COVID-19 and Global Economic Growth: Policy Simulations with a Pandemic-Enabled Neoclassical Growth Model
Ian M. Trotter, Luís A. C. Schmidt, Bruno C. M. Pinto, Andrezza L. Batista, Jéssica Pellenz, Maritza Isidro, Aline Rodrigues, Attawan G. S. Suela, Loredany Rodrigues
CCOVID-19 and Global Economic Growth: PolicySimulations with a Pandemic-Enabled NeoclassicalGrowth Model
Ian M. Trotter ∗ , Lu´ıs A. C. Schmidt , Bruno C. M. Pinto , Andrezza L.Batista , J´essica L. V. Pellenz , Maritza Rosales , Aline Rodrigues ,Attawan G. S. Suela , and Loredany C. C. Rodrigues Institute of Public Policy and Sustainable Development, Universidade Federal de Vi¸cosa Department of Agricultural Economics, Universidade Federal de Vi¸cosa
Working Paper – June 2020
Abstract
During the COVID-19 pandemic of 2019/2020, authorities have used temporary ad-hoc policy measures, such as lockdowns and mass quarantines, to slow its trans-mission. However, the consequences of widespread use of these unprecedented mea-sures are poorly understood. To contribute to the understanding of the economicand human consequences of such policy measures, we therefore construct a math-ematical model of an economy under the impact of a pandemic, select parametervalues to represent the global economy under the impact of COVID-19, and performnumerical experiments by simulating a large number of possible policy responses.By varying the starting date of the policy intervention in the simulated scenarios,we find that the most effective policy intervention occurs around the time when ∗ Corresponding author. E-mail: [email protected] a r X i v : . [ ec on . GN ] J un OVID-19 and Global Economic Growth
Working Paper – June 2020 the number of active infections is growing at its highest rate – that is, the resultssuggest that the most severe measures should only be implemented when the diseaseis sufficiently spread. The intensity of the intervention, above a certain threshold,does not appear to have a great impact on the outcomes in our simulations, due tothe strongly concave relationship that we identify between production shortfall andinfection rate reductions. Our experiments further suggest that the interventionshould last until after the peak established by the reduced infection rate, whichimplies that stricter policies should last longer. The model and its implementation,along with the general insights from our policy experiments, may help policymakersdesign effective emergency policy responses in the face of a serious pandemic, andcontribute to our understanding of the relationship between the economic growthand the spread of infectious diseases.
Keywords:
Economic growth, Pandemics, COVID-19, Policy.
Pandemics have caused death and destruction several times throughout human history,and have caused large and lasting impacts on society: for example the Black Death in14 th century Europe (Herlihy, 1997), the Spanish flu in 1918-1920 (Johnson and Mueller,2002), HIV in the 1980s (Pope and Haase, 2003), and H1N1 in 2009 (Trifonov et al.,2009). In 2019 and 2020, policymakers have struggled to design effective policy responsesto the COVID-19 pandemic, as authorities have resorted to unprecedented and widespreaduse of temporary emergency measures such as lockdowns, mass quarantines, and other“social distancing” measures, despite that the human and economic consequences of these ad-hoc interventions are poorly understood. It is clear that more research is neededon effective policy intervention during pandemics, and how to mitigate its human andeconomic impacts (Bauch and Anand, 2020).Therefore, we incorporate a model of disease transmission into a model of economicgrowth, considering that policymakers can implement temporary policies that simultane-ously slow the spread of the disease and lower the economic output. We then select model2OVID-19 and Global Economic Growth Working Paper – June 2020 parameters so as to represent the global economy under the impact of COVID-19, andperform numerical simulations of various policies to provide insights into the trade-offsbetween short- and long-term human and economic outcomes. The policy simulationswill help policymakers understand how altering the starting time, intensity, and durationof the policy intervention can impact the outcomes, and contribute to the understandingof how to design effective policies to confront rapidly spreading pandemics.A growing body of literature has been devoted to studying the economic impacts ofsocial distancing measures and the design of emergency policies during pandemics. Forinstance, Eichenbaum et al. (2020) constructed a mathematical model and ran numericalexperiments that resulted in important insights for policymakers during a pandemic, al-though the study abstracts from forces that affect the long-term economic development.Andersson et al. (2020) use a similar framework to study policy responses and the trade-off between output and health during a pandemic, operating with a welfare function thatonly includes the period of the epidemic, and will also not account for lasting impactsof the pandemic on the economy. Our study is similar to these studies, although ourmodel includes capital stock and population growth, which enables us to look at possiblelong-term effects beyond the duration of the pandemic. Guan et al. (2020) consider theglobal economic impacts of the COVID-19 pandemic in a model with multiple countriesand productive sectors, suggesting that the economic impacts propagate through sup-ply chains, and that earlier, stricter, and shorter lockdowns will minimise the economicdamages. Acemoglu et al. (2020) include a simplified evaluation of economic loss intoan SIR-type epidemiological model that distinguishes between age groups, and find thatlockdown policies that are tailored to the different age groups are more efficient. La Torreet al. (2020) model the social costs of an epidemic, in which a social planner must chooseto allocate tax revenues between prevention and treatment, showing that the optimalallocation depends on the infectivity rate of the epidemic. Alvarez et al. (2020) studiedthe optimal lockdown policy in a linear economy, where each fatality incurs a cost.However, even before the onset of the COVID-19 pandemic, several studies had focusedon the determinants and incentives of social distancing during epidemics (Fenichel, 2013;3OVID-19 and Global Economic Growth
Working Paper – June 2020
Perrings et al., 2014; Kleczkowski et al., 2015; Toxvaerd, 2020; Quaas et al., 2020). Themodels in these studies mainly apply economic insights to improve epidemiological modelsby examining and modelling individual contact decisions.On the relationship between epidemics and economic growth, Herlihy (1997) andHansen and Prescott (2002) argue that the increased mortality caused by the Black Deathresulted in great economic damage by decreasing labour supply, which induced the sub-stitution of labour for capital and triggered an economic modernisation that eventuallylead to greater economic growth. Delfino and Simmons (2000), however, argue that thereare causal links in both directions in the interaction between the economy and diseasetransmission, and proposed a model that combines disease transmission into a model ofeconomic growth, although the model is not used to explore temporary policy interven-tions during specific rapidly-spreading pandemics. Bonds et al. (2010), Ngonghala et al.(2014) and Ngonghala et al. (2017) show that poverty traps frequently arise when com-bining models of infectious diseases and of economic development, and that these mayhelp explain the differences in economic development between countries. Goenka et al.(2010), Goenka et al. (2014) and Goenka and Liu (2019) also integrate disease transmis-sion directly into an economic growth model, and allow investments in health – building health capital – to affect the epidemiological parameters. Optimal investments in healthand the accumulation of health capital, however, are different from designing temporarypolicies during a pandemic. Most of these studies are based on disease models that al-low the individual to contract the disease multiple times, which is consistent with somediseases that are common in the developing world, such as malaria or dengue fever, butmay not be applicable to COVID-19. Furthermore, most of these models do not includedisease-related mortality, one of the main channels through which serious pandemics affectthe economy (Hansen and Prescott, 2002). These studies, however, show that integratingdisease transmission in economic growth models can result in multiple steady states, andwe are therefore careful to choose an approach that is not sensitive to this: we use ourmodel to run numerical experiments, solving the numerical optimisation problem using That is, in this particular case, that a greater disease burden leads to more poverty, and more povertyleads to a greater disease burden, so as to create a self-perpetuating effect.
