A “mobilize and transition” strategy could reduce COVID-19 mortality while cushioning the economic decline.
Read the Reflection, written 30 July 2021, below the following original Transmission.
While the human toll of the COVID-19 pandemic has been apparent for some time, the economic picture is now starting to come into greater focus. Initial unemployment claims in the United States jumped from 280,000 to almost 3.3 million for the week ending March 21 [1], then doubled to over 6.6 million for the following week [2]. By way of comparison, weekly unemployment claims have never previously exceeded 700,000 in the history of the recorded data. The S&P 500 Index reached a record high in mid-February, then lost a third of its value in a month. Congress passed a $2 trillion stimulus bill on March 27, one quarter of which allows for loans and grants to firms under the discretion of the Secretary of the Treasury. And the Federal Reserve invoked the “unusual and exigent circumstances” clause of Section 13(3) of the Federal Reserve Act to break out of its usual shackles and channel credit to (non-bank) firms, states, and municipalities.
The plan seems to be to drastically scale back economic and social activity and wait for the pandemic to pass, in the hope that it will do so in about two or three months, with a rapid return to normalcy thereafter. A statement by Secretary of Labor Eugene Scalia [3] on the unemployment numbers exemplifies this thinking; he observes that the stimulus bill “provides incentives and funding for businesses to keep their workers on payroll, so that, as soon as possible, we can spring back to the strong economic conditions we enjoyed just weeks ago.”
But what if three months is not enough [4], and we see ebbs and flows in confirmed cases over one or two years, in concert with the relaxation and tightening of social distancing measures? The social, political, and economic implications of this would be dire. And what if changes in the composition of demand [5] for goods and services are enduring, with less expenditure on travel and lodging and more on public health and distance learning for years to come? Then a large-scale reallocation of workers and capital across sectors will be needed even after the threat has passed, and a return to the pre-crisis status quo will not be possible in any case.
An alternative paradigm [6], which we call mobilize and transition, allows for a return to active participation in economic life for some portion of the population long before the pandemic has been fully contained. The goal is to use an initial period of aggressive social distancing of up to three months in order to build out the infrastructure of pandemic preparedness and management that countries like Taiwan and Singapore have used to maintain far greater control over COVID-19 than we have managed. We mobilize in order to transition to being a society with the kind of pandemic resilience that permits maximal mobility for as large a portion of the population as possible even when the pandemic is ongoing.
This strategy involves large-scale testing, on the order of several million individuals per day, in order to partition the population into those who are believed to be safe and those whose status remains undetermined. It involves two regimes, which we call find the safe and find the virus, with transitions between regimes being contingent on epidemic trajectories and economic conditions.
In the find the safe regime, those with a recent negative disease test or a serological test indicating immunity can return to the workforce, entering occupations for which there is intense demand. This will require credible verification of safe status. Testing would be focused on those in the health and care professions, first responders, sanitation workers, and those connected with production and delivery of food and other essential goods and services. And it would include those who are willing and able to enter or re-enter these occupations without the need for extensive training, and those who could provide such training as is deemed necessary.
The find the virus regime involves broad-based testing in order to find and isolate those who are infected, and to warn and recommend testing for their contacts. In order to make best use of scarce testing resources and personnel, those with greatest likelihood of carrying the virus should be prioritized for testing, and this determination could be based on location, occupation, demographic characteristics, and proximity to others who have recently tested positive.
There is a useful analogy here to the literature on police stops and searches, where differences across groups in contraband recovery rates (or hit rates) are viewed as a diagnostic test [7] for discrimination. A non-discriminatory police department seeking to maximize, say, weapon recovery should conduct searches in such a manner as to equalize marginal hit rates across identifiable subgroups of the population. That is, the likelihood of weapon recovery should be independent of group membership among those who are close to the threshold for a search. Translating this to the case of testing, the targeting of individuals should be such that the likelihood of testing positive is roughly equalized across locations, occupations, and demographic groups, at least among those who have been recently tested. If a location is turning up more positives than another at the margin, resources would be better used by shifting to the former at the expense of the latter. Such adjustments require extensive mobile testing capability.
Innovative use of mobile technologies can facilitate more finely targeted testing, while preserving privacy and civil liberties. For example, an app developed by MIT researchers [8] collects location data every five minutes and stores it locally without any identifying information. Anyone testing positive can transfer this data to a health professional, who can upload it to a central server, again with all identifying information redacted. This allows intersections to be traced, and people who have crossed paths with those who have tested positive to be warned, even though their personal data never leaves their phone without their permission. Other related applications are under development.
