Modern Pandemics: Recession and Recovery

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1295 August 2020 Modern Pandemics: Recession and Recovery Chang Ma, John Rogers, and Sili Zhou Please cite this paper as: Ma, Chang, John Rogers, and Sili Zhou (2020). “Modern Pandemics: Recession and Recovery,” International Finance Discussion Papers 1295. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/IFDP.2020.1295. NOTE: International Finance Discussion Papers (IFDPs) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the International Finance Discussion Papers Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

Modern Pandemics: Recession and Recovery∗ Chang Ma† John Rogers‡ Sili Zhou§ First Draft: March 2020 This Version: June 2020 Abstract Weexaminetheimmediateeffectsandbounce-backfromsixmodernhealthcrises: 1968Flu,SARS(2003),H1N1(2009),MERS(2012),Ebola(2014),andZika(2016). Time-seriesmodelsforalargecross-sectionofcountriesindicatethatrealGDPgrowth falls by around three percentage points in affected countries relative to unaffected countriesintheyearoftheoutbreak. Bounce-backinGDPgrowthisrapid,butoutput isstillbelowpre-shocklevelfiveyearslater. Unemploymentforlesseducatedworkers is higher and exhibits more persistence, and there is significantly greater persistence infemaleunemploymentthanmale. ThenegativeeffectsonGDPandunemployment are felt less in countries with larger first-year responses in government spending, especially on health care. Affected countries’ consumption declines, investment drops sharply, and international trade plummets. Bounce-back in these expenditure categories is also rapid but not by enough to restore pre-shock trends. Furthermore, indirecteffectsonown-countryGDPfromaffectedtradingpartnersaresignificantforboth the initial GDP decline and the positive bounce back. We discuss why our estimates arealowerboundfortheglobaleconomiceffectsofCOVID-19andcomparecontours ofthecurrentpandemictothehistoricalepisodes. Keywords: Health crises; COVID-19; Output loss; Unemployment; Trade network;Fiscalpolicy JELClassification: I10,E60,F40 ∗We thank Valerie Cerra, Neil Ericsson, Huasheng Gao, Nils Gornemann, Yi Huang, Esa Jokivuolle, Hidehiko Matsumoto, Jun Qian, Shangjin Wei, Jonathan Wright, and seminar participants at the Bank of Finland (BOFIT), Federal Reserve Board, Fudan University, IMF and SUFE for helpful comments. Lexie Banham, Caitlin Dutta, and Jiqiao Gao provided superb research assistance. The views in this paper are solelytheresponsibilityoftheauthorsandshouldnotbeinterpretedasreflectingtheviewsoftheBoardof GovernorsoftheFederalReserveSystemorofanyotherpersonassociatedwiththeFederalReserveSystem. †FanhaiInternationalSchoolofFinance,FudanUniversity(changma@fudan.edu.cn). ‡InternationalFinanceDivision,FederalReserveBoard(John.Rogers@frb.gov). §FanhaiInternationalSchoolofFinance,FudanUniversity(silizhou@fudan.edu.cn).

“We’ve never had a coronavirus pandemic infection like this. It may have happenedcenturiesago,butwedidn’tseeit.” —MichaelOsterholm,PhD,MPH,DirectoroftheCenterforInfectious DiseaseResearchandPolicy,UniversityofMinnesota,29May2020 1 Introduction Epidemiologists, economists, and policymakers continue to devote considerable attention to forecasting the human ravages and economic toll of the coronavirus COVID-19. As worldwide deaths attributed to the pandemic approach half a million, prospects for economic activity and financial markets are equally funereal. Although economists have documented that many financial and political crises are associated with severe recessions (see Cerra and Saxena (2008), Reinhart and Rogoff (2009) and Jorda` et al. (2013)), until very recently little attention was paid to global health crises.1 This changed dramatically with theoutbreakofCOVID-19,afterwhichmanynewhealthcrisespapershavebeenwrittenin ashortperiodoftime. Mostpapershavefocusedonthecurrentcrisis,itseconomicimpact, and policy responses.2 The accuracy and usefulness of those analyses will be proven in time, of course, as at this early stage we still know little about the general features of a pandemiclikeCOVID-19andhowtodealwithit. This paper makes progress understanding COVID-19 by systematically documenting the global impact of previous pandemics and epidemics in a large set of countries. We analyzesixepisodesidentifiedbyglobalhealthexpertsinJamisonetal.(2017),beginning with the 1968 Flu up to Zika in 2016. We focus on estimating the effect of these health crisesonGDPgrowthandunemployment,bothintheonsetyearofthecrisisaswellasthe dynamic effects over time. The latter gives us insights into how quickly countries recover economically. Inthatvein,weexaminewhetherornoteconomicrecoveryisaidedbyfiscal policy. We also examine the effects of the health crises on the components of GDP and on international trade. With the latter, we examine spillover or network effects, asking for example,howmuchisanindividualcountry’seconomyaffectedbythefactthatitstrading partnersufferedfromthehealthcrisis? 1ExceptionsincludeJamisonetal.(2013),Fanetal.(2016),Jamisonetal.(2017)andreferenceswithin. 2Again,withexceptionslikeJorda` etal.(2020)andBarroetal.(2020),whoexaminepastcrises. 1

We primarily use local projections impulse responses as in Jorda` (2005). This gives us a flexible and widely used approach to estimate the effect of a health crisis shock on GDP growth or unemployment of affected countries relative to unaffected countries, including the dynamic effects. Identification relies on the dates that health organizations officially declaredacrisis. Wealsomakeuseofpanelregressions,whichfacilitaterobustnesschecks of our baseline results, including addressing concerns about endogeneity, which we do in a seemingly unrelated regressions framework. We allow for cross-sectional dependence by correcting standard errors around all of our estimates using the method of Driscoll and Kraay(1998).3 Wefindthattheeconomicimpactoftheaveragepasthealthcrisisissizeable. RealGDP falls by around three percentage points and unemployment rises by nearly one percentage point, in affected countries relative to unaffected countries, in the year the outbreak is officially declared. These effects are larger for affected countries that experience more severe health crisis shocks. Moreover, these effects are very persistent. Although GDP growth rebounds quickly in one year, output remains below its pre-shock level five years later. For unemployment, it takes two years for the effect to vanish. Our findings on the effectofhealthcrisesareconsistentwithpreviousanalysesoffinancialcrises,inparticular with respect to the persistence of the shock’s effects, as in Cerra and Saxena (2008), for example. As a basis for understanding the magnitude and persistence of our health crisis shocks, we show that they are similar to those from systemic banking crisis shocks, as identifiedbyLaevenandValencia(2013). Furthermore, we document heterogeneity in the effects of health crises. First, we show that there is a differential effect on workers based on education and gender. For example, less educated workers experience larger unemployment than those with higher levels of education. Inaddition,thepersistenceoffemaleunemploymentissignificantlygreaterthan of male unemployment. Second, the services and industry sectors are relatively hard hit, in terms of both GDP growth and unemployment, while agriculture is largely unaffected. Third, there is notable cross-country heterogeneity. For example, affected countries in the World Bank’s High Income Country (HIC) category experience a larger decline in GDP growth (increase in unemployment) relative to unaffected HICs than is the case with Low IncomeCountries(LIC).4 3ResultsfromestimatinganAR(4)asinCerraandSaxena(2008)aresimilartoJorda’slocalprojections. Theseareavailableonrequest. Anotherapproachwouldbetoestimateimpulseresponsesusingpanelvector autoregressions,anoptionweeschewinfavorofthesimplicityandflexibilityoflocalprojections. 4OnepotentialreasonisthatHICrelymoreonservicesand(or)industrysectorsthanLIC. 2

Thenegativeimpactofhealthcrisesisfeltinallcomponentsofnationalspending. Both consumption and investment decline, with the latter being especially large. International tradealsoplummets,andonceagain,bounce-backisrapidbutbyanamountinsufficientto restorethepre-crisistrend. Thedeclineintotalspendingcouldspillovertoothercountries, includingthoseunaffectedbythecrisis,throughatradelinkagechannel. Wefindthatthese indirecteffectsondomesticGDP—fromtradingpartnersaffectedbythedisease—arenot trivial, both in terms of magnifying the initial decline in GDP and in the positive bounceback. Our estimate of the indirect channel working through international trade is around 20%ofthetotaleffect,consistentwithstructuralmodelestimationinBonadioetal.(2020). Cangovernmentpolicymakeadifference,asproposedby,e.g.,Gourinchas(2020)and Drechsel and Kalemli-Ozcan (2020)? We find that countries that respond in the onset year with higher government expenditures, especially on health care, enjoy more bounce-back in output growth compared to countries with less of a fiscal expenditures response. Given that the health crises have a rather persistent effect on output, according to our estimation, a quicker and larger bounce-back resulting from a stabilizing fiscal policy could have a permanent impact on economic activity, consistent with Dupraz et al. (2019). In contrast, wedonotfindthatloweringtaxesiseffectiveinhasteningrecovery. Contribution to the Literature We contribute to several strands of the literature. First, our paper belongs to the literature that investigates the effect of financial and political crises as in Cerra and Saxena (2008), ReinhartandRogoff(2009),Jorda` etal.(2013)andLaevenandValencia(2013). Different from these papers, we investigate the effect of global health crises using several postwar pandemics and epidemics, in the similar spirit of Jorda` et al. (2011) who study financial crises using data from 14 developed countries over 140 years (1870–2008). Jorda` et al. (2020)alsoexaminelow-frequencyeconomicconsequencesofpandemicsbutfocusonthe real rates of return, while we examine GDP, unemployment, and international trade. Our work is also related to papers that look at the effect of the 1918 Spanish flu (Barro et al. (2020)andCorreiaetal.(2020))withimplicationsfortheCOVID-19pandemic. Second, our paper contributes to the large volume of new work investigating the economic impact and policy implications of COVID-19. Most of the work has been based on estimation or calibration of versions of the SIR model. For example, Atkeson (2020) analyzesdiseasescenariosthataredesignedtoprovideinputintocalculationsofeconomic 3

costs. Specifically, he works with a Markov model of epidemic spread in which the population is divided into three categories: susceptible, actively infected, and no longer contagious. How an epidemic plays out over time is determined by the transition rates between these three states. Eichenbaum et al. (2020) emphasize that the severity of the recession will be exacerbated by people’s decisions to cut back on economic activity in order to reduce the severity of the epidemic and save lives. As the authors emphasize, the optimal governmentcontainmentpolicysavesthousandsoflivesbutworsenstherecessionbecause infected people do not fully internalize the effect of their decisions on the spread of the virus. Berger et al. (2020) focus on testing and case-dependent quarantine during a period of asymptomatic infection, and find that testing can result in a pandemic with smaller economic losses while keeping the human cost constant. Glover et al. (2020) emphasize the distributional consequences of shutdown policies. Different from those papers, ours directlyestimatestheeconomicimpactandpolicyeffectivenessusinghistoricalevents. Third, our paper contributes to the literature that investigates the role of government policy in containing crises. For example, Gourinchas (2020) and Drechsel and Kalemli- Ozcan (2020) both propose a strong fiscal response to contain the impact of COVID-19. A large and growing literature studies different policy responses to contain the impact of COVID-19 such as Alvarez et al. (2020), Guerrieri et al. (2020), Fornaro and Wolf (2020) and Bethune and Korinek (2020). Our paper adds to this work by directly estimating the impactofdifferentpolicyresponsestopastcrises. Inthissense,ourpaperiscloselyrelated to the work by Cerra et al. (2013), which looks at different international policy responses tospurarecoveryfromrecessions. How much can we say about COVID-19 based on this paper? We believe that our estimates are likely a lower bound, for reasons of both “shock” and “propagation”. COVID- 19 is more widespread than the average crisis in our sample, and may have a higher kill rate. Travel bans, social distancing, and economic lock downs are without parallel. In the COVID-19 world with more substantial trade linkages, the indirect, trade network channel is likely to be more important than what we find for these historical episodes. The fact that today’s global value chains are more prevalent suggests that countries will go down, and perhaps rebound, more sharply from COVID-19. The early signs indeed point to COVID-19 being worse.5 Nevertheless, massive interventions by central banks and fiscalpolicymakers,ofthetypewefindhelpstospeeduprecovery,arenowbeingundertaken 5Accordingtoinitialdatareleases,GDPgrowthin2020Q1inChina,theU.S.,andEuroareawere-6.8%, -4.8%,and-14.5%,respectively,whileU.S.unemploymentskyrocketedintodoubledigitsinAprilandMay. 4

worldwide. Restoration of robust international trade linkages remains an open question, however. Ominous signs of prolonged backlash against China appear from policymakers and in the media. The sentiment for countries not to be so reliant on imports, especially in sensitivesectorslikemedicalsupplies,maywellproveanintractablefoeoftrade. Inthenextsection,wedescribeourdata. Section3describesoureconometricapproach, including how we address concerns about endogeneity. Section 4 documents the effect of health crises on GDP and unemployment, while section 5 presents the effects on spending and investigates propagation through trade linkages. Section 6 considers the effectiveness of fiscal policy responses. Section 7 concludes. We discuss the relevance of our results for the ongoing pandemic in Appendix C, including projections indicating how different “this time” is materializing in 2020-21 compared to estimates from past crises. Our online supplementcontainsadditionalinformationondatasourcesandtablesandfigures. 2 Data We combine data from several sources. For the annual country-level analyses, we rely mainly on the World Development Indicators (WDI) from the World Bank. We also get quarterly GDP data from OECD National Accounts Statistics. Forecasts of GDP growth are obtained from Consensus Economics Inc. and bilateral trade data from the World IntegratedTradeSolution(WITS)database. Toidentifythepandemicandepidemicevents,we manuallycollectdatafromtheWHOandotherpublicresources. Epidemic and Pandemic Events WefocusonsixpostwarpandemicandepidemiceventsidentifiedinJamisonetal.(2017)’s volume 9 of Disease Control Priorities, a book authored by well-known global health experts. The Disease Control Priorities Network (DCPN) was a multi-year project managed bytheUniversityofWashington’sDepartmentofGlobalHealthandtheInstituteforHealth Metrics and Evaluation.6 As of this writing, the book has received more than 3,000 citationsaccordingtoGoogleScholar. Threeeditionshavebeenpublished: DCP1in1993(by theWorldBank),DCP2in2006,andmostrecentlyDCP3in2017.7 Werelymainlyonthe 6Seehttp://dcp-3.org/about-projectfordetails. 7Contributorsincludeover500scholars, policymarkersandtechnicalexperts. Theeditorsincludewellknown economists and CDC experts, such as Dean Jamison, Hellen Gelband, Susan Horton, Prabhat Jha, 5

