Tentative lessons: 5 things the literature can tell us about battling global pandemics

In the context of continuing waves of Covid-19 infections and associated measures to fight the pandemic, the literature is starting to draw lessons from the various policies implemented. In this column, we highlight five lessons, while acknowledging that disentangling the effects of the different factors – among which are lockdown measures – is challenging since they have been at play simultaneously or following the same sequence across countries.

More stringent and earlier lockdowns – when the number of cases is low – are more efficient to curb infections. According to empirical findings in IMF (2020), the number of infections has been significantly lower for countries with early lockdowns. In addition, a more stringent lockdown has an immediate effect on curbing infections; otherwise, the effect is non-significant. Other empirical studies on US (Demirguc-Kunt et al. 2020) or European data (Dave et al. 2021b) confirm these results.

Theoretical models confirm that an early and stringent lockdown reduces the economic impact and death toll of the pandemics. Alvarez et al. (in press) conclude that it is optimal to implement a strict lockdown for only two weeks after the first Covid-19 cases. Other models support this finding both in the medical (Buckman et al. 2020, Vinceti et al. 2020) and economic (Eichenbaum et al. 2020, Farboodi et al. 2021) literatures. However, this might not indicate that returning to a lockdown would be unnecessary thereafter. Caulkins et al. (2021) show that it can be optimal to have two or three distinct lockdown periods, depending on local preferences regarding how to balance health and economic impacts.

Various studies have also highlighted the benefits of sanitary measures based on mass testing, generalized use of masks, and screening. Summers et al. (2020), Shaw et al. (2020) and Yalaman et al. (2021) document the efficiency of Asian countries’ strategies based on: (i) early and massive introduction of borders screening, (ii) stringent process to isolate suspect cases and virus-bearers, (iii) use of new technologies for efficient contact tracing, and (iv) generalised use of masks. Making masks compulsory is assessed to reduce the number of infections between 25% and 40% compared to optional mask-wearing (Mitze et al. 2020, Krishnamachari et al. 2021, Chernozhukov et al. 2021). Masks not only help prevent viral transmission but also reduce exposure to cold environments (Bubbico et al. 2021). These sanitary measures are more efficient, however, if combined with social distancing (Firth et al. 2020, Ando et al. 2021) and if the population’s civic sense is high (Barrios et al. 2021). As regards testing, Atkeson et al. (2020) find that the economic benefits of rapid screening programmes exceed their costs by a ratio of 4 to 15. But while most also point to a positive effect of extensive testing (Brotherhood et al. 2020, Hellmann and Thiele 2020, Su et al. 2021), Acemoglu et al. (2020b) report an ambiguous impact since it might lead the population into reducing its social interactions less, fostering the spread by undetected virus-bearers.

Disentangling empirically the marginal impact of each measure is complex as they have generally been implemented simultaneously or following the same sequence (Hsiang et al. 2020) with the low quality of data on infections being an additional challenge (Bonacini et al. 2021). Studies have nevertheless estimated their marginal impact. Among those, Deb et al. (2020) estimate not only the health benefits – i.e. how it slows the spread of the virus – but also the economic costs. They find that workplace closures are efficient in reducing infections but are also the costliest in terms of economic impact. They also report that school and public transport closures have a high economic cost but a limited effect on the outbreak. Finally, the authors find that international travel restrictions and, to a lesser extent, limits on the size of gatherings and cancelling of public events display the greatest benefit-to-cost ratios.

The impact of each measure remains, however, highly debated. Table 1 recapitulates how various studies estimate the health benefits of different measures. It highlights large discrepancies. A consensus seems to emerge, however, on the high impact of cancelling public events, and on the mild impact of public transport and non-essential business closures. On the latter, while Song et al. (2021) estimate that it has indeed been significantly protective for workers in this sector, it however translated into higher unemployment (Sjoquist and Wheeler 2021). Studies point to the benefits of an alternative more targeted closure for only high-contact places such as restaurants, gyms, and pubs (Courtemanche et al. 2020, Chang et al. 2021).

