The reactions to the COVID-19 pandemic have been widely heterogeneous across countries and subject to criticisms from one direction or another. As shown by Jinjarak et al. (2020) and Askitas et al. (2020), some countries have met with more apparent success than others. China took drastic measures initially – stopping all economic activity in the most affected areas, as did New Zealand – until the virus was suppressed. Japan and South Korea implemented policies to mitigate the virus’s diffusion without greatly affecting their economies. Sweden minimised its intervention to such a degree that we can name it a no-intervention policy. Since any intervention is economically costly, these different approaches raise questions: should countries follow China’s and New Zealand’s suppression strategies? Or is it better to do just enough to keep the affected population under the hospital capacity constraint? If so, how much is enough? Could a no-intervention policy be optimal? What role should testing play in this context?
The key factors
In Piguillem and Shi (2020), we stress three fundamental issues that we think should drive optimal interventions. First, whether to choose suppression or mitigation depends on the possibility of eliminating the virus. Many epidemiologists argue that this is not possible and that it would eventually be endemic. For instance, some believe that SARS (Severe Acute Respiratory Syndrome) is not extinct but dormant, waiting to come back. This is embedded in the multiplicative structure of the standard SIR models (Susceptible, Infectious, Recovered). By imposing restrictions, one can divide the number of virus carriers as many times as one wants by another number, but the result would never be zero. Once the division stops (the restrictions are lifted), the number starts to grow again. Thus, governments should seek only mitigation: flattening the curve. We think that previous experiences, such as those with SARS, imply that the virus could indeed be suppressed. To give suppression a chance, we assume that whenever the number of contagious carriers is below one person, the virus is eliminated.1 Second, it is important to consider the capacity of the health system to deal with a large inflow of patients. By many accounts, COVID-19 is not an extremely deadly illness when properly treated. Hence, considering hospital capacity is a first-order issue. Third, there is a behavioural response: knowing of the virus’s existence, people choose to be cautious and thereby slow the speed of the spread. This reaction could be significant, as shown by Farboodi et al. (2020) and Durante et al. (2020). The government must internalise behaviours such that any intervention only complements private individual decisions.
Last but not least, the main friction creating the need for indiscriminate quarantines is the government’s inability to distinguish the asymptomatically infected from the healthy population. If it could make this distinction, only the affected would quarantined, letting the unaffected population continue with their normal activities. As stressed by Galeotti et al. (2020), testing is valuable for many reasons, but it is also costly, and mass testing could be prohibitively expensive. This is a cost-benefit analysis that should be properly addressed in the current situation.
Mitigation, suppression, or no-intervention?
We use data arising from the outbreak in Italy. Since the official case number grossly underestimates the actual prevalence, we target the number of fatalities. But the fatalities are also underestimated in the official data, at least in the early days of the outbreak, so we use the excess deaths per day relative to previous years published by the national statistics institute. To capture the population’s reaction, we use the cell-phone mobility index constructed by Durante et al. (2020), which shows that Italians reacted sharply to the presence of COVID, reducing their movements by almost 25% even before any government intervention. Depending on the relative value of one life and the government’s aversion to recessions, we find three types of optimal policies: non-intervention, suppression, and mitigation. However, conditional on following one of these strategies, the intensity and duration of lockdowns is barely affected by either the aversion to recessions or the relative valuation of life.
There are two important components to understanding these results: the minimum critical mass and the endogenous reaction of the population. First, if one believes that the critical mass is zero, then suppression strategies are never optimal; governments should seek only to mitigate and flatten the curve. In this case, the proportion of economic activity that must be shut down is substantially lower than implemented by Italy and many other countries. This mitigation policy amounts to shutting down between 10% to 15% of economic activities for about 65 days to ensure that hospitals are not overwhelmed at the peak of the outbreak.
In contrast, when the government perceives that the virus can be suppressed and chooses the suppression strategy, the lockdown is substantially stricter: 60% of all economic activity must be shut down for 75 days. In either case, the intensity and duration of the lockdowns are independent of the relative value of a life versus foregone production, determined by the dynamics of the virus and only that.
Does this mean that the statistical value of life and cost in terms of output lost are irrelevant? The answer is clearly no. The value of life and output cost matter, but only in determining whether to seek mitigation, suppression, or no-intervention whatsoever. When the value of a life is not too large and the population’s behavioural response is large, it is optimal not to intervene, in line with the view of Maloney and Taskin (2020). As the value of life increases, intervening becomes optimal, but the intensity of the intervention does not increase gradually. Instead, the intensity jumps to suppression, when possible, or to mitigation when suppression is perceived as impossible. This could explain the apparently large dispersion in government reactions. Small variations in the perceived trade-off between lives and production and individual self-containing measures can generate large variations in public responses.
We want to emphasise the relevance of the endogenous social reaction. For instance, Walker et al. (2020) initially estimated 645,000 fatalities for Italy without any intervention. However, given the observed population’s reaction before the intervention, we estimate that the number of fatalities would have been 215,000. In contrast, with an optimal suppression policy, there would be 17,000 fatalities, while with optimal mitigation the number of fatalities is substantially larger, at least 180,000. When comparing both policies with the observed quarantines, we find that the implemented quarantine in Italy is too ‘soft’ to be the optimal suppression and too ‘harsh’ to be the optimal mitigation strategy. Nevertheless, conditional on the information available at the time, it is troublesome to deem Italy’s strategy as suboptimal.
Testing is a valuable yet expensive substitute for indiscriminate quarantines
What role should testing play in these strategies? The main complication when analysing testing policies is that its cost is uncertain. We may have good information about the cost of a small number of testing kits, but it is unclear how to extrapolate those values to mass testing strategies. It is not even clear if massive random testing is feasible. Can we test one million people in a day? Is it technologically feasible? If so, how costly would it be? We take the stance that mass testing is technologically feasible, in the sense that one could potentially rapidly increase testing capacity, but we assume that the cost grows exponentially with the number of tests. If NASA employees were asked in 1963 whether it would be possible to go to the moon in six years, they would have said no; but when the government decided to spend 3% of GDP on the project, it became technologically feasible.
Hence, we assume that the marginal cost of the first test is one day of daily output per worker and grows quadratically. The speed at which the marginal cost grows is chosen in such a way that it would be economically infeasible to test the entire population at once. We find that testing is intensively used as a substitute for indiscriminate quarantines and generates substantial welfare gains. The output gains are so large that lockdowns can be completely avoided. In our favourite scenario, testing is used intensively, with an average of 1% of the unidentified population tested every day for slightly under two months. This policy is very costly, amounting to 1.9% of annual GDP. But the cost is easily compensated by rendering the indiscriminate quarantine substantially milder. Instead of shutting down 60% of economic activities, it requires shutting down only 50%; instead of doing so for 75 days, it does so for only 55. Finally, we find that the implemented quarantine in Italy closely resembles the optimal suppression quarantine with testing.
Walker, P G T, C Whittaker et al. (2020), “The Global Impact of COVID-19 and Strategies for Mitigation and Suppression”, On behalf of the imperial college covid-19 response team, Imperial College London.
1 The standard SIR model allows for fractions of a person to be infected, which is what technically generates the impossibility of suppression.