Working Paper – June 2020 backwards induction.The macroeconomic impact of the HIV/AIDS epidemic has also received much researchattention, altough many studies assume the disease transmission is exogenous (Cudding-ton, 1993b,a; Cuddington and Hancock, 1994; Haacker, 2002; Arndt, 2003; Cuesta, 2010).Azomahou et al. (2016) allow mortality rate to depend on health expenditure, but thedisease transmission still remains exogenous to their model. Bell and Gersbach (2009)studied government investments in health and education in an overlapping generationsmodel of HIV/AIDS in Africa. However, the HIV/AIDS epidemic and the COVID-19 pan-demic are so different, that we do not expect the insights from these studies to transferdirectly to the COVID-19 pandemic.The COVID-19 pandemic has shown that many authorities are prepared to implementstrict and dramatic emergency policies at short notice in order to slow down the spread of aserious pandemic. However, our understanding of the economic and human consequencesof such measures is still incipient. At the same time, implementing such policies is delicate,and, if done improperly, authorities could damage the economy whilst still failing to lowerthe transmission rate of the disease. There is therefore an urgent need for research thatdevelops guidelines for the use of these emergency measures, and to help policymakersunderstand their impacts and consequences over time.We contribute to the study of the efficiency, impacts, and consequences of temporaryemergency measures during a pandemic by incorporating a policy parameter into a modelthat integrates disease transmission dynamics and economic growth. Our model providesa theoretical framework for understanding the impact of emergency policies on the tra-jectory of the pandemic as well as the main economic variables, in light of their mutualinteractions. To gain a deeper understanding of the emergency policies, we select param-eter values that are consistent with the global economy under the impact of COVID-19,and numerically simulate a large number of scenarios for possible emergency policies. Us-ing this simulation-based approach, we investigate the impact of altering the starting dateof policy intervention, the intensity of the policy intervention, and the duration of thepolicy intervention. Altering the simulated policies along these three dimensions provides5OVID-19 and Global Economic Growth
Working Paper – June 2020 insights into the impact of the emergency measures on the trajectory of the pandemic andthe development of the main economic variables. These insights could help policymakersdesign effective emergency responses to pandemics.Our work differs from earlier studies in some important aspects. Principally, the periodof interest in our study extends beyond the duration of the pandemic. Therefore, ourmodel includes the dynamics of capital accumulation, population growth and pandemicdeaths, which have not been jointly considered in previous studies. These components areimportant to study the impact of the pandemic on economic growth beyond the short-and medium term. Other novel aspects of our model include a relationship betweenthe reduction in economic output and the reduction in the infection rate in the shortrun, a mortality rate that depends on the infection rate, and explicitly modelled excesscosts of hospital admissions due to the pandemic. Although some of these features areincluded in previous models, they have not yet been combined into a single comprehensiveframework. In addition, our study contains some early estimates of the economic impactsof COVID-19 on Europe’s five largest economies, constructed by analysing changes inreal-time and high-frequency data on electricity demand. We also make our data andcustom computer code freely available, which will hopefully be useful to the researchcommunity and contribute to future developments in the area.In addition to this introduction, this paper consists of three sections. The followingsection presents a mathematical framework that incorporates a model of disease trans-mission into a model of economic growth, shows how model parameter values were chosento fit the model to the global economy during the COVID-19 pandemic, and details thenumerical experiments that were performed. In the third section, we present, interpret,and discuss the results of the numerical experiments and their implications. In the finalsection, we summarise the main findings of the study.
Here we detail the integration of an epidemiological model into a neoclassical modelof economic growth – known as one of the “workhorses” of modern macroeconomics6OVID-19 and Global Economic Growth
Working Paper – June 2020 (Acemoglu, 2011). We first modify the SIR (Susceptible-Infected-Recovered) model of thespread of an infection, pioneered by Kermack and McKendrick (1927), then incorporate itinto the a model setup similar to the classical Ramsey-Cass-Koopmans model in discretetime. We then explain how we select functional forms and parameters to represent theglobal economy and the global spread of COVID-19, before outlining a set of numericalexperiments designed to give insight into the economic and epidemiological impacts ofvarying the starting time, intensity and duration of the policy interventions.
Here we combine an epidemiological model with a model of economic growth, concentrat-ing on three main bridges between the models. First, we assume that the spread of apandemic reduces the labour force, since infected or deceased individuals will not work,and this reduces economic output. Second, we assume that society incurs additional directcosts, for instance due to the hospitalisation of infected individuals, and these costs mustbe covered with output that would otherwise have been consumed or invested. Third,we assume that governments may, through policy, simultaneously impact both the spreadof the pandemic and the efficiency of economic production. These interactions betweenthe spread of the pandemic and economic growth are the main focus of our model, whichjointly represents the dynamics of the spread of the pandemic and the dynamics of eco-nomic growth.The SIR model (Kermack and McKendrick, 1927; Brauer and Castillo-Chavez, 2012)is a simple Markov model of how an infection spreads in a population over time. Thismodel divides a population ( N ) into three categories: Susceptible ( S ), Infected ( I ), andRecovered ( R ). In each period, the number of susceptible individuals who become infectedis a product of the susceptible population, the number of individuals who are alreadyinfected, and an infection rate b . A given proportion of the infected individuals ( r ) alsorecover in each period.To incorporate the SIR model into a model of economic growth, we make two adap-7OVID-19 and Global Economic Growth Working Paper – June 2020 tations to the basic SIR model. First, we introduce a distinction between recoveredindividuals ( R ) and deceased individuals ( D ), since recovered individuals will re-enter thelabour force whereas deceased individuals will not: each period, infected individuals willrecover at a rate r and pass away at a rate m . Second, instead of considering the popula-tion to be of a fixed size, we allow the population to grow over time. Population growthis usually negligible at the timescale of interest for models of epidemics or pandemics, butit is significant in the timescales of economic growth. Therefore, we introduce a logisticmodel for population growth, and new individuals will be added to the number of suscep-tible individuals each period. Using two parameters, a and a , to describe the populationgrowth, we can describe the spread of the pandemic in the population as follows: N t +1 = a N t + a N t − mI t (1) S t +1 = S t + ( a − N t + a N t − bS t I t (2) I t +1 = I t + bS t I t − rI t − mI t (3) R t +1 = R t + rI t (4) D t +1 = D t + mI t . (5)Many variations of the basic SIR model already exist, and it would be possible to incor-porate more complex dynamics into the epidemiological model. However, this model willbe sufficient for our current purposes.The model of economic growth assumes that a representative household chooses whatquantity of economic output ( Y ) to consume ( C ) or save (invest) each period in orderto maximise an infinite sum of discounted utility, represented by a logarithmic utilityfunction. Output is produced by combining labour and capital ( K ) using a technologyrepresented by a Cobb-Douglas production function with constant returns to scale andtotal factor productivity A t . However, we allow pandemic policy, represented by p , toreduce the total output, and furthermore assume that only susceptible and recovered8OVID-19 and Global Economic Growth Working Paper – June 2020 individuals are included in the labour force: Y t = (1 − p ) A t K αt ( S t + R t ) − α , (6)in which α represents the output elasticity of capital, and total factor productivity A t grows at a constant rate g : A t +1 = (1 + g ) A t . (7)Assuming that physical capital ( K t ) depreciates at a rate of δ from one period to the nextand that the pandemic causes direct costs H ( · ) to society, the capital stock accumulatesaccording to the following transition equation: K t +1 = (1 − δ ) K t − C t − H ( · ) . (8)Given a utility discount factor β , we assume that a benevolent social planner choosesan infinite stream of consumption { C t } ∞ t =0 to optimise the discounted sum of logarithmicutility, solving the following maximisation problem:max { C t } ∞ t =0 ∞ (cid:88) t =0 β t N t ln (cid:18) C t N t (cid:19) , while respecting the restrictions represented by equations (1) through (8). Since ourperiod of interest is actually finite, this maximisation problem can be solved numericallyby backwards induction, provided that the terminal period is chosen so far in the futurethat it will not interfere with the period of interest . The model presented in the previous subsection relies on parameters for population dy-namics, the spread of a pandemic, production of economic output, and the accumulation The implementation of the model,
Macroeconomic-Epidemiological Emergency Policy InterventionSimulator , is available at https://github.com/iantrotter/ME3PI-SIM . Working Paper – June 2020 of physical capital. In order to perform computational experiments, we need to determinerealistic numerical values for these, as well as initial values for the state variables.One fundamental issue that we must address is that the study of epidemics and eco-nomic growth usually consider different timescales: whereas the spread of an epidemicis usually analysed at a daily or weekly timescale, economic growth is usually studiedat an annual, or even a decennial, timescale. To reconcile these differences, we choosea daily timescale for our model. A daily timescale is suitable for studying the spread ofa pandemic, since pandemics spread rapidly and their health effects pass almost entirelywithin a short timeframe. However, daily resolution is an unusual choice for a model ofeconomic growth, as capital accumulation, technological progress, and population growthare almost negligible from one day to the next. During a pandemic, however, daily move-ments of individuals in and out of the labour force could have a large impact on economicproduction, and indirectly on the accumulation of capital – and in order to adequatelycapture these effects, we choose a daily resolution for the model. Parameter values forboth the parameters belonging to the economic components, as well as the epidemiologicalcomponents, are therefore chosen to represent a daily timescale.
