative or absolute per capita terms. Our second set of findings examine implications of this increased adoption on market structure and financial intermediaries. This is done by drilling down on outcome heterogeneity across app products and provider types across time. In particular, our findings suggest that traditional incumbents did well in the early stages of the pandemic, perhaps leveraging factors related to initial market capture and trustworthiness (e.g., familiarity, longevity, reliability). However, over the protracted period, “Bigtech” and fintech startups were able to accelerate uptake of their digital services over−and−above traditional incumbents. As such, niche and unbundled products such as pure−play payment, lending, and investment apps see higher relative increases over time compared to general banking apps (i.e., bundled products). We posit that the latter provider types were able to respectively leverage their economies of scale and greater agility to meet evolving consumer needs and attract them to more niche products, as the pandemic prolonged. Indeed, we find that provider−level characteristics most associated with higher increases in app adoption emphasize the importance of innovation−related characteristics, such as patents. Furthermore, we examine heterogeneity in product adoption operating on the asset and liability side of the traditional bank balance sheet. On the asset side, we observe increased adoption of legitimate lending apps in advanced economies but, problematically, a sharp increased adoption of potentially predatory or fraudulent apps in developing countries. On the the liability side, we document increased standalone neo−bank adoption in developing economies but affiliated neo−bank and payment wallet adoption in advanced economies.

The documented digital acceleration will likely have important longer−term implications for involved consumers and providers. On the demand side, while the large−scale shifts in fintech adoption and particularly the use of alternative lending sources may have helped many households and small firms mitigate the expected short−term fall in productivity and economic growth stemming from COVID, they may also have implications for overindebtedness and consumer protection. It remains to be seen whether these newer fintech lending sources acted as complements or substitutes and whether the adopting borrowers suffer disproportionately from loan performance issues and other adverse outcomes over time. Moreover, it will be important to also measure the direct and indirect effects of the documented uptick in predatory and fraudulent lending apps on consumers and legitimate lenders. On the supply side, the trends we document have policy relevance for regulators and supervisors who have been closely watching the trajectory of fintech and interplay between traditional incumbents, “BigTech” and other fintech players, such as neobanks. Our evidence suggests that COVID has accelerated the ongoing changes in the financial intermediation landscape (e.g., the rise of BigTech and modularized financial services) and will require regulators to be proactive in monitoring and dealing with exposed regulatory gaps as a new equilibrium settles. For example, it would be important to analyze relevant provider−side outcomes post COVID (e.g., rates of closures, acquisitions, funding, changes to deposit base or credit rationing for universal banks, etc.) to see if there is indeed stronger evidence of movement towards a “better bank” or “disaggregated” bank scenarios. Future research could drill−down on these longer−term issues as the answers to these questions are in essence, empirical.

CRediT authorship contribution statement

Jonathan Fu: Project administration, Conceptualization, Methodology, Software, Validation, Formal analysis, Writing – original draft, Writing – review & editing, Visualization, Resources, Data curation. Mrinal Mishra: Conceptualization, Methodology, Software, Validation, Formal analysis, Writing – original draft, Writing – review & editing, Visualization.

Declaration of Competing Interest

None.

A Appendix

A.1 Have other industries been equally affected?

As an additional exercise, we examine heterogeneity in adoption across the major app categories on the Google Play and Apple iOS app stores. This allows us to shed light on what industries may have undergone the most significant digital transformation during the COVID−19 pandemic and benchmark how the rate of digital acceleration compares between the finance sector and other sectors of activity. To investigate this empirically, we similarly extract daily app download data at the country−category level and re−construct our main dependent and independent variables. Table A3

Table A.3.

Sensitivity tests on construction of main DV This table presents coefficient estimates for a panel regression model estimating the country−level relationship between the spread of COVID−19 on the relative (logarithmic) and per capita changes in daily downloads for finance category mobile apps. We rerun the same specification substituting in raw downloads and then using 5−day, 7−day, 10−day, and 14−day leading moving average to calculate downloads. Post−confirmed case denotes a dummy indicator that turns on after a given country saw its first confirmed COVID−19 case. In lockdown denotes a dummy indicator that turns on and off during each countries’ lockdown period(s). The data sample covers the android and iOS mobile finance app markets for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. All models include a general Post dummy indicator that turns on after the first confirmed COVID−19 case worldwide. We run these models on our preferred specification, which includes country X month−of−the−year interactions to control for both country−level characteristics and seasonality. Standard errors are clustered at the country level and in parentheses. *** p<0.01, ** p<0.05, * p<0.10. Data sources: AppTweak, OxCGRT, and aura vision.

Panel A. Pandemic spread	DV = Relative (Ln) daily app downloads					DV = # of daily app downloads per capita (103)
	Original	5DMA	7DMA	10DMA	14DMA	Original	5DMA	7DMA	10DMA	14DMA
Post−Confirmed case =1	0.206∗∗∗	0.190∗∗∗	0.160∗∗∗	0.128∗∗∗	0.080	0.531∗∗∗	0.516∗∗∗	0.497∗∗∗	0.481∗∗∗	0.456∗∗∗
	(0.034)	(0.036)	(0.041)	(0.044)	(0.049)	(0.099)	(0.099)	(0.099)	(0.100)	(0.101)
Observations	50,976	50,976	50,976	50,976	50,976	50,976	50,976	50,976	50,976	50,976
R2	0.980	0.975	0.959	0.947	0.931	0.891	0.894	0.893	0.891	0.886
Panel B. Government response	DV=Relative (Ln) daily app downloads					DV=# of daily app downloads per capita (103)
In Lockdown =1	0.067∗∗∗	0.055∗∗	0.047∗∗	0.039∗	0.028		0.183∗∗∗	0.173∗∗∗	0.163∗∗∗	0.153∗∗
	(0.022)	(0.022)	(0.022)	(0.022)	(0.022)		(0.060)	(0.060)	(0.059)	(0.059)
Observations	43,188	43,188	43,188	43,188	43,188		43,188	43,188	43,188	43,188
R2	0.981	0.976	0.959	0.947	0.930		0.860	0.864	0.864	0.862

Table A.4.

