The role of online platforms in weathering the COVID-19 shock

Abstract

This policy brief reports on the activity of online platforms during the COVID-19 crisis. Google Trends data for OECD and other G20 countries indicate that in some areas (such as retail sales, restaurant delivery, and mobile payments) online-platform use increased markedly during the first half of 2020, when most countries imposed lockdown and physical distancing measures. Thus, in this period, some economic transactions may have shifted to online marketplaces as people and businesses increasingly turned to online platforms to pursue economic and social activities. The rise in platform use was however highly heterogeneous across areas of activity and countries. Countries with higher levels of economic and technological development, easier access to infrastructure and connectivity, better digital skills, and wider Internet use tended to experience a larger increase in the use of online marketplaces, possibly mitigating the negative effects on output and jobs of the COVID-19 shock. This highlights the role of policies in strengthening countries’ digital preparedness and their resilience to future shocks.

This is an updated version of the Policy Brief dated 16 Dec 2020. The only changes concern the labels of Figure 2.

Introduction

The COVID-19 pandemic and the measures implemented to flatten the epidemic curve have jolted economies worldwide. In most OECD and G20 countries, data indicate that in the first half of 2020 lockdowns and mobility restrictions caused the largest fall of real GDP in recent history. Industries requiring physical proximity for product or service delivery or among co-workers – such as accommodation, restaurants and transport – were those hardest hit (OECD, 2020[1]).

Amidst the economic mayhem caused by the pandemic, digital technologies have emerged as a key element of economic and social resilience. By allowing businesses and governments to continue operating during lockdowns, digital technologies enabled faster economic responses to the emergency. Evidence is emerging on how the COVID-19 crisis triggered large changes in the use of digital technologies by people, businesses and governments. For instance, the demand for broadband communication services has soared, with some operators experiencing as much as a 60% increase in Internet traffic (OECD, 2020[2]). Online shopping, videoconferencing and telework have surged whereas governments have accelerated the shift towards the digitalisation of many public services. As a result, the profits and market capitalisation of digital companies whose services experienced a large increase in demand during lockdowns increased to record levels and far ahead of other companies.1

Among digital technologies, online marketplaces were already developing fast before the COVID-19 crisis (OECD, 2020[3]). Due to innovative use of data collected from users and large network externalities, they had been capturing a growing (though still limited) share of activity in many sectors – such as retail marketplaces, accommodation, transport, restaurants and increasingly in business-to-business (B2B) and professional services. Their fast rise raises several concerns spanning multiple policy areas, such as competition, labour market regulation, taxation, privacy and consumer protection. Understanding the increasing reliance of people and businesses on online platforms during the crisis is particularly important as there is no close parallel in recent history to draw insights from.2

This policy brief contributes to the debate on the diffusion of online platforms by investigating how the COVID-19 shock impacted on their activities. It relies on a comprehensive list of about 1400 online platforms covering nine economic areas (including accommodation, catering, marketplaces, B2B activities, transport and professional services) and all OECD and G20 countries (excluding China and Colombia). Online platform activity is proxied by Google Trends data, which provide real time information on the intensity of search for keywords linked to platform names. Studies have corroborated the reliability of Google Trends search data to track economic time series in real time (see Boxes 1 and 2). We look at the yearly change of online-platform activity in the first six months of 2020, thus covering the period in which most countries implemented the most severe lockdown and mobility restriction measures. We distinguish among online platforms in different activity areas and explore which pre-crisis country characteristics and policies may have eased the shift of activity towards online platforms.

Key findings

• The lockdown and physical distancing measures imposed in many countries during the first half of 2020 contributed to shift some economic activity towards online platforms.

• Changes in online-platform use (as proxied by Google Trends data) were very uneven across platforms’ activity areas. In areas not requiring physical proximity for product and service delivery (such as mobile payments, marketplace to consumers, professional services and restaurant delivery), online-platform use increased by about 20% as platforms allowed businesses and households to continue producing and working during lockdowns. In other areas, which require physical proximity (such as accommodation, restaurant booking and transport), platform activity declined markedly (by around 70%), reflecting the generalised economic disruption caused by the COVID-19 pandemic.

• The containment measures introduced to control the pandemic explain part of the large variation in online-platform activity across countries. Simple model simulations show that moving from no containment measures to the strictest ones would have increased the growth rate of online-platform activity (in those areas not requiring physical proximity) by 15 percentage points.

• The increase in online-platform use varied across countries depending on pre-existing structural conditions and policies. The increase tended to be larger in more developed and technologically-advanced countries, those with easier access to infrastructure and connectivity, higher skill levels, and more widespread use of the Internet.

• Online platforms can play a role in mitigating the negative effects of future shocks causing severe disruptions to physical economic activity. They can ensure, at least in some sectors, continuity in production and market exchanges while respecting physical distancing measures.

• Policies to improve access to infrastructure and connectivity, those directed at improving digital skills and spreading the use of internet can accelerate the diffusion of online platforms (along with other digital technologies) and enhance countries’ digital preparedness for future shocks.

Gauging online-platform activity during the COVID-19 crisis

Despite the rising importance of online platforms in developed and emerging economies, official measurement of their activities and operations is still incomplete and inadequate (OECD, 2018[4]; OECD, 2019[5]). Efforts to measure online-platform activity are still piecemeal, often focused on a specific subset of platforms, and the available data are often not comparable across countries because of different collection methods. Different initiatives at national and international levels are ongoing to overcome these problems and collect more comprehensive data that is comparable across countries (OECD, 2018[4]; OECD, 2020[6]; OECD, WTO and IMF, 2020[7]).

Previous studies focusing on online platforms have relied on a variety of approaches to deal with data limitations. These include case studies (OECD, 2019[8]), proprietary data (Farronato, Fong and Fradkin, 2020[9]) and the number of internet searches for online-platform names as a proxy for platform use (Bailin Rivares et al., 2019[10]). These approaches are complementary and have different pros and cons. Case studies allow for a detailed comparative analysis of the functioning of different types of online platforms, taking into consideration also qualitative information, but their coverage is necessarily limited (often to the largest platforms). Proprietary data obtained from online platforms allow for precise analyses of market transactions (effects on prices and quantities), but they ultimately reflect the activity of a single online platform and a narrow area; moreover these data are confidential and difficult to get. Internet search data allow for investigating the use of a large number of online platforms in a consistent way across countries and are now widely available; however, data are limited to the use of platforms (no other information concerning online platforms is available) and internet searches are just a proxy for platform use that also depend on the popularity of the search engine and its change over time. Also, internet searches do not capture the download and use of apps on mobile phones.3

In this policy brief we follow the third approach and use Google Trends data on a list of 1400 platforms covering OECD and G20 countries (see Box 1). We collect monthly data and focus on the year-on-year change for the first six months of 2020. Google Trends data have been used extensively to detect sudden changes in macroeconomic junctures (see Box 2).

Box 1. Data collection

We define online platforms as online services that facilitate interactions between two or more distinct but interdependent sets of users (either firms or individuals) (OECD, 2019[8]) and restrict our attention to platforms that facilitate the exchange of goods or services. We adopt the approach of OECD (2020[3]) and focus on platforms across nine areas of activity where the importance of platforms has been growing in the past years: accommodation, marketplace to consumers, marketplace business-to-business, mobile payments, personal services, professional services, restaurant booking, restaurant delivery and taxi services.