9thvolumeofedition3whichfocusesontheeconomicimpactofpandemics. Using this volume as our guide, the six episodes we analyze are: the 1968 Flu (aka “Hong Kong flu”), SARS (2003), H1N1 (2009), MERS (2012), Ebola (2014), and Zika (2016). We determine the timing of the event from the dates that the World Health Organization (WHO) officially declares a Public Health Emergency of International Concern (PHEIC).Inmostcases,therearesignificanttimelagsbetweentheinitialappearanceofan outbreakandofficialdeclaration.8 ReportinglagsandevendiscrepanciesbetweentheCentersforDiseaseControlandPrevention(CDC)andtheWHOdonotaffectourkeyidentificationvariable—adummythatequalsonewhenWHOdeclaresapandemic/epidemicfor an affected country and zero otherwise. In our matched sample, we have 287 country-year observationsfortheidentifiedshocks.9 DetailedinformationisinTableS.1. Having identified the epidemic/pandemic events and affected countries, we examine data on total cases and deaths from the official websites of the WHO, European Centre for Disease Prevention and Control (ECDC), CDC and from public news articles. Among the six events, the most widespread and deadly one is H1N1. It affected more than 200 countries, with more than 284,000 recognized deaths reported by the US CDC.10 The ECDC is theonlysourcecontainingdetailedinformationforallaffectedcountriesaroundtheworld. Figure A.1 depicts the global severity of those episodes, displaying the ECDC reported number of cases. Although the on-going crisis stands out for its severity, other episodes were large. For example, it is estimated that 500,000 infections occurred in Hong Kong in thefirsttwoweeksofthe1968Flu. Correspondingly,governmentshaverespondedquickly to contain the negative effect of those health crises. We provide details of each historical episodeintheonlinesupplementTableS.2. RamananLaxminarayan,CharlesN.MockandRachelNugent.TheprojectwasfundedbytheBill&Melinda GatesFoundation,andthevolumeincludesanintroductionbyLawrenceH.Summers. 8Forexample,HoffmanandSilverberg(2018)findthattheH1N1outbreakinitiallybeganonMarch15, 2009, was detected by officials on March 18, 2009, but was declared a PHEIC only on April 25, 2009. Similarly,theWestAfricanEbolaoutbreakbeganDecember26,2013,wasdetectedonMarch22,2014,but wasdeclaredaPHEIConlyonAugust8,2014. ForZika,themainconcernwasaboutidentificationbetween microcephalyandthetrueZikavirusinfections. SomeconsiderthisoutbreaktohavebegunonOctober22, 2015,whentheriseinmicrocephalycaseswasfirstidentified.Later,onNovember28,2015,therewasstrong evidenceforalinkbetweenthevirusandthemicrocephaly. Nevertheless,theZikaoutbreakwasdeclareda PHEIConlyonFebruary1,2016. 9Originally, we have 313 country-year observations for the identified shocks, with 287 of them having dataforgrowthrates. 10This amount is much larger than the number reported by WHO. The discrepancy exemplifies the challenges in finding reliable and complete coverage of cases and fatalities, a subject we return to below. Detailed information is at http://www.cidrap.umn.edu/news-perspective/2012/06/ cdc-estimate-global-h1n1-pandemic-deaths-284000. 6

Country-level Variables We mainly use annual country-level data from the World Bank’s World Development Indicators (WDI). This data set offers wide country coverage, containing the 210 countries (economies) listed in Table A.1. The data set contains annual observations from 1960 to 2018. TheWDIdatabaseisalsousefulinprovidingconsistentcoverageofmanyvariables we use for cross sectional comparison. This includes key controls for our GDP growth and unemployment regressions such as trade to GDP, domestic credit to GDP, population, andGDPpercapita. WealsousequarterlyrealGDPgrowth,fromtheOECDNationalAccountsStatistics. ThesystemicbankingcrisesareidentifiedbyLaevenandValencia(2013) (with an updated dataset in Laeven and Valencia (2020)) and a U.S. recession dummy is from the NBER. Forecasts of GDP growth are obtained from Consensus Economics Inc. The data are monthly, from a survey of analysts from large banks and financial firms. The data covers over 32 countries from January 1990 to February 2020. We take GDP growth expectations based the end of year t−1 on year t for each country-year. We also collect bilateral trade data from the World Integrated Trade Solution (WITS), which aggregates data from UN COMTRADE and UNCTAD TRAINS database. It provides bilateral trade exports and imports for more than 200 countries from 1988 to 2018 (see Table S.3). All continuous variables are trimmed at the top and bottom 1% to remove outliers. Summary statisticsareinTableS.4ofouronlinesupplement. GDP growth Around Health Crises AsummarylookattherelationshipbetweenthesehealthcrisesandannualrealGDPgrowth canbeseeninFigure1.11 Intheupperleftpanel,wedepictthedistributionofGDPgrowth in all country-years other than the health crisis episodes. The upper right panel is the equivalent for affected countries in the year the crisis was officially declared, while the panel just below it (lower right) is for unaffected countries in that same year. Finally, to gauge bounce-back, the lower left panel depicts the GDP growth distribution for affected countriesintheyearimmediatelyfollowingthecrisis. Thecoloredbarsdepictthevaluesin eachgroupingforthreecountries: Finland,UnitedStates,andChina.12 Average GDP growth for the non-disease sample is 3.8%. In the onset years of the health crises, average GDP growth falls noticeably for affected countries, to 1.4%, while 11WealsoinvestigatehigherfrequencyquarterlydatainOnlineSupplementSectionS.4. 12NotethattheU.S.wasnever“Unaffected”,hencenoobservationinthelowerrightpanel. 7

Figure1RealGDPGrowthDistributionsinDiseaseandNon-DiseaseYears ytisneD 51. 1. 50. 0 01- 5- 0 NIF ASU 5 anihC 01 51 Non-disease Years Mean = 3.83 China = 8.29 (red) USA = 3.02 (blue) Finland = 2.74 (yellow) ytisneD 51. 1. 50. 0 01- 5- NIF 0 ASU 5 anihC 01 51 Onset Years (Affected Countries) Mean = 1.41 (287 eposides) China = 9.10 (3 eposides) USA = 1.77 (6 eposides) Finland = -3.04 (2 eposides) ytisneD 51. 1. 50. 0 01- 5- 0 ASU 5 NIF anihC 51 Next Years (Affected Countries) Mean = 3.98 China = 9.51 USA = 2.72 Finland = 6.09 ytisneD 51. 1. 50. 0 01- 5- 0 NIF 5 anihC 01 51 Onset Years (Unaffected Countries) Mean = 3.72 China = 3.35 Finland = 0.72 NOTE: ThedistributionofrealGDPgrowthrate(%)for(i)normalperiods(includingallcountriesinallnon-diseaseyearsandall Unaffectedcountriesduringtheonsetyearsofdiseaseepisodes),(ii)and(iii)onsetyearsofthediseaseepisodesforallAffectedand Unaffectedcountries,respectively,and(iv)theyearsubsequenttoonsetyearforAffectedcountries.Theyellow,blueandredlinemarks thegrowthrateforFinland,USandChina. holdingat3.7%inunaffectedcountries(upperandlowerrightpanelsofFigure1). Average GDP growth among affected countries bounces back in the following year to just under 4.0% (lower left panel). In order to give a further idea about cross-country outcomes, we depict the location of Finland, the U.S., and China. Of the six crises, these countries were affected in 2, 6, and 3 episodes, respectively. Although the right panels indicate that bounce-backinGDPgrowthisrobustonaverageforaffectedcountries,differentcountries have different experiences. Growth in China continues practically unabated even through crises episodes in which it was affected. Finland and the U.S. are close to each other in non-crisisyears,withmeanGDPgrowthratesof2.7%and3.0%,respectively,butFinland ishitmuchharderbythecrises,with-3.0%averageGDPgrowthcomparedto1.8%forthe US.Finlandalsoenjoyshighergrowthinbounce-backyears,however,at6.1%versus2.7% fortheUS.Thedifferentcrosscountryoutcomessuchasthesearecrucialforidentification. 8

3 Methodology We use two approaches to study the effect of health crises on global macroeconomic outcomes such as GDP growth and unemployment. First is the local projections method of Jorda` (2005),whichweusetoestimateimpacteffectsanddynamicresponsestothehealth crisisshock. Thisapproachisflexible,robust,andverywidelyusedintheliterature.13 Second, we use panel regressions. These facilitate studying robustness of our baseline results to various adjustments, including addressing endogeneity concerns. We use the Driscoll andKraay(1998)correctionforallconfidencebandsandregressionstandarderrors. ImpulseResponseFunctionsWebeginwiththelocalprojectionsmethodofJorda` (2005) toestimateimpulseresponsefunctionsinthefullpanelofcountries. 4 4 y =αH+ ∑βHy +∑δHD +X +ε ,withH =0,1,···,5. (1) it+H i j it−j s it−s it it j=1 s=0 where y is alternatively real GDP growth or unemployment rate for country i in year t, it D is a shock dummy variable indicating a pandemic/epidemic disease hitting country i in it year t and X includes country-level controls for Trade/GDP, Domestic Credit/GDP, popit ulation and log GDP per capita. We include decade dummies and country fixed effects to control for unobserved cross section and cross time heterogeneity. To control for business cycles and financial crises, we also include a US recession dummy (from the NBER) and a systemic banking crisis dummy as in Laeven and Valencia (2013). We display impulse responses to an unexpected shock to D at time t, signifying the onset year of the crisis. it Specifically,weplotthedynamicsof{δH}5 forhorizonsuptofiveyearsaftertheshock, 0 H=0 alongwithonestandarderrorbands. Panel Regressions Our panel OLS regression is similar to the local projection estimation equationin(1)andgivenasfollows y =α +βD +X +ε (2) it i it it it where here we restrict y to be real GDP growth rate for country i in yeart, while D and it it 13With objectives related to ours, Jorda` et al. (2013) study the dynamics effects of financial crises using thetechnique,forexample. 9

X are the same as in equation (1).14 In some specifications, we replace D with measures it it ofcrisisseverity,suchasindividualcountries’mortalityratesorinfectionrates,aswellasa relativeseverity dummyapproach, as explainedin detaillater. Toestimate standard errors, we follow Driscoll and Kraay (1998), who note that traditional panel data techniques that fail to account for cross-sectional dependence will result in inconsistently estimated standard errors. This is especially a problem with relatively large cross sections but small time series samples. We implement their non-parametric covariance matrix estimation techniquewhichtheyshowyieldsstandarderrorestimatesthatarerobusttoverygeneralforms ofcross-sectionalandtemporaldependence. Exogeneity It is important to address concerns about endogeneity in our approach. The first concern is the assumption that the health crisis shock dummy D is exogenous to outit put growth and unemployment. Alternatively, one could conceive that output growth is exogenous, that recessions increase the probability of a health crisis, and that this reverse causality accounts for the associations that we document. Furthermore, it might be that third factors simultaneously affect GDP growth and the probability of a health crisis, including government expenditures on health care, the focus of section 6. Or it may be that (severityof)healthcrisesandgovernmentexpendituresareendogenous. Similar concerns are voiced (and dexterously addressed) by Cerra and Saxena (2008), in the case of financial and political shocks. Health crisis shocks are arguably more exogenous to country-level growth and employment than are financial crisis shocks,15 but nevertheless we investigate the empirical importance of the endogeneity concerns. First, we examine the role of expectations. We test if Consensus forecasts point to expected lower GDP growth simultaneously with the occurrence of a disease outbreak. Although this expectations channel is easier to see working through financial crises (investors foreseeingrecessionusherinacrisis),itisconceivablethatexpectedweakergrowthcouldsew the seeds for health crises via health preparedness channels. We show robustness of our baseline findings to controlling for consensus forecasts of GDP growth. We also test the pre-trendassumptionforourpanelregression,showingthatlaggedshocksareinsignificant forGDPgrowth(seeonlineSupplementTableS.5). Second,wejointlyestimateasystemofseeminglyunrelatedregressionsthattakesinto 14Tosavespace,wereportregressionswithGDPgrowthonly;resultsforunemploymentareconsistent. 15“Thevirusrespectsnoborders,”ChinesePresidentXiJinping,G20Leaders’SummitonCOVID-19,27 Mar2020. “TheCOVID-19outbreakisthecommonenemyoftheworld.” 10

Figure2EffectofHealthCrisesonGDPGrowthandUnemployment GDPGrowth Unemployment tnecreP 1 0 1- 2- 3- 4- 0 1 2 3 4 5 Years tnecreP 1 5. 0 5.- 0 1 2 3 4 5 Years NOTE: Impulse response functions (IRF) are estimated based on the local projection method as in Jorda` (2005): yit+H =αH i + ∑4 j=1 βH j yit−j+∑4 s=0 δH s Dit−s+Xit+εit,withH=0,1,···,5, whereyit istheannualrealGDPgrowthrate(unemploymentrate)for countryiatyeart,Ditisadummyvariableindicatingadiseaseeventhittingcountryiinyeart,withXitincludingcountry-levelcontrols suchasTrade/GDP,DomesticCredit/GDP,populationandlogGDPpercapita.Wealsoincludeadecadedummy,USrecessiondummy, abankingcrisisdummyandcountryfixedeffects. StandarderrorsarecorrectedusingDriscollandKraay(1998). Onestandarderror bandsareshown. account feedback between countries’ health expenditure, the probability (or severity) of a healthcrisisshock,andrealGDPgrowth. g =α1+θ D +µ D +β g +γ HealthExp +X +ε1 (3) it i 1 it 1 it−1 1 it−1 1 it−1 it it HealthExp =α2+θ D +µ D +β g +γ HealthExp +X +ε2 (4) it i 2 it 2 it−1 2 it−1 2 it−1 it it D =α3+µ D +β g +γ HealthExp +X +ε3 (5) it i 3 it−1 3 it−1 3 it−1 it it where g is annual real GDP growth for country i at year t, D is the shock dummy, it it HealthExp is current health expenditures (% GDP), and X includes the same countryit it level controls as in equation (1). All estimates include decade dummies, U.S. recession dummy, systemic banking crises dummy and country fixed effects as in the baseline panel OLS model. In the system of three equations, we allow for health crises to affect both real GDP growth and health expenditure contemporaneously, while assuming that growth and health expenditures affect health crises only with a lag. We alternatively estimate only the systemofequations(3)and(5).16 16Wealsoexaminereplacingtheshockdummyvariablewiththeexpostmortalityrate. 11

4 Effects on GDP and Unemployment 4.1 Impulse Response Functions Figure2displayslocalprojectionsestimatesofrealGDPgrowthandunemploymenttothe identifiedhealthcrisisshock. TheleftpanelrepresentsthepathofGDPgrowthinaffected countries relative to unaffected countries, following the health crisis shock. We display estimates for the crisis onset year and subsequent five years. On average, GDP growth in affected countries is 2.4% below that of unaffected countries in the onset year.17 Furthermore, bounce-back from health crises shocks appears quickly according to our estimates, with affected countries enjoying nearly a one percentage point higher growth rate than unaffected countries in the year following the crisis.18 Resumption in growth in affected countriesisnotsufficienttoovercometheinitialdecline,however,leavingthelevelofGDP persistentlylowerinaffectedcountriescomparedtounaffectedcountries. The right panel of Figure 2 indicates that in the onset year, unemployment is 0.7% higher in affected countries relative to unaffected countries. There is more persistence in unemployment than GDP growth, as unemployment remains 0.5% higher in affected countries in the year after onset. Disruptions to the labor market take longer to overcome thanthosetooutput. InFigure3andFigure4,wedisplayunemploymentimpulseresponses by gender, education level, and sector. The effect of the crisis is felt less strongly on those with a higher education level. Industrial workers (and output) are hit harder than workers in the service and agricultural sectors, as displayed in Figure 4. In addition, although the impact effect on unemployment is felt approximately equally between males and females, there is significantly greater persistence in female unemployment. Hardest hit of all are femaleworkerswithabasiceducation,asseeninthelowerrightpanelofFigure3. Health Crises and Systemic Banking Crises Forperspective,wejointlyestimatetheeffectonGDPgrowthofhealthcrisesandbanking crises, identified by Laeven and Valencia (2013), by augmenting our baseline estimation equation (1) with a dummy for the systemic banking crises and its four lags. As shown in Figure 5, the effects on GDP growth of a health crisis (in blue) are of the same magnitude 17Against this, note that the IMF forecasts -5% world GDP growth for 2020, down sizably from actual growthof+2.9%in2019(WorldEconomicOutlook,June2020). 18TheIMFforecastsahealthyrecoveryof5.4%inworldGDPgrowthin2021. 12