Effect of COVID−19 on finance app adoption by provider type, early pandemic. This table presents coefficient estimates for a panel regression model estimating the relationship between the spread of COVID− 19 on relative (logarithmic) change in daily app downloads for finance category mobile apps. Post−confirmed case is a dummy indicator that turns on after a given location saw its first confirmed COVID−19 case. In lockdown denotes a dummy indicator that turns on and off during each countries’ lockdown period(s). All models also include a general post dummy indicator that turns on after the first confirmed COVID−19 case worldwide. We include interaction terms by provider category to test for differential effects. The base level is traditional incumbents. We run separate specifications including all countries in the sample (All), subset to advanced economies (AE) and subset to emerging and developing economies (EMDE). Appendix Table A.1 provides the country categorization. The data sample covers top Android and iOS mobile finance apps for a globally representative sample of countries daily from January 1st, 2019 to August 9th, 2020. Standard errors are clustered at the country−application level and in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, Crunchbase, OxCGRT, and aura vision.

	DV = Relative (ln) daily app downloads			DV = Relative (ln) daily app downloads
Base level = Traditional incumbents	(All)	(EMDE)	(AE)	(All)	(EMDE)	(AE)
Post−Confirmed case = 1	0.17∗∗∗ (0.023)	0.23∗∗∗ (0.040)	0.13∗∗∗ (0.026)
Post−Confirmed case = 1 X Fintech incumbents	−0.05	−0.07	−0.02
Post−Confirmed case = 1 X Bigtech	(0.040) −0.21∗∗∗ (0.028)	(0.078) −0.29∗∗∗ (0.051)	(0.041) −0.14∗∗∗ (0.029)
Post−Confirmed case = 1 X Fintech startups	−0.06	−0.05	−0.07
In Lockdown = 1	(0.044)	(0.083)	(0.040)	0.13∗∗∗	0.18∗∗∗	0.05∗
				(0.028)	(0.047)	(0.027)
In Lockdown = 1 X Fintech incumbents				−0.08	−0.13	−0.01
In Lockdown = 1 X Bigtech				(0.049) −0.19∗∗∗ (0.033)	(0.091) −0.28∗∗∗ (0.058)	(0.046) −0.11∗∗∗ (0.031)
In Lockdown = 1 X Fintech startups				−0.08	−0.07	−0.10∗
				(0.057)	(0.101)	(0.044)
Observations	960,113	389,880	570,233	960,113	389,880	570,233
R2	0.936	0.932	0.941	0.935	0.930	0.941
Additional controls:Country−application fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Country X Month interactions	Yes	Yes	Yes	Yes	Yes	Yes

Table A.5.

Adoption patterns by provider type as time from COVID progresses. This table presents coefficient estimates for a panel regression model estimating the relationship between the spread of COVID−19 on relative (logarithmic) change in daily app downloads for finance category mobile apps. Days since first COVID case is a continuous variable capturing time elapsed (transformed into logarithmic form) since each countries’ first confirmed COVID case. First wave of lockdowns and Second wave of lockdowns denote dummy indicators that turn on and off during each countries’ lockdown periods. Post denotes a general dummy indicator that turns on after first confirmed COVID−19 case worldwide. The base level is traditional incumbents. We run separate specifications including all countries in the sample (All), advanced economies (AE) and emerging and developing economies (EMDE). Appendix Table A.1 provides the country categorization. The data sample covers top Android and iOS mobile finance apps for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. All models include a general Post dummy indicator that turns on after the first confirmed COVID−19 case worldwide, in addition to country X month−of−the−year interactions to control for both country−level characteristics and seasonality. Standard errors are clustered at the country−application level and in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, Crunchbase, OxCGRT, and aura vision.

Table A.6.

Adoption patterns by company characteristics as time from COVID progresses. This table presents coefficient estimates for a panel regression model estimating the relationship between the spread of COVID−19 on the relative (logarithmic) change in daily app downloads for finance category mobile apps. Days since first COVID case is a continuous variable capturing time elapsed (transformed into logarithmic form) since each countries’ first confirmed COVID case. The base level varies based on model specification and is labeled in the top row. The data sample covers top android and iOS mobile finance apps for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. All models include a general Post dummy indicator that turns on after the first confirmed COVID−19 case worldwide. We run these models on our preferred specification, which includes country X Month−of−the−Year interactions to control for both country−level characteristics and seasonality. Standard errors are clustered at the country−application level and in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, Crunchbase, OxCGRT, and aura vision.