We use the list of platforms assembled used in OECD (2020[3]) for 43 OECD and G20 countries. The list was created using Crunchbase (a company providing business information on private and public innovative companies on a global scale) to search for the top 30 competitors by country and area of activity using specific keywords for each area. The list was then adjusted after consultations with country experts, revisions of news articles and other nation-specific sources and inspection of all platforms’ websites, by deleting platforms that did not conform to the definition, and adding new ones that had originally been missed. The final list includes 1407 platforms, both large and international (such as Airbnb and Uber) but also smaller and domestic (like Leboncoin or Madeinportugal). A detailed description of the dataset is available in OECD (2020[3]).

As regards internet searches, we follow previous studies (Bailin Rivares et al. (2019[10])) and use the Google Trends API (Application Programming Interface) to get monthly indices of search volumes for each platform in a given geographic area and time period. This provides information for all countries where the Google search engine is used between 2004 and the present. Because Google provides all data on a normalised scale comparable only between search terms in the same query, we work with year-on-year growth rates of search intensities.

Box 2. Google Trends as a measure of socioeconomic outcomes

A growing body of research has used Google Trends data to forecast or nowcast socioeconomic indicators, bridge time lags and improve forecasting based on more standard datasets. Most analyses based on searches focus on labour market and consumption behaviours, since these areas are most strictly linked to household internet searches. Studies show that Google searches outperform survey-based data when forecasting US private consumption (e.g. Vosen and Schmidt (2011[11]) and Woo and Owen (2019[12])). Focusing on tourism flows in Switzerland, Siliverstovs and Wochner (2018[13]) find Google Trends to be an accurate predictor of tourism demand, and conclude that search-based indicators may serve as valuable real-time complements for the guidance of economic policy.

Several studies have also explored the forecasting power of Google Trends data on unemployment (Smith, (2016[14]) for the United Kingdom, and D’Amuri and Marcucci (2017[15]), and Maas (2020[16]) for the United States). Applications to GDP nowcasting include Woloszko (2020[17]) for OECD countries, Götz and Knetsch (2019[18]) for Germany, and Ferrara and Simoni (2019[19]) for the Euro area. Chen et al. (2015[20]) use search data to detect business cycle turning points and peak dates for the United States.

Additional areas of research using Google Trends data include finance. Predictions of stock prices and their volatility follow the seminal contribution by Da, Engelberg and Gao (2011[21]). More recently, Aouadi, Arouri and Roubaud (2018[22]) find international evidence in favour of the role of information demand (proxied by daily search volumes) for stock market liquidity. Castelnuovo and Tran (2017[23]) use Google Trends data for the United States and Australia to construct an uncertainty index. Based on these data and exploiting the COVID-19 shock, Amstad et al. (2020[24]) build a new index capturing investors’ attitudes toward pandemic-related risks. In the same spirit, Eichenauer et al. (2020[25]) propose a novel method for using Google Trends data to construct daily sentiment indicators for a set of German-speaking countries.

As regards the use of online platforms, Harris and Krueger (2015[26]) use Google Trends data to proxy for the number of workers in several platforms in three areas (taxi, personal services, and professional services) in the United States. Their estimates are broadly in line with estimates from McKinsey using different sources (McKinsey Global Institute, 2015[27]).

We combine these data with information from the OECD COVID-19 Policy Tracker, a database to track government responses to the COVID-19 crisis that is updated daily (Bulman and Koirala, 2020[28]). The Policy Tracker provides information on the measures taken by governments in response to epidemic outbreaks. We focus on containment policies introduced to limit human interactions through confinement or stay-at-home orders, school closures, and restrictions on international and/or within-country travel. The severity of these measures is summarised by an index ranging from zero (no measure taken) to one (maximum level of containment) in a consistent way across countries. It is composed by different sub-indexes: cancellation of public events, obligatory shutdown of the economy, confinement lockdowns, travel restrictions and closure of schools. As alternative measures of the severity of the epidemics, we use the Oxford COVID-19 Government Response Tracker’s Stringency Index (Petherick et al., 2020[29]), Google mobility data, and the number of new COVID-19 cases.

Online platforms during the COVID-19 crisis

During the time span considered (from January to June 2020), the containment index of the OECD Policy Tracker followed a similar pattern across most of countries in the dataset: it started increasing around late February and early March, peaked in April, and began to decline in May. The severity of containment measures was however rather heterogeneous across countries. Figure 1 shows for all countries in the dataset the maximum value and the average value of the containment index during these six months. We identify four groups of countries based on the maximum level attained by the containment index during the six-month period covered: soft containment, medium-soft, medium-strict, and strict containment measures.

Figure 1. Countries’ responses to the COVID-19 pandemic differ

OECD containment index, January – June 2020

Note: The OECD containment index captures the measures countries have undertaken to limit human interactions through confinement or stay-at-home orders, school closures, and restrictions on international and/or within-country travel. It ranges from zero to one.

Source: OECD.

The average yearly change of online-platform activity (as gauged by Google Trends monthly data) was heterogeneous across the nine activity areas considered and the four country groups (Figure 2). Online platforms operating in activity areas relying on physical proximity for product and service delivery (namely restaurant booking, accommodation and transportation) experienced a marked drop in activity. Platform activity started to drop as countries began to implement lockdown measures and reached an all-time low when the containment index peaked (see Figure 1.A.1 in the Annex also). For instance, as demand for hotel and restaurant services collapsed during lockdowns, online platforms active in accommodation and restaurant booking might have experienced similar large drops in activity to these businesses (many of which actually closed, at least temporarily). Online-platform activity started to recover gradually when containment measures were progressively eased over May and June 2020. Countries that imposed soft lockdown measures also experienced major adjustments in consumer behaviour, and large variation in online-platform use, pointing to the role that voluntary mobility restrictions may have played, in addition to government-mandated restrictions. However, in strict-containment countries the recovery in online-platform use seems to have been weaker than in soft-containment countries, reflecting probably the deeper and long-lasting economic disruptions these countries suffered.

Figure 2. During lockdowns online-platform activity expanded in some areas and declined in others

Note: The lines show the median of the year-on-year percentage change of activity (proxied by Google Trends data) of about 1 400 platforms (see Box 1); the period covered is January-June 2020. Activity areas not requiring physical proximity are mobile payments, marketplace to consumers, professional services and restaurant delivery. Activity areas requiring physical proximity are restaurant booking, accommodation, and transportation. Activity areas with mixed physical proximity requirements are personal services and marketplace for business-to-business transactions. See Figure 1A.1 in the Annex for a decomposition across areas.

Source: Google Trends data and OECD calculations.

The activity of online platforms operating in activity areas requiring less or no physical proximity for product or service delivery (mobile payments, marketplace to consumers, professional services and restaurant delivery) rose as lockdown measures became stricter. The increase in activity of online platforms in strict-containment countries appears to be more persistent than in soft-containment countries. Stricter containment measures may have induced more long-lasting changes in how people and businesses use online platforms and digital technologies more generally.

Online platform activities in the business-to-business and personal services areas were mostly flat during the first half of 2020 (Figure 2). This may be due to the heterogeneity of the related modes of product delivery, which may require vastly different degrees of physical proximity. Business-to-business activities cover in principle all industrial sectors and are difficult to classify. Online platform activity in this area is still small but it is evolving fast (Ifo, 2020[30]).Personal services platforms involve transactions that can be performed online (such as tutoring) and transactions for which physical proximity is necessary (such as hairdressing or house cleaning services).

Overall, focusing on the expanding areas (mobile payments, marketplace to consumers, professional services and restaurant delivery), the distribution of the change in online-platform activity in strict-containment countries is to the right of that of the others (Figure 3). This suggests that in these countries online-platform use increased more rapidly than in countries adopting softer containment measures.

