VoxEU Column COVID-19

Mandated and targeted social isolation policies flatten the COVID19 curve and can help mitigate the associated employment losses

China responded to the Covid-19 outbreak with draconian mandatory social distancing policies. Other countries have reacted more timidly, either by deliberate choice, as in the United States, or due to implementation constraints, as in some European countries, leaving individuals to choose for themselves an appropriate degree of self-isolation. This column, we argue that differences in mandated social distancing policies across countries have contributed to astonishingly different outcomes.

Voluntary social distancing and lack of compliance with mandated polices to halt the spread of Covid-19 have led to unnecessarily high infection rates and death tolls across many countries.  Individuals are likely to isolate voluntarily when the probability of contracting the disease is sufficiently high. However, people also weigh the contagion risk against losses of income and the inconvenience of living in isolation (Baldwin 2020a). In consequence, voluntary social distancing can start to keep people at home only when the infection risk becomes visible, and the epidemic has already taken off. This is too late to flatten the curve, with little or no effect during the accelerating stage of the epidemic or its peak. Politicians, internalising these preferences, are reluctant to impose mandated social distancing at the very early stages of the epidemic, based only on intelligence, scientific evidence and medical advice. Compliance with any advisory orders and warnings is low when the epidemic is not yet terrifying. 

In a new paper (Chudik et al. 2020), we consider and contrast government-mandated social distancing policies with voluntary self-isolation in a standard Susceptible-Infected-Recovered (SIR) model. We begin with a simple case in which the fraction of exposed population is set at the outset of the spread of the epidemic, for example close to what we believe China did after the start of the epidemic in Wuhan. We then consider a variation of the SIR model where this fraction can change over time due to the voluntary decision to isolate at the individual level. Using a standard individual decision framework, we show that the proportion of the population that self-isolates voluntarily increases with the probability of contracting the disease. However, voluntary self-isolation kicks in only as the epidemic curve peaks. As a result, it only marginally affects the steepness or the peak of the epidemic curve, and it has little or no effect during the initial stages of the epidemic (Figure 1). In this context, the role of government mandates and guidelines are of utmost importance.

Figure 1 Simulated COVID19 infections with and without voluntary social-distancing 

Note: The figure plots simulations of the COVID-19 curve with and without voluntary social distancing, labelled I_VARYING (red line) and I_FIXED (blue line), respectively.

We also model the short-term impact of the epidemic on employment, our proxy for the recession curve. This permits an assessment of the costs and benefits of alternative societal decisions on the degree and the nature of government-mandated social distancing policies. We consider the employment losses associated with alternative degrees of mandated social isolation in conjunction with an adjustment for the type of jobs or the sectors that are targeted, and how social distancing is managed in practice.

Within this set up, a given level of social distancing can have different employment consequences, depending on how it is implemented. In the extreme case in which social distancing is applied uniformly across all workers and sectors, social distancing results in a proportionate fall in employment (e.g. Atkeson 2020). But smart social distancing policies that enable individuals to work from home, avoid public transportation, and allow work outside the home but with adequate PPEs, together with widespread testing and isolation rules as discussed by Baldwin (2020b), permit a partial mitigation of the employment losses associated with social distancing. We simulate employment losses for a number of different combinations of the level of social distancing and the efficiency with which they are implemented. We find that, for the degrees of exposure that can flatten the epidemic curve (implying fewer hospitalisations and deaths from COVID-19), the employment losses can be moderated somewhat, but still remain substantial (Table 1). 

Table 1 Simulated employment loss under alternative social distancing scenarios and modalities of implementation

Notes: The losses are given in per cent per annum over 120 days which is the simulated length of the epidemic. The epidemic is simulated using a SIR model with R0 = 3 and the removal rate of 14 days. λ is the fraction of the population exposed to the virus. α is a parameter that determines the economic cost of the isolation measures. 

We also provide extensive simulation-based and empirical evidence on Chinese provinces on the duration of the epidemic. We find that the COVID-19 epidemic curve takes about 50 days to peak and 120 days to complete (Figure 1), with the most sizeable portion of infections occurring within a 60-day window centered on the epidemic peak (Figure 2). 

Figure 2 Actual and fitted COVID19 epidemic curve for Hubei 

Notes: The figure plots fitted (orange line) and actual Hubei infections (blue line). See Chudik et al. (2020) for the details of the econometric model and its estimation.  