Population Growth
The parameters for the logistic population growth model, a and a , were selected by first estimating a linear regression model on annual global populationdata from the World Bank between 1960 and 2018 . The estimation results are shownin Table 1, and the regression coefficients a y and a y – representing the parameters of anannual model – were converted into their corresponding daily values by calculating: a = 1 + a y − a = a y . The fitted values of the population model is shown in Panel (a) of Figure 1, and appearto match the historical data for global population closely. Available at https://data.worldbank.org/indicator/SP.POP.TOTL , accessed on 2020-05-04
Working Paper – June 2020
Table 1:
Estimated parameters for the Gordon-Schaefer population growth model. The modeluses ordinary least squares, on annual data from 1960 to 2018.
Dependent variable:
World Population t WorldPopulation t − ∗∗∗ (0.001)World Population t − − . × − ∗∗∗ (0.000)Observations 58R ∗∗∗ (df = 2; 56) Note: ∗ p < ∗∗ p < ∗∗∗ p < Capital Stock
We imputed the global physical capital stock by combining annual dataon global gross physical capital formation, available for the period between from theWorld Bank , with an assumed physical capital depreciation rate of δ = 4 . , whose distribution is shown in Panel (b) ofFigure 1. The resulting estimated level of global physical capital stock from 1990 to 2019is shown in Panel (c) of Figure 1. Production
Following Nordhaus (1992), we set the output elasticity of capital to α =0 .
3. We then combine annual data of global output from the World Bank between 1990and 2018 with the global population and the imputed global stock of physical capital, inorder to estimate the total factor productivity, A t , and its growth rate, g . This gives anannual total factor productivity growth rate of around 1 . g = 3 . × − over the period. The modelled global production Available at http://api.worldbank.org/v2/en/indicator/NE.GDI.TOTL.KD , accessed on 2020-05-04. Available for download at , accessed on 2020-05-04. Available at http://api.worldbank.org/v2/en/indicator/NY.GDP.MKTP.PP.KD , accessed on2020-05-04.
Working Paper – June 2020 P e o p l e (a) Population Modelled PopulationObserved Population 0.02 0.04 0.06 0.08 0.10
Depreciation Rate F r e q u e n c y (b) Depreciation Rate U S D ( ) (c) Capital Stock Imputed Capital Stock 1992 1996 2000 2004 2008 2012 2016 20201.52.02.53.0 U S D ( ) (d) Gross World Product Modelled Gross World ProductObserved
Figure 1:
Calibration of the economic parameters. Panel (a): Global population and one-step-ahead predicted population from the fitted Gordon-Schaefer population growth model. Panel(b): National capital depreciation rates from the Penn World Tables 9.1 for 2017, with thedashed black line marking the median value δ = 4 . is shown in Panel (d) of Figure 1, and fits the observed data relatively well, although themodelled production level slightly overestimates global production at some points. Utility
We select an annual utility discount rate that corresponds to an annual rate of ρ = 8%. This discount rate allows the simulated investment from the model to match theobserved gross physical capital formation in the period between 1990 and 2010, as shownin Panel (d) of Figure 7. Although this discount rate appears somewhat high, it is notunreasonable if we take into consideration that the model represents the global economy,and that large parts of the global population consists of low-income households with highdiscount rates. 12OVID-19 and Global Economic Growth Working Paper – June 2020
Excess Direct Pandemic Costs
A pandemic directly causes additional costs to soci-ety, which is captured by the function H ( · ) in our mathematical model. To model thiscost, we look to the literature which has estimated the excess hospital admission costsfor a recent similar pandemic, the H1N1 pandemic in 2009: for Spain (Galante et al.,2012), Greece (Zarogoulidis, 2012), Australia and New Zealand (Higgins et al., 2011),New Zealand (Wilson et al., 2012), and the United Kingdom (Lau et al., 2019). Fig-ure 2 shows the direct hospitalisation costs attributed to the H1N1 pandemic in variouscountries, along with the number of hospital admissions. Based on these previous costestimates for the H1N1 pandemic, we use a flat cost of u = 5 ,
722 USD per hospital ad-mission (see Table 2), corresponding to the solid red line in Figure 2. Although we assumea flat cost per admission, it may be more reasonable in other contexts – for instance whenapplying the model to specific regions – to consider a cost function with an increasingmarginal cost: as hospital capacity becomes constrained in the short-run during a surgeof admissions, one could expect the unit cost to increase. However, in the context of ourglobal model, we do not distinguish between the regions in which the cases occur, andtherefore cannot accurately capture such a saturation effect. Therefore, we choose a directcost function that is simply linear in the number of hospital admissions.We assume that h = 14 .
7% of the confirmed infected cases will be admitted to hospi-tal , and our direct cost function is given by: H t = uhbS t I t . Infection, Recovery and Mortality Rates
To estimate the mortality and recoveryrates, r and m , we solve the transition equations for the number of recovered R t (equation(4)) and the number of deceased D t (equation (5)) for their respective parameters: r = R t +1 − R t I t , m = D t +1 − D t I t . This hospitalisation rate corresponds to the median of USA state level hospitalisation rates reportedin the daily COVID-19 reports from the Center for System Science and Engineering at John HopkinsUniversity, on the 10 th of May 2020, available at https://github.com/CSSEGISandData/COVID-19 . Working Paper – June 2020
Table 2:
Esimated direct cost function.
Dependent variable:
Total Costs (USD)Admissions 5,722.078 ∗∗∗ (664.874)Observations 6R ∗∗∗ (df = 1; 5) Note: ∗ p < ∗∗ p < ∗∗∗ p < Admissions T o t a l C o s t , U S D SpainGreeceAustralia and New ZealandNew Zealand UK W1UK W2
Observed CostFitted Cost
Figure 2:
Direct costs of the H1N1 pandemic, based on data compiled by Lau et al. (2019).