One essential feature of the proposed strategy, and indeed any strategy under current conditions, is the universalization of mask use, subject to availability of supply. Widespread mask use has helped limit contagion in many Asian countries, but the practice remains far from universal in the United States. The meaning of mask use needs to be transformed through public messaging [9] by civic and political leaders, so that it is associated with altruism and civic responsibility instead of carrying the stigma of sickness or fearfulness [10].
We believe that the mobilize and transition strategy will result in lower mortality from the disease while cushioning the decline in output and employment and leading to a more rapid recovery with a very different allocation of workers and capital across sectors relative to the pre-pandemic period. But the economic hardship will nevertheless be extraordinary, with double-digit unemployment rates for several months, if not years. Social support therefore has to be an essential component of the strategy. The stimulus bill allows for some cash payments to households, but this could be routinized in the form of a basic income and distributed through individual accounts at the Federal Reserve [11]. The latter would be a radical departure from current practice, but these are indeed “unusual and exigent circumstances” under which the central bank has the power — and indeed the responsibility — to take steps that may have been unimaginable in quieter times.
Danielle Allen
Harvard University
E. Glen Weyl
Microsoft Office of the Chief Technology Officer; RadicalxChange Foundation
Rajiv Sethi
Barnard College; Columbia University; Santa Fe Institute
REFERENCES
- https://www.dol.gov/sites/dolgov/files/OPA/newsreleases/ui-claims/20200510.pdf
- https://www.dol.gov/sites/dolgov/files/OPA/newsreleases/ui-claims/20200551.pdf
- https://www.dol.gov/newsroom/releases/osec/osec20200326
- https://ethics.harvard.edu/when-can-we-go-out
- https://news.gallup.com/poll/306053/americans-hesitant-return-normal-short-term.aspx
- https://ethics.harvard.edu/mobilizing-political-economy
- https://journals.sagepub.com/doi/abs/10.3818/JRP.4.1.2002.131
- http://safepaths.mit.edu/
- https://www.nytimes.com/2020/03/31/health/cdc-masks-coronavirus.html
- https://asia.nikkei.com/Spotlight/Coronavirus/Asians-in-US-torn-between-safety-and-stigma-over-face-masks
- https://internationalbanker.com/finance/social-support-financial-architecture-proposal-reform/
T-007 (Allen, Weyl, Sethi) PDF
Read more posts in the Transmission series, dedicated to sharing SFI insights on the coronavirus pandemic.
Listen to SFI President David Krakauer discuss this Transmission in episode 27 of our Complexity Podcast.
Reflection
July 30, 2021
PREDICTION AND POLICY IN A COMPLEX SYSTEM
A number of events occurred in 2020 that were interconnected in complex and subtle ways that no single academic discipline is well-equipped to understand.
Two of these—initial claims for unemployment benefits (measured weekly) and new confirmed deaths from COVID-19 (measured daily and smoothed using a seven-day average)—are shown in figure 1, both for the United States. The two series track each other closely initially, with deaths lagging claims, and then start to diverge from July onward.
These two phenomena are clearly connected, and we could better predict and respond to each of them if their linkages were more deeply understood. However, it would have been virtually impossible to use historical data to build and calibrate models that provide an integrated analysis of this kind. For one thing, initial weekly unemployment claims reached a peak of almost seven million in April, after never having exceeded 700,000 in earlier years, even at the depths of the global financial crisis of 2008-2009. That is, the labor market data in figure 1 lie outside the range of all observed historical experience. Similarly, one needs to go back a century for a pandemic with comparable levels of mortality in the United States.
Figure 1. Daily new confirmed COVID-19 deaths per million people (seven-day average) and weekly initial claims for unemployment insurance (millions of persons) in the United States throughout 2020. Sources: Our World in Data (https://ourworldindata.org/explorers/coronavirus-data-explorer) and Federal Reserve Economic Data (https://fred.stlouisfed.org/graph/), respectively.
There were many other events in 2020 that were connected to these two, but in ways that are difficult to grasp based on traditional scientific methods and boundaries. The mass actions following the killing of George Floyd were quite possibly the largest in American history1 for a time; at peak in June there were a half million people protesting in more than five hundred cities. Mississippi removed the confederate battle emblem2 from its state flag after a century and a quarter, which few would have imagined possible a year earlier. And Congress voted to override3 a presidential veto of a defense bill that began the process of renaming military bases that honored Confederate leaders.