Figure3EffectonUnemployment(%): EducationandGenderBreakdown PanelA:BasicEducation PanelB:IntermediateEducation PanelC:AdvancedEducation tnecreP 5.1 1- 0 1 2 3 4 5 Years tnecreP 5.1 1- 0 1 2 3 4 5 Years tnecreP 5.1 1- 0 1 2 3 4 5 Years PanelD:Male PanelE:Female PanelF:FemalewithBasicEducation tnecreP 5.1 1- 0 1 2 3 4 5 Years tnecreP 5.1 1- 0 1 2 3 4 5 Years tnecreP 5.1 1- 0 1 2 3 4 5 Years NOTE: Impulseresponsefunctions(IRF)areestimatedbasedonthelocalprojectionmethodasinJorda` (2005)yit+H =αH i +∑4 j=1 βH j yit−j+∑4 s=0 δH s Dit−s+Xit+εit,withH=0,1,···,5, whereyitistheannualunemploymentrateforcountryiatyeart,Ditisadummyvariableindicatingadiseaseeventhittingcountryiinyeart,withXitincludingcountry-levelcontrolssuchas Trade/GDP,DomesticCredit/GDP,populationandlogGDPpercapita. Wealsoincludeadecadedummy,USrecessiondummy,abankingcrisisdummyandcountryfixedeffects. Standard errorsarecorrectedusingDriscollandKraay(1998). Onestandarderrorbandsareshown. PanelsA,BandCpresentIRFsofunemploymentforworkerswithbasiceducation,intermediate education,andadvancededucation,respectively.PanelsDandEpresentIRFsofunemploymentformaleandfemaleworkers,respectively.PanelFpresentsunemploymentforfemaleworkers withbasiceducation. 13

Figure4EffectonGDPgrowthandUnemployment(%): SectorBreakdown GDPgrowth PanelA:AgriculturalSector PanelB:IndustrySector PanelC:ServiceSector tnecreP 4 2 0 2- 4- 6- 0 1 2 3 4 5 Years tnecreP 4 2 0 2- 4- 6- 0 1 2 3 4 5 Years tnecreP 4 2 0 2- 4- 6- 0 1 2 3 4 5 Years Unemployment PanelD:AgriculturalSector PanelE:IndustrySector PanelF:ServiceSector tnecreP 7. 3.- 0 1 2 3 4 5 Years tnecreP 7. 3.- 0 1 2 3 4 5 Years tnecreP 7. 3.- 0 1 2 3 4 5 Years NOTE:Impulseresponsefunctions(IRF)areestimatedbasedonthelocalprojectionmethodasinJorda`(2005)yit+H=αH i +∑4 j=1 βH j yit−j+∑4 s=0 δH s Dit−s+Xit+εit,withH=0,1,···,5,where yitistherealGDPgrowthrateorannualunemploymentrateforcountryiatyeart,Ditisadummyvariableindicatingadiseaseeventhittingcountryiinyeart,withXitincludingcountry-level controlssuchasTrade/GDP,DomesticCredit/GDP,populationandlogGDPpercapita. Wealsoincludeadecadedummy,USrecessiondummy,abankingcrisisdummyandcountryfixed effects. StandarderrorsarecorrectedusingDriscollandKraay(1998). Onestandarderrorbandsareshown. PanelA(D),B(E)andC(F)presentIRFsforrealGDPgrowth(unemployment) rateatagricultural,industryandservicesectors. 14

Figure5EffectsofHealthCrisesandBankingCrisesonGDPGrowth tnecreP 1 0 1- 2- 3- Health Crises Banking Crises 0 1 2 3 4 5 Years NOTE: Impulse response functions (IRF) are estimated based on the local projection method as in Jorda` (2005) git+H =αH i + ∑4 j=1 βH j git−j+∑4 s=0 δH s DH it− ea s lthCrises+∑4 s=0 γH s DB it− an s kingCrises+Xit+εit,withH=0,1,···,5,wheregit istheannualrealGDPgrowth (cid:16) (cid:17) rateforcountryiatyeart,DHealthCrises DBankingCrises isadummyvariableindicatingadiseaseevent(bankingcrisis)hittingcountry it it iinyeart,withXitincludingcountry-levelcontrolssuchasTrade/GDP,DomesticCredit/GDP,populationandlogGDPpercapita.We alsoincludeadecadedummy,USrecessiondummyandcountryfixedeffects. StandarderrorsarecorrectedusingDriscollandKraay (1998).Thebluelinerepresents (cid:8) δH(cid:9)5 andtheredlinerepresents (cid:8) γH(cid:9)5 .Onestandarderrorbandsareshown. 0 H=0 0 H=0 as banking crises (in red), although the dynamics are different. In the onset year, there is a fall in real GDP, by 2.2% for health crisis and 1.3% for banking crises. However, one year later, GDP growth bounces back after a health crisis to 0.7% but continues to fall after a bankingcrisis. Althoughthemagnitudefromahealthcrisisintheonsetyeariscomparable tothatofabankingcrisis,itfeaturesfasterbounce-backofgrowththanbankingcrises. 4.2 Panel Regressions Inordertoexaminevariousadjustmentstothebaselineresultsdisplayedabove,weestimate panel regressions for real GDP growth in Table 1. Column (1) displays results for the full sample period 1960-2018, while the remaining columns are for 1990-2018 due to our use of consensus forecasts, which are available for 32 countries beginning in 1990. Specifications with the forecasts control for expectations, which essentially entirely account for the effects of the economic control measures. Table 1 includes all pandemic/epidemic events in the shock dummy while Table 2 utilizes separate shock dummies for each episode. The coefficientsinTable1ontheshockdummyrangefrom-1.4%to-3.4%,statisticallysignificant and economically large. In Table 2, with separate crisis event shock dummies, H1N1 15

hasthelargesteffect,consistentwithH1N1havingthelargestnumberofdeathsandcases. Butstill,theeffectoftheotherdiseaseepisodesisnotnegligible. We devote special attention to the H1N1 crisis, given its simultaneous occurrence with the 2009 Global Financial Crisis, with three elements of the estimation. First, we examine robustness to excluding the episode. Second, we include in our impulse response function estimation equation and panel regressions a recession dummy for the U.S. economy and a systemicbankingcrisisdummy. Thosedummyvariablesshouldabsorbthecontemporaneous effect from the global financial crisis on GDP and unemployment. Third, we examine robustnesstoweightingourshockdummybymeasuresoftheseverityofeachhealthcrisis, asinTableS.6. Eventhoughtheglobalfinancialcrisisaffectedmostcountriesin2009,the crosscountryheterogeneityinH1N1exposureisarguablyexogenoustothefinancialcrisis. As seen in the table, the coefficients on our severity proxies are significantly negative for GDPgrowth: moreseverehealthcrisesportendgreatereconomicdamage. Note two caveats about our severity estimation. First, there might be non-negligible measurement error for individual country reports of deaths and infection cases.19 For example,thereportingdiscrepancy(bothcasesanddeaths)betweentheCDCandWHOcould be systematically biased and incomplete. This consideration does not affect identification of the shock itself, but might contaminate interpretation of the severity panel regression estimates. Second, weighting the shock dummy by the individual country cases or deaths measure (however mis-estimated) assumes that, e.g., a 2% death rate in Ebola creates the sameeconomicimpactasa2%deathrateinH1N1. Itismorereasonabletocomparedeath ratesandthus(cross-sectional)severitywithinthesamehealthcrisis. Tothisend,andtobeconsistentwiththeonlyforminwhichseveritydataareavailable for the 1968 Flu (“isolated”, “regional”, and “widespread”), we form three dummy variables that capture the relative severity for affected countries in each episode.20 We label affected countries as high, medium or low severity, using their ex-post mortality or case rate for each episode.21 With this, our severity analysis groups countries into four categories: unaffected countries, low affected countries, medium affected countries and high affected countries (see Online Supplement Table S.7 for country-episode category assign- 19Inourmatched287country-yearsampleforthehealthcrisesdummy,wehaveinformationoncasesfor 265ofthemandondeathsfor259ofthem. Wedonothaveexactcasesanddeathsforthe1968Flu. 20We still use the individual country’s data for either mortality or case rates to form our new dummy variables. Althoughtheremightbemeasurementerrorforanindividualcountry’sdata,therelativemeasure weconstructshouldcontainlessofit. 21Thethresholdispercentiles30and70. Theresultsremainunchangedifweusethe1/3and2/3cutoff. 16

Table1TheEffectofHealthCrisesonGDPGrowth GDPgrowthrate% (1) (2) (3) (4) SamplePeriod: 1960-2018 1990-2018 AllEvents AllEvents AllEvents WithoutH1N1 Shock -2.60** -2.60** -3.44*** -1.40*** (1.18) (1.21) (0.96) (0.28) ConsensusForecast 0.52*** 0.64*** (0.12) (0.11) Trade/GDP 0.15 0.42 0.51 0.34 (0.20) (0.34) (0.43) (0.38) DomesticCredit/GDP -0.67* -0.71 -1.42 -0.99 (0.38) (0.45) (0.99) (0.84) Log(Population) 0.17*** 0.12* 0.08 0.07 (0.04) (0.06) (0.06) (0.05) Log(GDPpercapita) -0.31*** -0.20 -0.04 -0.03 (0.07) (0.12) (0.14) (0.13) Recession -0.35 -0.52 -0.33 0.16 (0.21) (0.37) (0.43) (0.25) BankingCrisis -1.34*** -1.43*** 0.40 -0.10 (0.32) (0.37) (0.67) (0.56) Constant 4.63*** 4.47*** 2.11*** 1.70*** (0.42) (0.47) (0.46) (0.40) Observations 6536 4303 531 502 WithinR2 0.056 0.064 0.229 0.179 DecadeFE Yes Yes Yes Yes CountryFE Yes Yes Yes Yes NOTE: ThedependentvariableisrealannualGDPgrowth. Thesampleperiodforcolumn(1)is1960-2018whilethesampleperiod forcolumns(2)-(4)is1990-2018. Theshockdummyequalsoneforcountryihitbyahealthcrisisinonsetyeart,andzerootherwise. Incolumns(1)-(3),weincludesixhealthcriseswhilecolumn(4)excludesH1N1. Inallspecifications,weincludebothcountryand decadefixedeffects.AllstandarderrorsarecorrectedusingDriscollandKraay(1998)andreportedinparentheses.∗,∗∗and∗∗∗indicate statisticalsignificanceatthe10%,5%,and1%level,respectively. ments). We expect that all affected country severity dummy variables in the GDP growth regressions will be negative and have an average magnitude that is approximately equal to thecoefficientontheshockdummyinTable1. Furthermore,weexpectthatthecoefficient onhigherseveritydummiesshouldbelargerthanforlowerseveritydummies. Table3reportsourpanelregressionwiththeseveritydummyvariables. Thecoefficients onalldummiesarenegative,consistentwithourmainregressioninTable1. Theeconomic magnitude is much larger for high and medium severity countries than for low severity countries. The coefficients are highly significant and vary between -3.1% and -4.8% for the high and medium severity dummies, while they vary from -0.9% to -1.8%, sometimes insignificantly, for the low severity dummies. Interestingly, the high and medium severity 17

Table2TheEffectofHealthCrisesonGDPGrowth,byCrisis GDPgrowthrate% (1) (2) (3) (4) SamplePeriod: 1960-2018 1990-2018 AllEvents AllEvents AllEvents WithoutH1N1 EBOLA 0.83*** 0.57* -0.29 -0.34 (0.28) (0.29) (0.26) (0.28) H1N1 -4.31*** -4.27*** -5.14*** (0.55) (0.61) (0.38) MERS -1.11*** -0.88** -1.51*** -1.40*** (0.33) (0.33) (0.31) (0.30) SARS 0.08 0.05 -1.02*** -1.06*** (0.49) (0.45) (0.24) (0.24) Zika -0.51* -0.47 -2.15*** -2.15*** (0.27) (0.30) (0.28) (0.28) Hkflu 0.26 (0.27) ConsensusForecast 0.54*** 0.63*** (0.12) (0.11) Trade/GDP 0.13 0.40 0.41 0.34 (0.20) (0.33) (0.39) (0.38) DomesticCredit/GDP -0.62* -0.66 -1.24 -0.99 (0.36) (0.43) (0.91) (0.85) Log(Population) 0.18*** 0.12** 0.09 0.07 (0.03) (0.06) (0.05) (0.05) Log(GDPpercapita) -0.31*** -0.20 -0.06 -0.03 (0.07) (0.12) (0.14) (0.13) Recession -0.17 -0.20 0.13 0.17 (0.19) (0.31) (0.27) (0.24) BankingCrisis -1.36*** -1.49*** 0.02 -0.10 (0.32) (0.38) (0.55) (0.56) Constant 4.64*** 4.36*** 2.02*** 1.71*** (0.44) (0.46) (0.41) (0.40) Observations 6536 4303 531 502 WithinR2 0.067 0.081 0.264 0.181 DecadeFE Yes Yes Yes Yes CountryFE Yes Yes Yes Yes NOTE:ThedependentvariableisrealannualGDPgrowth.Thesampleperiodforcolumn(1)is1960-2018whilethesampleperiodfor columns(2)-(4)is1990-2018.Countryanddecadefixedeffectsareincluded.AllstandarderrorsarecorrectedusingDriscollandKraay (1998)andreportedinparentheses.∗,∗∗and∗∗∗indicatestatisticalsignificanceatthe10%,5%,and1%level,respectively. 18