Working Paper – June 2020
Using daily data for the number of confirmed, recovered and deceased cases, made avail-able by John Hopkins University , we can calculate the recovery and mortality rates foreach day, as shown in the bottom two rows of Figure 3.To estimate the infection rate, b , we solve equation (3) for the parameter b : b = I t +1 − (1 − r − m ) I t S t I t . Taking into account that I t in the model refers to the number of active cases, whereas thedata reports the accumulated number of cases, and using the population growth modelto help estimate the number of susceptible individuals, we calculate the daily infectionrates, shown in the top row of Figure 3. As the infection rate b varies over time, wechoose a relatively high infection rate to represent the infection rate in the absence ofpolicy intervention, b = 2 . × − , which equals the upper quartile (75%) of theobserved infection rates. As we simulate different intervention policies, this base infectionrate b will be modified.We notice from Figure 3 that the mortality rate appears to rise and fall together withthe infection rate . This might reflect that lack of capacity in a health system increasesmortality rate, and this is therefore a feature that we would like to capture. We thereforeestimate m as a function of b : m = k b k , in which k and k are constants. Table 3 shows the regression for estimating the param-eters k and k , and the fitted function is shown in Figure 4, together with the observedvalues of daily infection and mortality rates.For the recovery rate in the simulations, we select the median of the daily recoveryrates calculated from the data, r = 0 . b , however, will bedetermined individually for each scenario, and will reflect the pandemic policy simulated Available at https://github.com/CSSEGISandData/COVID-19 , accessed on 2020-05-06. Alvarez et al. (2020) have also made this observation, and included the effect in their model.
Working Paper – June 2020
Infection Rate (b) - Timeseries
Infection Rate (b) - Histogram
Recovery Rate (r) - Timeseries
Recovery Rate (r) - Histogram
Mortality Rate (m) - Timeseries
Mortality Rate (m) - Histogram
Figure 3:
Calibration of the parameters for the SIR model. Panel (a) and (b): the developmentand distribution of the SIR infection rate, b . Panel (c) and (d): the development and distributionof the SIR recovery rate, r . Panel (e) and (f): the development and distribution of the SIRmortality rate, m . Table 3:
Mortality rate model.
Dependent variable: ln(Mortality Rate)ln(Infection Rate) 0.717 ∗∗∗ (0.065)Constant 12.561 ∗∗∗ (1.642)Observations 104R ∗∗∗ (df = 1; 102) Note: ∗ p < ∗∗ p < ∗∗∗ p < Working Paper – June 2020
Infection Rate
1e 100.0000.0050.0100.0150.020 M o r t a li t y R a t e Observed ValuesFitted Model
Figure 4:
Global mortality rate and infection rate. in each of the scenarios. The mortality rate m will be determined by the infection rate b ,according to the relationship between them we estimated earlier. The Production-Infection Trade-Off
Our model assumes that pandemic policy mainlyimpacts the spread of the pandemic through manipulating the infection rate b , and mainlyinfects economic growth by affecting production of economic output, Y t . Our model con-tains a single parameter, p , to represent pandemic policy, which directly represents theshortfall in global production. In order to analyse the trade-off between production andthe infection rate, however, we must establish how the infection rate, b , is impacted bythe policy parameter p – that is, we must quantify how the infection rate responds toforegone production.We expect the relationship between the infection rate, b , and the GDP shortfall, p , toexhibit two specific characteristics. Firstly, we expect a reduction in the infection rate asthe GDP shortfall increases, because we assume pandemic policies are designed to reducethe infection rate, which result in a shortfall in the economic production as a side-effect.Secondly, we expect the reduction in the infection rate to be greater at first, because weexpect measures to be enacted in order from more to less effective, and from less to moredisruptive. That is, the infection rate reductions exhibit a form of decreasing returns inthe GDP shortfall. We suggest that the percentage reduction in the infection rate, ∆ b (%),responds to the percentage reduction in GDP, ∆GDP(%), as follows:∆ b (%) = q ∆GDP(%) q , Working Paper – June 2020 in which q and q are constants, and q ∈ (0 , , and we utilise data from Europe’s five biggesteconomies, which have all been significantly impacted by COVID-19: France, Germany,Italy, Spain, and the United Kingdom. To measure the daily shortfall in electricity con-sumption, however, we must compare the observed electricity consumption to what itwould have been under normal conditions. Therefore, we first need to create a coun-terfactual representing the electricity consumption under normal conditions. We use theautomated forecasting procedure by Taylor and Letham (2017) to calibrate models on thedaily national electricity consumption (load) data from 2015 until March 1, 2020. Thisperiod does not include the main impacts of the pandemic, such that forecasts from themodels for the period from March 1, 2020 to May 10, 2020 can act as counterfactuals –how electricity consumption would have been expected to develop under normal condi-tions. These counterfactuals may then be compared to the observed values in the sameperiod, and allows us to calculate the daily electricity consumption shortfall in terms ofpercentages.The relationship between electricity consumption and production has been the subjectof many studies (often as variants of “income elasticity of electricity consumption”), and, Available at https://transparency.entsoe.eu/ , accessed on 2020-05-13.
Working Paper – June 2020 E s t . G D P S h o r t f a ll ( % ) Estimated GDP Shortfall
FranceGermanyItalySpainUnited Kingdom2020-03-032020-03-102020-03-172020-03-242020-03-312020-04-072020-04-142020-04-212020-04-282020-05-05100755025025 I n f e c t i o n R a t e Infection Rate
FranceGermanyItalySpainUnited Kingdom
Figure 5:
Weekly estimated GDP shortfall for France, Germany, Italy, Spain, and the UnitedKingdom. The GDP shortfall is based on the difference between actual and expected electricityconsumption. The average weekly infection rate is based on the number of confirmed cases. synthesising the studies into a useful heuristic, we assume that a 1% decrease in electricityconsumption is associated with a 1.5% reduction in GDP. The estimated GDP shortfallfor France, Germany, Italy, Spain and the United Kingdom are illustrated in Panel (a) ofFigure 5, which shows a clear increase in the GDP shortfall throughout March 2020, and astable shortfall of around 10%-20% throughout April and the start of May 2020, appearingto correspond closely to the lockdown periods of these countries. Panel (b) of Figure 5shows the reduction in the infection rate for the five countries over the same time period,with the base period for the infection rate considered to be the first seven days in March,and it is clear that the infection rate has decreased as the GDP shortfall has dropped,which is consistent with our expectations. The scatter plot of the estimated GDP shortfalland the reduction in infection rates, shown in Figure 6, shows a clear relationship betweeninfection rate reductions and GDP shortfall. The red line in Figure 6 shows the model,with constants estimated as in Table 4. The model fits the observations well, and theestimated values for the constants, q and q , conform to our expectations.19OVID-19 and Global Economic Growth Working Paper – June 2020
Est. GDP Shortfall (%) I n f e c t i o n R a t e R e d u c t i o n ( % ) FranceGermanyItalySpainUnited KingdomModel
Figure 6:
Weekly estimated GDP shortfalls for France, Germany, Italy, Spain, and the UnitedKingdom, based on the difference between actual and expected electricity consumption. Theaverage weekly infection rate is based on the number of confirmed cases. The red line representsthe model estimated on the data.
Table 4:
Model of reductions in the infection rate as a function of reductions in economicproduction.