Studies also bring forward heterogeneities associated with duration, geographical factors, and government efficiency. Li et al. (2021) find evidence that the impact depends on time horizon with for example international travel restrictions efficient after seven days but not after 28 days. Burlig et al. (2021) model a same non-linear impact in time for domestic travel bans. More generally, Bakker and Goncalves (2021) show that the impact of measures on infections declined over time. As regards geographical heterogeneities, Russell et al. (2021) show that international travel restrictions might have little impact on pandemics except in countries with low Covid-19 incidence and large numbers of arrivals from abroad. Bennett (2021) show a significant efficiency of lockdown measures in high-income areas but non-significant in the low-income ones while Becchetti et al. (2020) find them to be more effective in highly polluted areas. Pan et al. (2020) also report heterogeneities associated with racial composition and poverty. Finally, Bakker and Goncalves (2021) find that measures have been more efficient in countries with higher government’s effectiveness.

Table 1 Relative impact of various measures on containing infections

Whatever their individual impact, however, most studies converge on their combined efficiency – even though voluntary social distancing also naturally reduces infections. Flaxman et al. (2020) estimate that comprehensive lockdowns (a mix of workplaces and school closures, cancelling of public events, stay-at-home orders, and limits on the size of gatherings) in Europe have reduced the reproduction rate by 80%. Santeramo et al. (2021) for Italy and Ferguson et al. (2020) for the UK and the US reach the same conclusion. Voluntary social distancing taken spontaneously by the population might however explain a share of the reduction in infections. Agrawal et al. (2021) and Berry et al. (2021) fail to find that places that implemented lockdowns earlier or for longer have lower excess deaths, while Singh et al. (2021) find only a modest effect. In the same vein, studies have documented that lockdown measures account for a relatively small share of the change in individuals’ behaviours (Gupta et al. 2020b, Cronin and Evans 2020).

On top of a negative short-term economic impact, stricter measures might entail long-term detrimental effects on inequalities, mental health, and human capital. The impact of lockdowns is disproportionate on vulnerable groups such as low-skilled workers (Cajner et al. 2020), whose jobs are less likely to be able to be performed remotely (Dingel and Nieman 2020). School closures and lack of access to reliable childcare have taken a higher toll on young parents (Papanikolaou and Schmidt 2020), and notably women (Del Boca et al. 2020, Albanesi and Kim 2021). This has even been the case in academia where women, particularly those who have children, report a disproportionate reduction in time dedicated to research relative to others (Deryugina et al. 2021). In the longer run, job losses might have hysteresis effects with workers falling into long-term unemployment. In addition to this immediate destruction of human capital, school closures can also weigh on future generations’ capacity to accumulate human capital (Fuchs-Schündeln et al. 2021). Isolation measures such as stay-at-home orders also affect mental health (Béland et al. 2020b, Sibley et al. 2020). Finally, the aggregate effect on the death toll might be more ambiguous. Mulligan (2020) and Faust et al. (2021) show that the pandemics and associated recession may lead to a significant increase in the number of deaths from suicide, substance abuse, and murder – in particular among disadvantaged populations (Chen et al. 2020b, Krieger et al. 2020). Lin et al. (2021) also show that in low-income countries, recessions coming with lockdowns increase child mortality, leading to an inter-generational trade-off with the Covid-related deaths avoided mostly for seniors.

Targeting might seem a priori relevant. Studies have documented heterogeneous impacts depending on population density (Dave et al. 2021b), age and dependency ratio which particularly influence the mortality rate (Levin et al. 2020, Bürgi and Gorgulu 2020), and workers categories (Akbarpour et al. 2020).