Figure 3. Online-platform activity expanded more strongly in countries adopting stricter containment measures

Probability density of year-on-year percentage change of online platform activity (proxied by Google Trends data), January-June 2020

Note: The horizontal axis shows the average year-on-year percentage change of activity (proxied by Google Trends data) of online platforms for the period in which containment measures have been in place (i.e. the OECD containment index was above zero). The following activity areas are considered: mobile payments, marketplace to consumers, professional services and restaurant delivery. The vertical axis shows the probability density of the distribution of the year-on-year percentage change of online-platform activity.

Source: Google Trends data and OECD calculations.

The descriptive statistics above suggest that in some activity areas (those requiring little or no physical proximity for product and service delivery) the use of online platforms rose with the start of lockdowns and mobility restrictions. Businesses and people increasingly turned to online platforms to pursue economic and social activities that could no longer take place in the same form as before. In other activity areas (those requiring physical proximity for product and service delivery) online platforms experienced a large decline in activity, in line with the traditional businesses operating in the same areas. Here, online platforms could not offer businesses and people alternative ways of accessing these services while complying with government COVID-19-related restrictions. The response of platform use to the crisis seems to be related to the severity of containment measures.

The change in online-platform traffic was however highly heterogeneous even across countries that adopted similar containment measures. This warrants investigation on the role possibly played by differences in the intensity of lockdown measures (and the extent to which they were enforced) while accounting for country characteristics and policies that may be correlated with online-platform adoption and uptake. We therefore turn to regression analysis, which we perform in two steps. First, we look at the link between the strictness of lockdown measures and changes in platform use (Box 3). To capture the behavioural changes due to perceived health risks and the degree of enforcement of lockdown measures, we also use, as alternatives to the OECD containment index, the stringency index of the Oxford COVID-19 Government Response Tracker, the number of new COVID-19 cases and the Google Mobility index. Second, we explore the influence of pre-existing countries’ characteristics and policies on changes in online-platform use during the pandemic.

Box 3. Linking lockdown and country characteristics to platform use during the pandemic: empirical methodology

Our analysis relies on monthly data, from January to June 2020, and covers 43 countries. To investigate the link between online-platform use and COVID-19-related lockdown measures, we estimate the following model:

${∆PlatformUse}_{p,s,c,t}={{\beta }_{1}Lockdown}_{c,t}+ {\theta }_t+{\delta }_{s,c}+{\epsilon }_{p,s,c,t}$ (1)

Where ${∆PlatformUse}_{p,s,c,t}$ is the log change in the search volume index for platform p between 2020 and 2019, in area s, country c, and in month t. ${Lockdown}_{c,t}$ is our main variable of interest, and measures the stringency of the containment in country c in month t. The containment index is based on measures taken by countries to limit human interaction, such that a larger value indicates more physical distancing and less mobility, and estimated by the OECD, as described above.

As an alternative measure of the stringency of lockdowns we use the Oxford COVID-19 Government Response Tracker’s Stringency Index (Petherick et al., 2020[29]). The index is available at the daily level, and we aggregate it to monthly values. As we are interested in actual mobility restrictions due to the severity of the epidemics, we also use the change in the Google mobility index, which captures the change in the number of visits to different locations and calculates the difference relative to a baseline before the pandemic outbreak.4 The index is available daily at the country level, and we aggregate it to obtain monthly values. Finally, we also use the monthly number of new COVID-19 cases in each country (per million of people), collected by the OECD from several country-specific sources. As these three variables are highly correlated (Annex Table 1.A.4), we use them as substitutes in regressions.

We control for cross-country shocks, such as news of international outbreaks or scientific knowledge of the virus, with monthly fixed-effects, ${\theta }_t$, and for time-invariant characteristics of areas and countries, like existing levels of income, health care, or infrastructure, with area-country fixed effects, ${\delta }_{s,c}$. The latter are important, as many of these characteristics are likely to be correlated with platform use across countries and areas. Further confounders that may vary at the country-month level, such as increased broadband capacity, are difficult to control for, as data are not usually available, and country-month fixed effects do not allow to estimate the coefficient for the containment index (as they would wipe out all of its variation). However, most of these omitted drivers tend to evolve slowly; so we expect the effect of confounding factors of this sort to be minimal. Finally, ${\epsilon }_{p,s,c,t}$ is the error term.

We estimate model (1) through the methodology for linear models with multi-way fixed effects developed by Correia (2016[31]). We adjust standard errors by clustering at the platform-sector-country level.

Overall, considering all online platforms together, containment measures were negatively associated with the use of online platforms (Table 1 column 1). We investigate the differential effects of the COVID-19 shock on platforms operating in different areas below. Using alternative measures capturing the severity of the COVID-19 crisis and the behavioural changes it caused yields similar results (Table 1 columns 2-4). As described in Box 1, we use the Oxford stringency index, the change in the Google mobility index (where positive changes in the Google mobility index imply higher mobility and vice-versa), and the number of new COVID-19 cases per million people. These measures are highly correlated (Annex Table 1.A.4) and are capturing to a large extent the same phenomenon: the behavioural changes of people and businesses during the COVID-19 pandemic. From now on, we focus on the OECD containment index, but using the alternative measures produces similar results.

Table 1. The COVID-19 shock and online-platform use

	(1)	(2)	(3)	(4)
Dependent variable	Change in platform use
Containment index (OECD)	-0.150***
	(0.0339)
Stringency indicator (Oxford)		-0.000718**
		(0.000352)
Google mobility change			0.00128***
			(0.000360)
Number of cases				-0.00606**
				(0.00256)
Constant	-0.0332***	-0.0519***	-0.0851***	-0.0776***
	(0.0116)	(0.0154)	(0.00695)	(0.0112)
Observations	31,083	31,083	25,495	25,646
R-squared	0.148	0.147	0.181	0.183

Note: The dependent variable is the year-on-year logarithmic change of the Google Trends searches for online platforms. The table shows the results of regression (1) in Box 3 ${∆PlatformUse}_{p,s,c,t}={\beta }_1{Lockdown }_{c,t}+ {\theta }_t+{\delta }_{s,c}+{\epsilon }_{p,s,c,t}$ with the Lockdown variable being in turn the OECD containment index, the Oxford stringency indicator, the google mobility change and the logarithmic of the number of new cases. Increases in the Google mobility change means higher mobility and vice-versa. All regressions include month and country-area fixed effects. Standard errors are clustered at the level of the platform-sector-country.

Source: OECD estimations based on data from the OECD, Google Trends, Google Mobility, Oxford Covid-19 Government Response Tracker, and Crunchbase.

The point estimates reported above may hide the heterogeneous dynamics of online-platform use across different areas. Table 2 shows the estimated coefficients of the OECD containment measures distinguishing between three distinct types of platforms: those that operate in areas requiring physical proximity for product and service delivery; those that do not have such a requirement; and those that have mixed physical proximity requirements because of the great heterogeneity of products they provide.5

Results indicate that online-platform activity increased in areas not requiring physical proximity whereas it decreased in areas requiring physical proximity, even after controlling for country and area fixed effects (Table 2). The point estimates suggest the effects are sizeable. Moving from no containment measures to the strictest measures might have increased the yearly growth rate of online-platform activity in areas not requiring physical proximity by about 15 percentage points and lowered it by 70 percentage points in areas requiring physical proximity. As regards platforms operating in areas with mixed physical proximity requirements (i.e. personal services and business to business), the impact of the containment measures on activity is close to zero and not statistically significant. This is likely to reflect the large heterogeneity of platforms operating in these sectors. Finally, Figure 4 shows the changes in platform use associated with changes in the containment index for each activity area separately (see Table 1.A.1 in Annex 1 for the full results)6. These confirm the results at a finer level, pointing out that the increase in platform activity is driven by three areas: mobile payments, marketplace to consumers and professional services.