Table3TheEffectofHealthCrisesonRealGDPGrowth,bySeverity GDPgrowthrate% (1) (2) (3) (4) (5) (6) SamplePeriod: 1960-2018 1990-2018 1960-2018 1990-2018 HighMortalityRate -3.56*** -3.72*** -4.04*** (1.04) (1.06) (0.87) MediumMortalityRate -3.68*** -3.59*** -4.48*** (0.96) (1.06) (0.60) LowMortalityRate -0.90 -0.90 -1.41** (0.82) (0.79) (0.58) HighCases/Pop -3.01** -3.07** -4.79*** (1.31) (1.42) (1.19) MediumCases/Pop -3.18** -3.14** -3.84*** (1.41) (1.41) (0.63) LowCases/Pop -1.32* -1.31* -1.84* (0.74) (0.73) (0.94) ConsensusForecast 0.51*** 0.52*** (0.12) (0.12) Trade/GDP 0.16 0.45 0.49 0.15 0.44 0.48 (0.20) (0.35) (0.45) (0.20) (0.34) (0.42) DomesticCredit/GDP -0.66* -0.70 -1.31 -0.67* -0.71 -1.28 (0.37) (0.45) (0.97) (0.37) (0.45) (0.95) Log(Population) 0.18*** 0.12** 0.08 0.17*** 0.12* 0.07 (0.03) (0.06) (0.06) (0.03) (0.06) (0.05) Log(GDPpercapita) -0.32*** -0.21* -0.06 -0.31*** -0.19 -0.04 (0.07) (0.12) (0.14) (0.07) (0.12) (0.14) Recession -0.33 -0.48 -0.23 -0.36* -0.55 -0.39 (0.20) (0.35) (0.39) (0.21) (0.37) (0.44) BankingCrisis -1.33*** -1.43*** 0.28 -1.34*** -1.44*** 0.42 (0.32) (0.38) (0.65) (0.32) (0.37) (0.65) Constant 4.61*** 4.45*** 2.16*** 4.61*** 4.46*** 2.21*** (0.42) (0.46) (0.45) (0.42) (0.47) (0.46) Observations 6536 4303 531 6536 4303 531 WithinR2 0.059 0.069 0.237 0.057 0.066 0.231 DecadeFE Yes Yes Yes Yes Yes Yes CountryFE Yes Yes Yes Yes Yes Yes NOTE: ThedependentvariableisrealannualGDPgrowth. Thesampleperiodforcolumns(1)and(4)is1960-2018whilethesample periodforcolumns(2)-(3)and(5)-(6)is1990-2018. Countryanddecadefixedeffectsareincluded. Allstandarderrorsarecorrected usingDriscollandKraay(1998)andreportedandreportedinparentheses.∗,∗∗and∗∗∗indicatestatisticalsignificanceatthe10%,5%, and1%level,respectively. 19

dummies, both large and highly statistically significantly negative, are not significantly differentfromeachother. Thisindicatesthattherelationshipbetweenhealthcrisisseverity andeconomiclossisnon-monotonic: atsomepointalongtheseverityspectrum,additional severity doesn’t bring any more economic losses. For comparison, we also estimate local projection impulse response functions for real GDP growth using these three new dummy variablesanddisplaytheminFigureS.1oftheOnlineSupplement. Table4PlaceboTest GDPgrowthrate% (1) (2) (3) (4) SamplePeriod: 1960-2018 1990-2018 AllEvents AllEvents AllEvents WithoutH1N1 Shock 0.18 0.01 -0.05 -0.05 (0.25) (0.28) (0.51) (0.51) ConsensusForecast 0.57*** 0.66*** (0.15) (0.11) Trade/GDP 0.19 0.53 0.86 0.37 (0.22) (0.39) (0.69) (0.39) DomesticCredit/GDP -0.72 -0.75 -1.73 -0.98 (0.44) (0.51) (1.15) (0.85) Log(Population) 0.17*** 0.12* 0.05 0.07 (0.03) (0.06) (0.06) (0.05) Log(GDPpercapita) -0.33*** -0.23* -0.04 -0.04 (0.08) (0.13) (0.15) (0.13) Recession -0.55* -0.91* -0.89 0.23 (0.30) (0.49) (0.74) (0.22) BankingCrisis -1.25*** -1.27** 1.33 -0.07 (0.39) (0.49) (1.06) (0.55) Constant 4.62*** 4.61*** 1.91*** 1.63*** (0.46) (0.51) (0.51) (0.39) Observations 6536 4303 531 502 WithinR2 0.040 0.036 0.118 0.169 DecadeFE Yes Yes Yes Yes CountryFE Yes Yes Yes Yes NOTE: The dependent variable in columns (1)-(4) is real annual GDP growth rate. The sample period for column (1) is 1960-2018 while the sample period for columns (2)-(4) is 1990-2018. The shock variable is randomlygenerated. Countryanddecadefixedeffectsareincluded. Allstandarderrorsarecorrectedusing Driscoll and Kraay (1998) and reported in parentheses. ∗, ∗∗ and ∗∗∗ indicate statistical significance at the 10%,5%,and1%level,respectively. Finally,asarobustnesscheckontheidentificationofdiseaseepisodeevent-years,wedo aplacebotestbyrandomlypickingacountry-yearobservationasourshockdummyandreestimatingthepanelregression. TheresultsareinTable4. Thecoefficientonthisrandomly 20

constructed variable is statistically insignificant, suggesting that our shock dummy indeed capturestheeffectofhealthcrisesonrealGDPgrowth. Feedback among Growth, Health Crises, and Health Expenditures AsdiscussedinSection3,ourbaselineestimationassumesthatthehealthcrisisshockisexogenous to contemporaneous GDP growth. Although this is arguably reasonable, one may wonder whether lower past economic growth reduces health-related expenditures, making the country more vulnerable to a health crisis. Here we allow GDP growth, health expenditures, and the health crisis to be jointly determined in a system of equations (3), (4) and (5). We estimate this using seemingly unrelated regressions (SUR), modeling the determination of the shock dummy linearly, and report results in Table 5. Our key messages from the baseline regression are robust: GDP falls by 2.4 % in the onset year, according to the SUR estimates, and bounces back by 1.2% in the following year. Moreover, higher past growth does lower the probability of a health crisis. Somewhat anomalously, higher past healthexpendituresincreasetheprobabilityofahealthcrisis.22 4.3 Geographic, Sector, and Episode Breakdowns We explore heterogeneity in the effects of health crises along multiple additional dimensions.23 Panel A in Figure 6 displays estimates for the H1N1 crisis alone. Consistent with the results above, the effect of H1N1 is larger than our full sample estimates. In the onset year,thegrowthrateforaffectedcountriesis4.2%lowerthanforunaffectedones. Thereis stillbounce-backoneyearlater—thegrowthrateforaffectedcountriesis1%higherthan thatforunaffectedones. PanelBinFigure6considersHigh-incomecountries(inblue)and Low-income countries (in red), as classified by the World Bank.24 High income countries affected by the crisis have a GDP growth rate in the onset year that is 2.4% less than the GDP growth for high income countries unaffected by the crises. Bounce-back for these affected high-income countries is quick, however, as seen by the fact that growth is 0.8% 22The sample size is reduced to around 2,500 because we add health expenditures, which is unavailable forsomecountries. 23Tosavespace,wedisplayimpulseresponsefunctionsonlyforrealGDPgrowth. Thoseforunemployment,whichareavailableuponrequest,areconsistentwiththeGDPgrowthinthesenseofOkun’slaw. 24The WorldBank groups countries intofour categories based on2018 GNI per capita— High-income, Upper-middle-income, Lower-middle-income and Lower-income economies. We estimate the impulse responsefunctionsforHigh-incomeandLower-incomecountrygroupsseparately. 21

Table5SeeminglyUnrelatedRegressions: Growth,HealthCrises,andHealthExpenditure System1 Shock Shock(t−1) GDPGrowth(t−1) HealthExp(t−1) Obs R2 GDPgrowth -2.38*** 1.18*** 0.21*** 0.29*** 2,523 0.42 (0.21) (0.22) (0.02) (0.07) HealthExp 0.24*** -0.06* -0.00 0.77*** 2,523 0.96 (0.04) (0.04) (0.00) (0.01) Shock -0.10*** -0.01*** 0.02** 2,523 0.16 (0.02) (0.00) (0.01) System2 GDPgrowth -2.36*** 1.04*** 0.24*** 0.18*** 2,676 0.40 (0.21) (0.21) (0.02) (0.07) Shock -0.09*** -0.01*** 0.01** 2,676 0.15 (0.02) (0.00) (0.01) NOTE: System1reportsestimatesfromthejointestimationofsystemofequations(3),(4)and(5). System2 reportsestimatesfromthejointestimationofsystemofequations(3)and(5). ∗,∗∗and∗∗∗indicatestatistical significanceatthe10%,5%,and1%level,respectively. higher in affected countries in the year after the crisis was declared. According to the red lineinthefigure,affectedlow-incomecountrieshaveGDPgrowthratesthatarenotsignificantly different from unaffected low-income countries. Note that these are within-group comparisons, and hence do not speak to the issue of whether high income or low income countriesaremoreaffectedbyhealthcrises.25 Panel C and Panel D show the effects on advanced and emerging market economies according to the IMF classification. In the onset year, the growth rate among advanced economiesfallsby2.7%inaffectedcomparedtounaffectedcountries. Oneyearlater,there isabouncebackto1.1%fortheadvancedcountrygroup. Foremergingmarketeconomies, the growth rate falls by 2.1% for affected countries compared to unaffected ones, with a bounce back at 0.5% one year after the shock. One potential reason for a larger effect of healthcrisesonadvancedcountrygroupsisduetotheeconomicstructure. Asnotedabove, in Figure 4, we divide GDP into three sectors and find that industry and service sectors areaffectedmorebyhealthcrises,whileagriculturaloutputisnotsignificantlydifferentin affectedandunaffectedcountries. Panel E and Panel F consider geographic regions. The decline in growth for affected 25The IMF growth forecasts for Low Income Developing countries is -1% in 2020, down from 5.2% in 2019. Thiscomparestoaforecastof-8.1%in2020forAdvancedEconomies. TheIMFprojectsarebound to5.2%forthelowincomecountriesin2021. 22

Figure6EffectonGDP:EpisodeandGeographicBreakdowns PanelA:H1N1 PanelB:Highvs.LowIncomeCountry tnecreP 2 1 0 1- 2- 3- 4- 5- 0 1 2 3 4 5 Years tnecreP 2 1 0 1- 2- 3- 4- 5- HIC LIC 0 1 2 3 4 5 Years PanelC:AdvancedEconomies PanelD:EmergingMarketEconomies tnecreP 2 1 0 1- 2- 3- 4- 0 1 2 3 4 5 Years tnecreP 2 1 0 1- 2- 3- 4- 0 1 2 3 4 5 Years PanelE:EastAsiaandSouthAsia PanelF:EuropeandCentralAsia tnecreP 2 1 0 1- 2- 3- 4- 5- 6- 0 1 2 3 4 5 Years tnecreP 2 1 0 1- 2- 3- 4- 5- 6- 0 1 2 3 4 5 Years NOTE: Impulse response functions (IRF) are estimated based on the local projection method as in Jorda` (2005) git+H =αH i + ∑4 j=1 βH j git−j+∑4 s=0 δH s Dit−s+Xit+εit,withH=0,1,···,5,wheregit istheannualrealGDPgrowthrateforcountryiatyeart,Dit isadummyvariableindicatingadiseaseeventhittingcountryiinyeart,withXit includingcountry-levelcontrolssuchasTrade/GDP, DomesticCredit/GDP,populationandlogGDPpercapita. Wealsoincludeadecadedummy,USrecessiondummy,abankingcrisis dummyandcountryfixedeffects.StandarderrorsarecorrectedusingDriscollandKraay(1998).Onestandarderrorbandsareshown. PanelAre-definesthedummyDit toflagtheH1N1shockonly. PanelBpresentsIRFsforthesampleof“HighIncomeCountry”and “LowIncomeCountry”accordingtoWorldBankClassification. PanelC(D)presentsIRFsforthesampleofadvancedeconomies (emergingmarketeconomies).PanelE(F)isforEastAsiaandSouthAsia(EuropeandCentralAsia). 23

East and South Asia countries relative to the unaffected ones is 1.1% in the onset year, with a 1.8% bounce-back one year later. For the Europe and Central Asia group, affected countries have a 4.3% decrease in GDP growth compared to unaffected countries in the onset year, with a 0.9% bounce-back one year later. One potential explanation may be due totheroleoffiscalpolicy,whichisexploredbelow. 5 International Trade and Cross-Country Propagation Inadditiontotheeffectontheproductionside,wealsoinvestigatetheeffectofpasthealth crises on different components of total spending. In the Online Supplemental Appendix Section S.3, we show the negative and significant effects on consumption and investment. Here we focus on international trade (exports plus imports). Declines in spending may explain why the effect of health crises on output is very persistent. Furthermore, the drop inspendingcouldalsospillovertoothercountries,includingunaffectedcountries,through an international trade channel. To this end, we decompose the effect from health crisis shocksintoadirectchannelandanindirectchannelthroughaffectedtradingpartners. Being involved in a global value chain through trade could be a mixed blessing for a country during a pandemic. On the one hand, the negative impact of health crises on the trading partner can spillover to the domestic economy through a trade channel, making health crises economically more contagious. On the other hand, the bounce-back effect from a health crisis for the affected trading partner can also benefit the domestic country. Moreover, being more integrated into global value chains can help firms diversify risks whenthecountryitselfishitbythehealthcrisis(seeHuang(2017)). Toestimatetheeffect of such a channel, we decompose the impact of health crises into a direct channel and an indirectchannelthatcapturestheeffectofthecrisisontradingpartners. Inthissection,weestimatethetradenetworkeffectandcompareitsimportanceacross episodes. To understand this, consider the “trade network heat maps” of Figure A.2. This depicts the severity of a health crisis episode for each country by using infection cases from each of that country’s trading partners and weighting case numbers by their bilateral trade share with the domestic country. In other words, for each country the map depicts: how much do we trade with other countries and how badly were those trading partners affected?26 AsseeninFigureA.2,thistradelinkagechannelvariesfromepisodetoepisode 26Recallthatthetradedataisavailableonlyfrom1988-2018,hencenoheatmapforthe1968Flu. 24

Figure7HealthCrisesandInternationalTrade PanelA:EffectonTradegrowth(exports+imports) PanelB:EffectonGDPgrowth:thetradechannel tnecreP 01 0 01- 02- 03- 0 1 2 3 4 5 Years tnecreP 1 0 1- 2- 3- Direct Indirect 0 1 2 3 4 5 Years NOTE: Impulse response functions (IRF) are estimated based on the local projection method as in Jorda` (2005): git+H =αH i + ∑4 s=1 βH s git−s+∑4 s=0 δH s Dit−s+Xit+εit,withH=0,1,···,5, wheregit istheannualrealgrowthrateoftotaltrade(export+import) inPanelAandisGDPgrowthinPanelBforcountryiatyeart,Dit isadummyvariableindicatingahealthcrisishittingcountryi inyeart,withXit includingcountry-levelcontrolssuchasTrade/GDP,DomesticCredit/GDP,populationandlogGDPpercapita. We alsoincludeadecadedummy,U.S.recessiondummy,abankingcrisisdummyandcountryfixedeffects. InpanelB,wealsoincludea controlvariableDijt intheregression,whereDijt=1ifcountryi’stradingpartnercountryjhasbeenhitbythehealthcrisisatyeart. Thebluelineisthedirecteffect(coefficientonDit)whilethereddashedlineistheindirecteffect(coefficientonDijt).Standarderrors arecorrectedusingDriscollandKraay(1998).Onestandarderrorbandsareshown. and varies across countries during any given episode. Clearly, the trade network effect is potentiallymuchmoresevereduringCOVID-19thantheotherepisodes. With our health crises–trade network proxies in hand, we start by estimating the effect ofhealthcrisesonthegrowthrateofinternationaltrade,thesumofacountry’smultilateral exportsplusimports. Crisescanlowertradethroughbothanextensivemarginandintensive margin,asnotedbyFernandesandTang(2020)wholookattheeffectofSARSonChinese trade. Potential lock-downs and travel bans could amplify the negative impact. In Panel A ofFigure7,wedisplayourresults,derivedfromthecustomarylocalprojectionsestimator. Internationaltradeofaffectedcountriesplummetsintheonsetyear,byaround19.0%. This is on par with the U.S. trade collapse in 2008-09 (see Levchenko et al. (2010) and Novy and Taylor (2014)). Affected country trade rebounds quickly, growing relative to the trade ofunaffectedcountriesby7.2%oneyearlater. To capture the propagation effects to other countries through trade networks, we begin by separately estimating the direct effect of the health crisis, captured by our shock dummy, and the indirect effect, captured by an indicator function that flags whether the tradingpartnerisaffectedbythehealthcrisis. Toimplementthis,weaugmentourbaseline estimation equation (1) with a dummy variable that indicates whether any of one’s trading 25