Dependent variable: ln(∆ b (%))ln(∆GDP(%)) 0.238 ∗∗∗ (0.045)Constant 3.677 ∗∗∗ (0.114)Observations 45R ∗∗∗ (df = 1; 43) Note: ∗ p < ∗∗ p < ∗∗∗ p < Working Paper – June 2020
Table 5:
Parameter values.
Parameter Description Value a , a Logistic population growth (annual) 1.028, -2.282 × − δ Capital depreciation rate (annual) 4.46% α Output elasticity of capital 0.3 g Growth rate of total factor productivity (annual) 1.3% ρ Utility discount rate (annual) 8% u Cost per hospital admission 5,722 USD h Hospital admissions per confirmed case 14.7% r Daily recovery rate per active infection 2.1% b Base infection rate (no intervention) 2 . × − k , k Mortality rate parameters, m = k b k q , q Infection rate parameters, ∆ b (%) = q ∆GDP(%) q Table 5 summarises the chosen values for the parameters in the model. Having definedparameter values such that the model represents the global economy under the impact ofCOVID-19, we can now define scenarios that can be simulated numerically, and provideinsight into the impact of policy on both economic growth and the spread of the pandemic.
To have a basis for comparison, we first simulate two baseline scenarios: the
No Pandemic scenario, in which no pandemic occurs, and the
No Intervention scenario, in which thepandemic occurs with no direct intervention ( p = 0). Comparing the remaining scenariosto the first baseline scenario ( No Pandemic ) will help us understand the impact of thepandemic. In addition, the baseline scenario will provide the initial conditions for popula-tion and capital stock at the start of the pandemic, as observations are not yet available.Comparing the remaining scenarios to the second baseline scenario (
No Intervention ) willhelp us understand the impact of the simulated policy intervention. The initial values forthese simulations are shown in Table 6.Having established the baseline scenarios, we run a series of simulations to investigatethree fundamental aspects of the policy intervention. First, we alter the timing of thestart of the intervention, to explore the advantages and disadvantages of starting the in-tervention early or late. Second, we alter the intensity of the intervention, to investigatethe differences in the impacts between light and severe interventions. And, finally, we21OVID-19 and Global Economic Growth
Working Paper – June 2020
Table 6:
Parameters used for the baseline scenarios.
Scenario Start date N I R D b A K No Pandemic × × No Intervention ×
510 28 17 2.041 × − × alter the duration of the intervention. That is, by running numerical experiments thatvary the policy interventions in commencement, intensity and duration, we answer threefundamental policy questions: “When?”, “How much?”, and “For how long?”. Whentaken together, these experiments will provide insight into the economic and health im-pacts of varying policies along these three dimensions, and highlight the trade-offs thatpolicymakers must consider: When to intervene?
Holding the intervention intensity and duration fixed at 10% and26 weeks, we run simulations altering the start of the policy intervention betweenApril 09, 2020, and June 02, 2020.
How much?
Holding the starting date of the intervention fixed at March 12, 2020 – thedate when the WHO declared COVID-19 to be a pandemic – and the duration fixedat 26 weeks, we alter the intensity of the intervention from 5% to 25%, in steps of10 percentage points.
For how long?
Keeping the starting date of the intervention fixed at March 12, 2020,and the intervention intensity fixed at 10%, we alter the duration of the interventionbetween 4 weeks and 76 weeks.The initial values used in all these simulations are the same as in the
No Intervention scenario, specified in Table 6. Taken together, these three sequences of simulations willprovide important and actionable insights into the impacts of policy intervention on botheconomic growth and on the spread of the pandemic that will help policymakers under-stand the relevant trade-offs. 22OVID-19 and Global Economic Growth
Working Paper – June 2020 U S D ( ) (a) Capital Stock Simulated Capital StockImputed Capital Stock 1990 1994 1998 2002 2006 20105.255.505.756.006.256.506.757.00 P e o p l e (b) Population Simulated PopulationObserved Population1990 1994 1998 2002 2006 20101.41.61.82.02.22.42.6 U S D ( ) (c) Production Simulated ProductionObsered GWP 1990 1994 1998 2002 2006 20102.53.03.54.04.5 U S D ( ) (d) Investment Simulated Gross Capital FormationObserved Gross Capital Formation
Figure 7:
Model backtest results. Panel (a): The simulated daily development of the globalphysical capital stock and the daily imputed global physical capital stock. Panel (b): Globaldaily simulated and observed population. Panel (c): Simulated and observed daily gross worldproduction. Panel (d): Simulated and observed daily global gross physical capital formation.
Before presenting the main simulation results, we first present the results of a backtest.This is shown in Figure 7, and shows that the model captures the main features of theobserved historical data. Although the backtest in this case is not an out-of-sample test,due to lack of data, the backtest provides strong support for the economic components ofthe model.
The results of simulating the baseline scenarios –
No Pandemic and
No Intervention –are shown in Figure 8. As expected, the
No Pandemic scenario is characterised by steadyeconomic growth, and no infected or deceased individuals. The
No Intervention scenario,however, shows a large and abrupt drop of around 45% in production during the first23OVID-19 and Global Economic Growth
Working Paper – June 2020 half of 2020, as the pandemic spreads through the population. The number of activeinfections peaks in mid-June, 2020. As the pandemic subsides, a large proportion of thelabour force never returns as the mortalities reach 1.75 billion people, and productionrecovers only to 85% of its pre-pandemic value before 2021. Although growth in economicproduction resumes after the pandemic, production remains 20%-25% below the produc-tion in the
No Pandemic scenario until the end of the simulation in 2030. In summary,the
No Intervention scenario shows substantial loss of human life, as well as a lasting andsignificant negative impact on production, and we expect that shrewd policy interventioncould partially mitigate these impacts.In the following, we run model simulations to gain insight into when to start the policyintervention, to what degree to intervene, and for how long the intervention should last.
We first run a series of simulations to examine the question of when a possible policyintervention should start. In this series of simulations, the intervention intensity is heldfixed at 10% (that is, the intervention causes a 10% decline in production), and theduration of the intervention is held fixed at 26 weeks. Multiple simulations are run withdiffering starting dates for the policy intervention. This series of simulations is shown inFigure 9, with three possible starting dates for the policy intervention: April 9, May 21,and July 2.Examining Panel (f) of Figure 9, we note that intervening on July 2 allows the pan-demic to spread almost identically to the
No Intervention scenario – that is, July 2 istoo late for effective intervention because the peak in active infections has passed, mostof the damage is already done, and the pandemic is decelerating by itself. However, byintervening on July 2 when the number of active infections is near its highest, many mor-talities are avoided, and the human and economic damage is somewhat lower than in the
No Intervention scenario. Further, we note that intervening on April 2 does not appearto significantly alter the course of the pandemic or mitigate its effects – intervening soearly in the pandemic only serves to delay the main wave of infections.24OVID-19 and Global Economic Growth
Working Paper – June 2020 U S D ( ) (a) Capital Stock No PandemicNo Intervention 2020 2022 2024 2026 2028 2030345678 P e o p l e (b) Labour Force No PandemicNo Intervention U S D ( ) (c) Production No PandemicNo Intervention U S D ( ) (d) Investment No PandemicNo Intervention U S D ( ) (e) Pandemic Direct Costs No PandemicNo Intervention 2020 2022 2024 2026 2028 203001234 P e o p l e (f) Active Infections No PandemicNo Intervention P e o p l e Recovered Individuals
No PandemicNo Intervention P e o p l e (h) Deceased Individuals No PandemicNo Intervention
Figure 8:
Simulation results for the baseline scenarios.
