Some Basic Economics of COVID-19 Policy

The costs of the COVID-19 crisis come in two primary forms. The first is the direct impact in terms of health and lives lost. The second is the indirect impact that comes from efforts by individuals, private institutions, and governments to mitigate those health impacts, such as social distancing, stay-at-home orders, and mandatory business closures. It is imperative that we keep in mind that both are costs, and that less of one typically means more of the other. Like it or not, the first lesson of economics is that there are trade-offs, and choices are inevitable. 

Regardless of how we choose to bear them, the costs of the pandemic will be large. Some very rough estimates provide perspective. Based on our earlier work on the value of mortality reductions and improved health, we estimate that an unrestricted pandemic infecting 60 percent of the US population and with an infection fatality rate (IFR) below 1 percent would result in roughly 1.4 million deaths, heavily concentrated among the elderly, with a total value of lost lives of about $6 trillion.[1] For comparison, that is equivalent to about 30 percent of annual US GDP, suggesting that even small progress against the spread of the disease can be quite valuable. 

Against this, we estimate that efforts to slow the pandemic via a nationwide shutdown of “non-essential” economic activities would carry a cost approaching $7 trillion per year (roughly $20 billion per day), even ignoring other long-run costs from reduced values of human and physical capital and any intrinsic value of reduced civil liberties.[2] Of course, an unrestricted pandemic is implausible even in the absence of government interventions, as individuals have powerful incentives to engage in self-protection once the risks are even partially known. Even so, these are big numbers.

If we fully understood the trade-offs, the economic problem would be difficult but entirely standard. While the required choices would not be pleasant or easy, they amount to a far simpler problem than the one we actually face. If an extensive shutdown of economic activity costs $7 trillion, largely in terms of economic hardship, and a limited response would lead to a $6 trillion loss of life, then an intermediate solution could, in principle, achieve a great deal. The value of an intermediate solution is that by focusing on those actions that provide the largest benefits in terms of lives saved while imposing the least economic costs, we will be able to do better, maybe much better, than the trade-off of $7 trillion in costs for $6 trillion in benefits would imply. 

Take another policy arena involving externalities and decisions of life or death: highway safety. At one extreme we could consider the lives lost from fully unregulated driving behavior. At the other extreme we could take the costs of shutting down all highways. Neither extreme is among the serious alternatives, which include setting speed limits and requiring special licenses for operating the most dangerous types of vehicles. While it is helpful to know what it would cost to do nothing or to have a complete shutdown, the real policy work is on the intermediate solutions.[3]

In the current pandemic the benefits of reducing exposure vary substantially across individuals, while the costs of disease transmission differ substantially across both individuals and circumstances. These conditions imply that targeted policies are highly productive. For example, there is strong evidence that the risks of serious illness and death vary by an order of magnitude between the young and the old. At all ages, individuals with comorbidities, such as heart disease and diabetes, are much more vulnerable. Likewise, some activities (such as an individual driving alone in a long-haul truck, or working with familiar and careful colleagues) involve far less risk of infection than individuals greeting guests as they enter a hotel or interacting with strangers in a bar or at a sporting event.

The magnitude of the costs likely to be imposed by the pandemic, the gaps in the world's collective knowledge about COVID-19, and the dearth of experience in confronting similar events may contribute to the impression of a disorienting landscape for making policy decisions. This paper lays out the trade-offs involved with regulating the behavior of the general population during the COVID-19 pandemic.  It summarizes some of what we already know and highlights some key unknowns that remain. It compares relative advantages and disadvantages of large-scale social distancing (LSSD) regulation to a policy of screen, test, trace and quarantine (STTQ). A key feature of LSSD is high fixed cost—the impact on economic activity is roughly independent of the level of infection—while many of the the costs of STTQ scale with the level of infection. While some aspects of an optimal strategy are fairly general, we find that the optimal type and timing of strategies depends critically on whether we expect to contain the pandemic long-term or until a vaccine or cure arises or whether we focus on minimizing the long-run costs of a pandemic that will run its course. In particular we examine strategies for (1) limiting the impact of the disease for a given level of infection—a policy that makes sense regardless of the end-game scenario; (2) buying time to limit congestion of the health-care system or find a vaccine or effective treatment; (3) limiting the long-run impact of a disease that runs its course; and (4) adopting a long-run solution to contain infections indefinitely. It concludes with robust regulatory principles, including favoring decentralized mechanisms over direct control, distinguishing marginal from average effects, treating information as a public good, and ensuring that the chosen regulations deliver more net benefits than less-stringent alternatives.

Things we know with some confidence

In thinking about trade-offs, it is useful to begin with a few premises in which we, as economists, have a fair degree of confidence. 