Table 2. The COVID-19 shock impacted online-platform use differently across activity areas

Dependent variable	Change in platform use
OECD containment index * (areas not requiring physical proximity)	0.1385***
	(0.031)
OECD containment index * (mixed physical-proximity areas)	0.0020
	(0.039)
OECD containment index * (areas requiring physical proximity)	-1.1360***
	(0.048)
Constant	-0.048***
	(0.012)
Observations	31,083
R-squared	0.20

Note: The dependent variable is the year-on-year logarithmic change of the Google Trends searches for online platforms. The table shows the results of regression (1) in Box 3 distinguishing between activity areas requiring physical proximity or not. Areas requiring physical proximity are accommodation, restaurants and transport. Areas not requiring physical proximity are marketplace to consumers, mobile payments, professional services, and restaurant delivery. Mixed physical-proximity areas are personal services and business to business marketplaces. All regressions include month and country-area fixed effects. Standard errors are clustered at the level of the platform-sector-country.

Source: OECD estimations based on Google Trends and Crunchbase data.

The evidence therefore suggests that behavioural changes during the COVID-19 shock shifted some economic activity towards online platforms operating in areas not requiring physical proximity. This may have then mitigated the negative impact of the COVID-19 shock on output and jobs as online platforms allowed economic and social activities to continue while complying with lockdown and physical distancing rules. Anecdotal evidence supports this hypothesis: e.g., visits to Amazon’s worldwide marketplaces rose during the pandemic, reaching a record of 5.4 billion in August 2020 and its marketplace was the fastest growing of its businesses7; in 2020Q2, the number and value of Paypal transactions recorded the largest increase since 2014.8

However, the activity of online platforms operating in areas requiring physical proximity were strongly depressed by the COVID-19 shock. For instance, bookings through Airbnb dropped by more than 90% in January-March 2020 compared with one year before9. While Uber’s revenues from mobility services (i.e. taxi) dropped by more than 60% in 2020Q2 compared with the previous year, those from restaurant delivery rose by 170%10.

Figure 4. Lockdown measures have impacted platform traffic differently across activity areas

Note: The bars represent the estimated coefficient of the lockdown indicator interacted with area dummies using regression (1) in Box 3:${∆PlatformUse}_{p,s,c,t}={{\beta }_{1}Lockdown}_{c,t}+{\beta }_2{Lockdown}_{c,t}*{Area}_s+ {\theta }_t+{\delta }_{s,c}+{\epsilon }_{p,s,c,t}$. The dependent variable is the year-on-year logarithmic change of the Google Trends searches for online platforms. Full results are in column (3) Table 1A.1. The lines in the graph are the 95% confidence interval. The x-axis is in logarithmic scale (-1 is equivalent to a fall of online platform activity, as proxied by Google Trends searches, of 63%). Lockdown indicates the OECD containment index. The estimation includes country-area and month fixed effects, and standard errors are clustered at the platform-sector-country level.

Source: OECD estimations based on Google Trends and Crunchbase data.

Which structural and policy characteristics are associated with increases in online-platform activity during the COVID-19 crisis?

The response of online-platform use to the COVID-19 shock may vary across countries depending on their degree of digital preparedness. In digitally advanced countries, past policies and pre-crisis conditions might have made it easier for businesses and people to start using online platforms (and more generally to adopt digital technologies to support business activity) when lockdowns were imposed.

Several country-specific structural characteristics and policies are associated with the diffusion of online platforms, by increasing both platform usage and the number of active platforms (OECD, 2020[3]). Here we report preliminary findings for three main areas: 1) technological and economic development (proxied by an index of the availability of latest technologies, a measure of digital adoption, and the level of GDP per capita); 2) access to digital infrastructure (proxied by a measure of mobile broadband penetration and an indicator of restrictions on connectivity and access to infrastructure); 3) digital literacy (proxied by a digital skills readiness index and the share of individuals using the Internet to pay bills). Annex Table 1.A.2 provides details on the measurement and coverage for these variables. The characteristics and policies considered are tentative and highly selective. Moreover, as stated above, the measure of platform activity used in this policy brief (i.e. the number of Google keyword searches for platform names) is imperfect (e.g. it does not reflect searches via mobile apps).

Therefore, the analysis is explorative in nature and its purpose is to start identifying only some of the determinants that may be related with countries’ digital preparedness. Also, the results are purely descriptive and cannot be interpreted in a causal way. However, conditional on these data and statistical limitations, they may provide prima facie evidence on how countries’ digital preparedness has facilitated the shift of some economic activities towards online marketplaces, possibly helping their economies to weather the COVID shock.

We investigate these issues by estimating the specification reported in Box 3 and adding an interaction term between the containment index and the selected country-specific structural and policy factors. The coefficient of the interaction term captures the extent to which each of these factors may have accelerated or hindered the increase in the use of online platforms during the COVID-19 shock. The graphs below report the impact of containment measures on the activity of platforms operating in areas that recorded an increase in the use of online platforms in the first half of 2020 (i.e. mobile payments, marketplace to consumers, professional services and restaurant delivery). We are interested in the factors that may have accelerated or hindered the increase in online-platform use during the height of the pandemic. Full results and confidence intervals of estimated effects are presented in Annex Table 1.A.3.

The level of technological development

The increase in the use of online platforms during the COVID-19 crisis was stronger in countries with a higher level of technological and economic development (Figure 5). The estimated impact of the containment index on online-platform activity rises with the availability of latest technologies and the rate of adoption of digital technologies. The same seems to be true for real GDP per capita, as platform traffic growth tends to be larger in richer countries (Figure 5), though differences across countries are smaller. Thus, less developed countries could be less resilient to shocks involving severe disruptions to physical economic activity as people and firms find shifting activities to online marketplaces more difficult.11

Figure 5. The use of online platforms during the COVID-19 crisis rises with the level of technological and economic development

Effect of containment measures on online-platform use (% increase in Google keyword searches)

Note: The panels show the marginal effect of the containment index estimated through$\mathrm{ }{∆PlatformUse}_{p,s,c,t}={{\beta }_1{Lockdown}_{ct}+ {\beta }_2 Lockdown}_{c,t}\mathrm{*}{Characteristic}_c+\mathrm{ }{\theta }_t+{\delta }_{s,c}+{\epsilon }_{p,s,c,t}$, on three country-level policy/structural characteristics. Full results are in columns (1)-(3) Table 1A.3. The policy/structural characteristics considered are: the availability of latest technologies measured by an index 1-7 by the World Economic Forum (Panel A); the digital adoption index, which is a World Bank 0-1 index of digital adoption measured across people, government, and business (Panel B); GDP per capita levels in 2015 USD PPP (Panel C). Each panel shows the effect of the containment index estimated at different values of the country policy/characteristics considered: average value of bottom 3 countries; average value computed considering all countries, average value of top 3 countries. In Panel A, bottom 3 countries refer to Argentina, India, and Russia, and top 3 countries refer to Norway, United States, and Finland. In Panel B, bottom 3 countries refer to Indonesia, India, and Mexico, and top 3 countries refer to Korea, Austria, and Lithuania. In Panel C, bottom 3 countries refer to India, Indonesia, South Africa, and top 3 countries refer to Switzerland, Ireland, and Luxembourg.

Source: OECD estimations based on data from Google Trends, Crunchbase, World Economic Forum, and the World Bank.