A veto-proof majority of the Minneapolis City Council pledged to dismantle4 the city’s police force, then began to reverse course5 a few months later. New York City disbanded its plainclothes anticrime units,6 which were responsible for a disproportionate share of abuse complaints and deadly force incidents, but also for significant numbers of gun confiscations. Many police departments responded to threats of defunding with slowdowns and pullbacks.7 There was a rise in shootings and homicides8 across towns and cities nationwide, the causes of which remain poorly understood.
Scientific understanding of such interconnected phenomena is hampered by the fact that they are addressed by largely separate research communities. Sociologists and criminologists look at crime and policing, epidemiologists at the spread and containment of disease, economists at financial and labor market movements, and so on. This specialization has both costs and benefits. It allows for deep focus on a key set of interactions and causal chains, at the cost of omitting factors that are (or can quickly become) highly relevant.
How might understanding, prediction, and policy design be improved by taking such interactive complexity into account?
One approach is to leverage big data, computational capacity, and recent advances in machine learning to build versatile and flexible models. The reliance on historical data is a limitation, however, because accurate prediction is most needed when one is outside the range of prior experience. Unique policy initiatives can make inference from past experience unreliable. The two series shown in figure 1 start to diverge in part because of the effects of large-scale policy interventions like the CARES Act,9 which was signed into law in late March, and which had a much stronger impact on the economy over time than on the spread of the disease.
Purely data-driven technocratic approaches sometimes fail spectacularly because a lot depends on the data ontology, and relevant issues need to emerge through a bottom-up rather than top-down process. This suggests the need to explore participatory and democratic mechanisms that can be used to augment and improve upon top-down technocratic data analysis.
One approach to harnessing such decentralized knowledge, including information and insight held by nonexperts, is through market mechanisms. Prediction markets are exchanges on which state-contingent contracts can be traded; these contracts pay a specified amount if a referenced event occurs and nothing otherwise. Contracts that reference election outcomes have been traded on the Iowa Electronic Markets10 for over thirty years, and other venues such as PredictIt11 in the United States and Betfair12 in the United Kingdom list heavily traded prediction market contracts. Kalshi,13 a major new regulated exchange for the trading of such securities, launched in 2021.
With suitable caveats, the price of a prediction market contract (relative to the payment promised if the referenced event occurs) can be interpreted as the event’s probability of occurrence, as judged by the market. Any feature that a trader considers relevant influences the contract price, so there is no ex ante restriction on the sources of information used for any particular predictive task.
But markets are just one mechanism for participatory information aggregation, and they have certain drawbacks. They tend to create a competitive—rather than participatory and communicative—dynamic and thus somewhat reduce the incentive to engage in nonlinear conversation. In addition, prices depend on the distribution of trader wealth, which need not be closely associated with good judgment. There can also be incentives for manipulation14 when beliefs about an event can affect its objective probability of realization.
There exist alternative mechanisms for the aggregation of information and judgment that rely on persuasion and consensus building rather than strict competition. The pol.is15 digital tool, for example, allows for the posting of policy statements by any user, which can be voted on by others, with the various positions being displayed on an interactive real-time map showing clusters of opinion. Users are incentivized to offer statements that can attract broad support, and the map can reveal the evolution of changing positions over time as the population of users moves toward agreement. This platform has been used in Taiwan to inform legislative action16 on a number of issues.
During the early stages of the pandemic, the use of innovative digital tools17 for information sharing in Taiwan helped to identify and reduce exposure without onerous restrictions, allowing the country to achieve lower case and death rates18 compared to most democracies at comparable levels of prosperity and life expectancy.
The tools used to aggregate information for policy purposes could also, in principle, be used to improve scientific forecasting. Like prediction markets, such crowdsourced forecasts are not constrained by preselected variables or historical data in the way that most models are. Integrating conventional models, prediction markets, and nonmarket crowdsourced aggregation mechanisms remains an interesting challenge19 for researchers.
But no matter how much better or well-integrated with other mechanisms the models get, there will always be spectacular failures of prediction. For this reason, it is important to design institutions and policies in such a manner as to be robust to limited understanding of the underlying science.
In the case of police use of deadly force, for instance, Lawrence Sherman has argued20 that focusing on the individual culpability of officers is much less likely to lead to effective change than attention to the “complex organizational processes that recruited, hired, trained, supervised, disciplined, assigned, and dispatched the shooter before anyone faced a split-second decision to shoot.” That is, most police homicides ought to be handled more like failures of air traffic control than crimes, resulting in an evaluation of organizational systems21 alongside prosecution for unlawful conduct where appropriate.
And in the case of pandemics, a program of large-scale testing, tracing, and supported isolation22 could allow for better suppression and economic resilience even if the connections between contagion, behavior, and economic conditions continue to remain poorly understood.