• The costs of disease and premature death are high. Living longer is a good thing, and empirical evidence shows life and health are valued highly, but they are not the only thing. People’s behavior reveals that they are willing to bear greater risks to life and health in order to have more of other goods and services. This willingness to substitute leads to what economists call the Value of a Statistical Life (VSL)—what individuals would be willing to pay for a small reduction in their own mortality risk that would save one life in a large population. Estimates vary, but the current VSL used by US government agencies for cost-benefit analysis is around $13 million. That’s an average across all ages. The VSL for very old individuals is lower because they have fewer years of remaining life to lose, and because they are in generally poorer health than younger people. The value of a statistical life is a powerful tool because it allows us to assess some fundamental trade-offs between health and other aspects of people’s lives. It is critical to remember that the trade-off here is not between “lives” and GDP—it is the trade-off between two things that people themselves value: health and other aspects of their lives.
• The costs of a widespread shutdown of economic activity are also high, and unlikely to be sustainable over long periods. In recent research, one of the authors of this analysis, Casey B. Mulligan, puts the costs of a widespread shutdown like the current one in the US at $7 trillion per year—e.g. $1.75 trillion for a three-month shutdown. This includes some long-run costs but ignores others—such as those due to business failures (lost value of capital) and worker displacement (lost value of firm-specific human capital)—that will occur. It does not include any costs of deaths from the disease, some of which occur even with a shutdown. “Shutting off” the economy is not like shutting off the lights in your house—the economy will not return to its pre-pandemic productivity for some time, and many individuals and businesses will experience a permanent loss, so some of these costs represent future losses generated by the current economic distress.[4]  
• The fact that individuals put great value on their own health and longevity means that there are strong individual incentives to engage in self-protection. These incentives are particularly strong for the elderly and individuals with comorbidities, who are much more vulnerable to the disease. Similar to what is observed with seasonal influenza, we should then expect much lower incidence of infection in vulnerable populations, but higher rates of hospitalization and death for those infected. Early evidence indicates that is the case with COVID-19. Absent an end to the pandemic, self-protection via social distancing and other efforts would exist in the absence of government-imposed restrictions, and they will persist after those restrictions are relaxed. 
• Transmission of the virus is a two-sided affair involving an infected individual transmitting the virus to someone who is not yet infected. These interactions can be reduced by affecting the incentives of either side, but the costs of such policies will vary greatly across situations. When there are few infected individuals circulating in the population, quarantining the uninfected is not cost effective—a broad quarantine involves restricting the activity of many individuals to prevent relatively few infections. Under such conditions it is far more cost effective to work on the other side of the transmission mechanism, with a regime focused on testing and quarantining the infected and thereby limiting situations where one infected individual can infect many others. Broad-based lockdowns have a strong element of economies of scale—the high costs of implementing them is essentially unrelated to the level of infection in the population, while the benefits in terms of infections prevented are roughly proportional to the infection rate. This means that broad lockdowns make the most sense when the level of infection is high. In the language of economists, the marginal product of mandatory social distancing is greatest when there are many infected individuals circulating. 
• If recovered individuals have immunity (which is not yet established), then recovered individuals represent a substantial resource. They can interact with both the uninfected and the infected with no direct costs to themselves or follow-on costs to others. Identifying these individuals is important for helping them make good individual decisions as well as for recruiting them for tasks that would be not suitable for others. Conscription of the immune is a national security possibility, but in our view conscription would be inappropriate and create poor incentives for voluntary testing.  Compensating immune individuals for the valuable services they provide would create far better incentives.
• The nature of transmission via person-to-person proximity creates various “externalities”—individuals’ decisions to interact do not fully incorporate the costs imposed on others. In general, people with the disease (symptomatic or not) will have too many infection-causing contacts, and those without the disease will not have the proper incentive to avoid getting it. These “negative externalities” are the main reason for government-imposed social-distancing measures of the type now in effect. 
• An infected individual can also generate externalities on the treatment side. Existing capacity of intensive-care facilities and other health infrastructure is inadequate to effectively deal with massive illness within a narrow window of time. “Flattening the curve” (spreading a given incidence of disease over time) will improve health outcomes and save lives when congestion reduces the effectiveness of care. Individuals will internalize some of these costs because they anticipate their own care will be of lower quality if they become infected when hospitals are highly congested, but they will not internalize the costs they impose on others.
• The gain from reducing the incidence of infections in the presence of externalities may seem obvious, but there are important caveats. First, the externality created by an infected individual can actually benefit others (a positive externality), which would call for less social distancing. Suppression of the disease delays the development of “herd immunity”—if an individual’s recovery from the disease creates immunity and so the inability to transmit the disease to others, it might be better if those at low risk (such as the young) were quickly infected, while vulnerable groups were isolated. Moreover, when the pandemic will persist and the chance of future infection is large, a reduction in current contacts and infections can actually increase future infections. This can reverse the direction of the externality—individuals may have too much incentive to avoid infection early on if the primary impact of that avoidance is to increase infections later, when the externalities are larger. For example, if the highly vulnerable are isolated now but are difficult to isolate indefinitely, or if the externality from hospital congestion is less today than in the future, reducing infections today might make things worse.
• A vaccine that will stop the disease is possible, but it is unlikely to be widely available for a year or more. This implies that any plan to buy time in anticipation of a vaccine or cure will require a cost profile that makes a multiyear investment in mitigation worthwhile.
• Finally, we know that we will know more in a month or in six months than we do now. As we argue below, this is a key fact for designing policy responses, which can evolve from a more restrictive approach while the enemy is poorly understood, to a managed pandemic strategy characterized by more targeted restrictions as knowledge and information accumulate.