These findings are in line with studies highlighting the role of digital technologies in disaster risk management (Madhavaram et al., 2017[32]). Our results suggest that the limited capacity of businesses and people to swiftly shift activity to online marketplaces may curtail their capacity to continue working and producing during shocks such as COVID-19. Anecdotal evidence and business surveys conducted during the COVID-19 crisis underline the clear link between the degree of digital preparedness and business resilience (KPMG, 2020[33]).

Infrastructure and connectivity

Quality access to communication networks and services is key to the adoption of digital technologies and the diffusion of digital services, including those delivered by online platforms. This depends on two main factors: past investment to build and upgrade communication networks and a sound regulatory framework ensuring easy, fair and affordable access to the network so as to meet the rising use of and demand for data.

Our analysis suggests that the increase in the use of online platforms during lockdowns was larger in countries with better communication infrastructure (proxied by mobile broadband penetration) and a better regulatory framework relating to infrastructure and connectivity – measured by the sub-component of the Digital Services Trade Restrictiveness Index concerning restrictions to digital infrastructure and connectivity (Figure 6).12

Better communication infrastructures and regulatory framework relating to infrastructure and connectivity are critical to allow all parts of society to benefit from digital technologies. Rural populations often experience significantly worse access to communications networks than urban populations (OECD, 2019[34]). The impact of COVID-19 in rural areas (and in developing countries) may have been especially hard because of effects of mobility restrictions combined with the poor access to communication infrastructures and connectivity (OECD, 2018[35]).13 Several good practices to promote digital connectivity exist, such as subsidising national and rural broadband networks, promoting municipal networks, designing competitive tenders for private-sector network deployment, and managing or implementing open access arrangements. Our results suggest that improving the regulatory framework relating to infrastructure and connectivity can strengthen countries’ resilience to shocks such as COVID-19.

Education and Skills

In a world increasingly permeated by digital technologies, the skills needed in the labour market are continuously evolving, with an increasing demand for workers with high skill levels. Workers – either dependent employees, self-employed or in managerial roles – increasingly need to be skilled in a range of digital technologies. The wide diffusion and effective use of digital technologies (including online platforms) also depends on the general digital skills – such as the ability to use the Internet – of the entire population, including those not in the workforce, such as the elderly. The OECD’s Survey of Adult Skills (PIAAC) suggests that more than 50% of the adult population on average in 28 OECD countries can only carry out the simplest set of computer tasks, such as writing an email and browsing the web, or have no ICT skills at all. Overall, the digital divide, which initially focused on gaps in Internet access, increasingly concerns the different ways people are able to use the Internet and the benefits they derive from using digital technologies (OECD, 2019[36]). The effective use of online platforms is likely to require a minimum level of general digital skills that is relatively high.

Figure 6. Broadband penetration and good network regulation have underpinned the increase in platform use during the COVID-19 crisis

Effect of containment measures on online-platform use (% increase in Google keyword searches)

Note: The panels show the marginal effect of the containment index estimated through$\mathrm{ }{∆PlatformUse}_{p,s,c,t}={{\beta }_1{Lockdown}_{ct}+ {\beta }_2 Lockdown}_{c,t}\mathrm{*}{Characteristic}_c+\mathrm{ }{\theta }_t+{\delta }_{s,c}+{\epsilon }_{p,s,c,t}$, on three country-level policy/structural characteristics. Full results are in columns (4)-(5) Table 1A.3. The policy/structural characteristics considered: are: the levels of mobile broadband penetration, measured as mobile broadband connections per 100 inhabitants (Panel A) and based on the levels of restrictions relating to the infrastructure and connectivity sub-component of the Digital Services Trade Restrictiveness Index. Each panel shows the effect of the containment index estimated at different values of the country policy/characteristics considered: average value of bottom 3 countries; average value computed considering all countries, average value of top 3 countries. In Panel A, bottom 3 countries refer to India, Hungary, and Mexico, and top 3 countries refer to Estonia, Finland, and Japan. In Panel B, top 3 values are for Japan, Austria, Australia, Switzerland plus other countries with the same value, and bottom 3 values are for South Africa, Russia, Indonesia, and Argentina, Chile, and Poland.

Source: OECD estimations based on data from Google Trends, OECD, and World Bank.

Our results show that during lockdowns online-platform activity might have risen more in countries where digital skills are high and where the percentage of individuals that used the Internet to pay bills in the previous year is largest (Figure 7). To facilitate the take-up of online platforms (along with the adoption of digital technologies more generally) policies need to go beyond the fundamental objective of ensuring that the education system equips all students with basic digital skills (as well as solid literacy, numeracy and problem-solving skills). Other important initiatives involve building and improving life-long learning systems so as to ensure all adults can acquire and maintain the skills required to handle modern technologies effectively. Many of these skills can be acquired also outside formal education and training institutions – for instance, in the workplace. Boosting funds for on-the-job training and establishing a certification system for skills acquired outside formal education channels can encourage training in the workplace (OECD, 2016[37]).

Figure 7. The use of online platforms during COVID-19 crisis rises with skill levels

Effect of containment measures on online-platform use (% increase in Google keyword searches)

Note: Note: The panels show the marginal effect of the containment index estimated through$\mathrm{ }{∆PlatformUse}_{p,s,c,t}={{\beta }_1{Lockdown}_{ct}+ {\beta }_2 Lockdown}_{c,t}\mathrm{*}{Characteristic}_c+\mathrm{ }{\theta }_t+{\delta }_{s,c}+{\epsilon }_{p,s,c,t}$, on three country-level policy/structural characteristics. Full results are in columns (6)-(7) Table 1A.3. The policy/structural characteristics considered: the skills readiness index from the WEF Networked Readiness Index (Panel A) and the percentage of individuals over 15 years of age that used the Internet to pay bills in the previous year, taken from the World Bank (Panel B). Each panel shows the effect of the containment index estimated at different values of the country policy/characteristics considered: average value of bottom 3 countries; average value computed considering all countries, average value of top 3 countries. In Panel A, bottom 3 countries refer to India, South Africa, and Mexico, and top 3 countries refer to Belgium, Switzerland, and Finland. In Panel B, bottom 3 countries refer to India, Indonesia, and Mexico, and top 3 countries refer to Denmark, Finland, and Norway.

Source: OECD estimations based on data from Google Trends, World Bank, and World Economic Forum.

Conclusion and policy insights

This note uses Google Trends data to understand the change in online-platform activity throughout the first six months of 2020 and how it relates to lockdown and mobility restriction measures imposed over the same period. Changes in online-platform activity occurred differently across activity areas. Online platforms operating in areas requiring physical proximity for service delivery – such as accommodation and restaurant booking – suffered marked decreases in activity; instead, activity of platforms operating in areas less affected by physical proximity requirements – such as mobile payments, marketplace to consumers and restaurant delivery –increased by around 20% on average.

The increase in online-platform use varied across countries depending on pre-existing structural conditions and policies. The increase tended to be larger in more technologically advanced countries, those with better communication networks and higher levels of digital skills. These results suggest that policies may play an important role in strengthening the digital preparedness of countries to face shocks causing severe disruptions of traditional market and social activities. When COVID-19 hit, some countries were better prepared than others and in these countries people and businesses may have found it easier to shift activities towards online platforms.

The COVID-19 shock has added urgency to policies aiming at accelerating the digitalisation of public and private sector activities. During the crisis many OECD and G20 countries implemented a range of such policies including, for instance, improving broadband connectivity, helping firms adopt online business models, promoting online payments, and enhancing digital skills (G20, 2020[38]; OECD, 2020[2]; OECD, 2020[39]).