	DV = Relative (ln) # of daily app downloads
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Base levels (for categorical interaction variables)=		Outside region	Outside country
Log (Days since 1st COVID case)	0.039∗∗∗	0.035∗∗∗	0.028∗∗∗	0.017∗∗	0.025∗∗∗	0.017∗∗∗	0.019∗∗∗	0.023∗
	(0.007)	(0.004)	(0.004)	(0.007)	(0.004)	(0.004)	(0.007)	(0.013)
Log (Years since company founded)	0.017
	(0.015)
Log (Days since 1st COVID case) X Log (Years since company founded)	−0.005∗∗
	(0.002)
Provider from same region		0.067∗∗
		(0.034)
Log (Days since 1st COVID case) X Provider from same region		−0.013∗∗∗
		(0.005)
Provider from same country			0.081∗∗
			(0.032)
Log (Days since 1st COVID case) X Provider from same country			−0.004
			(0.004)
Log (Number of trademarks)				0.032∗∗
				(0.013)
Log (Days since 1st COVID case) X Log (Number of trademarks)				0.002
				(0.002)
Log (Number of articles)					0.008
					(0.006)
Log (Days since 1st COVID case) X Log (Number of articles)					−0.001
					(0.001)
Log (Number of patents)						0.012∗∗
						(0.006)
Log (Days since 1st COVID case) X Log (Number of patents)						0.003∗∗
						(0.001)
Log (Number of apps)							−0.005
							(0.022)
Log (Days since 1st COVID case) X Log (Number of apps)							0.003
							(0.003)
Log (Number of products active)								0.037
								(0.032)
Log (Days since 1st COVID case) X Log (Number of products active)								−0.002
								(0.004)
Observations	782,430	881,942	881,942	392,912	690,423	392,912	501,098	511,664
R2	0.568	0.543	0.544	0.658	0.573	0.657	0.604	0.579

Table A.7.

Effect of COVID−19 on finance app adoption by provider size this table presents coefficient estimates for a panel regression model estimating the relationship between the spread of COVID− 19 on relative (logarithmic) number of daily app downloads for finance category mobile apps. Post−confirmed case denotes a dummy indicator that turns on after each country's first confirmed COVID case. Post denotes a general dummy indicator that turns on after first confirmed COVID−19 case worldwide. The data sample covers top Android and iOS mobile finance apps for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. All models include a general Post dummy indicator that turns on after the first confirmed COVID−19 case worldwide, in addition to country X month−of−the−year interactions to control for both country−level characteristics and seasonality. Standard errors are clustered at the country−application level and in parentheses *** p < 0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, OxCGRT, and Aura Vision.

Panel A. employment size	DV = Relative (ln) daily app downloads
Base level = Less than 50 employees	(All)	(EMDE)	(AE)
Post−Confirmed case = 1	0.118∗∗∗	0.146∗∗	0.055
	(0.005)	(0.016)	(0.296)
Post−Confirmed case = 1 × 51–500 employees	0.018	0.059	0.056
	(0.713)	(0.484)	(0.337)
Post−Confirmed case = 1 ×  501–1000 employees	−0.037	0.072	−0.009
	(0.595)	(0.655)	(0.898)
Post−Confirmed case = 1 × 1001–5000 employees	−0.091∗	−0.069	−0.037
	(0.069)	(0.421)	(0.530)
Post−Confirmed case = 1 × 5001–10,000 employees	0.030	−0.036	0.115
	(0.577)	(0.631)	(0.123)
Post−Confirmed case = 1 × 10,001 + employees	−0.039	0.023	−0.019
	(0.403)	(0.756)	(0.734)
Observations	749,983	273,182	476,801
R2	0.913	0.907	0.913
Panel B. Revenue Size	DV=Relative (ln) daily app downloads
Base level = Less than USD 1M	(All)	(EMDE)	(AE)
Post−Confirmed case = 1	0.209∗∗∗	0.204∗	0.193∗∗∗
	(0.001)	(0.071)	(0.003)
Post−Confirmed case = 1 X USD 1 M – 10M	−0.115	−0.061	−0.136∗
	(0.113)	(0.634)	(0.063)
Post−Confirmed case = 1 X USD 10 M – 50M	−0.150∗	−0.038	−0.208∗∗∗
	(0.060)	(0.784)	(0.009)
Post−Confirmed case = 1 X USD 50 M – 100M	−0.195∗∗	0.297	−0.211∗∗∗
	(0.043)	(0.499)	(0.006)
Post−Confirmed case = 1 X USD 100 M – 500M	−0.174∗∗	−0.136	−0.182∗
	(0.047)	(0.328)	(0.075)
Post−Confirmed case = 1 X USD 500 M – 1B	−0.017	0.047	−0.123∗
	(0.846)	(0.750)	(0.099)
Post−Confirmed case = 1 X USD 1B – 10B	−0.134∗	−0.057	−0.128∗
	(0.051)	(0.648)	(0.066)
Post−Confirmed case = 1 X USD 10B+	−0.080	−0.058	−0.068
	(0.360)	(0.734)	(0.433)
Observations	481,831	167,576	314,255
R2	0.918	0.908	0.925

Table A.8.

Adoption patterns by product type as time from COVID progresses This table presents coefficient estimates for a panel regression model estimating the relationship between the spread of COVID− 19 on relative (logarithmic) change in daily app downloads for finance category mobile apps. Days since first COVID case is a continuous variable capturing time elapsed (transformed into logarithmic form) since each countries’ first confirmed COVID case. First wave of lockdowns and Second wave of lockdowns denote dummy indicators that turn on and off during each countries’ first and second lockdown periods. We include interaction terms by product category to test for differential effects. The base level is general banking apps. All models include a general Post dummy indicator that turns on after the first confirmed COVID−19 case worldwide. We run separate specifications including all countries in the sample (All), subset to advanced economies (AE) and subset to emerging and developing economies (EMDE). Appendix Table A.1 provides the country categorization. The data sample covers top android and iOS mobile finance apps for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. Standard errors are clustered at the country−application level and in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, Crunchbase, OxCGRT, and aura vision.