Table6TheEffectofHealthCrisesonGDPGrowth: TradeLinkages GDPgrowthrate% (1) (2) (3) (4) (5) (6) SamplePeriod: 1988-2018 Shock -2.22** -1.98** (1.03) (0.97) HighMortalityRate -3.28*** -3.02*** (0.86) (0.83) MediumMortalityRate -3.13*** -2.87*** (0.88) (0.86) LowMortalityRate -0.55 -0.40 (0.61) (0.56) HighCases/Pop -2.62** -2.36** (1.21) (1.15) MediumCases/Pop -2.71** -2.45** (1.20) (1.11) LowCases/Pop -0.92 -0.71 (0.55) (0.49) ShocktoTradePartner -0.52** -0.55* -0.56** (0.23) (0.27) (0.26) TradeWeightedbyIndirectShock -1.00** -0.99** -1.07** (0.38) (0.48) (0.44) Trade/GDP 0.19 0.17 0.21 0.19 0.20 0.18 (0.33) (0.33) (0.34) (0.34) (0.34) (0.33) DomesticCredit/GDP -0.73 -0.73 -0.72 -0.72 -0.73 -0.73 (0.46) (0.46) (0.45) (0.45) (0.45) (0.45) Log(Population) 0.12** 0.11** 0.12** 0.12** 0.11** 0.11** (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) Log(GDPpercapita) -0.20** -0.21** -0.20** -0.22** -0.19** -0.21** (0.09) (0.09) (0.09) (0.09) (0.09) (0.09) Recession -0.56 -0.57 -0.52 -0.52 -0.58 -0.59 (0.38) (0.38) (0.36) (0.36) (0.38) (0.38) BankingCrisis -1.54*** -1.54*** -1.54*** -1.54*** -1.55*** -1.55*** (0.37) (0.36) (0.37) (0.37) (0.36) (0.36) Constant 4.76*** 4.99*** 4.75*** 4.97*** 4.76*** 5.01*** (0.46) (0.51) (0.45) (0.52) (0.45) (0.51) Observations 4502 4502 4502 4502 4502 4502 WithinR2 0.065 0.066 0.070 0.070 0.066 0.067 DecadeFE Yes Yes Yes Yes Yes Yes CountryFE Yes Yes Yes Yes Yes Yes NOTE: The dependent variable is annual real GDP growth. Shock dummy equals one for country i in the onset yeart, and zero otherwise. Shock to trade partner equals 1 if one of the country’s trading partners is hitbyahealthcrisis,and0otherwise. Theweighttradenetworkincolumns(2),(4),and(6)isconstructed bymultiplyingtheshocktoacountry’stradingpartnerdummybytheshareofbilateraltradebetweenthese twocountriesinthecountry’stotaltrade(Tradeweightedbyindirectshock). Standarderrorsarecorrected usingDriscollandKraay(1998)andreportedinparentheses. ∗,∗∗ and∗∗∗ indicatestatisticalsignificanceat the10%,5%,and1%level,respectively. 26

partners has been hit by the health crisis in the same year. As seen in Panel B of Figure 7, indirect effects are not trivial, contributing approximately -0.5% to GDP growth in the onset year (versus direct effects of -1.8%) and +0.4% in the bounce-back year, or about halfthemagnitudeoftherecovery’sdirecteffect.27 We also use panel regressions to test the importance of trade linkages, as in Table 6.28 In column (1), we have a dummy capturing whether the trading partner was affected, as in the IRF estimation. In column (2), we add a continuous variable, labelled trade weighted byindirectshock,whichmultipliestheshockdummy(toacountry’stradingpartner)bythe bilateraltradebetweenthesetwocountries,asashareofthecountry’stotaltrade. Columns (3) and (4) use the ex-post high, medium and low mortality rate dummies, while columns (5) and (6) use the equivalent case rate dummies, and so is akin to column (1) and column (2). The estimates indicate that the indirect effect of health crises through trade linkages is largeandsignificant. Accordingtocolumn(1),theimpactthroughtradeisaroundonethird ofthedirecteffect. Whentakingintoaccounttheimportance(weights)ofdifferenttrading partners, the effect becomes larger, especially for countries with high severity. The effects ofhealthcrisesondomesticGDPgrowtharesignificantlymagnifiedbytradelinkages. 6 Fiscal Policy In response to COVID-19, finance ministries have undertaken a variety of spending and tax-related policies designed to support households and businesses, and soften the impact on economic activity. According to the standard Keynesian logic, fiscal stimulus in a time ofcrisis,eitherbyincreasinggovernmentspendingorcuttingtaxes,canspeedupeconomic recovery (see Gourinchas (2020)). More generally, fiscal policy has been proposed as an effective way to address crises, such as during the zero-lower bound period and in times of secular stagnation (see Eggertsson (2011), Eggertsson and Krugman (2012), Eggertsson et al. (2016), Benigno and Fornaro (2018), Fata´s and Summers (2018), Fornaro and Wolf (2020)). Furthermore, Dupraz et al. (2019) find a permanent effect from stabilizationpolicyindampeningeconomicfluctuationsandraisingtheaveragelevelofactivity. A 27Our estimation of the indirect trade channel is very similar to the work by Bonadio et al. (2020), who findthatonethirdoftheaveragerealGDPdownturnduetotheCOVID-19shockisthroughglobalsupply chains,usinganestimatedstructuralmodel. 28Theseusetradingpartner’sshockdummiestomeasuretheindirecttradechannel. TableS.8oftheonline supplement shows robustness to using individual countries mortality or case rates to construct the indirect trademeasure. 27

Figure8EffectonGDPGrowthandUnemployment ConditionalonImmediateHealthSpendingResponse GDPgrowth PanelA:HighHealthExpenditureResponse PanelB:LowHealthExpenditureResponse tnecreP 2 1 0 1- 2- 3- 4- 0 1 2 3 4 5 Years tnecreP 2 1 0 1- 2- 3- 4- 0 1 2 3 4 5 Years Unemployment PanelC:HighHealthExpenditureResponse PanelD:LowHealthExpenditureResponse tnecreP 1 0 1- 0 1 2 3 4 5 Years tnecreP 1 0 1- 0 1 2 3 4 5 Years NOTE: Impulse response functions (IRF) are estimated based on the local projection method as in Jorda` (2005): yit+H =αH i + ∑4 s=1 βH s yit−s+∑4 s=0 δH s Dit−s+Xit+εit,withH=0,1,···,5,whereyit istheannualrealGDPgrowthrateorunemploymentratefor countryiatyeart,Ditisadummyvariableindicatingadiseaseeventhittingcountryiinyeart,withXitincludingcountry-levelcontrols suchasTrade/GDP,DomesticCredit/GDP,populationandlogGDPpercapita.Wealsoincludeadecadedummy,U.S.recessiondummy, abankingcrisisdummyandcountryfixedeffects. StandarderrorsarecorrectedusingDriscollandKraay(1998). Onestandarderror bandsareshown.Eachrowdividescountriesbasedontheaverageof Zit−Zit−1 acrossallsixhealthepisodeswheretistheonsetyearof GDPit−1 eachepisode.Zreferstohealthexpenditure.Highreferstocountriesinthe75percentileandabovewhilelowreferstocountriesinthe 25percentileandbelow. well-designedfiscalpolicyshouldreducethepersistentnegativeeffectfromhealthcrises. In this section, we analyze the effects of fiscal policy during past health crises. Our key indicator is a measure of countries’ fiscal adjustment in the onset year, the change in government spending or revenues, divided by the previous year’s GDP. We focus on government health care spending, defined by the World Bank as “including healthcare goods andservicesconsumedbutnotincludingcapitalhealthexpendituressuchasbuildings,machinery, IT and stocks of vaccines for emergency or outbreaks”.29 As Chang et al. (2019) 29We also conducted the exercise on the basis of total government expenditures, in addition to health 28

note, government spending on health care is an important input for health policy globally. To study the effectiveness of such spending, we separate countries into high adjustment countries, defined as the 75th percentile and above, and low adjustment countries, defined asthe25thpercentileandbelow.30 Wethenre-estimatethemodelontheseparategroups. Figure 8 shows the impulse response functions for real GDP growth and unemployment for high and low adjustment countries. Both groups experience equally large impact declines in GDP growth. However, high expenditure countries clearly bounce back more robustly (Panel A) than low adjustment countries (Panel B). Those differential effects also appear in unemployment. As seen in Panel C, the effect on unemployment in high health expenditure adjustment countries is relatively small on impact, less than 1%, and not persistent. In contrast, Panel D indicates that unemployment in low-adjustment countries is persistentlyelevatedaftertheshock. The results above could be spurious if, for example, high adjustment countries also happen to be low severity countries, in terms of cases or deaths. To investigate this, we calculate the correlation between a country’s severity measure and its health spending adjustment, by episode. We report these results in the supplemental appendix Panel B of Table S.7 and scatter plot of Figure S.4. The underlying data are displayed in Panel A of TableS.7. Wefindaslightnegativecorrelation,insignificantlydifferentfromzero. What about government debt sustainability? Surely, debt to GDP will rise during a health crisis, as GDP falls and fiscal policy expands. But we have found that by spending more (perhaps through a higher debt), the economy can bounce-back more quickly than it otherwise would. A faster recovery is thus likely to enhance rather than weaken debt sustainability in the medium run. This argument is further strengthened in a low interest rate environment. To examine the past responses of fiscal variables to health crisis shocks, we generate impulse response functions for central government debt, the government budget surplus, government spending, and government revenue in Figure S.3.31 Following the shock, government revenue falls and spending increases, resulting in a negative fiscal surplus and increase in debt. However, the negativefiscal surplus converges to an insignificantlypositiveleveltwoyearsaftertheshock,whilethedebtslowlyadjuststozero. carespending, andfindsimilarresults. Resultsforthesameexercisebasedonhighversuslowtaxrevenue collectioncountriesdonotindicatesignificantdifferences. SeeFigureS.2intheOnlineSupplement. 30Thegroupingisbasedontheaveragefiscaladjustmentmeasureacrosssixepisodes. Thisincludesboth affectedcountriesandunaffectedcountries. 31Duetodataavailability,oursamplesizeforthisexperimentiscutsignificantly,to1,277observations. 29

7 Conclusion We study the economic impact of modern pandemics and epidemics. We estimate that the typical health crisis lowers GDP growth in affected countries by nearly three percentage points in the onset year and that this effect persists for at least five years. Unemployment risespersistentlytoo,withlargereffectsonfemalesandthelesseducated. Healthcrisesnot only lower output but also decrease consumption and investment spending. Furthermore, internationaltradeplummets,andthisnegativelyaffectsothercountriesthroughtradelinkages. Nevertheless, trading networks also benefit countries when there is bounce-back one year after the onset of a health crisis. We also show that fiscal policy helps to mitigate the effect of health crises. Increasing government spending, in particular on health care, significantlyspeedsupGDPgrowthrecoveryandreducesunemploymentafterthecrisis. Althoughtherearemanyparallelsbetweenthesepost-wardiseaseepisodesandCOVID- 19, there is a lot to suggest that this pandemic will have a much larger toll on human lives. Theunprecedentedscaleoflockdownsinseveralcountrieswillhampereconomicactivity even for countries that have lower severity or thwart the virus more quickly. There are alsoreasonstothinkthatCOVID-19willbeconsiderablymorerecessionary. Forone,U.S. fiscal space is relatively limited now. If fiscal policy does not move enough, or with the right mix, COVID-19 could have an even more persistent effect on output. Furthermore, a restoration of robust international trade linkages remains an open question. Ominous signs of backlash against China already appear. The sentiment for countries not to be so reliant on imports, especially in sensitive sectors like medical supplies, may well prove an intractablefoeoftrade. TheseconsiderationsarefleshedoutwithestimatesinAppendixC assessinghowdifferentisthistimewithCOVID-19. References ALVAREZ, F. E., D. ARGENTE, AND F. LIPPI (2020): “A simple planning problem for covid-19lockdown,”NBERWorkingPaperNo.26981. ATKESON, A.(2020): “WhatwillbetheeconomicimpactofCOVID-19intheUS?Rough estimatesofdiseasescenarios,”NBERWorkingPaperNo.26867. 30

BAKER, S. R., N. BLOOM, AND S. J. DAVIS (2016): “Measuringeconomicpolicyuncertainty,”QuarterlyJournalofEconomics,131,1593–1636. BAKER, S. R., N. BLOOM, S. J. DAVIS, AND S. J. TERRY (2020): “Covid-induced economicuncertainty,”NBERWorkingPaperNo.26983. BARRO, R. J., J. F. URSU´A, AND J. WENG (2020): “The coronavirus and the great influenzapandemic: Lessonsfromthe“SpanishFlu”forthecoronavirus’spotentialeffects onmortalityandeconomicactivity,”NBERWorkingPaperNo.26866. BENIGNO, G. AND L. FORNARO (2018): “Stagnationtraps,”ReviewofEconomicStudies, 85,1425–1470. BERGER, D. W., K. F. HERKENHOFF, AND S. MONGEY (2020): “An SEIR infectious diseasemodelwithtestingandconditionalquarantine,”NBERWorkingPaperNo.26901. BETHUNE, Z. A. AND A. KORINEK (2020): “Covid-19 infection externalities: Trading offlivesvs.livelihoods,”NBERWorkingPaperNo.27009. BONADIO, B., Z.HUO, A.A.LEVCHENKO, ANDN.PANDALAI-NAYAR(2020): “Global supplychainsinthepandemic,”CEPRDiscussionPaperNo.14766. CARROLL, C. D., E. CRAWLEY, J. SLACALEK, AND M. N. WHITE (2020): “Modeling theconsumptionresponsetotheCARESact,”WorkingPaper. CERRA, V., U.PANIZZA, ANDS.C.SAXENA(2013): “Internationalevidenceonrecovery fromrecessions,”ContemporaryEconomicPolicy,31,424–439. CERRA, V. AND S. C. SAXENA (2008): “Growthdynamics: themythofeconomicrecovery,”AmericanEconomicReview,98,439–57. CHANG, A. Y., K. COWLING, A. E. MICAH, A. CHAPIN, C. S. CHEN, G. IKILEZI, N. SADAT, G. TSAKALOS, J. WU, T. YOUNKER, ET AL.(2019): “Past,present,andfutureofglobalhealthfinancing: areviewofdevelopmentassistance,government,out-ofpocket,andotherprivatespendingonhealthfor195countries,1995–2050,”TheLancet, 393,2233–2260. CHANG, C., K. ORTIZ, A. ANSARI, AND M. E. GERSHWIN (2016): “The Zikaoutbreak ofthe21stcentury,”JournalofAutoimmunity,68,1–13. 31

