The rapid global spread of the SARS-CoV-2 virus and the associated COVID-19 disease has sparked urgent questions about the impact of epidemics on the economy. The objective of this blog post is to provide a non-exhaustive overview of what we – from an academic research perspective – currently know about this impact. Where I see them, I also highlight some unanswered questions in this “coronomics” literature, and point to possible weaknesses of the current frameworks.

Macroeconomic Models with Contagion Effects

A rapidly growing body of research merge SIR or SEIR models of contagion into standard macroeconomic models (see, e.g., Alvarez, Argente, and Lippi, 2020, Bodenstein, Corsetti, and Guerrieri, 2020, Eichenbaum, Rebelo, and Trabandt, 2020, Glover, Heathcote, Krueger, and Ríos-Rull, 2020, Jones, Philippon, and Venkateswaran, 2020, and Kaplan, Moll, and Violante, 2020). SIR/SEIR models are differential-equation models used within epidemiology to study the spread of an infectious disease. The models divide the population into compartments, such as susceptible (S), exposed (E), infectious (I), and recovered (R), and illustrate the flows between each compartment. The models can be used to predict the prevalence (total number of infected) or the duration of an epidemic, or to evaluate the effect of, say, mitigation and vaccination measures.[1]

In the macro-SIR models, an epidemic has both aggregate demand and aggregate supply effects. The supply effects appear because people reduce their labor supply, in order to avoid being infected, and because sick people are less productive. (The strength of these effects is unclear empirically. Many hourly employees have neither the right to paid sick leave nor the savings necessary for an unpaid leave. Thus, it seems unlikely that they will voluntarily self-quarantine, even if they fall sick.) In Jones, Philippon, and Venkateswaran (2020), supply effects also appear because people are less productive when working from home. The demand effects emerge since people reduce their consumption, in order to avoid exposure to the virus by limiting the time shopping and because of precautionary saving. In the models, the supply and demand effects generate large recessions even in the absence of public health interventions.

A cornerstone of the macro-SIR frameworks is that the competitive equilibrium is not Pareto optimal, because (infected) people do not internalize that their consumption and work activity may infect other people. Thus, there is an externality in the disease spread. This opens up the possibility of welfare-improving social distancing, implemented by the government by reducing economic activity. In Jones, Philippon, and Venkateswaran (2020), interventions also work through working-from-home technologies that front-load the optimal intervention, since working from home involves learning by doing.

Eichenbaum, Rebelo, and Trabandt (2020) make three interesting points about the economic effects of medical preparedness, treatments, and vaccines. First, in the case of too little medical preparedness, the competitive equilibrium involves a more severe recession, because people internalize the higher fatality rates caused by crowded hospitals providing substandard treatments. Thus, the public cuts back more aggressively on its consumption and work, to reduce the probability of being infected. Second, if people expect a treatment to arrive, they become more willing to engage in market activities, since the expected cost of an infection is smaller. This limits the severity of the recession, but also increases the number of people infected until the treatment arrives. Thirdly, with vaccines as a possibility, it is optimal to introduce more front-loaded and draconian mitigation measures to minimize infections. This causes a deeper, but hopefully also shorter, recession, as the measures may be lifted once the vaccine arrives.

Heterogeneity

The costs associated with COVID-19 are unevenly distributed. Health risks accrue disproportionally to the elderly. Yet, to the extent that this group has stable pensions, it is unaffected by the economic hardship brought about by mitigation measures. These economic costs instead accrue directly to workers who cannot work from home or have jobs requiring much social interaction (e.g., waiters or shop assistants). These workers also have disproportionally low incomes and low liquid wealth. Using heterogeneous-agent macro-SIR models, Glover, Heathcote, Krueger, and Ríos-Rull (2020) and Kaplan, Moll, and Violante (2020) show that this heterogeneity in life situations adds an unpleasant distributional dimension to the trade-off between the severity of a recession and the health consequences of an epidemic, already present in representative-agent macro-SIR models. (Kaplan, Moll, and Violante, 2020 assume homogeneity in health risks, and instead focus on profession-contingent wealth inequality. Heterogeneity in health risks should generally accentuate the policy trade-off also in their model.)

In countries with strong social protection, the heterogeneity in consumption responses following the COVID-19 outbreak may not be so large. For instance, Andersen, Hansen, Johannesen, and Sheridan (2020) find only a modest (3 p.p.) excess reduction in spending by workers in a shutdown sector, using Danish credit card data. A possible explanation for this is the government furlough scheme, implemented by the Danish government with the goal of preventing mass layoffs. The scheme replaces 75 pct. of workers’ salaries at qualifying firms. About 5 pct. of the Danish labor force is currently participating in the scheme.

Shutdowns: Demand or Supply Shocks?

Eichenbaum, Rebelo, and Trabandt (2020) effectively represent mitigation measures that shut down sectors of the economy as demand disruptions, in that these measures enter into households’ demand first-order conditions. Guerrieri, Lorenzoni, Straub, and Werning (2020) instead argue that shutdowns are “Keynesian supply shocks”, namely supply disruptions that trigger changes in aggregate demand larger than the disruptions themselves. Thus, prices fall rather than rise, as in the case of a regular adverse supply shock. The mechanism works as follows: When workers lose their income due to the shutdown, they reduce their spending. In one-sector models, this always causes a reduction in demand smaller than or equal to the reduction in income. In multi-sector models, however, the reduction in demand may be larger if the complementarity between goods from different sectors and the intertemporal elasticity of substitution are both sufficiently high. (Intuitively, I do not demand hotel accommodation unless I travel to another country, something than I am currently banned from doing.) Guerrieri, Lorenzoni, Straub, and Werning (2020) also argue that fiscal stimuli may be less effective than usual, because the stimuli, per construction, cannot help shutdown sectors. The authors instead advocate policies (e.g., lax monetary policy, loan forbearance, grace periods on tax payments, or rent subsidies) that reduce the fixed costs incurred by temporarily closed businesses, since such cost reductions may forestall bankruptcies.