This new urgency is welcome but a co-ordinated approach is needed to exploit complementarities and minimising trade-offs across policy areas. The breadth of policies and interventions (spanning for instance connectivity, skills, finance) and the time it takes for them to realise their effects require building synergies among the various policy initiatives (OECD, 2019[40]). For example, investment in information and communication technologies (ICT), such as broadband, is necessary but not sufficient for the effective use of digital tools. Complementary investment in knowledge-based capital, such as skills, is needed. At the same time, the general digital skills of the population – acquired through the education system – are the foundation on which workers in some industries can build more specialised and advanced ICT skills throughout their careers and continuous learning. Policies to protect the privacy of personal data and strengthen cybersecurity can enhance trust in digital technologies and online platforms, accelerating their adoption. Incentives linked to the use of electronic payments can be conducive to a more intense use of online payments. Establishing or updating Industry 4.0 plans in view of the new challenges the COVID-19 crisis has created would be a good way to build such a co-ordinated approach.

The increase in online platform use during the COVID-19 has also heightened concerns about the quality and safety of jobs intermediated by online platforms. Platforms often rely on flexible work arrangements, which could lead to an increase in “false” self-employment, poor quality jobs and career prospects, and contribute to a segmented labour market (Mira d’Ercole and MacDonald, 2018[41]). In response to the crisis, platform companies have taken measures to protect the health and incomes of platform workers, including: introducing contactless delivery or temporarily ceasing high-risk services; providing personal protective equipment or hygiene products; providing full or partial pay for sick or self-isolating workers (generally up to a maximum period of two weeks) (OECD, 2020[42]). At the same time, many governments have exceptionally extended social protection schemes to platform (and other self-employed) workers as the traditional social safety net does not cover them (or covers them only partially). These are temporary solutions, but they could be the start of a process to durably improve the quality and safety of platform-mediated jobs. This is likely to require policy changes in different areas including labour market regulation, social protection and workers’ representation (Lane, 2020[43]).

References

[24] Amstad, M. et al. (2020), “Investors’ risk attitudes in the pandemic and the stock market: new evidence based on internet searches”, BIS Bulletins 25, Bank for International Settlements, https://www.bis.org/publ/bisbull25.htm (accessed on 2 October 2020).

[22] Aouadi, A., M. Arouri and D. Roubaud (2018), “Information demand and stock market liquidity: International evidence”, Economic Modelling, Vol. 70, pp. 194-202, http://dx.doi.org/10.1016/j.econmod.2017.11.005.

[10] Bailin Rivares, A. et al. (2019), “Like it or not? The impact of online platforms on the productivity of incumbent service providers”, OECD Economics Department Working Papers, No. 1548, OECD Publishing, Paris, https://dx.doi.org/10.1787/080a17ce-en.

[28] Bulman, T. and S. Koirala (2020), The OECD COVID-19 Policy Tracker: What are governments doing to deal with the COVID-19 pandemic?, https://oecdecoscope.blog/2020/05/15/the-oecd-covid-19-policy-tracker-what-are-governments-doing-to-deal-with-the-covid-19-pandemic/ (accessed on 11 September 2020).

[23] Castelnuovo, E. and T. Tran (2017), “Google It Up! A Google Trends-based Uncertainty index for the United States and Australia”, Economics Letters, Vol. 161, pp. 149-153, http://dx.doi.org/10.1016/j.econlet.2017.09.032.

[20] Chen, T. et al. (2015), “The 2007-2008 U.S. recession: what did the real-time google trends data tell the United States?”, Contemporary Economic Policy, Vol. 33/2, pp. 395-403, http://dx.doi.org/10.1111/coep.12074.

[31] Correia, S. (2016), “A Feasible Estimator for Linear Models with Multi-Way Fixed Effects”, Unplublished manuscript.

[15] D’Amuri, F. and J. Marcucci (2017), “The predictive power of Google searches in forecasting unemployment”, International Journal of Forecasting, Vol. 33/4, pp. 801-816, https://www.sciencedirect.com/science/article/pii/S0169207017300389 (accessed on 23 March 2020).

[21] Da, Z., J. Engelberg and P. Gao (2011), “In Search of Attention”, Journal of Finance, Vol. 66/5, pp. 1461-1499, http://dx.doi.org/10.1111/j.1540-6261.2011.01679.x.

[25] Eichenauer, V. et al. (2020), “Constructing Daily Economic Sentiment Indices Based on Google Trends”, IIMB Management Review Available online, https://www.sciencedirect.com/science/article/pii/S0970389617302719 (accessed on 2 October 2020).

[9] Farronato, C., J. Fong and A. Fradkin (2020), Dog Eat Dog: Measuring Network Effects Using a Digital Platform Merger, NBER Working Paper 28047.

[44] Ferencz, J. (2019), “The OECD Digital Services Trade Restrictiveness Index”, OECD Trade Policy Papers, No. 221, OECD Publishing, Paris, https://dx.doi.org/10.1787/16ed2d78-en.

[19] Ferrara, L. and A. Simoni (2019), “When Are Google Data Useful to Nowcast GDP? An Approach Via Pre-Selection and Shrinkage”, SSRN Electronic Journal, http://dx.doi.org/10.2139/ssrn.3370917.

[38] G20 (2020), Policy Options to Support Digitalisation of Business Models during COVID-19, https://g20.org/en/media/Documents/Policy%20Options%20to%20Support%20Digitalization%20of%20Business%20Models%20during%20COVID-19.pdf.

[18] Götz, T. and T. Knetsch (2019), “Google data in bridge equation models for German GDP”, International Journal of Forecasting, Vol. 35/1, pp. 45-66, http://dx.doi.org/10.1016/j.ijforecast.2018.08.001.

[26] Harris, S. and A. Krueger (2015), A Proposal for Modernizing Labor Laws for Twenty-First Century Work: The “Independent Worker” | The Hamilton Project, The Hamilton Project, https://www.hamiltonproject.org/papers/modernizing_labor_laws_for_twenty_first_century_work_independent_worker (accessed on 2 October 2020).

[30] Ifo (2020), Industrial digital economy – B2B platforms, https://english.bdi.eu/publication/news/industrial-digital-economy-b2b-platforms/ (accessed on  November 2020).

[33] KPMG (2020), Enterprise Reboot: Scale Digital Technologies to Grow adn Thrive in the New Reality, https://home.kpmg/xx/en/home/insights/2020/08/enterprise-reboot.html.

[43] Lane, M. (2020), “Regulating Platform Work in the Digital Age”, Going Digital Toolkit Policy Note, No. No 1, OECD, https://goingdigital.oecd.org/toolkitnotes/regulating-platform-work-in-the-digital-age.pdf (accessed on 21 September 2020).

[16] Maas, B. (2020), “Short‐term forecasting of the US unemployment rate”, Journal of Forecasting, Vol. 39/3, pp. 394-411, http://dx.doi.org/10.1002/for.2630.

[32] Madhavaram, S. et al. (2017), “ICTs in the Context of Disaster Risk Management, stakeholders, and implications”, Journal of Information, Communication and Ethics in Society, Vol. 15/1, pp. 32-52, https://doi.org/10.1108/JICES-11-2015-0038.

[27] McKinsey Global Institute (2015), Connecting talent with opportunity in the digital age | McKinsey, https://www.mckinsey.com/featured-insights/employment-and-growth/connecting-talent-with-opportunity-in-the-digital-age# (accessed on 2 October 2020).

[41] Mira d’Ercole, M. and M. MacDonald (2018), Measuring platform and other new forms of work: Issues pape.