Numerous theoretical models advocate for targeted measures on seniors and employees whose job can be performed remotely. Acemoglu et al. (2020a), Alon et al. (2020), and Gollier (2020) conclude that applying more stringent measures to those aged 65+ reduces the economic cost while maximizing health benefits. Focusing on deaths and ICU (intensive care unit) bed occupancy, Ferguson et al. (2020) estimate that social distancing only for those aged 70+ has two to three times the effect of social distancing for the entire population. In addition, Aum et al. (2020) show that locking-down only employees whose job can be performed remotely reduces by half the economic cost compared to a situation where all workers are required to stay at home – for the same health benefits. Another possibility is the implementation of alternate slots in firms and schools to reduce social interactions (Akbarpour et al. 2020). A counterargument, however, comes from studies such as Checo et al. (2021), who find that targeted measures have a higher macroeconomic cost as they remain in place for longer. Another is given by Singh et al. (2021), who empirically find that only measures targeting the general population have a statistically significant impact.

The literature provides mixed evidence regarding geographical targeting. Li et al. (2020) and Lin and Meissner (2020) conclude that local lockdowns have a limited impact on the spread of the virus. Elenev et al. (2021) also provide evidence of spillovers from stay-at-home orders. Dave et al. (2021c) also show how a ‘super-spreader’ event in a US state with a loose lockdown can impact infections in other states with more stringent measures in place. On the opposite, Fang et al. (2020) empirically show how locking down 63 cities in Hubei had successfully contained the spread across China. On a more theoretical perspective, Fajgelbaum et al.’s (2020) model demonstrates that stringent measures only to selected boroughs in large metropolis could be as efficient as a generalised lockdown while significantly reducing the economic impact. Similarly, in a model separating cities and rest of the state, Bisin and Moro (2021) find that a city-only lockdown does not induce a much larger fraction of infected persons than a general lockdown. Finally, Crucini and O'Flaherty (2020) suggest that local restrictions are optimal in a fiscal union as a national policy would be too restrictive for mildly infected areas, weighing therefore too much on the local economic activity.

At the European level, some studies show the benefits of a coordinated approach not only when implementing lockdowns but also when relaxing them. Ash (2020) estimates that relaxing together would delay the resurgence of the virus by five weeks. Symmetrically, she shows that a coordinated implementation of lockdowns across Europe has a stronger impact on infections – in line with the findings of Ruktanonchai et al. (2020). This is particularly due to strong health spillovers across Europe (Costa-i-Font 2020).

Countries with more stringent lockdowns have experienced sharper GDP contractions. This relation remains valid for other macroeconomic indicators such as households’ consumption (Baker et al. 2020b, Carvalho et al. 2020), employment (Béland et al. 2020a, Schotte et al. 2021) or industrial production (Deb et al. 2020). Theoretical models back such a correlation (e.g. Baqaee and Farhi 2020).

However, even in the absence of lockdown measures, the spread of the virus affects economic activity. Voluntary social distancing has a major impact on activity while heightened uncertainty (Baker et al. 2020a) and deteriorating economic prospects (Baek et al. 2020) also weigh on it. Studies recapitulated in Table 2 show that lockdown measures are estimated to account for around 10 to 60% of the total economic impact of Covid-19. The literature provides vast evidence for an impact of the pandemics in the absence of lockdowns. Chetty et al. (2020) note a contraction of activity before the start of lockdowns in the US. Rojas et al. (2020) and Kahn et al. (2020) observe that the surge in unemployment claims has been homogeneous across the US, notwithstanding local measures. Going forward, Chen et al. (2020a) and Berry et al. (2021) find even no robust empirical evidence for a significant effect of lockdown measures on economic activity.