We then provide estimates on the number of days to recover or die (the ‘removal rate’) and exposure rates using daily data on confirmed, recovered and death cases for Chinese provinces, and other countries from the Johns Hopkins University database. To do so, we discretise our SIR model and run regressions in confirmed recoveries and the number of active cases, respectively. To address the likely biases associated with limited and non-random testing, we allow for both systematic and random measurement errors. We show that the degree of social distancing can be identified up to a scaling factor, which is linked to the fraction of asymptomatic cases in the total number of tested individuals. Using the quasi-experimental data from the well-known case of the Diamond Princess cruise ship, we estimate this fraction to be around 1/2, implying that for every confirmed case there is most likely another infected case that remains undetected.

Conditional only on a weak assumption for the basic reproduction rate, we report three important results on the removal and exposure rates. First, using daily data on the complete epidemic history of Chinese provinces, we estimate recovery rates to be around 21 days, substantially higher than the 14 days typically assumed in designing quarantine policies. The individual estimates are quite precise with little variation across the Chinese provinces. We also find that the random component of the measurement error in the underlying data is relatively unimportant for the estimation of recovery rates. 

Table 2 Estimated removal rates: Chinese provinces and other countries 

Notes: The table reports mean group (MG) estimates of the number days it took to recover or die from COVID19, its inverse (the so-called removal rate) together with the standard error in parenthesis, for Chinese provinces excluding and other countries. The MG estimator is the simple average of the unit-specific estimates (Pesaran and Smith, 1995). The full sample for the Chinese provinces is Jan-22 to March-31, 2020 with 70 observations in total. The sample for other countries varies, depending on when the epidemic started, from a particular day at the end-February to April 12, with 40-50 observations.  

Second, for Chinese provinces excluding Hubei, the epicentre of the epidemic, we estimate that the share of exposed population is very small even after allowing for a 50% under-recording of asymptomatic infections and errors in measuring recovered cases. Excluding the Hubei province, we find the share of exposed population across China to be less than two individuals per 100,000! This is an astonishingly low rate and is consistent with dramatically falling estimates of the effective reproduction rate at the onset of the epidemic in China, reported in the medical literature. In contrast, in the case of Hubei, we estimate an exposure rate of 40 to 60 times larger. 

Figure 3 Exposure rate across Chinese provinces (per 100,000 population)

Notes: The figure plots the exposure rates for Chinese provinces estimated through January 22- April 12, 2020 (Full sample, yellow bars) and through January 22-February 20, 2020 (Subsample, blue bars).

Third, and most unfortunately, our estimates of the exposure rate in other countries are much higher: between three and six times larger than in Hubei, despite the time-lead enjoyed by these countries to prepare for the epidemic. The country-specific estimates in other countries are much more heterogeneous, with Iran, South Korea and Austria doing relatively better, and Italy, Spain and Belgium displaying the highest exposure rates. However, these estimates are based on limited data and should be approached with caution, and viewed as preliminary and indicative. 

Figure 4 Comparing exposure rates across countries (per 100,000 population)

Notes: The figure plot estimated exposure rates for other countries for which there is sufficient data, calibrating the recovery rate to the estimated country-specific removal rate (yellow bars) or the MG estimate of the removal rate (blue bars).  

We conclude that voluntary social isolation is ineffective and that mandated social distancing is required from the early outset of the epidemic to flatten the curve. We find the exposure rates to be similar across Chinese provinces ex-Hubei, and highly heterogeneous across other countries, largely reflecting the mandated level of social distancing. Whilst social distancing is costly in terms of employment and lost output, its smart and targeted design and implementation, applied consistently within and across countries, can mitigate the short-term costs and allow a more orderly return to normality.

Authors’ note: The views expressed in this column are those of the authors and do not necessarily reflect those of the Federal Reserve Bank of Dallas or the Federal Reserve System.


Atkeson, A (2020), "What will be the economic impact of COVID-19 in the US? Rough estimates of disease scenarios", NBER Working Paper 26867.

Baldwin, R (2020a), “COVID, remobilisation and the ‘stringency possibility corridor’: Creating wealth while protecting health”,,10 April.

Baldwin, R (2020b), “COVID-19 testing for testing times: Fostering economic recovery and preparing for the second wave”,, 25 March.

Chudik, A, M H Pesaran, and A Rebucci (2020), “Voluntary and Mandatory Social Distancing: Evidence on COVID-19 Exposure Rates from Chinese Provinces and Selected Countries”, CEPR Discussion Paper 14646.

Pesaran, M H and R Smith (1995), “Estimating long-run relationships from dynamic heterogeneous panels”, Journal of Econometrics 68(1): 79-113.

1,155 Reads