[6] OECD (2020), A Roadmap Toward a Common Framework for Measuring the Digital Economy: Report for the G20 Digital Economy Task Force, OECD Publishing, Paris, http://www.oecd.org/sti/roadmap-toward-a-common-framework-for-measuring-the-digital-economy.pdf (accessed on 2 October 2020).

[3] OECD (2020), Are platforms killing the offline star? Platform diffusion and the productivity of traditional firms, ECO/CPE/WP1(2020)6.

[2] OECD (2020), Keeping the Internet up and running in times of crisis, http://www.oecd.org/coronavirus/policy-responses/keeping-the-internet-up-and-running-in-times-of-crisis-4017c4c9/ (accessed on 9 October 2020).

[1] OECD (2020), OECD Economic Outlook, Volume 2020 Issue 1, OECD Publishing, Paris, https://dx.doi.org/10.1787/0d1d1e2e-en.

[39] OECD (2020), The potential of online learning for adults: Early lessons from the COVID-19 crisis, http://www.oecd.org/coronavirus/policy-responses/the-potential-of-online-learning-for-adults-early-lessons-from-the-covid-19-crisis-ee040002/ (accessed on 9 October 2020).

[42] OECD (2020), What have platforms done to protect workers during the coronavirus (COVID-19) crisis?, http://www.oecd.org/coronavirus/policy-responses/what-have-platforms-done-to-protect-workers-during-the-coronavirus-covid-19-crisis-9d1c7aa2/#boxsection-d1e28 (accessed on 4 Dicember 2020).

[8] OECD (2019), An Introduction to Online Platforms and Their Role in the Digital Transformation, OECD Publishing, Paris, https://dx.doi.org/10.1787/53e5f593-en.

[34] OECD (2019), “Enhancing Access and Connectivity to Harness Digital Transformation”, OECD Going Digital Policy Note, OECD, Paris, http://www.oecd.org/going-digital/enhancing-access-digital-transformation.pdf.

[40] OECD (2019), Going Digital: Shaping Policies, Improving Lives, OECD Publishing, Paris, https://doi.org/10.1787/9789264312012-en.

[5] OECD (2019), Measuring the Digital Transformation: A Roadmap for the Future, OECD Publishing, Paris, https://doi.org/10.1787/9789264311992-en.

[36] OECD (2019), OECD Skills Outlook 2019 : Thriving in a Digital World, OECD Publishing, Paris, https://dx.doi.org/10.1787/df80bc12-en.

[35] OECD (2018), “Bridging the Rural Digital Divide”, OECD Digital Economy Papers, No. 265, OECD Publishing, Paris, https://doi.org/10.1787/852bd3b9-en (accessed on 5 August 2020).

[4] OECD (2018), Measuring online platforms and cloud computing in National Accounts, Workshop: Working Party on Measurement and Analysis of the Digital Economy, https://www.oecd.org/going-digital/mdt-roadmap-cloud-computing.pdf.

[37] OECD (2016), “Skills for a Digital World: 2016 Ministerial Meeting on the Digital Economy Background Report”, OECD Digital Economy Papers, No. 250, OECD Publishing, Paris, https://dx.doi.org/10.1787/5jlwz83z3wnw-en.

[7] OECD, WTO and IMF (2020), Handbook on Measuring Digital Trade: Volume 1, https://www.oecd.org/sdd/its/handbook-on-measuring-digital-trade.htm (accessed on 2 October 2020).

[29] Petherick, A. et al. (2020), “Variation in government responses to COVID-19”, Blavatnik School Working Paper 32, Blavatnik School of Government, Oxford University, https://www.bsg.ox.ac.uk/research/publications/variation-government-responses-covid-19.

[13] Siliverstovs, B. and D. Wochner (2018), “Google Trends and reality: Do the proportions match?: Appraising the informational value of online search behavior: Evidence from Swiss tourism regions”, Journal of Economic Behavior and Organization, Vol. 145, pp. 1-23, http://dx.doi.org/10.1016/j.jebo.2017.10.011.

[14] Smith, P. (2016), “Google’s MIDAS Touch: Predicting UK Unemployment with Internet Search Data”, Journal of Forecasting, Vol. 35/3, pp. 263-284, http://dx.doi.org/10.1002/for.2391.

[11] Vosen, S. and T. Schmidt (2011), “Forecasting private consumption: Survey-based indicators vs. Google trends”, Journal of Forecasting, Vol. 30/6, pp. 565-578, http://dx.doi.org/10.1002/for.1213.

[17] Woloszko, N. (2020), “Tracking activity in real time with google trends”, Unpublished manuscript.

[12] Woo, J. and A. Owen (2019), “Forecasting private consumption with Google Trends data”, Journal of Forecasting, Vol. 38/2, pp. 81-91, http://dx.doi.org/10.1002/for.2559.

Annex 1.A. Data and regression results

Figure 1.A.1. Online-platform activity during lockdowns by area of activity

Note: The lines show the median of the year-on-year percentage change of online-platform activity (as proxied by Google Trends data) for the different activity areas; the period covered is January-June 2020.

Source: Google Trends and OECD computations

Table 1.A.1. Detail by area of activity and testing different clustering approaches

	(1)	(2)	(3)	(4)
Dependent variable	Change in platform use
Lockdown Indicator	-0.150***	-0.150**
	(0.0339)	(0.0680)
Lockdown Indicator x Accommodation			-1.223***	-1.223***
			(0.0523)	(0.0565)
Lockdown Indicator x Marketplace B2B			0.0477	0.0477
			(0.0781)	(0.0656)
Lockdown Indicator x Marketplace to consumers			0.124***	0.124***
			(0.0363)	(0.0423)
Lockdown Indicator x Mobile payments			0.179***	0.179***
			(0.0352)	(0.0405)
Lockdown Indicator x Personal services			-0.00758	-0.00758
			(0.0406)	(0.0412)
Lockdown Indicator x Professional services			0.110***	0.110**
			(0.0375)	(0.0497)
Lockdown Indicator x Restaurant booking			-0.737***	-0.737***
			(0.140)	(0.0966)
Lockdown Indicator x Restaurant delivery			0.106*	0.106
			(0.0582)	(0.0685)
Lockdown Indicator x Transport			-0.964***	-0.964***
			(0.0968)	(0.106)
Constant	-0.0332***	-0.0332	-0.0472***	-0.0472***
	(0.0116)	(0.0224)	(0.0108)	(0.0117)
Observations	31,083	31,083	31,083	31,083
R-squared	0.148	0.148	0.201	0.201
Clustering	Platform-sector-country	Country-area	Platform-sector-country	Country-area

Note: Estimations of equation (1) in columns (1) and (2) (${∆PlatformUse}_{p,s,c,t}={{\beta }_{1}Lockdown}_{c,t}+ {\theta }_t+{\delta }_{s,c}+{\epsilon }_{p,s,c,t})$and equation (1) with the lockdown indicator interacted with every area dummy (${∆PlatformUse}_{p,s,c,t}={{{\beta }_{1}Lockdown}_{c,t}+{\beta }_{2}Lockdown}_{c,t}*{Area}_s+ {\theta }_t+{\delta }_{s,c}+{\epsilon }_{p,s,c,t})$. Regressions include month and country-area fixed effects. Standard errors are clustered at the platform-country-sector level (columns 1 and 3) and country-area level (columns 2 and 4). Significance level for which the null hypothesis is rejected: *** 1%, ** 5%, * 10%.

Source: Google Trends, Crunchbase, and OECD computations.