	DV = Relative (ln) daily app downloads			DV = Relative (ln) daily app downloads
Base level = General banking apps	(All)	(EMDE)	(AE)	(All)	(EMDE)	(AE)
Panel A. Time from COVID cases within country ln(Days since 1st COVID case)	0.015∗∗∗	0.033∗∗∗	0.003
	(0.003)	(0.005)	(0.003)
ln(Days since 1st COVID case) X Payment	0.015∗∗∗	0.009	0.021∗∗∗
ln(Days since 1st COVID case) X Lending	(0.005) 0.048∗∗∗ (0.015)	(0.010) 0.063∗∗ (0.027)	(0.005) 0.027∗∗∗ (0.010)
ln(Days since 1st COVID case) X Insurance	0.010		0.028∗∗∗
ln(Days since 1st COVID case) X Investment	(0.010) 0.034∗∗∗	0.011	(0.010) 0.050∗∗∗
	(0.008)	(0.014)	(0.008)
ln(Days since 1st COVID case) X Government	0.056∗	0.048	0.033∗∗
	(0.030)	(0.041)	(0.015)
ln(Days since 1st COVID case) X Miscellaneous	−0.003	0.008	0.006
Panel B. Patterns across successive lockdown waves	(0.011)	(0.020)	(0.012)
First wave of lockdowns = 1				−0.006	−0.025	−0.014
				(0.016)	(0.026)	(0.019)
First wave of lockdowns = 1 X Payment				0.051∗	0.013	0.092∗∗∗
				(0.030)	(0.060)	(0.029)
First wave of lockdowns = 1 X Lending				0.012	−0.011	0.011
				(0.073)	(0.125)	(0.044)
First wave of lockdowns = 1 X Insurance				−0.077		−0.012
First wave of lockdowns = 1 X Investment				(0.074) 0.097∗∗	−0.011	(0.075) 0.190∗∗∗
				(0.048)	(0.073)	(0.062)
First wave of lockdowns = 1 X Government				0.242∗	0.269	0.043
				(0.145)	(0.189)	(0.086)
First wave of lockdowns = 1 X Miscellaneous				−0.094∗∗	−0.082	−0.050
				(0.043)	(0.072)	(0.051)
Second wave of lockdowns = 1				−0.197∗∗∗	−0.030	−0.155∗∗∗
				(0.020)	(0.050)	(0.020)
Second wave of lockdowns = 1 X Payment				0.223∗∗∗	0.131	0.225∗∗∗
				(0.053)	(0.128)	(0.056)
Second wave of lockdowns = 1 X Lending				0.203∗∗∗	0.062	0.219∗∗∗
Second wave of lockdowns = 1 X Insurance Second wave of lockdowns = 1 X Investment				(0.074) 0.170∗ (0.091) 0.238∗∗∗	(0.383) −0.093∗	(0.076) 0.185∗∗ (0.088) 0.278∗∗∗
				(0.034)	(0.054)	(0.038)
Second wave of lockdowns = 1 X Government Second wave of lockdowns = 1 X Miscellaneous				0.117∗ (0.065) 0.278∗∗∗ (0.078)	0.000 (.) 0.000 (.)	0.138∗∗ (0.065) 0.291∗∗∗ (0.079)
Observations	865,031	338,032	526,999	865,031	338,032	526,999
R2	0.909	0.905	0.912	0.907	0.902	0.910

Fig. 10 provides a descriptive plot of absolute downloads by app category across our study period. We observe that the app categories roughly cluster into three different tiers in terms of levels of downloads. First, the most popular app categories in terms of downloads are “Entertainment & Music” and “Lifestyle & Social”, which see daily adoption rates several times higher than the others. However, while these categories do show signs of also experiencing a notable uptick in the lockdown period, it is worth highlighting that they return to a more similar level in both absolute and relative terms by the end of the sample period. Second, apps falling in the “Business & Productivity”, “Shopping”, “Finance”, “Navigation & Travel” fall in a middle tier of popularity and start off at a similar level of daily downloads prior to the onset of COVID−19 (roughly between 10–15 million per day). Within this group, the “Business & Productivity” and “Finance” app categories see large upticks around the onset of COVID−19. More importantly, they also exhibit fairly sustained increases in daily downloads by the end of the sample period. By contrast, the “Shopping” and “Navigation & Travel” categories see notable absolute and relative decreases during the COVID−19 period and appear to struggle to return to pre−COVID levels. Finally, apps in the “Education”, “Health & Medicine”, “Food & Drink”, and “News” categories comprise a bottom tier. The “Education” and “Health & Medicine” app categories also show signs of considerable upticks and sustained higher levels. Meanwhile, we observe negligible changes for the “Food & Drink”, and “News” app categories. In other words, a general takeaway is that productive sector apps dramatically gain in popularity with respect to more leisure−focused apps, which had dominated prior to the onset of COVID. (Similar trends exist if we alternatively depict the downloads in terms of per capita rates.)

Fig. 10.

Daily rate of downloads for mobile apps across major app store categories. This figure depicts the daily rate of downloads per capita for mobile applications across the major app store categories. The data includes mobile apps from both the Android and iOS platforms from October 1st, 2019 to December 9th, 2020. We calculate a 14−day leading moving average on downloads to smooth day−to−day fluctuations.

To make formal estimations and account for the app categories’ different pre−COVID levels of downloads, we re−run Eq. (2) with app category interaction dummies to test for differential effects of COVID−19 on relative downloads. Table A.9 depicts results from this app category−level analysis. We use the “Finance” app category as the base level to facilitate interpretation. The empirical results largely confirm the descriptive evidence. During the lockdown periods, as measured using our In Lockdown dummy, we observe that “Business and Productivity”, “Education”, and “Health & Medicine” apps experience the largest increase in relative terms. However, in terms of overall change in adoption since the onset of COVID (i.e., as measured by the Post−Confirmed case dummy), “Finance” apps actually appear to have had the highest sustained growth with the exception of “Business and Productivity” apps.