Working Paper – June 2020 U S D ( ) (a) Capital Stock No PandemicNo Intervention2020-04-092020-05-212020-07-02 2020 2022 2024 2026 2028 2030345678 P e o p l e (b) Labour Force No PandemicNo Intervention2020-04-092020-05-212020-07-02 U S D ( ) (c) Production No PandemicNo Intervention2020-04-092020-05-212020-07-02 U S D ( ) (d) Investment No PandemicNo Intervention2020-04-092020-05-212020-07-02 U S D ( ) (e) Pandemic Direct Costs No PandemicNo Intervention2020-04-092020-05-212020-07-02 2020 2022 2024 2026 2028 203001234 P e o p l e (f) Active Infections No PandemicNo Intervention2020-04-092020-05-212020-07-02 P e o p l e Recovered Individuals
No PandemicNo Intervention2020-04-092020-05-212020-07-02 P e o p l e (h) Deceased Individuals No PandemicNo Intervention2020-04-092020-05-212020-07-02
Figure 9:
Simulation results when varying the starting date of the policy intervention.
Working Paper – June 2020
The intervention starting on May 21 – about one month before the peak of the
No In-tervention scenario – appears to be the most effective of our simulations, both consideringthe economic impacts and the final mortality rate. May 21 appears to be just before theinflection point of the
No Intervention scenario, and the number of infections is growingat its highest rate. Between the three simulated scenarios, this is by far the preferredoption.It seems that timing the policy intervention is of great importance to mitigate both thehuman and the economic impacts. Although we do not believe that the exact dates holdfor the COVID-19 pandemic in particular, these simulations lead us to interesting insights:policy intervention appears to be most effective when the number of active infections isapproaching its inflection point, and is growing at its highest rate. An intervention that istoo early will only serve to delay the critical phase of the pandemic, and an interventionafter the peak has occurred will obviously do nothing to lower the peak. Although itmay be difficult to know beforehand when a pandemic will enter its critical phase, thetiming of the policy intervention is of paramount importance, and our results suggestthat authorities should implement emergency policies only when the disease is sufficientlyspread. However, we expect that the exact specification of sufficiently spread to varysignificantly from place to place, depending on local conditions.This finding contradicts the claims of Guan et al. (2020), who argue that interventionsshould be “earlier, stricter and shorter”: instead, our results show that starting the inter-vention before the disease is sufficiently spread will either simply delay the critical phaseof the pandemic (if the intervention indeed is kept “short”), or prolong the intervention (ifthe intervention is extended). The study by Eichenbaum et al. (2020) focuses on “startingtoo late” and “ending too early”, yet our results suggest that policymakers also need toavoid starting too early.
In this series of simulations, we keep the starting date of the policy intervention fixed atMarch 12 and the duration of the policy intervention fixed at 26 weeks, whilst varying27OVID-19 and Global Economic Growth
Working Paper – June 2020 the intervention inensity. We simulate policies that reduce production by 5%, 15%, and25%, and the simulation results are shown in Figure 10.The three simulations with different intervention intensity, shown in Figure 10, suggestthat varying the intensity of the intervention mainly alters the timing of the pandemic, butdoes little to mitigate the economic and human impacts: apart from a delay in the mainphase of the pandemic, most variables behave similar to the
No Intervention scenario.This indifference between the intensities of the interventions is likely related to therelationship we identified between the GDP shortfall and the infection rate reduction, asshown in Figure 6. There are strong diminishing returns, such that even an interventionof a low intensity (5%) already reduces the infection rate substantially (60%), and thatadditional measures have a lower effect on the infection rate.Essentially, the intensity of the intervention – above a certain minimum level – appearsto be less important than the timing of the intervention. Again, this finding contradictsthe study of Guan et al. (2020), our results indicating that intervention should perhapsbe less strict, as the intervention intensity faces strong diminishing returns.
To analyse the impact of the duration of the policy intervention, we vary the duration ofthe policy intervention, whilst maintaining the intervention intensity fixed at 10% and thestarting date fixed at March 12. Figure 11 shows the results of simulating interventiondurations of 4 weeks, 28 weeks, 52 weeks and 76 weeks. It is clear from the figure thatthe duration of the policy intervention can have a large impact on the trajectory of thepandemic, and its human and economic aftermath.From Figure 11, it seems that policies with longer durations clearly lead to lower humanand economic impacts, with a 76-week duration – the longest of our simulations – showingdramatically lower number of total mortalities, as well as a much quicker post-pandemicrecovery of production. The key appears to be that 76 weeks is sufficient to include thepeak in active infections of the trajectory determined by the reduced infection rate. Thisobservation suggests that policies with a lower intensity would require a shorter duration,28OVID-19 and Global Economic Growth
Working Paper – June 2020 U S D ( ) (a) Capital Stock No PandemicNo Intervention05%15%25% 2020 2022 2024 2026 2028 2030345678 P e o p l e (b) Labour Force No PandemicNo Intervention05%15%25% U S D ( ) (c) Production No PandemicNo Intervention05%15%25% U S D ( ) (d) Investment No PandemicNo Intervention05%15%25% U S D ( ) (e) Pandemic Direct Costs No PandemicNo Intervention05%15%25% 2020 2022 2024 2026 2028 203001234 P e o p l e (f) Active Infections No PandemicNo Intervention05%15%25% P e o p l e Recovered Individuals
No PandemicNo Intervention05%15%25% P e o p l e (h) Deceased Individuals No PandemicNo Intervention05%15%25%
Figure 10:
Simulation results when varying the intensity of the policy intervention.
Working Paper – June 2020 U S D ( ) (a) Capital Stock No PandemicNo Intervention04 weeks28 weeks52 weeks76 weeks 2020 2022 2024 2026 2028 2030345678 P e o p l e (b) Labour Force No PandemicNo Intervention04 weeks28 weeks52 weeks76 weeks U S D ( ) (c) Production No PandemicNo Intervention04 weeks28 weeks52 weeks76 weeks U S D ( ) (d) Investment No PandemicNo Intervention04 weeks28 weeks52 weeks76 weeks2020 2022 2024 2026 2028 2030010000200003000040000500006000070000 U S D ( ) (e) Pandemic Direct Costs No PandemicNo Intervention04 weeks28 weeks52 weeks76 weeks 2020 2022 2024 2026 2028 203001234 P e o p l e (f) Active Infections No PandemicNo Intervention04 weeks28 weeks52 weeks76 weeks P e o p l e Recovered Individuals
No PandemicNo Intervention04 weeks28 weeks52 weeks76 weeks P e o p l e (h) Deceased Individuals No PandemicNo Intervention04 weeks28 weeks52 weeks76 weeks
Figure 11:
Simulation results when varying the duration of the policy intervention.
Working Paper – June 2020 whereas policies with a higher intensity would require a longer duration – something whichmay, at first, appear counter-intuitive.Our results partially support the findings of Eichenbaum et al. (2020), who warnof ending the intervention “too early”, but contradict the claims of Guan et al. (2020)that intervention should be “short” – although the duration of the intervention shouldnaturally be as short as possible. Our results suggest that stricter policies will requirelonger durations.