CORREIA, S., S. LUCK, AND E. VERNER (2020): “Pandemics depress the economy, publichealthinterventionsdonot: Evidencefromthe1918Flu,”SSRNWorkingPaper. DRECHSEL, T. AND S. KALEMLI-OZCAN(2020): “Arestandardmacroandcreditpolicies enough to deal with the economic fallout from a global pandemic? A proposal for a negativeSMEtax,”WorkingPaper. DRISCOLL, J. AND A. KRAAY (1998): “Consistent covariance matrix estimation with spatiallydependentpaneldata,”ReviewofEconomicsandStatistics,80,549–560. DUPRAZ, S., E. NAKAMURA, AND J. STEINSSON(2019): “Apluckingmodelofbusiness cycles,”NBERWorkingPaperNo.26351. EGGERTSSON, G. B.(2011): “Whatfiscalpolicyiseffectiveatzerointerestrates?” NBER MacroeconomicsAnnual,25,59–112. EGGERTSSON, G. B. AND P. KRUGMAN (2012): “Debt, deleveraging, and the liquidity trap: A Fisher-Minsky-Koo approach,” Quarterly Journal of Economics, 127, 1469– 1513. EGGERTSSON, G. B., N. R. MEHROTRA, S. R. SINGH, AND L. H. SUMMERS(2016): “A contagious malady? Open economy dimensions of secular stagnation,” IMF Economic Review,64,581–634. EICHENBAUM, M. S., S. REBELO, AND M. TRABANDT (2020): “The macroeconomics ofepidemics,”NBERWorkingPaperNo.26882. FAN, V. Y., D. T. JAMISON, AND L. H. SUMMERS (2016): “The inclusive cost of pandemicinfluenzarisk,”NBERWorkingPaperNo.22137. FATA´S, A. AND L. H. SUMMERS(2018): “Thepermanenteffectsoffiscalconsolidations,” JournalofInternationalEconomics,112,238–250. FERNANDES, A.ANDH.TANG(2020): “Howdidthe2003SARSepidemicshapeChinese trade?” SSRNWorkingPaper3603699. FORNARO, L. AND M. WOLF(2020): “Covid-19coronavirusandmacroeconomicpolicy,” CEPRDiscussionPaperNo.14529. 32

GLOVER, A., J. HEATHCOTE, D. KRUEGER, AND J.-V. R´IOS-RULL (2020): “Health versus wealth: On the distributional effects of controlling a pandemic,” NBER Working PaperNo.27046. GOURINCHAS, P. O. (2020): “Flattening the pandemic and recession curves,” Working Paper, https://drive.google.com/file/d/1mwMDiPQK88x27JznMkWzEQpUVm8Vb4WI/ view?usp=sharing. GUERRIERI, V., G. LORENZONI, L. STRAUB, AND I. WERNING (2020): “Macroeconomic implications of COVID-19: Can negative supply shocks cause demand shortages?” NBERWorkingPaperNo.26918. HOFFMAN, S. J. AND S. L. SILVERBERG (2018): “Delays in global disease outbreak responses: Lessons from H1N1, Ebola, and Zika,” American Journal of Public Health, 108,329–333. HUANG, H. (2017): “Germs, roads and trade: Theory and evidence on the value of diversificationinglobalsourcing,”SSRNWorkingPaper3095273. JAMISON, D., R. NUGENT, H. GELBAND, S. HORTON, P. JHA, R. LAXMINARAYAN, AND C. MOCK (2017): Disease Control Priorities, vol. 9, World Bank: Washington, DC,3ed. JAMISON, D. T., L. H. SUMMERS, G. ALLEYNE, K. J. ARROW, S. BERKLEY, A. BI- NAGWAHO, F. BUSTREO, D.EVANS, R.G. FEACHEM, J.FRENK, ETAL.(2013): “The Lancetcommissions,”Lancet,382,1898–955. JORDA`, O`. (2005): “Estimation and inference of impulse responses by local projections,” AmericanEconomicReview,95,161–182. JORDA`, O`., M. SCHULARICK, AND A. M. TAYLOR (2011): “Financial crises, credit booms, and external imbalances: 140 years of lessons,” IMF Economic Review, 59, 340–378. ———(2013): “Whencreditbitesback,”JournalofMoney,CreditandBanking,45,3–28. JORDA`, O`., S. R. SINGH, AND A. M. TAYLOR (2020): “Longer-run economic consequencesofpandemics,”NBERWorkingPaperNo.26934. 33

LAEVEN, L. AND F. VALENCIA (2013): “Systemic banking crises database,” IMF EconomicReview,61,225–270. ———(2020): “SystemicbankingcrisesdatabaseII,”IMFEconomicReview,1–55. LEVCHENKO, A. A., L. T. LEWIS, AND L. L. TESAR (2010): “The collapse of international trade during the 2008–09 crisis: In search of the smoking gun,” IMF Economic review,58,214–253. MALMENDIER, U. AND L. S. SHEN (2020): “Scarred consumption,” NBER Working PaperNo.24696. MATEUS, A. L., H. E. OTETE, C. R. BECK, G. P. DOLAN, AND J. S. NGUYEN-VAN- TAM (2014): “Effectiveness of travel restrictions in the rapid containment of human influenza: a systematic review,” Bulletin of the World Health Organization, 92, 868– 880D. NOVY, D. AND A. M. TAYLOR(2014): “Tradeanduncertainty,”ReviewofEconomicsand Statistics,1–50. REINHART, C. M. AND K. S. ROGOFF (2009): “The aftermath of financial crises,” AmericanEconomicReview,99,466–72. SAUNDERS-HASTINGS, P. R. AND D. KREWSKI (2016): “Reviewing the history of pandemic influenza: understanding patterns of emergence and transmission,” Pathogens, 5, 66. WILLIAMS, H. A., R. L. DUNVILLE, S. I. GERBER, D. D. ERDMAN, N. PESIK, D. KUHAR, K. A. MASON, L. HAYNES, L. ROTZ, J. ST. PIERRE, ET AL. (2015): “CDC’s early response to a novel viral disease, middle east respiratory syndrome coronavirus (MERS-CoV), September 2012–May 2014,” Public Health Reports, 130, 307– 317. 34

A Data Sources TableA.1ListofCountriesfromWDI(TotalNumber: 210) Aruba Bolivia Dominica Grenada Kiribati Malta PapuaNewGu SlovakRepub Venezuela,R Afghanistan Brazil Denmark Greenland St.Kittsan Myanmar Poland Slovenia BritishVirg Angola Barbados DominicanRe Guatemala Korea,Rep. Montenegro PuertoRico Sweden VirginIslan Albania BruneiDarus Algeria Guam Kuwait Mongolia Korea,Dem. Eswatini Vietnam Andorra Bhutan Ecuador Guyana LaoPDR Mozambique Portugal Seychelles Vanuatu UnitedArab Botswana Egypt,Arab HongKongSA Lebanon Mauritania Paraguay SyrianArab Samoa Argentina CentralAfri Eritrea Honduras Liberia Mauritius WestBankan TurksandCa Yemen,Rep. Armenia Canada Spain Croatia Libya Malawi FrenchPolyn Chad SouthAfrica AmericanSam Switzerland Estonia Haiti St.Lucia Malaysia Qatar Togo Zambia Antiguaand Chile Ethiopia Hungary Liechtenstei Namibia Romania Thailand Zimbabwe Australia China Finland Indonesia SriLanka NewCaledoni RussianFede Tajikistan Austria Coted’Ivoir Fiji India Lesotho Niger Rwanda Turkmenistan Azerbaijan Cameroon France Ireland Lithuania Nigeria SaudiArabia Timor-Leste Burundi Congo,Dem. FaroeIsland Iran,Islami Luxembourg Nicaragua Sudan Tonga Belgium Congo,Rep. Micronesia, Iraq Latvia Netherlands Senegal Trinidadand Benin Colombia Gabon Iceland MacaoSAR,C Norway Singapore Tunisia BurkinaFaso Comoros UnitedKingd Israel Morocco Nepal SolomonIsla Turkey Bangladesh CaboVerde Georgia Italy Monaco Nauru SierraLeone Tuvalu Bulgaria CostaRica Ghana Jamaica Moldova NewZealand ElSalvador Tanzania Bahrain Cuba Gibraltar Jordan Madagascar Oman SanMarino Uganda Bahamas,The CaymanIslan Guinea Japan Maldives Pakistan Somalia Ukraine BosniaandH Cyprus Gambia,The Kazakhstan Mexico Panama Serbia Uruguay Belarus CzechRepubl Guinea-Bissa Kenya MarshallIsl Peru SouthSudan UnitedState Belize Germany EquatorialG KyrgyzRepub NorthMacedo Philippines SaoTomeand Uzbekistan Bermuda Djibouti Greece Cambodia Mali Palau Suriname St.Vincent 35

TableA.2MainVariableConstruction Variable Description Source Pandemics-relatedMeasures HealthShock Anindicatorequalstooneifacountryisaffectedbysixpan- HandCollected demicsathealthcrisisyeartandzerootherwise. MortalityRate Theratiooftotaldeathstototalaffectedcases(inpercentage) HandCollected foreachaffectedcountriesathealthcrisisyeartandzerofor thoseunaffectedcountries. Cases/Pop Theratiooftotalaffectedcasestonationalpopulation(in10 HandCollected thousand)foreachaffectedcountriesathealthcrisisyeartand zeroforunaffectedcountries. Country-levelMeasures GDPgrowthrate AnnualpercentagegrowthrateofGDPbasedonconstantlocal WDI currency. Unemploymentrate Theshareofthelaborforcethatiswithoutworkbutavailable WDI forandseekingemployment(InternationalLabourOrganizationEstimate). TaxRevenue(%GDP) Taxrevenuereferstocompulsorytransferstothecentralgov- WDI ernment for public purposes. Certain compulsory transfers suchasfines,penalties,andmostsocialsecuritycontributions areexcluded. Expense(%ofGDP) Expenseiscashpaymentsforoperatingactivitiesofthegov- WDI ernmentinprovidinggoodsandservices.Itincludescompensationofemployees(suchaswagesandsalaries),interestand subsidies,grants,socialbenefits,andotherexpensessuchas rentanddividends. CurrentHealthExpenditure(%ofGDP) Level of current health expenditure expressed as a percent- WDI ageofGDP.Estimatesofcurrenthealthexpendituresinclude healthcaregoodsandservicesconsumedduringeachyear.This indicatordoesnotincludecapitalhealthexpendituressuchas buildings,machinery,ITandstocksofvaccinesforemergency oroutbreaks. CentralGovernmentDebt(%ofGDP) Debtistheentirestockofdirectgovernmentfixed-termcon- WDI tractualobligationstoothersoutstandingonaparticulardate. Itincludesdomesticandforeignliabilitiessuchascurrencyand moneydeposits,securitiesotherthanshares,andloans.Itisthe grossamountofgovernmentliabilitiesreducedbytheamount ofequityandfinancialderivativesheldbythegovernment. GDPConsensusForecast ConsensusforcastsofpercentagegrowthrateofGDPatyeart ConsensusEconnomicsInc. basedontheendofyeart−1. Trade/GDP Thesumofexportsandimportsofgoodsandservicesmea- WDI suredasashareofGDPatyeart. DomesticCredit/GDP Domesticcredittoprivatesectorbybanksmeasuredashareof WDI GDPatyeart. Log(Population) Thenaturallogarithmoftotalpopulationbasedonthedefacto WDI definitionofpopulationatyeart. Log(GDPpercapita) ThenaturallogarithmofGDPpercapita(measuredasGDP WDI dividedbymidyearpopulation)inconstant2010U.S.dollarat yeart. RecessionDummy Anindicatorequalstooneifyeartiswithinthecontractions NBER ofU.S.businesscycleandzerofortheexpansions. BankingCrisisDummy Anindicatorequalstooneifacountryatyeartisidentifiedas LaevenandValencia(2013) systematicbankingcrisisandzerootherwise. QuarterlyGDPgrowthrate QuarterlypercentagegrowthrateofGDP(seasonaladjusted) OECDNationalAccountsStatistics basedonsamequarteratyeart−1(YoYchange). 36

B Figures FigureA.1SeverityofSixModernHealthCrisesandCOVID-19: TotalAffectedCases COVID-19inJune1,2020 0~10 11~100 101~1000 1001~10000 10001~100000 100001+ 1968Flu SARS Isolated R W e i g d i e o s n p a r l ead 0 1 ~ 1~ 10 100 101~1000 1001~10000 10001~100000 100001~1000000 H1N1 MERS 011111 ~10000 ~1000 1 ~100 01 ~10 01 ~1 001 ~ 001 0001 000 000000000 011111 ~10000 ~1000 1 ~100 01 ~10 01 ~1 001 ~ 001 0001 000 000000000 Ebola Zika 0~10 0~10 11~100 11~100 101~1000 101~1000 1001~10000 1001~10000 10001~100000 10001~100000 100001~1000000 100001~1000000 NOTE: ThisfiguredepictstheseverityofhealthcrisisepisodesinoursampleperiodandCOVID-19. Weclassifyeconomiesintosix groupsbasedonthereportedcases.Thedatafor1968Fluisavailableonlybyseveritygroupings:isolated,regionalandwidespread. 37

FigureA.2TradeNetworkIntensityinHealthCrisisYears COVID-19 SARS 0 1 1 1 1 1 ~ 1 0 0 0 0 ~ 1 0 0 0 1 ~ 1 0 0 0 1 ~ 1 0 0 1 ~ 1 0 0 1 + 0 0 1 0 0 0 0 0 0 000 0 1 1 1 1 1 ~ 1 0 0 0 0 ~ 1 0 0 0 1 ~ 1 0 0 0 1 ~ 1 0 0 1 ~ 1 0 0 1 + 0 0 1 0 0 0 0 0 0 000 H1N1 MERS 0~10 0~10 11~100 11~100 101~1000 101~1000 1001~10000 1001~10000 10001~100000 10001~100000 100001+ 100001+ Ebola Zika 0~10 0~10 11~100 11~100 101~1000 101~1000 1001~10000 1001~10000 10001~100000 10001~100000 100001+ 100001+ NOTE: Thisfiguredepictsthetradenetworkintensitymeasureusingbothex-postcasesandbilateraltradedata. Foreachcountry’s severity,weweightitstradingpartners’casenumberusingthebilateraltradeshare.Duetodatalimitation,weusethetradedatain2018 andthereportednumberofcasesforCOVID-19asofJune1,2020toconstructtheCOVIDpanel. 38