Table A.9.

The effect of COVID−19 on adoption of other major app categories this table presents coefficient estimates for a panel regression model estimating the country−app category relationship between the spread of COVID−19 on relative (logarithmic) number of app downloads. Post−confirmed case is a country−level dummy indicator that turns on after a given location saw its first confirmed COVID−19 case. In lockdown denotes a dummy indicator that turns on and off during each countries’ lockdown period(s). We include interaction terms by broader app category to test for differential effects. We run separate specifications including all countries (ALL) in the sample, subset to advanced economies (AEs) and subset to emerging and developing economies (EMDEs). The data sample covers the Android and iOS mobile finance app markets for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. Standard errors are clustered at the country application level and in in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, OxCGRT, and aura vision.

	DV = Relative (ln) daily app downloads			DV = Relative (ln) daily app downloads
Base level = Finance category apps	(All)	(EMDE)	(AE)	(All)	(EMDE)	(AE)
Post−Confirmed case = 1	0.32∗∗∗	0.41∗∗∗	0.26∗∗∗
	(0.035)	(0.042)	(0.050)
Post−Confirmed case = 1 X Business & productivity	0.09∗	−0.00	0.15∗
	(0.041)	(0.051)	(0.058)
Post−Confirmed case = 1 X Education	−0.09∗	−0.12∗	−0.08
	(0.041)	(0.051)	(0.057)
Post−Confirmed case = 1 X Entertainment & music	−0.32∗∗∗	−0.31∗∗∗	−0.33∗∗∗
	(0.045)	(0.053)	(0.064)
Post−Confirmed case = 1 X Food & drink	−0.36∗∗∗	−0.44∗∗∗	−0.30∗∗∗
	(0.046)	(0.065)	(0.062)
Post−Confirmed case = 1 X Health & medicine	−0.03	−0.09	0.00
	(0.039)	(0.051)	(0.054)
Post−Confirmed case = 1 X Lifestyle & social	−0.16∗∗∗	−0.12∗	−0.19∗∗
	(0.047)	(0.056)	(0.064)
Post−Confirmed case = 1 X Maps, navigation, & travel	−0.54∗∗∗	−0.58∗∗∗	−0.50∗∗∗
	(0.049)	(0.071)	(0.066)
Post−Confirmed case = 1 X News	−0.35∗∗∗	−0.56∗∗∗	−0.21∗∗
	(0.054)	(0.088)	(0.064)
Post−Confirmed case = 1 X Shopping	−0.34∗∗∗	−0.56∗∗∗	−0.19∗∗
	(0.056)	(0.092)	(0.065)
In Lockdown = 1				0.23∗∗∗	0.28∗∗∗	0.19∗∗∗
				(0.027)	(0.033)	(0.042)
In Lockdown = 1 X Business & productivity				0.21∗∗∗	0.14∗∗	0.27∗∗∗
				(0.032)	(0.046)	(0.044)
In Lockdown = 1 X Education				0.16∗∗∗	0.08	0.21∗∗∗
				(0.035)	(0.053)	(0.046)
In Lockdown = 1 X Entertainment & music				−0.08∗	−0.10	−0.06
				(0.033)	(0.051)	(0.044)
In Lockdown = 1 X Food & drink				−0.14∗∗∗	−0.16∗	−0.12∗
				(0.042)	(0.067)	(0.053)
In Lockdown = 1 X Health & medicine				0.24∗∗∗	0.17∗∗	0.30∗∗∗
				(0.038)	(0.061)	(0.046)
In Lockdown = 1 X Lifestyle & social				0.04	0.06	0.02
				(0.035)	(0.053)	(0.044)
In Lockdown = 1 X Maps, navigation, & travel				−0.57∗∗∗	−0.55∗∗∗	−0.59∗∗∗
				(0.039)	(0.060)	(0.050)
In Lockdown = 1 X News				−0.02	−0.14	0.07
				(0.044)	(0.072)	(0.052)
In Lockdown = 1 X Shopping				−0.19∗∗∗	−0.35∗∗∗	−0.07
				(0.047)	(0.078)	(0.046)
Observations	509,760	205,320	304,440	509,760	205,320	304,440
R2	0.901	0.887	0.903	0.899	0.884	0.901
Additional controls:
Country−category X Month−of−the−Year interactions	Yes	Yes	Yes	Yes	Yes	Yes

In line with our earlier analysis in Section 3.2, we examine whether the end of the first lockdown periods coincided with differential adoption trends across the various app categories. We again restrict our time series to the period after the start dates of all countries in our sample who had a first lockdown wave and omit any countries that have had a second lockdown wave from this particular analysis to avoid confounding effects. We create a dummy indicator End lockdowns that turns on after the end date of the respective countries’ first lockdowns. We then run a model specification that includes interaction terms for the different app categories at the country level. The results are depicted in Table A.10 , and show that the finance app category emerges from the end of the first lockdown period with a positive differential relative to almost all other app categories. The main exception are apps in the “Maps, Navigation, and Travel” category, which had notably seen a very strong decline during the 1st lockdown period but seem to be coming back up to pre−lockdown levels. By contrast, finance category apps appear unique in having maintained large and steady growth both during and after the first lockdown period.

Table A.10.

Change in app adoption across lockdown waves, by broader category this table presents coefficient estimates for a panel regression model estimating the country−level relationship between the end of government lockdowns and the relative (logarithmic) number of daily app downloads. End of first lockdown is a country−level dummy indicator that turns on after the end date of each countries’ first lockdown period, where applicable. For this analysis, we restrict to the period after the start date of our sample countries’ first lockdowns. Countries which have had a second lockdown period as of December 9th, 2020 are excluded from this analysis to avoid confounding effects. We include interaction terms by broader app category to test for differential effects. We run separate specifications including all countries (ALL) in the sample, subset to advanced economies (AEs) and subset to emerging and developing economies (EMDEs). Standard errors are clustered at the country application level and in parentheses *** p < 0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, OxCGRT, and Aura Vision.