We have tried to make sensible modelling choices in this study, but, like all models,our model is a simplification that focuses only on certain aspects and ignores others. Thesimulation results should not be understood literally : the intention of our model has neverbeen to provide numerically accurate predictions, but to generate insights into the impactsof policy interventions by analysing the dynamics of the system as a whole. Although theinsights from the numerical simulations can contribute to improving policy interventionsduring pandemics, it is important to appreciate the limitations of our model and results.We are concerned about the quality of the parameters used for the epidemiological partof the model: there is a great deal of doubt and uncertainty about the quality of the officialdatasets on the spread of the pandemic – that is, the number of confirmed cases, recoveredcases and mortalities. There is a general sense that the number of confirmed cases isnot representative for the number of infections, as testing is severely lacking in manyregions. The lack of testing also affects the number of mortalities due to the pandemic,and number of recovered cases. Although we have used the data that is available withoutmuch discrimination, we share the concerns of many other researchers as to the qualityof this data.It is also unusual for a model of economic growth to operate at a daily resolution. Wedo not think this directly invalidates our resulting insights, although it means that theparameter values may appear unusual to researchers and practicioners, and that specialcare must be taken in the interpretation of the results. An alternative would be to31OVID-19 and Global Economic Growth
Working Paper – June 2020 develop the model in continuous time, which might be more familiar to some. However,in that case it would be necessary to discretise the model later for performing numericalexperiments – the model would, in the end, be the same, so presenting the model directlyin discrete time appears to be a simpler alternative.The parameter values were chosen for the model to represent the global economy andthe global spread of COVID-19. There are, however, large differences between regionsin the world. For instance, the five countries used for estimating the economic impactof policy measures – France, Germany, Italy, Spain, and the United Kingdom, whichwere chosen for their data availability – are probably not entirely representative for therest of the world. There is also no global central government that implements globalpolicy, and our insights are therefore not directly applicable by any specific authority.The purpose of the study, however, was not to generate recommendations for specificactions, but to generate insights into the impacts and trade-offs that policy interventionsmust consider. Regional, national and local policy can differ from “global” policy – andprobably should, as policy can be optimised to local conditions – but the insights onintervention timing, intensity and duration may nevertheless be useful at these levelsalso. It would be possible to adapt the model for use at regional, national or local scales.In this case, we would recommend considering replacing the linear admission costs with aspecification that allows for increasing marginal costs of admissions, which might betterreflect increasing costs in the short run due to capacity saturation.For calibrating the model, we estimated the economic impacts using an estimate ofthe shortfall in electricity consumption. Although we believe this approach is valid, it isdifficult to assess the accuracy of the economic impact estimates. The approach basedon electricity consumption could also be complemented by other near-real-time, high-resolution data sources that are believed to correspond well to economic activity, such assatellite observations, mobile phone movement and activity data, and urban traffic data.However, these data are not usually as widely available as electricity consumption data.The model does not incorporate any demographic heterogeneity. Since some pan-demics appear to affect people with certain demographic characteristics differently, this32OVID-19 and Global Economic Growth
Working Paper – June 2020 may bias the results. For instance, the mortality rate of COVID-19 appears to differgreatly between old and young people: if the disease has a greater impact on groupsthat were not originally included in the work force anyway, the model could exaggeratethe economic impact of the pandemic by disconsidering demographic heterogeneity. Wedo not believe this to affect the main insights derived from our model simulations, sincewe do not think it substantially alters the dynamics of the system. However, it wouldcertainly be an issue for the “predictive accuracy” of the simulations.Since the model is deterministic, agents in the model have perfect foresight from thevery start of the simulation. This is, naturally, not true in the real world, in which there arelarge uncertainties about future developments. This gives the model agents an unrealisticability to plan for the future, and the economic portion of the model should therefore beconsidered an optimistic path. Another detail that also may positively bias the outcomes,is that the model does not include structural damages – such as bankruptcies, institutionalchange, changing habits, and so forth – and affords the model agents much more flexibilitythan economic agents may have in reality, where they may be facing additional restrictions.Finally, we only simulated very simple policies for the purpose of understanding theimpact of altering the policy in a very specific way. For instance, superior policies canbe made relatively easily by allowing the intensity of the intervention to vary during thepandemic. Our examples, in which starting dates, intensity and duration are fixed, onlyserved for illutration and to understand some of the dimensions of policy intervention.We reiterate that our purpose has not been to provide numerically accurate predictions– nor the means to generate accurate predictions – of the evolution of the COVID-19pandemic or the global economy. We have only explored particular aspects of effectivepolicy responses to a pandemic, using a very high-level and theoretical approach, and it iswith this in mind that our results are most appropriately appreciated. Our research doesnot aim to offer specific guidance for world authorities on the handling of the COVID-19pandemic, but to analyse how a pandemic interacts with the global economy and thushelp establish a set of of general guidelines.33OVID-19 and Global Economic Growth
Working Paper – June 2020
We have presented a mathematical model for the joint evolution of the economy anda pandemic, based on incorporating the dynamics of the SIR model that describes thespread of epidemics into a neoclassical economic growth model framework. This model issubsequently adapted to represent the global economy under the impact of the COVID-19 pandemic by selecting appropriate functional forms and parameter values. The modelincludes a parameter that represents policy, by which economic production can be loweredin exchange for a reduced infection rate.Using the calibrated model, we simulate the joint evolution of the economy and thepandemic for a series of policy assumptions in order to discover what is the most effectivetiming of a policy intervention, what intensity of policy intervention is most effective, andhow long policy intervention should last.Our experiments suggest that it is most effective to start the policy intervention slightlybefore the number of confirmed cases grows at its highest rate – that is, to wait until thedisease is sufficiently spread. Not only does this help lower the peak in active infections,it also reduces the economic impact and the number of mortalities. Starting too early candelay the pandemic, but does not otherwise significantly alter its course, whereas startingafter the peak in active infections can obviously not impact the peak.Furthermore, altering the intensity of the intervention does not appear to greatly in-fluence the evolution of the pandemic nor the economy, other than cause minor delays.We ascribe the lack of effect to the concave relationship that we estimated between inter-vention intensity and infection rate reduction, as a large reduction in infection rate canbe achieved by sacrificing a modest proportion of economic production, and appears toshow strong decreasing returns thereafter. Our estimates suggest that a 60% reductionin the infection rate can be achieved by sacrificing only 5% of production, whereas a 70%reduction in infection rate could be achieved for a 10% reduction in production.Altering the duration of the intervention showed that interventions with a longerduration lead to significantly lower mortalities and a quicker post-pandemic recovery ineconomic production. The key observation is that the policy must include the peak of the34OVID-19 and Global Economic Growth
Working Paper – June 2020new path set out by the reduced infection rate: in short, policy intervention should lastuntil the peak has passed. Therefore – somewhat counter-intuitively – stricter policiesshould last longer, and less strict policies should last shorter.Although the scenarios we present are not necessarily numerically accurate as predic-tions – mostly due to generalisations made for modelling purposes, large regional varia-tions, and large uncertainties in the parameters – our conclusions are based mainly onthe dynamics revealed by the policy experiments, and not specifically on their numericalvalues. As such, we hope that our model can serve as a tool for enhancing our under-standing of the design of effective policies against the spread of pandemics, and that ourinsights can contribute to this discussion and provide general guidelines for policymakers.