Conceptually, what separates supply and demand shocks is the co-movement of prices and quantities. However, when a sector of the economy is shut down, the quantity produced turns to zero, while prices are indeterminate. This is the point that Guerrieri, Lorenzoni, Straub, and Werning (2020) also make. However, because prices are indeterminate, I believe that the reference to shutdowns as adverse supply shocks (or as demand shocks, for that matter) in macroeconomic models is imprecise. Shutdowns are something else, exactly because prices are indeterminate.

The Spanish Flu

A key take-away from the theoretical literature is that externalities arising from the interactions between epidemics and economic activity naturally point to a role for policy interventions. The last pandemic with severe economic repercussions was the 1918 Flu or the Spanish Flu.[2] A group of papers empirically investigates the economic effect of mitigation measures taken during this emergency. Correia, Luck, and Verner (2020) regress manufacturing employment on proxies for the timing and intensity of mitigation measures across U.S. cities, along with additional control variables. They rule out interventions amid the pandemic as a source of economic decline. Moreover, their point estimates suggest that interventions might support employment by suppressing the pandemic. Whether the results generalize to COVID-19 is unclear. The Spanish Flu was chiefly fatal for working-age people. By contrast, the mitigation measures taken today are mainly aimed at protecting the elderly – a low-productive part of the population – from the disease.

Using international variation, Barro, Ursúa, and Weng (2020) regress measures of real activity on information about the intensity of Spanish Flu-related deaths, while statistically controlling for World War I-related losses. They find that GDP and consumption fell by 6 pct. and 8 pct., respectively, in the typical country, due to the direct and indirect effects of the flu. They argue that this is likely an upper-bound estimate of the medium-term effect of COVID-19 on the economy, again because this disease predominantly is fatal for the elderly.

A concern I have with both studies is their reliance on case fatality rates, defined as the number of diagnosed deaths relative to the diagnosed population. Diagnostic practices vary widely across regions. For instance, it could be that poor regions under-diagnose deaths, due to limited testing capacity. The estimated economic effect of an epidemic will then be biased upward if the economic effect can also be (partially) captured by other variables, such as time dummies. For this reason, epidemiologists often use all-cause excess mortality rates, which is the current all-cause mortality rate relative to its historical level. The idea behind using this statistic is that, as long as no other public health emergencies unfold contemporaneously, all excess mortality derives from the epidemic.

Looking Ahead

While an impressive amount of research has been done in a short period, many questions remain unanswered. Foremost among them is that the above-mentioned theoretical frameworks all assume that once governments lift their interventions, households’ consumption and labor-supply decisions will follow the same behavioral rules as before the crisis. Long-termed psychological effects related to, say, renewed travel-patterns, fear of restaurants, or social anxiety are thereby excluded by assumption. Such psychological effects could keep households’ demand low for a protracted period, particularly for services.

Standard macroeconomic frictions relating to demand still remain to be built into to macro-SIR models. For instance, falling real estate prices tend to propagate via the housing market, by eroding wealth, leading technical insolvency to proliferate, and causing banks to incur losses on mortgage lending. Such a development could lead to a credit squeeze, or, worse, a financial crisis. Moreover, as argued by Mian, Straub, and Sufi (2020), expansionary fiscal policies taken today boost households’ and governments’ debt levels and lead to smaller aggregate demand in the future, thus further lowering natural interest rates and supporting debt growth.

A retrenchment of the supply side could also slow down the recovery. Total factor productivity is likely to be persistently low, on account of disruptions of supply chains and the destruction of organizational capital associated with shutdowns and firm bankruptcies. Labor supply may too be smaller in the coming years, because of unemployment hysteresis effects, in addition to disease-related fatalities. Finally, labor productivity may fall because of insufficient business investment (e.g., due to strained balance sheets, demand-uncertainty, and debt-overhang) or deficient human capital investment (e.g., due to cancellations of apprenticeships and supplementary training).

Finally, further empirical evidence is needed to disentangle the mutual relevance of the theoretical mechanisms outlined above. This would provide a crucial step toward, not just an understanding of which research questions warrant each set of assumptions, but also a unification of the approaches already existing. I am convinced that this will remain a principal task of economists in the decade to come.

I would like to thank Adam Sheridan, Kasper Roszbach, Leif Brubakk, Mikkel Gandil, Nicolai Ellingsen, Norman Robert Spencer, Ragnar Juelsrud, SeHyoun Ahn, and Søren Hove Ravn for comments on the post. I would also like to thank Kåre Mølbak for introducing me to epidemiology.

[1]Atkeson (2020) and Berger, Herkenhoff, and Mongey (2020) provide introductions for economists to SIR/SEIR models of COVID-19 spread, without and with testing, respectively.

[2]The only thing particularly Spanish about the “Spanish Flu” is its name. Neutral Spain did not impose censorship during World War I. Widespread stories about the flu in Spanish newspapers therefore lead to the false impression that Spain was hit especially hard.

Bankplassen er en fagblogg av ansatte i Norges Bank. Synspunktene som uttrykkes her representerer forfatternes syn og kan ikke nødvendigvis tillegges Norges Bank. Har du spørsmål eller innspill, kontakt oss gjerne på bankplassen@norges-bank.no.