	DV = Relative (ln) daily app downloads			DV = Relative (ln) daily app downloads
Base level = Finance category apps	(All)	(EMDE)	(AE)	(All)	(EMDE)	(AE)
End of first lockdown = 1	0.07∗∗∗	0.08∗∗	0.06∗	0.07∗∗∗	0.08∗∗	0.06∗
	(0.017)	(0.024)	(0.025)	(0.017)	(0.024)	(0.025)
End of first lockdown = 1 X Business and productivity	−0.16∗∗∗	−0.15∗∗∗	−0.16∗∗∗	−0.16∗∗∗	−0.15∗∗∗	−0.16∗∗∗
	(0.020)	(0.027)	(0.028)	(0.020)	(0.027)	(0.028)
End of first lockdown = 1 X Education	−0.20∗∗∗	−0.17∗∗∗	−0.23∗∗∗	−0.20∗∗∗	−0.17∗∗∗	−0.23∗∗∗
	(0.029)	(0.041)	(0.039)	(0.029)	(0.041)	(0.039)
End of first lockdown = 1 X Entertainment and music	−0.14∗∗∗	−0.10∗∗∗	−0.18∗∗∗	−0.14∗∗∗	−0.10∗∗∗	−0.18∗∗∗
	(0.021)	(0.028)	(0.028)	(0.021)	(0.028)	(0.028)
End of first lockdown = 1 X Food and drink	−0.13∗∗∗	−0.13∗∗	−0.13∗∗∗	−0.13∗∗∗	−0.13∗∗	−0.13∗∗∗
	(0.030)	(0.050)	(0.036)	(0.030)	(0.050)	(0.036)
End of first lockdown = 1 X Health and medicine	−0.33∗∗∗	−0.34∗∗∗	−0.32∗∗∗	−0.33∗∗∗	−0.34∗∗∗	−0.32∗∗∗
	(0.022)	(0.031)	(0.031)	(0.022)	(0.031)	(0.031)
End of first lockdown = 1 X Lifestyle and social	−0.15∗∗∗	−0.14∗∗∗	−0.16∗∗∗	−0.15∗∗∗	−0.14∗∗∗	−0.16∗∗∗
	(0.020)	(0.028)	(0.028)	(0.020)	(0.028)	(0.028)
End of first lockdown = 1 X Maps, navigation, and travel	0.24∗∗∗	0.21∗∗∗	0.27∗∗∗	0.24∗∗∗	0.21∗∗∗	0.27∗∗∗
	(0.025)	(0.031)	(0.038)	(0.025)	(0.031)	(0.038)
End of first lockdown = 1 X News	−0.13∗∗∗	−0.16∗∗∗	−0.10∗∗∗	−0.13∗∗∗	−0.16∗∗∗	−0.10∗∗∗
	(0.026)	(0.042)	(0.029)	(0.026)	(0.042)	(0.029)
End of first lockdown = 1 X Shopping	−0.06∗∗	−0.08∗	−0.05	−0.06∗∗	−0.08∗	−0.05
	(0.024)	(0.035)	(0.033)	(0.024)	(0.035)	(0.033)
Observations	142,780	62,920	79,860	142,780	62,920	79,860
R2	0.993	0.987	0.996	0.993	0.987	0.996
Additional controls:
Country−app category X Month−of−the−Year interactions	Yes	Yes	Yes	Yes	Yes	Yes

Taken together, our result do indicate that the digital acceleration during COVID−19 was clearly not limited to the finance sector. Many other sectors of activity, notably including business and productivity, education, and health and medicine also saw large−scale increases in adoption. That having been said, apps in the financial sector show early signs of having a more sustained transformation in daily download rates relative to most other categories.

A.2 What types of countries experienced higher adoption due to COVID−19?

We examine how country−level characteristics explain differential finance mobile app adoption to explore other drivers of uptake during the pandemic period. We draw on known country−level determinants of fintech adoption from the literature to provide a benchmark for comparison. Specifically, we merge country−level characteristics on country development level (log of GDP per capita in $PPP), market size (log of country population), demographics (percentage of population over 65), access to ICT (percentage of the population with a mobile phone subscription), and ex−ante adoption of digital banking (as a proxy, we use the percentage of a country's population that reported regularly receiving wages using their mobile phones). We re−run Eq. (1) including these factors as interaction terms to test how they may explain country−level differences in fintech adoption during COVID. We focus on the relative results to avoid having countries with larger population size from purely driving the results.

Table A.11 presents results from these model specifications. In line with a number of prior studies and general intuition, we observe that country demographics appears to play a critical role in predicting dissemination and uptake of fintech during the pandemic as aging populations are strongly negatively associated with decreased adoption. Meanwhile, in contrast with other country−level studies on fintech adoption, we do not ultimately find that countries with higher economic development level (as proxied by GDP per capita) exhibit lower rates of adoption, nor is larger market size (proxied by country population) found to a strong predictor of fintech adoption during the pandemic. Moreover, we do not find particularly strong evidence that ICT access or ex−ante digital finance adoption is strongly associated with differential rates of adoption due to COVID−19.

Table A.11.