Table 1.A.2. Policy variables: sources and coverage

Area	Indicator	Source	Country coverage
Technological development	Availability of latest technologies - Index (1-7)	World Economic Forum	All countries
	Digital adoption index (0-1)	World Bank	All countries apart from the Czech Republic
	GDP per capita levels in 2015 USD PPP	OECD Statistics	All countries
Infrastructure and connectivity	Broadband connections per 100 inhabitants	OECD Going Digital	All countries apart from Argentina and Saudi Arabia
	Level of the Infrastructure and connectivity sub-component of the DSTRI	OECD	All countries
Education and skills	Skills readiness index	World Economic Forum	All countries
	Population over 15 that used the Internet to pay bills in the past year (%)	World Bank	All countries apart from Iceland

Note: All values used for the year 2016, as the year with best coverage across all variables.

Table 1.A.3. Lockdown and change in platform use: variation with country policies and characteristics

	Country policy/characteristics
	(1) Availability of latest technologies	(2) Digital Adoption Index	(3) GDP per capita (log)	(4) Infrast. & connec. restrictions	(5) Broadband penetration (mobile)	(6) Skills readiness (1-7)	(7) % Used internet pay bills
Dependent Variable	Change in platform use
Lockdown Indicator	-1.589***	-1.557***	-2.479***	-1.026***	-1.332***	-1.622***	-1.290***
	(0.299)	(0.305)	(0.722)	(0.0723)	(0.156)	(0.367)	(0.0890)
Lockdown Indicator x Policy/Characteristics	0.0893*	0.605	0.132*	-0.886	0.00263*	0.0915	0.437***
	(0.0526)	(0.419)	(0.0693)	(0.565)	(0.00155)	(0.0652)	(0.161)
Lockdown Indicator x (Area not requiring physical proximity)	1.455***	1.665***	2.309***	1.231***	1.484***	1.687***	1.434***
	(0.316)	(0.325)	(0.770)	(0.0685)	(0.165)	(0.389)	(0.0890)
Lockdown Indicator x (Area not requiring physical proximity) x Policy/Characteristics	-0.0328	-0.535	-0.100	0.416	-0.00232	-0.0750	-0.376**
	(0.0556)	(0.447)	(0.0739)	(0.597)	(0.00166)	(0.0693)	(0.171)
Constant	-0.0613***	-0.0522***	-0.0542***	-0.0539***	-0.0626***	-0.0526***	-0.0581***
	(0.0117)	(0.0118)	(0.0116)	(0.0117)	(0.0118)	(0.0116)	(0.0119)
Observations	24,383	23,785	24,383	24,383	23,381	24,383	24,137
R-squared	0.247	0.249	0.247	0.247	0.244	0.246	0.248
Marginal effects of lockdown indicator for different values of the policy/characteristics (for areas not requiring physical proximity)
Average value of bottom three countries	0.092**	0.144***	0.121***	0.087**	0.153**	0.136**	0.148***
	(0.041)	(0.042)	(0.045)	(0.043)	(0.061)	(0.042)	(0.038)
Average value	0.187***	0.158***	0.162***	0.163***	0.183**	0.157***	0.174***
	(0.034)	(0.035)	(0.033)	(0.034)	(0.035)	(0.03)	(0.034)
Average value of top three countries	0.237***	0.167***	0.189***	0.193***	0.202***	0.170***	0.197***
	(0.040)	(0.044)	(0.042	(0.038)	(0.057)	(0.041)	(0.043)

Note: Estimations of equation (1) interacting the lockdown indicator with each of the structural and policies characteristics above (${∆PlatformUse}_{p,s,c,t}={{{\beta }_{1}Lockdown}_{c,t}+{\beta }_{2}Lockdown}_{c,t}*{Characteristic}_c+{{{\beta }_{3}Lockdown}_{c,t}*{NoPhysicalProx}_s+{\beta }_{3}Lockdown}_{c,t}*{Characteristic}_c*{NoPhysicalProx}_s+ {\theta }_t+{\delta }_{s,c}+{\epsilon }_{p,s,c,t}$. Regressions include month and country-area fixed effects. Standard errors are clustered at the platform-sector-country level. Significance level for which the null hypothesis is rejected: *** 1%, ** 5%, * 10%. The marginal effects are computed only for activity areas that recorded an increase in the use of online platforms in the first half of 2020: mobile payments, marketplace to consumers, professional services, and restaurant delivery.

Source: Google Trends, Crunchbase, and OECD computations.

Table 1.A.4. Correlation between main explanatory variables

	Lockdown indicator (OECD)	Stringency indicator (Oxford)	Google mobility change	Total number of cases (log)
Lockdown indicator (OECD)	1
Stringency indicator (Oxford)	0.9341*	1
Google mobility change	-0.7151*	-0.7043*	1
Total number of cases (log)	0.6987*	0.801*	-0.4649*	1

Note: A star indicates significance level of 0.05.

Source: Data from OECD, Oxford Covid-19 Government Response Tracker, and Google mobility, and OECD computations.

Notes

← 1.  For instance, the market capitalisation of the mobile payment companies Paypal and Square has exceed that of most of the largest banks in the United States (www.cnbc.com/2020/09/04/disruptors-paypal-and-square-surpass-wall-street-giants-including-goldman-sachs-in-market-cap.html).

← 2.  The closest historic parallel in terms of mortality to the current crisis was the Spanish Flu pandemic from 1918 to 1920, which took place in a markedly different economic, social, and demographic context.

← 3.  The underlying assumption is that the volume of searches for a platform name is correlated with the volume of activity of the platform, either because users are directed to the platform’s website for direct use, or because users are looking for information or tips when using the platform through mobile apps.

← 4.  Baseline days represent a normal value for that day of the week, given as median value over the five‑week period from January 3rd to February 6th 2020.

← 5.  The areas requiring physical proximity are: accommodation, restaurant booking, and transport/taxi. The areas not requiring physical proximity are: mobile payments, marketplace to consumers, professional services and restaurant delivery. The areas having mixed physical proximity requirements include personal services and business to business marketplace platforms.

← 6.  We have also investigated the different sub-components of the OECD containment index: cancellation of public events, obligatory shutdown of the economy, confinement lockdowns, travel restrictions and closure of schools. This part of the analysis confirms the general picture described so far but does not provide much additional insights due to high correlation among the different sub-components.

← 7.  See https://www.marketplacepulse.com/articles/marketplace-was-amazons-fastest-growing-business-in-q2 and https://www.marketplacepulse.com/articles/amazons-fastest-growing-markets-during-the-pandemic.

← 8.  See https://www.statista.com/statistics/218495/paypals-net-number-of-payments-per-quarter.

← 9.  See https://www.businessinsider.fr/us/airbnb-city-bookings-drop-coronavirus-2020-3.

← 10.  See https://investor.uber.com/news-events/news/press-release-details/2020/Uber-Announces-Results-for-Second-Quarter-2020/default.aspx.

← 11.  Marginal effects’ confidence intervals are large. Differences in the effects of the countries’ characteristics on the use of platforms are statistically significant only for very large and very low value values of the countries’ characteristics.

← 12.  The Digital Services Trade Restrictiveness Index captures cross-cutting barriers that hamper the supply of services using electronic networks. It catalogues and quantifies different types of regulatory barriers (Ferencz, 2019[44]). Those related to infrastructure and connectivity reflect the adoption of best practice regulations on interconnections among network operators to ensure seamless communication. It also captures measures limiting or blocking the use of communications services, including Virtual Private Networks or leased lines. Lastly, this area covers policies impacting on connectivity such as measures on cross-border data flows and data localisation.

← 13.  Other types of infrastructure, such as roads and public transportation, can also affect the speed and availability of delivery services, especially in rural areas.