To stop the diffusion of the novel coronavirus, governments around the world have shut down their economies. The strategy has been effective, but also incredibly costly. The apparent trade-off between lives and livelihoods has been socially, politically, and economically contentious and therefore difficult to sustain.
In important new research from the Minneapolis Fed, senior economists have sought to ease this trade-off through “smart policies which are effective in reducing the spread of the disease while at the same time minimizing economic costs.”
“Smart policies … are effective in reducing the spread of the disease while at the same time minimizing economic costs.”
They develop a model of human networks, virus diffusion, and economic activity, and use it to analyze which networks are most proficient at generating economic value and which are best at spreading disease. To protect both health and jobs, they argue, policies should shut down interactions that contribute little economic output relative to their virus diffusion and maintain those that boost the economy without spreading disease.
Implementing a smart mitigation policy in early March 2020 would have prevented 300,000 COVID-19 cases and increased economic output by 1 percent by early May.
In a simulation based on the New York metropolitan area, the economists estimate that implementing a smart mitigation policy in early March 2020 would have prevented 300,000 COVID-19 cases and increased economic output by 1 percent by early May.
“Pandemic Control in ECON-EPI Networks,” a staff report by Minneapolis Fed senior monetary advisors Alessandra Fogli and Fabrizio Perri, with co-authors Marina Azzimonti of Stony Brook University and Mark Ponder of the University of Minnesota, focuses on the human networks that enable both economic activity and virus diffusion. Networks have numerous layers, the economists observe. “Individuals in the network differ in several dimensions,” they write, “and interact with each other through different network layers.”
Some, like families and factories, are quite stable, with repeated interactions among a fixed set of people. Other network layers are relatively unstable, with many random interactions between workers and customers—think of shopping malls, bars, or public transport. These unstable layers are more effective in spreading disease. Layers also differ in their economic productivity, some contributing more and some less to the nation’s output.
Cutting off some layers can have a large impact on infections, at relatively little economic cost. Closing down others will be economically costly but do little to save lives.
The disease and economic consequences of shutdown policies thus depend strongly on the type of layer interactions that are severed. Cutting off some layers can have a large impact on infections, at relatively little economic cost. Closing down others will be economically costly but do little to save lives. The economists’ model carefully differentiates among these layers, enabling strategic design to reduce viral diffusion at minimum cost.
The multilayered network is the first of three components to the economists’ model. To successfully estimate cost and disease consequences of various networks, the economists integrate it with two other elements. The second is the epidemiological (EPI) component, which describes disease dynamics. Economists have become adept this year at using a model first developed by epidemiologists in 1927. It describes the progression of a contagious disease through a population from those susceptible to infection (S), to those actually infected (I), and then to those recovered (R) or dead.
In the standard SIR model, the probability of infection is the same for all susceptible individuals and depends on the aggregate infection rate. In a network model, by contrast, individuals differ widely in infection probability, depending on their interactions with others. Do they live and work in stable layers with relatively few people or in unstable, random contact layers? The SIR model and the network model therefore have very different infection dynamics.
The third component is the economy (ECON). It describes the details of production and of connections between workers and shoppers. These details are central to likelihood of disease transmission, so the economists painstakingly classify actual business sectors into two types: high-contact (H), where workers have close, frequent interaction with different customers every day (retail, food, accommodation, schools, and health), and low-contact (L), where contact with customers is remote and/or infrequent (factory and some office work). (Many workers in L-sectors can work from home.) Because of this difference in proximity and frequency of customer interaction, production in H-sectors is more likely to spread infection than that in L-sectors.
The economists also estimate the economic value created by each worker—her or his marginal product—to gauge the output cost of different shutdown strategies. And they emphasize that in the H-sector, customer behavior is crucial. While the absence of producers affects supply in the economy, the absence of customers—due to not only shutdown policies, but fear—can drastically shrink demand: People stop spending, a phenomenon widely observed in the United States, with major economic consequences.