Table 2 Share of the economic impact attributed to lockdown measures

Studies regarding the Spanish flu also tend to find no significant effect of lockdowns on economic activity, both in the short and medium run. Some studies have documented an economic impact of the Spanish flu at the local (Dahl et al. 2020) or the global level (Barro et al. 2020) – in contrast however with Velde (2020) attributing the contraction to uncertainty surrounding the end of WWI. On US data, Correia et al. (2020) and Bodenhorn (2020) find no evidence for a significant economic impact of lockdowns in the short term, while Lilley et al. (2020) and Chapelle (2020) reach a similar conclusion regarding medium-term GDP growth. Such results should, however, be taken cautiously given the poor quality of the data and the fact that lockdown measures were far less stringent back then (Beach et al. in press).

Lesson 5: Relaxing the lockdown should be done gradually even during vaccine roll-out as the absence of an epidemic resurgence relies on rigorous sanitary measures.

Relaxing the lockdown should be done gradually and, where appropriate, differently depending on age and sectors. Gradualness is particularly essential if herd immunity has not been reached (Toda 2020). Dave et al. (2021d) and Singh et al. (2021) have documented anyway the persistence of individuals’ stickiness to pandemic behaviour, suggesting that even rapid and broad-based reopening may have muted impacts on mobility or economic activity. As such, the effects of restrictions and re-openings might be asymmetric depending on the phase of the pandemic (Dave et al. 2021a) with, for example, a smaller role for information shocks and lower demand for mitigation behaviours in the late-pandemic period. In addition, Favero et al. (2020) advocate for relaxing by age groups and by sectors to foster a quicker recovery. Baqaee et al. (2020) and Chang et al. (2021) also call for maintaining restrictions for ‘super-spreaders’ places (e.g. restaurants, gyms, pubs) and large public events.

Once lockdowns are relaxed, studies show the importance of sanitary measures to limit the spread of the virus even during vaccine roll-out. Renardy et al. (2020) find that delaying the re-opening does not reduce the magnitude of the following infectious wave, but only delays it; on the contrary, reducing levels of social interactions both delays and lowers it. Courtemanche et al. (2021) also document how the re-opening of schools in the midst of high virus spread and with only superficial social distancing rules substantially accelerated the spread of Covid-19 – in line with others putting this finding in perspective with the fact that, if associated with sanitary measures, school reopening has only modest effects on Covid-19 cases (Bravata et al. 2021, Goldhaber et al. 2021). Finally, Cot et al.’s (2021) model finds that vaccinations alone are not enough, and strict social distancing measures are still required until sufficient immunity is achieved. Agarwal et al. (2021) also find that relaxing restrictions during vaccine rollout significantly increase mortality.

While vaccination curbs infections and severe cases, fighting vaccine hesitancy might be key to achieve herd immunity and might require adequate policies. Broad-based vaccination campaigns have been shown to effectively limit the spread of the virus but above all the emergence of severe cases (Moghadas et al. 2021). However, building confidence in vaccines remains central to achieve a sufficient level of immunity in the population (Dror et al. 2020, Harrison and Wu 2020). General beliefs about vaccination and the own risk of infection can explain variations in this confidence (Wang et al. 2020, Sherman et al. 2021), but there is evidence that vaccine hesitancy is higher among low-income and low-educated people (Khubchandani et al. 2021). For this disadvantaged population, Alsan and Eichmeyer (2021) point to the benefits of relying on non-experts since the lower socioeconomic proximity of experts to them may undermine their trust in the vaccine. In addition, Gans (2021) finds that reducing the costs of access to the vaccine – for example, by providing vaccines at low-priced general merchandise retailers as suggested by Chevalier et al. (2021) – can improve its adoption. He finds, however, that this is not the case with policies that target the utility of un-vaccinated agents – for example, and most notably, vaccine passports – which either lead to equivalent or lower vaccine adoption. Finally, in terms of vaccination strategy, Więcek et al. (2021) show that fractional dosing would substantially reduce infections and mortality if it enables to increase the rate of vaccination.

Authors’ note: Finally, this column reflects the opinions of the authors and does not necessarily express the views of the Banque de France, LEO, LIEPP, or AMSE.