C Covid-19: This time is different, but how different? TherearemanyreasonstothinkthatCOVID-19willhavelargereffectsontheworldeconomy than our historical disease episodes. The current pandemic has more infection cases anddeathsthanthetypicalhistoricalepisode,andCOVID-19hasspreadtomorecountries. For example, the total confirmed case number of nearly 10 million, to date, exceeds total cases in all other episodes combined (total cases in 1968 Flu are not known but are estimated to 1 million worldwide as in Jorda` et al. (2020)). Moreover, there is a worldwide lock-down policy designed to contain COVID-19 (termed the “Great Lockdown” by the IMF).Althoughregionaltravelbanshavebeenusedinprevioushealthcrises,accordingto Mateus et al. (2014), national lockdowns as under COVID-19 are unprecedented (detailed informationisdisplayedinOnlineSupplementTableS.2). Theserestrictionshavenodoubt “flattenedthecurve”butalsocrippledeconomicactivityworldwide,atleastintheshortrun (Gourinchas(2020)). Fromtheperspectiveofeconomicstructure,thereareseveralreasons whytheimpactofCOVID-19mightbelarger. First,manycountrieshaveshiftedfromagriculturetotheindustryandservicessectors. Second,tradelinkagesbetweencountrieshave increased (see Figure A.2 for an illustration). A more intertwined world through global value/supplychainsmakesCOVID-19economicallymorecontagious. We use our estimates from the historical episodes, juxtaposed against current forecasts of the effects of COVID-19 from the IMF, World Bank, OECD, Consensus Forecasts, and the FOMC’s Summary of Economic Projections, to gauge just how different “this time” might be. We begin with a simple projection for GDP growth in 2020 and 2021 using our estimates from the historical episodes. We report three cases of GDP growth for: world, advanced economies, and the United States. Because COVID is more severe than the average historical episode, we base the projections on estimates of the high severity dummy fromourseverityspecifications. Weuseestimatesfortheonsetyearandoneyearlater(see FigureS.1andTableA.3). Wemakeprojectionsfortheworldandtheadvancedeconomies separately, from estimates using our full sample and advanced economies sample, respectively. WeuseestimatesfromtheadvancedeconomiessampletoformourU.S.projection. Theprojectionfor2020worldGDPgrowthratebasedonthehistoricaldiseaseepisodes, labeledMRZ,is-3.5%. Asseeninthetable,thisisnon-triviallylessgloomythantheIMF, World Bank, CF, and especially the OECD forecast of -7.1%. For the bounce-back year of 2021, and again under the assumption that this crisis were the same as the average ”severe”historicalepisode,GDPgrowthwouldbearoundonepercent,asseenfromtheMRZ 39

TableA.3Institutions’GDPGrowthForecastsfor2020-21and ProjectionsBasedonSixHistoricalCrisisEpisodes(MRZ) World MRZ IMF WorldBank OECD CF 2020 -3.5 -5.0 -5.5 -7.1 -4.1 (1.1) 2021 0.9 5.4 4.2 4.2 5.1 (0.5) Advanced MRZ IMF WorldBank OECD CF 2020 -3.3 -8.1 -7.0 -10.0 -7.0 (1.2) 2021 1.6 4.8 3.9 4.4 5.2 (0.4) U.S. MRZ IMF WorldBank OECD CF FRB 2020 -2.8 -8.0 -6.1 -7.6 -5.4 -6.5 (1.2) [-10.0,-4.2] 2021 1.4 4.5 4.00 2.1 4.3 5.0 (0.4) [-1.0,7.0] NOTE: Theinstitutionalforecastsaretakenfrom: June2020WorldEconomicOutlook(IMF);June2020GlobalEconomicProspects (WorldBank);May2020EconomicOutlook(OECD);May2020issueofConsensusForecasts(CF);Jun10SummaryofEconomic ProjectionsfollowingtheFOMCmeeting(FRB),wherewereportboththemedianestimateandtherange.TheMRZestimatesaretaken fromthecoefficientonthehighmortalityratedummy,definedasthetop30%ofcountrymortalityratesineachepisode.Fortheworld (advancedeconomies)estimates,weusethefullsample(advancedeconomysample)ofcountries. FortheU.S.,weadjustthepoint estimatefromtheadvancedeconomiessamplebytheirrelativeaveragegrowthrates,U.S.versusadvancedeconomies,whilekeeping thestandarddeviationthesame. projection. Recovery in world GDP growth is projected to be much higher by the institutional forecasters. This optimism could come from the assumption that policymakers are doing whatever it takes to contain COVID-19, or at least much more than in past crises, as the current low interest rate environment may make governments more willing to increase spending. As noted above, bounce-back that is stronger than in previous episodes could also be the result of the magnification effects of stronger trade linkages. The projection for advanced economies and U.S. based on the historical episodes is around -3% for 2020 and +1.5% for 2021. These history-based projections are around one-half to one-third of the magnitude, both up and down, being predicted for COVID-19, though the range of estimates from the FRB is rather wide. Overall, consistent with intuition, this time is seen to bedifferentbyprominentforecasters. 40

Online Supplement to ‘Modern Pandemics: Recession and Recovery’ byC.Ma,andJ.Rogers,andS.Zhou June2020

S.1 Figures FigureS.1EffectsofHealthCrisesonGDPbySeverity tnecreP 2 4- High Case Rate Medium Case Rate Low Case Rate 0 1 2 3 4 5 Years tnecreP 2 4- High Mortality Rate Medium Mortality Rate Low Mortality Rate 0 1 2 3 4 5 Years NOTE: Impulse response functions (IRF) are estimated based on the local projection method as in Jorda` (2005) git+H =αH i + ∑4 j=1 βH j git−j+∑4 s=0 δH s DH it−s +∑4 s=0 γH s DM it−s +∑4 s=0 µH s DL it−s +Xit+εit,withH=0,1,···,5,wheregit istheannualrealGDPgrowth rateforcountryiatyeart,DH(cid:0) DM,DL(cid:1) isadummyvariableindicatingahigh(medium,low)mortalityrateorcasesperpopulationrate it it it foranaffectedcountryiinyeart,withXit includingcountry-levelcontrolssuchasTrade/GDP,DomesticCredit/GDP,populationand logGDPpercapita.Wealsoincludeadecadedummy,USrecessiondummy,abankingcrisisdummyandcountryfixedeffects.Standard errorsarecorrectedusingDriscollandKraay(1998). Thebluelinerepresentslow,thegreendash-dottedlinerepresentsmediumand thereddashedlinerepresentshigh.Onestandarderrorbandsareshown. S.1

FigureS.2EffectonGDPGrowthConditionalonImmediateFiscalResponse: ResultsforGeneralExpendituresandTaxRevenues PanelA:HighExpenditureResponse PanelB:LowExpenditureResponse tnecreP 3 2 1 0 1- 2- 3- 4- 0 1 2 3 4 5 Years tnecreP 3 2 1 0 1- 2- 3- 4- 0 1 2 3 4 5 Years PanelC:HighTaxResponse PanelD:LowTaxResponse tnecreP 3 2 1 0 1- 2- 3- 4- 0 1 2 3 4 5 Years tnecreP 3 2 1 0 1- 2- 3- 4- 0 1 2 3 4 5 Years NOTE: Impulse response functions (IRF) are estimated based on the local projection method as in Jorda` (2005): git+H =αH i + ∑4 s=1 βH s git−s+∑4 s=0 δH s Dit−s+Xit+εit,withH=0,1,···,5,wheregit istheannualrealGDPgrowthrateforcountryiatyeart,Dit isadummyvariableindicatingadiseaseeventhittingcountryiinyeart,withXit includingcountry-levelcontrolssuchasTrade/GDP, DomesticCredit/GDP,populationandlogGDPpercapita. Wealsoincludeadecadedummy,U.S.recessiondummy,abankingcrisis dummyandcountryfixedeffects.StandarderrorsarecorrectedusingDriscollandKraay(1998).Onestandarderrorbandsareshown. Eachrowdividescountriesbasedontheaverageof Zit−Zit−1 acrossallsixhealthepisodeswheretistheonsetyearofeachepisode. Z GDPit−1 referstofiscalspendinginPanelAandB,andtaxrevenueinPanelCandD.Highreferstocountriesinthe75percentileandabove whilelowreferstocountriesinthe25percentileandbelow. S.2

FigureS.3EffectonGovernmentBudget PanelA:CentralGovernmentDebt(%GDP) PanelB:FiscalSurplus(%GDP) tnecreP 8 6 4 2 0 0 1 2 3 4 5 Years tnecreP 1 0 1- 2- 3- 0 1 2 3 4 5 Years PanelC:GovernmentSpending(%GDP) PanelD:GovernmentRevenue(%GDP) tnecreP 5.1 1 5. 0 5.- 0 1 2 3 4 5 Years tnecreP 5. 0 5.- 1- 0 1 2 3 4 5 Years NOTE: Impulse response functions (IRF) are estimated based on the local projection method as in Jorda` (2005): yit+H =αH i + ∑4 s=1 βH s yit−s+∑4 s=0 δH s Dit−s+Xit+εit,withH=0,1,···,5,whereyit istheannualcentralgovernmentdebt(%GDP),fiscalsurplus (%GDP),governmentspending(%GDP)orgovernmentrevenue(%GDP)forcountryiatyeart,Ditisadummyvariableindicatinga diseaseeventhittingcountryiinyeart,withXitincludingcountry-levelcontrolssuchasTrade/GDP,DomesticCredit/GDP,population andlogGDPpercapita. Wealsoincludeadecadedummy,U.S.recessiondummy,abankingcrisisdummyandcountryfixedeffects. StandarderrorsarecorrectedusingDriscollandKraay(1998).Onestandarderrorbandsareshown. S.3

FigureS.4HealthSpendingandCrisisSeverity Panel A: Health Spending Adjustment and Mortality Rate )PDG %( tnemtsujdA gnidnepS htlaeH 40. 30. 20. 10. 0 10.- 0 20 40 60 80 100 Mortality Rate (%) Panel B: Health Spending Adjustment and Case Rate )PDG %( tnemtsujdA gnidnepS htlaeH 40. 30. 20. 10. 0 10.- 0 1 2 3 Case/Pop Rate NOTE: PanelAplotstherelationshipbetweenhealthspendingadjustment(definedasthechangeofhealthspendingintheonsetyear normalizedbythepreviousyear’sGDP)andthemortalityrate,forallepisodesinaffectedcountries.Theregressionlinehasaslopeof -0.000012witht-statat-0.59.PanelBplotstherelationshipbetweenhealthspendingadjustmentandthecaserateforalltheepisodesin affectedcountries.Theregressionlinehasaslopeof0.0009witht-statat0.55. S.4

S.2 Data Sources TableS.1ListofGlobalPandemicandEpidemicEvents AnnouncementTime EventName AffectedCountries(Economies) #ofAffectedCountries(inmatchedsample) TotalDeaths TotalCases AverageMortalityRate 1968/07 HongkongFlu ARG,AUS,CHL,DNK,FIN,FRA,GBR,GRC,HKG, 18 N.A. N.A. N.A. ITA,JAM,JPN,NLD,NOR,PRT,SWE,USA,ZAF 2003/02 SARS AUS,CAN,CHE,CHN,DEU,ESP,FRA,GBR,HKG, 28 737 7750 9.51% IDN,IND,IRL,ITA,KOR,KWT,MAC,MNG,MYS, NZL,PHL,ROU,RUS,SGP,SWE,THA,USA,VNM, ZAF 2009/04 H1N1 AGO,ALB,AND,ARE,ARG,ASM,AUS,AUT,AZE, 167 14390a 526353 2.73% BDI,BEL,BGD,BGR,BHR,BHS,BIH,BLR,BLZ, BMU,BOL,BRA,BRB,BRN,BTN,BWA,CAN,CHE, CHL,CHN,CIV,CMR,COD,COG,COL,CPV,CRI, CUB,CYM,CYP,CZE,DEU,DMA,DNK,DOM,DZA, ECU,EGY,ESP,ETH,FIN,FJI,FRA,FSM,GAB,GBR, GEO,GHA,GRC,GRD,GTM,GUM,GUY,HND,HRV, HTI,HUN,IDN,IND,IRL,IRN,IRQ,ISL,ISR,ITA, JAM,JOR,JPN,KAZ,KEN,KHM,KIR,KNA,KOR, KWT,LAO,LBN,LBY,LCA,LIE,LKA,LSO,LUX, MAR,MDA,MDG,MDV,MEX,MHL,MKD,MLI, MLT,MMR,MNE,MNG,MOZ,MUS,MWI,MYS, NAM,NGA,NIC,NLD,NOR,NPL,NRU,NZL,OMN, PAK,PAN,PER,PHL,PLW,PNG,POL,PRI,PRT,PRY, PSE,QAT,ROU,RUS,RWA,SAU,SDN,SGP,SLB, SLV,SRB,STP,SUR,SVK,SVN,SWE,SWZ,SYC, TCD,THA,TJK,TLS,TON,TTO,TUN,TUR,TUV, TZA,UGA,URY,USA,VCT,VEN,VNM,VUT,WSM, YEM,ZAF,ZMB,ZWE 2012/03 MERS ARE,AUT,CHN,DEU,DZA,EGY,FRA,GBR,GRC, 26 498 1289 38.63% IRN,ITA,JOR,KOR,KWT,LBN,MYS,NLD,OMN, PHL,QAT,SAU,THA,TUN,TUR,USA,YEM 2014/08b Ebola ESP,GBR,GIN,ITA,LBR,MLI,NGA,SEN,SLE,USA 10 11323 28646 39.53% 2016/02c Zika ABW,ARG,ATG,BHS,BLZ,BOL,BRA,BRB,CAN, 38 20 197689 0.01% CHL,COL,CRI,CUB,CYM,DMA,DOM,ECU,GRD, GTM,GUY,HND,HTI,JAM,KNA,LCA,NIC,PAN, PER,PRI,PRY,SLV,SUR,TCA,TTO,URY,USA,VCT, VIR a ThisestimatesarefromEuropeanCenterforDiseasePreventionandControls(ECDC).Weusetheirestimatessincetheyprovidesdetailedcoverageandmortalityrateforeachcountry. Detailedinformationcanbefoundhere:https://en.wikipedia.org/wiki/2009_flu_pandemic_by_country.However,theestimatefromUSCentersforDiseaseControlandPrevention (CDC)forglobaldeathtrollis284,000,about15timesmorethanthenumberoflaboratory-confirmedcases. Seedetailsinhttp://www.cidrap.umn.edu/news-perspective/2012/06/ cdc-estimate-global-h1n1-pandemic-deaths-284000. b TheWestAfricanEbolaoutbreakbeganDecember26,2013andwasdeclaredaPHEICAugust8,2014. c TheZikavirusoutbreakoccurredatOctober,2015butwasdeclaredaPHEICFebruary1,2016 S.5