What we don’t know (yet):

We know much more about the nature of the pandemic today than we did a month or two ago. Many key features remain unknown, but they are knowable. 

• The extent of infection in the general population remains mysterious. Commonly discussed mortality rates are mainly based on “cases” of the disease, which occur when a person with more-or-less severe symptoms seeks medical care and tests positive. Limited testing capacity in the short run typically means that only those with severe symptoms and those known to have been exposed are tested. Evidence from several sources indicates that many or even most infected individuals are either asymptomatic or experience only mild symptoms—like a cold—that do not result in medical treatment or a test. Rough estimates of the share of unrecorded infections range from about half up to nearly 90 percent. Thus we have little knowledge of how lethal the disease really is, and how lethality differs across groups. Nor do we know how many carriers are in the population, the flow of new carriers, and the stock of individuals who have recovered from the disease and are possibly immune. All of these uncertainties can and hopefully will be resolved via random testing, both for the presence of disease and for antibodies that indicate recovery from it.
• On top of our ignorance of the extent of infection, we have limited evidence on the effectiveness of screening and testing at mitigating transmission. This is especially true at the beginning of the pandemic. As discussed below, this is a key reason for initially blunt policies such as mandatory social distancing and the forced closure of businesses. Yet limited evidence suggests that targeted policies informed by widespread testing can be very effective in arresting the spread of disease. As this is written, Iceland has (mostly non-randomly) tested more than 6 percent of its population and, using this information, has quarantined infected individuals and their contacts, and isolated the elderly and other vulnerable groups, all while imposing fewer society-wide restrictions than most other countries. Perhaps as a result, Iceland’s case mortality rate (currently below 0.4 percent) is among the lowest in the world, and its stock of known active infections has steadily declined by 60 percent from its early-April peak. Importantly, less than 5 percent of known infections in Iceland are among individuals aged 70 or above, compared to nearly 40 percent in Italy and Spain and over 20 percent in the US.[5] Random testing in Iceland has started in early April, and results indicate that less than 1 percent of the population is currently infected.[6]
• We are largely ignorant of the relative transmission rates of symptomatic and non-symptomatic individuals, and how many of each type there are. Evidence indicates that virtually all infected individuals are asymptomatic for at least a number of days, but they are nevertheless infectious to some greater or lesser degree. Some may never experience symptoms, and many have only mild symptoms that do not require medical attention. These individuals have little or no private incentive to be tested, and in most places limited testing resources have confined tests to those exhibiting severe symptoms and/or those known to be exposed. And because test results and symptoms are recorded at a point in time, we have very little information on how many asymptomatic and mildly symptomatic remain so.

What is in the toolbox?

There are two primary types of tools that have been proposed to fight the COVID-19 virus in the near-term:

1. Social distancing policies that attempt that mitigate the spread of the disease by limiting the extent to which individuals interact. Social distancing comes in many sizes. In our discussion we will focus mostly on large scale social distancing (LSSD) policies that are applied to the general population and cover a wide range of “non-essential” activities; examples of these policies include the stay-at-home orders currently in place in many countries. LSSD policies will typically have large costs that are independent of the level of infection because they restrict the activities of the general population rather than the activities of the much-smaller sub-group of the currently infected. Mulligan’s estimates imply current lockdown costs in the US of roughly $20 billion per day. Costs aside, early evidence indicates that LSSD has been successful at slowing the spread of the disease. Social distancing does not have to be uniformly applied nationwide; it can be applied to specific areas or populations where disease is most prevalent. We focus on LSSD policies due to their current prominence, but we also discuss the role that more targeted and more limited distancing policies can play.
2. The second widely proposed tool is a system we refer to as screen, test, trace, and quarantine (STTQ). Under STTQ individuals are screened on observable characteristics such as health condition (fever and other symptoms) and, based on screening criteria, tested for the presence of the disease. Those found to be infected are quarantined. Known contacts of the infected are traced, monitored, and quarantined as needed. When coupled with large-scale testing—such as random testing of the general population or testing of sub-populations suspected of having high levels of infection—these methods attempt to limit the spread of the disease by focusing on the potentially infected rather than those not infected. Importantly, unlike broad lockdown policies, the costs of STTQ are not purely fixed; instead they vary directly with the number of infected. The first component, screening, is an important part of limiting the costs of such a protocol since an effective screening policy reduces the population that must be tested.