Read more thoughts on the COVID-19 pandemic from complex-systems researchers in The Complex Alternative, published by SFI Press.
Reflection Footnotes
1 L. Buchanan, Q. Bui, and J.K. Patel, “Black Lives Matter May Be the Largest Movement in US History,” The New York Times, July 3, 2020, https://www.nytimes.com/interactive/2020/07/03/us/george-floyd-protests-crowd-size.html
2 R. Rojas, “Mississippi Governor Signs Law to Remove Flag with Confederate Emblem,” The New York Times, June 30, 2020, https://www.nytimes.com/2020/06/30/us/mississippi-flag.html
3 P. Ewing, “Congress Overturns Trump Veto on Defense Bill after Political Detour,” National Public Radio, January 1, 2021, https://www.npr.org/2021/01/01/952450018/congress-overturns-trump-veto-on-defense-bill-after-political-detour
4 D. Searcey and J. Eligon, “Minneapolis Will Dismantle Its Police Force, Council Members Pledge,” The New York Times, June 7, 2020, https://www.nytimes.com/2020/06/07/us/minneapolis-police-abolish.html
5 A.W. Herndon, “How a Pledge to Dismantle the Minneapolis Police Collapsed,” The New York Times, September 26, 2020, https://www.nytimes.com/2020/09/26/us/politics/minneapolis-defund-police.html
6 A. Watkins, “NYPD Disbands Plainclothes Units Involved in Many Shootings,” The New York Times, June 15, 2020, https://www.nytimes.com/2020/06/15/nyregion/nypd-plainclothes-cops.html
7 A. MacGillis, “How to Stop a Police Pullback,” The Atlantic, September 3, 2020, https://www.theatlantic.com/ideas/archive/2020/09/how-stop-police-pullback/615730/
8 D. Barrett, “2020 Saw an Unprecedented Spike in Homicides from Big Cities to Small Towns,” The Washington Post, December 30, 2020, https://www.washingtonpost.com/national-security/reoord-spike-murders-2020/2020/12/30/1dcb057c-4ae5-11eb-839a-cf4ba7b7c48c_story.html
9 Congress.gov, “Text - S.3548 - 116th Congress (2019-2020): CARES Act,” June 3, 2020, https://www.congress.gov/bill/116th-congress/senate-bill/3548/text.
10 https://iem.uiowa.edu/iem/
11 https://www.predictit.org
12 https://www.betfair.com
13 https://kalshi.com
14 D. Rothschild and R. Sethi, 2016, “Trading Strategies and Market Microstructure: Evidence from a Prediction Market,” The Journal of Prediction Markets 10(11), doi: 10.5750/jpm.v10i1.1179
15 https://pol.is/home
16 A. Tang, “A Strong Democracy is a Digital Democracy,” The New York Times, October 15, 2019, https://www.nytimes.com/2019/10/15/opinion/taiwan-digital-democracy.html; C. Horton, “The Simple but Ingenious System Taiwan Uses to Crowdsource its Laws,” MIT Technology Review, August 21, 2018, https://www.technologyreview.com/2018/08/21/240284/the-simple-but-ingenious-system-taiwan-uses-to-crowdsource-its-laws/; C. Miller, “Taiwan is Making Democracy Work Again. It’s Time We Paid Attention,” WIRED, November 26, 2019, https://www.wired.co.uk/article/taiwan-democracy-social-media
17 J. Lanier and E.G. Weyl, “How Civic Technology Can Help Stop a Pandemic: Taiwan’s Initial Success is a Model for the Rest of the World,” Foreign Affairs, March 20, 2020, https://www.foreignaffairs.com/articles/asia/2020-03-20/how-civic-technology-can-help-stop-pandemic.
18 See, for example, Our World in Data (https://ourworldindata.org/explorers/coronavirus-data-explorer), comparing Taiwan’s case counts and confirmed deaths to those of other nations.
19 R. Sethi, J. Seager, et al., “Models, Markets, and the Forecasting of Elections,” Social Science Research Network, January 16, 2021, doi: 10.2139/ssrn.3767544
20 L.W. Sherman, 2018, “Reducing Fatal Police Shootings as System Crashes: Research, Theory, and Practice,” Annual Review of Criminology 1, doi: 10.1146/annurev-criminol-032317-092409
21 B. O’Flaherty and R. Sethi, “Policing as a Complex System,” The Bridge: Linking Engineering and Society (National Academy of Engineering) Winter 2020.
22 “Roadmap to Pandemic Resilience,” Edward J. Safra Center for Ethics at Harvard University, April 20, 2020, https://ethics.harvard.edu/Covid-Roadmap