Country−level characteristics predicting differential effects of COVID on finance app adoption this table presents coefficient estimates for a panel regression model estimating the country−level relationship between the spread of COVID−19 on the relative (logarithmic) daily change in downloads for finance category mobile apps. Post−confirmed case denotes a dummy indicator that turns on after a given country saw its first confirmed COVID−19 case. In lockdown denotes a dummy indicator that turns on and off during each countries’ lockdown period(s). The data sample covers the android and iOS mobile finance app markets for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. All models include country X month−of−the−year interaction dummies to control for local seasonality. Standard errors are clustered at the country level and in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. Data sources: AppTweak, OxCGRT, Aura Vision, and WDI.

	DV = Relative (Ln) daily app downloads
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
Post−Confirmed case = 1	1.148∗	0.607	0.620∗∗∗	0.284∗	0.289∗∗∗
	(0.591)	(0.438)	(0.070)	(0.143)	(0.052)
Post−Confirmed case = 1 X ln(GDP per capita)	−0.084
	(0.059)
Post−Confirmed case = 1 X ln(Country pop.)		−0.019
		(0.026)
Post−Confirmed case = 1 X Pop. % age 65 and up			−0.025∗∗∗
			(0.005)
Post−Confirmed case = 1 X Pop. % w. mobile subscription				0.000
				(0.001)
Post−Confirmed case = 1 X Pop. % w ex−ante received wages via mobile phone					−0.003
					(0.007)
In−Lockdown = 1						0.595	0.261	0.623∗∗∗	0.188	0.287∗∗∗
						(0.675)	(0.455)	(0.079)	(0.173)	(0.046)
In−Lockdown = 1 X ln(GDP per capita)						−0.029
						(0.068)
In−Lockdown = 1 X ln(Country pop.)							0.002
							(0.026)
In−Lockdown = 1 X Pop. % age 65 and up								−0.024∗∗∗
								(0.005)
In−Lockdown = 1 X Pop. % w. mobile subscription									0.001
									(0.002)
In−Lockdown = 1 X Pop. % w ex−ante received wages via mobile phone										−0.003
										(0.008)
Observations	47,436	48,144	48,852	48,144	47,436	47,436	48,144	48,852	48,144	47,436
R2	0.979	0.979	0.981	0.979	0.979	0.974	0.974	0.975	0.974	0.974

Taken together, we interpret these overall results to suggest that, from a country−level perspective and for the time being, COVID−19 does not appear to have significantly exacerbated existing differences in digital finance adoption between wealthier or poorer countries nor those with higher or lower ex−ante adoption. This may be in part because the shock from COVID has been truly at a global scale, and forced major policy and behavioral changes in almost all countries. For these reasons, we focus most of our main analyses on exploring product− and provider−level factors that help explain the adoption of apps.

A.3 Does the spread of COVID−19 or government policy drive adoption?

Our overall main results are consistent in showing that a positive and significant relationship exists between the spread of COVID−19 and finance−related app downloads. However, the shock due to COVID−19 is quite unique with many different behavioral and policy effects timed quite closely and likely overlapping each other. The effects we observe may be due to the direct spread of COVID−19 (first−order effects), the lockdowns instituted by various governments (second−order effects), or more protracted economic shocks from job losses, business closures, etc. arising from the lockdowns over time (third−order effects).

Table A.12 depicts results for country−level specifications where we try to disentangle the first− and second−order effects—i.e., the spread of the pandemic and the government lockdowns—and, furthermore, to what extent the severity of cases or policies matters. To do so, we add in combinations of the various pandemic and government spread and intensity variables, to see what insights we can gain. There is considerable overlap between the outcomes due to the often close timing of COVID−19 and government lockdowns and related spikes in cases and policies. Hence, any evidence presented, is suggestive at best. Nevertheless, our results do suggest that the adoption in fintech, in relative and absolute per capita terms, was both driven both as a behavioral response to the onset of COVID−19 cases / deaths and separately as a reaction to government policies. For example, in our models in columns 4 and 9, which include separate intensity variables for the onset of lockdowns and the severity of confirmed deaths, we see signs of independent effects for the respective variables that are both statistically and economically meaningful. More generally, our other specifications yield significant effects for both government policy stringency and lockdown variables (columns 1, 2, 5, 6, 7, and 10) and confirmed cases and deaths (columns 3, 4, 8, and 9). That is, as might be expected, the variables capturing intensity of either government policies, cases, or deaths, typically dominate the effect in terms of economic and statistical significance, which highlights that the adoption is sensitive to the severity of COVID in a given location.

Table A.12.

Was COVID−19′s effect on finance app adoption driven by the public health or government policy shock? This table presents coefficient estimates for a panel regression model estimating the country−level relationship between the spread of COVID−19 on the relative (logarithmic) and per capita changes in daily downloads for finance category mobile apps. Post−confirmed case, Post−confirmed death and in lockdown denote dummy indicators that are turned on after the first confirmed COVID−19 case, death, and during government−initiated lockdowns, respectively. Meanwhile, we estimate whether the intensity of COVID−19 cases, deaths or government policy also factors into rates of adoption through variables measuring the number of confirmed cases, deaths, and index score for government policy (transformed to logarithmic form). The data sample covers the android and iOS mobile finance app markets for a globally representative sample of countries daily from January 1st, 2019 to December 9th, 2020. All models include country X month−of−the−year interactions to control for local seasonality. Standard errors are clustered at the country level and in parentheses. *** p < 0.01, ** p< 0.05, * p < 0.10. Data sources: AppTweak, OxCGRT, and Aura Vision.