Each method has a cost advantage. Their relative advantages will vary with the level of infection in the population. The LSSD “lockdown” policy is relatively cost effective (though still quite costly) when the level of infection is high. This is because the main component of costs—lost income from economic activity—do not increase with the level of infection. Further, LSSD reduces the chance of infection for a large number of individuals when many or all are at risk. In contrast, the STTQ policy is more cost effective, in relative terms, when the level of infection is low, because the main components of cost are only incurred when infections are found or when individuals pass some initial screen such as the existence of symptoms. Basic economics tells us that we should focus more on the broad-quarantine LSSD policy when infection rates are high and on STTQ when infection rates are low.

What are the objectives?

The broad policy objective is to save lives at the least cost. How that is achieved depends on circumstances and the nature of the disease, so it can be helpful to focus on more specific objectives as a means to reaching that overall goal. There are several specific objectives that a mitigation strategy might seek to achieve. It is helpful to outline these first since, as we argue below, different strategies will have different levels of effectiveness under each. As explained below, we believe specific objectives fall into four main categories:

1. Limit the impact of the disease for a given level of infection
2. Buy time
3. Limit the long-run impact of the disease while achieving population immunity
4. Adopt a long-run solution to contain infections indefinitely

We note up front that these objectives need not be mutually exclusive. For example, the first, limiting the impact of the disease for a given level of infection, is surely efficient regardless of other objectives. In contrast, achieving population immunity may be impossible, or simply unacceptable if the case mortality rate is very high, which means that long-run containment is the only viable strategy. For example, the CMR for the Ebola virus is around 50 percent—public policy is unlikely to accept that level of risk in order to achieve population immunity. 

For each objective we ask which of the two general strategies—LSSD or STTQ—would be most effective. What follows is a brief analysis of policies applied to each specific objective.

Limit the impact of the disease for a given level of infection

Both LSSD and STTQ limit the impact of the disease, though with different costs depending on the scale of infections at any point in time. Like STTQ—which is selectively applied by definition—more refined versions of LSSD can be targeted at segments of the at-risk population where the costs of infection are highest. Thus LSSD is likely to be an effective strategy in a densely populated area, such as New York City, when the disease is rampant and spreads easily via typical person-to-person contact, but it is less productive in a low-density area where infection rates remain low. And the two strategies can be applied in combination. Evidence shows that elderly individuals and those with existing comorbidities are far more vulnerable to the disease, so targeted LSSD for them is likely to be an element of any rational strategy, both because it mitigates infection and reduces health-care system congestion. STTQ can be applied to less-vulnerable groups and groups for which the cost of LSSD are high. Preliminary evidence from countries that followed forms of this mixed strategy—such as Iceland and South Korea--indicates large benefits in terms of limiting overall mortality, protecting vulnerable groups, and reducing strain on the health-care system. Since minimizing impact can be done for any level of overall infection, this objective can be combined with the others listed below.

Buy time

Potential discovery of a vaccine or cure implies that slowing the progression of the disease can have high value if doing so pushes the time of widespread infection past the date where a vaccine or cure arrives. This delay would be valuable even if it has no effect on the overall level of infection if a vaccine or cure fails to arrive. The key “output” we need to produce in this end-game scenario is time. In most models of disease progression the disease is slowed by reducing either the number of infected individuals at a given point in time—so interactions result in fewer future infections—or by limiting the rate at which a given number of infected can infect others. Both LSSD and STTQ can do this, but with much different costs. 

Consider a strategy that is applied for a length of time T, say a month, reducing the rate of transmission by X percent, say 50 percent. During the early exponential growth phase of the disease for given T and X, it doesn’t matter whether the strategy is LSSD or STTQ—the intervention will postpone the date at which any level of infection occurs by the same amount, which can be larger than T if the infected population declines during the quarantine.[7] We can then ask: when should the delay be applied, and by which method, so as to achieve delay at the lowest cost?  

Figures 1a and 1b illustrate the effects of “earlier” and “later” interventions during the disease’s exponential growth phase. Using a standard Susceptible-Exposed-Infected-Recovered (SEIR) model of disease progression, the figures compare the impact of a T=30 and X=50 percent intervention that begins at either day 30 or day 60 of the pandemic. By day 90, the point at which the later intervention ends, both interventions yield the same level of infections. But active infections under the early intervention are lower from day 30 to day 90. Further, cumulative infections under early intervention remain lower for even longer (Figure 1b)—the early intervention “buys more time”.[8] Another advantage of acting early, when active infections are low, is that the costs of achieving any delay will be lower under the more scalable STTQ method.