References
Acemoglu, D., 2011. Introduction to Modern Economic Growth. Princeton UniversityPress, Princeton.Acemoglu, D., Chernozhukov, V., Werning, I., Whinston, M., May 2020. Optimal Tar-geted Lockdowns in a Multi-Group SIR Model. Tech. Rep. w27102, National Bureau ofEconomic Research, Cambridge, MA.Alvarez, F. E., Argente, D., Lippi, F., Apr. 2020. A Simple Planning Problem for COVID-19 Lockdown. Working Paper 26981, National Bureau of Economic Research, Cam-bridge, Massachusetts.Andersson, T., Erlanson, A., Spiro, D., ¨Ostling, R., May 2020. Optimal Trade-Off BetweenEconomic Activity and Health During an Epidemic. arXiv:2005.07590 [econ].Arndt, C., Feb. 2003. HIV-AIDS, human capital, and economic growth prospects forMozambique. Tech. Rep. 26310, The World Bank.Azomahou, T. T., Boucekkine, R., Diene, B., May 2016. HIV/AIDS and Development: AReappraisal of the Productivity and Factor Accumulation Effects. American EconomicReview 106 (5), 472–477. 35OVID-19 and Global Economic Growth
Working Paper – June 2020
Bauch, C. T., Anand, M., May 2020. COVID-19: When should quarantine be enforced?The Lancet Infectious Diseases, S147330992030428X.Bell, C., Gersbach, H., Jan. 2009. The macroeconomics of targeting: The case of anenduring epidemic. Journal of Health Economics 28 (1), 54–72.Bonds, M. H., Keenan, D. C., Rohani, P., Sachs, J. D., Apr. 2010. Poverty trap formedby the ecology of infectious diseases. Proceedings of the Royal Society B: BiologicalSciences 277 (1685), 1185–1192.Brauer, F., Castillo-Chavez, C., 2012. Mathematical Models in Population Biology andEpidemiology. Vol. 40 of Texts in Applied Mathematics. Springer New York, New York,NY.Cuddington, J. T., Sep. 1993a. Further results on the macroeconomic effects of AIDS :The dualistic, labor - surplus economy. The World Bank economic review 7 (14078),403–417.Cuddington, J. T., May 1993b. Modeling the macroeconomic effects of AIDS, with anapplication to Tanzania. The World Bank economic review 7 (14101), 173–189.Cuddington, J. T., Hancock, J. D., Apr. 1994. Assessing the impact of AIDS on the growthpath of the Malawian economy. Journal of Development Economics 43 (2), 363–368.Cuesta, J., Sep. 2010. How much of a threat to economic growth is a mature AIDSepidemic? Applied Economics 42 (24), 3077–3089.Delfino, D., Simmons, P. J., Nov. 2000. Positive and Normative Issues of Economic Growthwith Infectious Disease. Tech. Rep. 00/48, Department of Economics, University ofYork.Eichenbaum, M. S., Rebelo, S., Trabandt, M., Apr. 2020. The Macroeconomics of Epi-demics. Working Paper.Fenichel, E. P., Mar. 2013. Economic considerations for social distancing and behavioralbased policies during an epidemic. Journal of Health Economics 32 (2), 440–451.36OVID-19 and Global Economic Growth
Working Paper – June 2020
Galante, M., Garin, O., Sicuri, E., Cots, F., Garc´ıa-Alt´es, A., Ferrer, M., Dominguez,`A., Alonso, J., Feb. 2012. Health Services Utilization, Work Absenteeism and Costs ofPandemic Influenza A (H1N1) 2009 in Spain: A Multicenter-Longitudinal Study. PLoSONE 7 (2), e31696.Goenka, A., Liu, L., Jul. 2019. Infectious diseases, human capital and economic growth.Economic Theory.Goenka, A., Liu, L., Nguyen, M.-H., May 2010. Infectious Diseases, Optimal Health Ex-penditures and Growth. Working Paper, National University of Singapore, Singapore.Goenka, A., Liu, L., Nguyen, M.-H., Jan. 2014. Infectious diseases and economic growth.Journal of Mathematical Economics 50, 34–53.Guan, D., Wang, D., Hallegatte, S., Huo, J., Li, S., Bai, Y., Lei, T., Xue, Q., Davis, S. J.,Coffman, D., Cheng, D., Chen, P., Liang, X., Xu, B., Lu, X., Wang, S., Hubacek, K.,Gong, P., Apr. 2020. Global economic footprint of the COVID-19 pandemic. Preprint,In Review.Haacker, M., 2002. Modeling the Macroeconomic Impact of HIV. International MonetaryFund.Hansen, G. D., Prescott, E. C., Aug. 2002. Malthus to Solow. American Economic Review92 (4), 1205–1217.Herlihy, D., 1997. The Black Death and the Transformation of the West. Harvard Uni-versity Press, Cambridge, Mass.Higgins, A. M., Pettil¨a, V., Harris, A. H., Bailey, M., Lipman, J., Seppelt, I. M., Webb,S. A., May 2011. The Critical Care Costs of the Influenza A/H1N1 2009 Pandemic inAustralia and New Zealand. Anaesthesia and Intensive Care 39 (3), 384–391.Johnson, N. P. A. S., Mueller, J., 2002. Updating the Accounts: Global Mortality of the1918-1920 ”Spanish” Influenza Pandemic. Bulletin of the History of Medicine 76 (1),105–115. 37OVID-19 and Global Economic Growth
Working Paper – June 2020
Kermack, W. O., McKendrick, A. G., Aug. 1927. A contribution to the mathematicaltheory of epidemics. Proceedings of the Royal Society of London. Series A, ContainingPapers of a Mathematical and Physical Character 115 (772), 700–721.Kleczkowski, A., Maharaj, S., Rasmussen, S., Williams, L., Cairns, N., Dec. 2015. Spon-taneous social distancing in response to a simulated epidemic: A virtual experiment.BMC Public Health 15 (1), 973.La Torre, D., Malik, T., Marsiglio, S., Apr. 2020. Optimal control of prevention and treat-ment in a basic macroeconomic–epidemiological model. Mathematical Social Sciences,S0165489620300433.Lau, K., Hauck, K., Miraldo, M., Feb. 2019. Excess influenza hospital admissions andcosts due to the 2009 H1N1 pandemic in England. Health Economics 28 (2), 175–188.Ngonghala, C. N., De Leo, G. A., Pascual, M. M., Keenan, D. C., Dobson, A. P., Bonds,M. H., Aug. 2017. General ecological models for human subsistence, health and poverty.Nature Ecology & Evolution 1 (8), 1153–1159.Ngonghala, C. N., Pluci´nski, M. M., Murray, M. B., Farmer, P. E., Barrett, C. B., Keenan,D. C., Bonds, M. H., Apr. 2014. Poverty, Disease, and the Ecology of Complex Systems.PLoS Biology 12 (4), e1001827.Nordhaus, W. D., 1992. An optimal transition path for controlling greenhouse gases.Science 258 (5086), 1315–1319.Perrings, C., Castillo-Chavez, C., Chowell, G., Daszak, P., Fenichel, E. P., Finnoff, D.,Horan, R. D., Kilpatrick, A. M., Kinzig, A. P., Kuminoff, N. V., Levin, S., Morin,B., Smith, K. F., Springborn, M., Dec. 2014. Merging Economics and Epidemiologyto Improve the Prediction and Management of Infectious Disease. EcoHealth 11 (4),464–475.Pope, M., Haase, A. T., Jul. 2003. Transmission, acute HIV-1 infection and the quest forstrategies to prevent infection. Nature Medicine 9 (7), 847–852.38OVID-19 and Global Economic Growth
Working Paper – June 2020
Data and Code Availability
All the data used in this study is available to the public, and the various data sourceshave been referenced at the appropriate places along the study.The custom computer code for running the simulations is available at https://github.com/iantrotter/ME3PI-SIMhttps://github.com/iantrotter/ME3PI-SIM