	DV = Relative (log) of # daily app downloads					DV = # daily app downloads per capita
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
Post−Confirmed case	−0.029					0.00001
	(0.043)					(0.000)
Post−Confirmed deaths		0.034					−0.00008
		(0.032)					(0.000)
ln(Government stringency index)	0.082∗∗∗	0.068∗∗∗			0.06189∗∗∗	0.00016∗∗∗	0.00018∗∗			0.00006
	(0.011)	(0.011)			(0.013)	(0.000)	(0.000)			(0.000)
In Lockdown			0.072∗∗	0.118∗∗∗				0.00016∗∗	0.00027∗∗∗
			(0.032)	(0.033)				(0.000)	(0.000)
ln(Number of confirmed cases)			0.027∗∗∗		0.022			0.00006∗∗∗		0.00012∗∗
			(0.004)		(0.016)			(0.000)		(0.000)
ln(Number of confirmed deaths)				0.036∗∗∗	−0.026				0.00008∗∗∗	−0.00012∗
				(0.006)	(0.020)				(0.000)	(0.000)
Observations	50,976	50,976	50,976	50,976	50,976	50,976	50,976	50,976	50,976	50,976
R2	0.980	0.980	0.979	0.979	0.980	0.893	0.893	0.891	0.886	0.895

Taken as a whole, we thus interpret the results to confirm that there were different mechanisms simultaneously influencing the increased uptake. While we are left with somewhat ambiguous findings regarding whether responses to the pandemic's actual spread or government policies ultimately dominate the effect, we do observe that the intensity of the pandemic's spread and government responses intuitively matter in either case.

A.4 Additional figures

Fig. 4, Fig. 5, Fig. 6, Fig. 7, Fig. 8, Fig. 9, Fig. 10

Fig. 4.

Daily per capita downloads of finance category mobile apps for countries most affected by COVID−19 in terms of per capita deaths. This figure depicts the daily rate of per capita downloads for finance category mobile applications for countries with the most confirmed per capita deaths from COVID−19 (i.e., as of December 9th, 2020). This includes Belgium, Peru, Italy, Slovenia, Spain, United Kingdom, United States, Argentina, Czech Republic, and Mexico. (Note: Bosnia and Herzegovina, North Macedonia, and Moldova rank among the top 10 countries in per capita deaths from COVID, but we do not have their download data.) Country codes are listed in Appendix Table A.1. The data includes mobile apps from both the Android and iOS platforms from October 1st, 2019 to December 9th, 2020. We use a 14−day leading moving average on downloads to smooth day−to−day fluctuations.

Fig. 5.

Daily per capita downloads of finance category mobile apps for countries most affected by COVID−19 in terms of absolute deaths. This figure depicts the daily rate of per capita downloads of finance category mobile applications for countries with the most confirmed deaths from COVID−19 in absolute terms (i.e., as of December 9th, 2020). This includes the United States, Brazil, India, Mexico, Great Britain, Italy, France, Spain, Russia, and Argentina. (Note: Iran also ranks among the top 10 countries in terms of absolute number of deaths from COVID, but we do not have their download data.) Country codes are listed in Appendix Table A.1. The data includes mobile apps from both the android and iOS platforms from October 1st, 2019 to December 9th, 2020. We use a 14−day leading moving average on downloads to smooth day−to−day fluctuations.

Fig. 6.

Scatterplot of finance category mobile app downloads per capita vs. confirmed COVID− 19 cases per capita. This figure depicts the relationship between countries’ average daily confirmed new cases of COVID−19 per capita and average daily number of downloads per capita for finance category mobile applications. We roughly restrict to the period after the first signs of the COVID−19 outbreak (i.e., since January 1st, 2020) and then take country−level averages for both. The underlying data includes a globally representative sample of mobile apps from both the android and iOS platforms. Country codes are listed in appendix Table A.1.

Fig. 7.

“Placebo” scatterplots of finance mobile app downloads vs. COVID−19 spread. Fig. 2 and 6 show a preliminary but intuitive positive relationships between the intensity of COVID's spread and levels of finance app downloads (even adjusting for per capita terms). In this figure, we set up a placebo analysis to support that our findings capture a real relationship. We replace the y−axis with the same variable on the corresponding date of the year 2019, i.e., we use the download levels on February 1st, 2019 to replace February 1st, 2020, February 2nd, 2019 to replace February 2nd, 2020, and so forth. We then calculate the average downloads per capita for the countries across time and rerun the scatterplots against indicators of COVID spread per capita also averaged across time. The underlying data includes a globally representative sample of mobile apps from both the android and iOS platforms. Country codes are listed in appendix Table A.1. Data sources: AppTweak and OxCGRT.

Fig. 8.

Event study model of finance category mobile app downloads relative to time of con− firmed COVID cases. This figure depicts results from an event study model estimating levels of finance mobile app downloads (in logarithmic form) relative to the time of 1st confirmed COVID cases. We set the date of the first confirmed case as t = 0 for each country and then calculate their average download levels by week from t = −52 to t = +36. We collapse the data at the country−week level and run a regression based model similar to those depicted in our analysis in Table 2. That is, it includes a general post dummy indicator that turns on after first confirmed COVID−19 case worldwide, as well as country and month−of−the−year interaction terms. The weekly dummy from before the date of the first confirmed case is omitted as the base level. The underlying data includes a globally representative sample of mobile apps from both the android and iOS platforms.

Fig. 9.

Event study model of finance mobile app downloads relative to end of first lockdowns. This figure depicts results from an event study model testing for trend breaks before and after the end of the first lockdown waves. Specifically, we set as t=0 the respective end dates of the first lockdowns for each country and then calculate their average download levels by week from t = −12 to t = +24. We collapse the data at the country−week level and run a regression based model similar to those depicted in our analysis in Table 3, Panel a, including country and month−of−the−year interaction terms. The weekly dummy from before the date of the first lockdown end date is omitted as the base level. The underlying data includes a globally representative sample of mobile apps from both the android and iOS platforms.