The model remains abstract until the economists bring it to the data, setting its parameters to match key features of the New York-Newark-Jersey City metropolitan area (New York MSA), a region of roughly 20 million people. The authors are meticulous in this procedure, devoting nearly a quarter of their paper to description and selection of model parameters.
Once they’ve designed and calibrated the model, the economists run several simulations. The first task is, simply but crucially, to validate the model. Can it match the data? In a word, yes.
They focus on the early pandemic period, from March 8 to May 25, and look at disease progression in the New York MSA. To ensure that the “agents” in the model (that is, people) interact at levels similar to New York reality at the time, they use Google mobility data, which measure population movement over time. Did people stay home? Did they travel to retail centers? Did they go to factories or office buildings?
These data reveal three stages: a sharp decline in mobility from March 8 to April 3, a subsequent lockdown period when mobility remained low, and a reopening period starting April 26, when mobility picked up again. The economists calibrate changes in network contacts in the model to match these mobility patterns and then examine the consequences for infection dynamics.
The simulation shows that the model can generate very accurate infection dynamics. While a standard SIR epidemiological model in which all people interact randomly projects geometric disease progression, the ECON-EPI network model matches New York MSA disease rates in exhibiting slower virus diffusion after the initial spike.
“In the network model,” the economists explain, “the infection naturally slows down, as the reduction in the number of contacts is enough to keep the infection local and prevent the disease from reaching the entire population.” Networks make the difference. In an SIR model, every individual is equally likely to contact every other individual, but that doesn’t reflect the reality of human networks. “In the network model, contacts are more clustered and less random.”
Smart mitigation and reopening
With the model validated, the economists run two policy experiments: the first to design “smart” mitigation policies, the second to investigate reopening strategies.
“Smart” means minimizing the economic cost of protecting lives from COVID-19. And the economists find that by increasing the proportion of H-sector workers put on furlough while sending more L-sector laborers back to work, policymakers can achieve a double gain: reducing disease spread and output loss.
In the New York metro, they estimate, “a policy involving a substantial double gain (reduction in infection cases equal to 1.5% of the population and 1% increase in output) could have been implemented.” (See figure.) Adding workers back to the L-sector would raise output with little change in infection level, while reducing workers in the H-sector (particularly smaller businesses) would substantially reduce infections at little sacrifice in output.
"Smart mitigation" policy would lower infection and increase output
Loading chart 1...
Loading chart 2...
Note: Figures indicate, respectively, percent of population that is infected, and economic output relative to prepandemic level of standard epidemiological model (benchmark) and “smart” policy that puts more H-sector workers on furlough and sends more L-sector laborers back to work.
Source: "Pandemic Control in ECON-EPI Networks," Azzimonti, Fogli, Perri, and Ponder (2020)
Their final set of simulations relates to strategies for reopening after lockdown. The key finding of this work is that the timing and extent of reopening are crucial. Broadly reopening high-contact sectors and/or schools while infection levels are still high will, in their model, inevitably generate a large second wave. “Our work suggests that if pools are open the Fourth of July, it is impossible to safely reopen school by Labor Day,” Perri commented.
By adding the crucial component of networks that account for how humans actually interact, [the model] generates accurate projections of infection diffusion and enables policy design to minimize the economic cost.
A wiser course would be to begin reopening the economy fairly early on, but only in the L-sector. That would generate significant gains in economic output with relatively little increase in infection growth. But even this modest reopening wouldn’t create conditions sufficient to allow school reopenings. This highlights the complementarity among different social layers, the economists observe. Even though infection rates wouldn’t dramatically rise with L-sector reopening, it would be sufficient to lead to a pandemic rebound if schools also opened.
The economists’ model is a significant achievement. By adding the crucial component of networks that account for how humans actually interact, it generates accurate projections of infection diffusion and enables policy design to minimize the economic cost of curbing disease progression. Saving lives and saving jobs becomes feasible, and policy choices less draconian.