TableS.2DetailsofSixPandemicandEpidemicEvents Episodes Vaccine/Cure GovernmentResponse 1968Flu “Splitvaccine”developedin1968 The1968Fluspreadwidelyasaresultofinternationalairtravel,buttheeffectssurfaceddifferentlyindifferentregions —theUSandCanadaexperiencedasevereinitialwavewithlessaseveresubsequentwave,whilethereverseheldtrue forEuropeandAsia. InNorthAmerica,wheretheburdenofthefluwasrelativelysmallincomparisontoinEurope andAsia,governmentreliedonvaccination,hospitalization,andantibioticstotreatsecondarypneumonia.Quarantines, closures,andothernon-pharmaceuticalmeansofinterventionwerenotquitenecessarytocurbthedisease. SARS Nocure EffortstosuppressSARSincludedisolationofsymptomaticpatientsandrigidhospitalinfectioncontrolpractices. The latterprovedtobeparticularlyeffectiveinthe2003SARSpandemicinhospitalsinHongKongSAR,China,inwhich noneofthehealthcareworkerswearingproperPPEevercontractedSARS.Governmentsmainlyutilizedcontainment measureswhichmirroredthoseusedtoridofbubonicplagues—casetracking,quarantiningthoseinfected,bansonlarge gatherings,examinationoftravelers,improvedPPEandbarrierprotection. Thesemeasures,workingintandemwith travelrestrictions, successfullycurbedSARSlikelybecauseSARSischaracterizedbyaninsignificantasymptomatic carrierstateandrelativelyshorterincubationperiods. H1N1 VaccinereleasedinOctoberof2009 In response to the outbreak of the Swine Flu, several countries’ governments focused on restricting travel amongst infectedregions.Additionally,privateandpublicsectorworkerswereadvisedtoimplementpreventativemeasures,and schoolswereclosedinareasofoutbreak. ChinarevertedtousingthesamemeasuresitusedtofightSARS,notably quarantininganyandallpersonswhowerepossiblyinfectedbyH1N1. Moreover,manycountriesplacedembargoson importsofporkfromMexicoandtheUS.Airportscreeningwasalsoimplementedduringthistime.However,ithasbeen shownthattravelrestrictionswithregardstocurbinginfluenzaareonlyeffectiveindelayingthespreadandpeakofthe disease.Extensivetravelrestrictionsarerequiredtohavesignificantimpactoncurbinginfluenza. Mers Noavailablevaccineorspecifictreatment TheCDCcollaboratedwiththeWorldHealthOrganization,andbeganrespondingtotheMerscrisisbeforeitreached theUS.Keyareasoffocusincludedepidemiology,laboratoryscience,travelers’health,andinfectioncontrol. Another wascollaborationwithincountriesandbetweencountries. TheCDCbroughtaboutdata-sharingagreementsbetween countriesandpromotedglobalsharingofspecimensandreagentstodeliveraneffectiveresponsetothedisease. Ebola Noknownvaccine/treatment Thehardest-hitcountriesimposedcertainmeasurestocurbthedevastationofEbola. Ingeneral,healthagenciesand hospitalsreliedonisolationofsymptomaticpatients,quarantining,andbolsteringofhospitalinfectioncontrolpracticesto combatEbola.Somecountrieswerebetterequippedthanotherstoexecutediseaseprevention–—Nigeriahadexperience runninganemergencyoperationscenterandutilizingglobalpositioningsystemsforcontacttracingduringpreviouspolio eradicationefforts.Ultimately,puttinganendtoEbolarequiredamultinationaleffort,withtheWorldBank’sPandemic EmergencyFinancingFacility(PEF)contributingUS$3.8billiontohelpwiththecostsofEbola,andtheWorldBank GrouppoolingUS$1.6billionfromtheInternationalDevelopmentAssociationandtheInternationalFinanceCorporation toputtowardseconomicrecoveryinGuinea,Liberia,andSierraLeone. Zika Novaccine/specifictreatment Inresponsetotheoutbreak,governmentsincludingthoseoftheUSandtheUKdeclaredtravelprecautions,advising pregnantwomen,inparticular,toavoidtravellingtocountriesaffectedbyZika. Controlmeasuressuchasinsectbite precautionsandremovalofpossiblebreedinggroundsformosquitoswereimplemented,aswellasregulatoryreporting onrecommendationsregardingZikaandpharmaceuticalintervention. NOTE: ThenotereliesoninformationmainlyfromJamisonetal.(2017),Mateusetal.(2014),Changetal.(2016),Williamsetal.(2015),Saunders-HastingsandKrewski(2016)andonline informationfromhttps://graduateinstitute.ch/communications/news/brief-international-history-pandemics. S.6

TableS.3QuarterlyGDPCountryCoverage CountryCode CountryName StartQuarter EndQuarter CountryCode CountryName StartQuarter EndQuarter ARG Argentina 1994Q1 2018Q4 ISL Iceland 1961Q1 2018Q4 AUS Australia 1961Q1 2018Q4 ISR Israel 1996Q1 2018Q4 AUT Austria 1961Q1 2018Q4 ITA Italy 1961Q1 2018Q4 BEL Belgium 1961Q1 2018Q4 JPN Japan 1961Q1 2018Q4 BGR Bulgaria 1996Q1 2018Q4 KOR Korea,Rep. 1961Q1 2018Q4 BRA Brazil 1997Q1 2018Q4 LTU Lithuania 1996Q1 2018Q4 CAN Canada 1962Q1 2018Q4 LUX Luxembourg 1961Q1 2018Q4 CHE Switzerland 1961Q1 2018Q4 LVA Latvia 1996Q1 2018Q4 CHL Chile 1996Q1 2018Q4 MEX Mexico 1961Q1 2018Q4 CHN China 2011Q1 2018Q4 NLD Netherlands 1961Q1 2018Q4 COL Colombia 2006Q1 2018Q4 NOR Norway 1961Q1 2018Q4 CZE CzechRepubl 1995Q1 2018Q4 NZL NewZealand 1988Q2 2018Q4 DEU Germany 1961Q1 2018Q4 POL Poland 1996Q1 2018Q4 DNK Denmark 1961Q1 2018Q4 PRT Portugal 1961Q1 2018Q4 ESP Spain 1961Q1 2018Q4 ROU Romania 1996Q1 2018Q4 EST Estonia 1996Q1 2018Q4 RUS RussianFede 2004Q1 2018Q4 FIN Finland 1961Q1 2018Q4 SAU SaudiArabia 2010Q1 2018Q4 FRA France 1961Q1 2018Q4 SVK SlovakRepub 1994Q1 2018Q4 GBR UnitedKingd 1960Q1 2018Q4 SVN Slovenia 1996Q1 2018Q4 GRC Greece 1961Q1 2018Q4 SWE Sweden 1961Q1 2018Q4 HUN Hungary 1996Q1 2018Q4 TUR Turkey 1999Q1 2018Q4 IDN Indonesia 1991Q1 2018Q4 USA UnitedState 1960Q1 2018Q4 IND India 1997Q2 2018Q4 ZAF SouthAfrica 1961Q1 2018Q4 IRL Ireland 1961Q1 2018Q4 TableS.4Country-levelData: SummaryStatistics Variables N mean p50 sd p75 p25 GDPgrowthrate% 8,991 3.76 3.80 4.11 1.00 6.00 Unemploymentrate% 5208 8.19 6.65 6.32 11.16 3.59 GDPConsensusForecast% 612 2.57 2.44 2.02 1.55 3.38 QuarterlyGDPgrowthrate% 7,876 3.33 3.24 3.51 1.49 5.22 QuarterlyGDPConsensusForecast% 1,552 2.93 2.61 1.78 1.93 3.42 Trade/GDP% 8,261 67.43 59.00 49.72 36.96 87.77 DomesticCredit/GDP% 7,605 33.78 23.00 39.23 12.00 45.00 Log(Population) 12,202 8.26 4.29 5.95 3.37 15.06 Log(GDPpercapita) 9,172 5.97 5.51 2.82 3.40 8.46 RecessionDummy 12,272 0.27 0.00 0.44 0.00 1.00 BankingCrisisDummy 12,272 0.01 0.00 0.11 0.00 0.00 TaxChange% 3,680 8.06 1.80 16.79 0.54 5.08 ExpenditureChange% 3,464 8.58 2.36 16.95 0.84 5.92 HealthChange% 2,947 0.63 0.50 0.61 0.24 0.90 S.7

TableS.5Pre-trendAnalysis GDPgrowthrate% (1) (2) (3) SamplePeriod: 1960-2018 1990-2018 1960-2018 Shock(-1) -0.18 -0.12 -0.17 (0.37) (0.43) (0.49) Shock -2.56** -2.55* -2.60* (1.22) (1.27) (1.30) Shock(+1) 0.49* 0.47* 0.63* (0.25) (0.25) (0.31) Shock(+2) 0.55*** 0.59*** 0.59*** (0.14) (0.13) (0.20) HealthExpenditure(Lagged) 0.16 (0.11) Trade/GDP 0.17 0.54 2.82*** (0.22) (0.46) (0.48) DomesticCredit/GDP -0.66* -0.70 -0.50 (0.38) (0.46) (0.43) Log(Population) 0.19*** 0.15*** 0.97 (0.04) (0.05) (2.57) Log(GDPpercapita) -0.36*** -0.30*** 2.72** (0.08) (0.10) (1.24) Recession -0.35 -0.50 -1.08** (0.25) (0.44) (0.48) BankingCrisis -1.28*** -1.38*** -2.22** (0.32) (0.37) (1.02) Constant 4.80*** 4.71*** -36.28 (0.49) (0.52) (45.87) Observations 6348 4158 2708 WithinR-square 0.058 0.067 0.131 DecadeFE Yes Yes Yes CountryFE Yes Yes Yes NOTE: This table estimates a panel regression with four dummy variables that flags one year before the healthcrises,theonsetyear,oneyearafterandtwoyearsafterthehealthcrises. Wealsoaddalaggedhealth expenditure (% GDP ) as a control in column (3). ∗, ∗∗ and ∗∗∗ indicate statistical significance at the 10%, 5%,and1%level,respectively. S.8

TableS.6TheEffectofHealthCrisesonRealGDPGrowth: WeightedbyDiseaseSeverity GDPgrowthrate% (1) (2) (3) (4) (5) (6) SamplePeriod: 1960-2018 1990-2018 1960-2018 1990-2018 MortalityRate -3.62* -3.42* -5.85*** (1.94) (1.86) (1.42) Cases/Pop -3.36*** -3.20*** -5.46*** (1.11) (1.13) (0.98) ConsensusForecast 0.52*** 0.57*** (0.14) (0.16) Trade/GDP 0.19 0.52 0.83 0.18 0.50 0.75 (0.22) (0.38) (0.67) (0.21) (0.38) (0.63) DomesticCredit/GDP -0.71* -0.75 -1.64 -0.71* -0.75 -1.45 (0.42) (0.50) (1.10) (0.40) (0.48) (1.04) Log(Population) 0.17*** 0.12* 0.06 0.17*** 0.11* 0.05 (0.03) (0.06) (0.05) (0.03) (0.06) (0.05) Log(GDPpercapita) -0.33*** -0.23* -0.05 -0.32*** -0.22* -0.03 (0.08) (0.13) (0.14) (0.07) (0.12) (0.14) Recession -0.51* -0.83* -0.75 -0.48* -0.77* -0.53 (0.27) (0.45) (0.66) (0.25) (0.41) (0.58) BankingCrisis -1.26*** -1.30*** 1.12 -1.27*** -1.31*** 0.97 (0.37) (0.47) (0.97) (0.36) (0.46) (0.92) Constant 4.62*** 4.58*** 1.98*** 4.61*** 4.54*** 1.80*** (0.45) (0.51) (0.51) (0.45) (0.50) (0.49) Observations 6522 4296 530 6525 4299 530 WithinR2 0.042 0.039 0.134 0.045 0.044 0.156 DecadeFE Yes Yes Yes Yes Yes Yes CountryFE Yes Yes Yes Yes Yes Yes NOTE: ThedependentvariableisrealannualGDPgrowthrate. Thesampleperiodforcolumns(1)and(4) is 1960-2018 while the sample period for columns (2)-(3) and (5)-(6) is 1990-2018. Country and decade fixedeffectsareincluded. AllstandarderrorsarecorrectedusingDriscollandKraay(1998)andreportedin parentheses. ∗,∗∗and∗∗∗indicatestatisticalsignificanceatthe10%,5%,and1%level,respectively. S.9

TableS.7DiseaseSeverityandHealthExpenditureResponseDummy PanelA:DiseaseSeverityandHealthExpenditureResponseDummy 1968Flu SARS H1N1 MERS Ebola Zika CountryName Country Mortality Case/Pop Health Expendi- Mortality Case/Pop Health Expendi- Mortality Case/Pop Health Mortality Case/Pop Health Mortality Case/Pop Health Mortality Case/Pop Health Code Rate ture Rate ture Rate Expenditure Rate Expenditure Rate Expenditure Rate Expenditure Aruba ABW 0 0 N.A. 0 0 N.A. 1 3 N.A. 0 0 N.A. 0 0 N.A. 1 2 N.A. Afghanistan AFG 0 0 N.A. 0 0 2 2 2 2 0 0 2 0 0 2 0 0 2 Angola AGO 0 0 N.A. 0 0 2 1 1 1 0 0 1 0 0 1 0 0 2 Albania ALB 0 0 N.A. 0 0 2 3 1 1 0 0 1 0 0 2 0 0 1 Andorra AND 0 0 N.A. 0 0 2 0 1 1 0 0 1 0 0 1 0 0 2 UnitedArab ARE 0 0 N.A. 0 0 1 3 1 1 2 3 1 0 0 1 0 0 1 Argentina ARG 1 1 N.A. 0 0 2 3 3 2 0 0 2 0 0 2 1 1 2 Armenia ARM 0 0 N.A. 0 0 2 0 0 1 0 0 2 0 0 2 0 0 1 AmericanSam ASM 0 0 N.A. 0 0 N.A. 1 3 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. Antiguaand ATG 0 0 N.A. 0 0 1 1 2 1 0 0 1 0 0 2 1 2 1 Australia AUS 3 3 N.A. 1 2 1 2 3 2 0 0 2 0 0 2 0 0 1 Austria AUT 0 0 N.A. 0 0 1 1 2 1 1 2 2 0 0 1 0 0 2 Azerbaijan AZE 0 0 N.A. 0 0 2 3 1 2 0 0 2 0 0 2 0 0 2 Burundi BDI 0 0 N.A. 0 0 1 1 1 2 0 0 1 0 0 1 0 0 2 Belgium BEL 0 0 N.A. 0 0 2 1 1 2 0 0 1 0 0 1 0 0 1 Benin BEN 0 0 N.A. 0 0 1 0 0 1 0 0 2 0 0 1 0 0 1 BurkinaFaso BFA 0 0 N.A. 0 0 1 0 0 2 0 0 1 0 0 1 0 0 2 Bangladesh BGD 0 0 N.A. 0 0 1 2 1 1 0 0 1 0 0 1 0 0 1 Bulgaria BGR 0 0 N.A. 0 0 2 3 1 1 0 0 2 0 0 2 0 0 2 Bahrain BHR 0 0 N.A. 0 0 2 2 3 1 0 0 2 0 0 1 0 0 1 Bahamas,The BHS 0 0 N.A. 0 0 1 3 2 1 0 0 1 0 0 1 1 2 2 BosniaandH BIH 0 0 N.A. 0 0 2 3 2 1 0 0 1 0 0 1 0 0 1 Belarus BLR 0 0 N.A. 0 0 2 0 0 1 0 0 2 0 0 2 0 0 2 Belize BLZ 0 0 N.A. 0 0 2 1 2 2 0 0 1 0 0 1 1 2 2 Bermuda BMU 0 0 N.A. 0 0 N.A. 3 2 N.A. 0 0 N.A. 0 0 N.A. 1 2 N.A. Bolivia BOL 0 0 N.A. 0 0 2 2 3 2 0 0 2 0 0 2 1 1 2 Brazil BRA 0 0 N.A. 0 0 2 3 2 2 0 0 2 0 0 2 3 3 2 Barbados BRB 0 0 N.A. 0 0 2 2 3 1 0 0 2 0 0 1 1 2 1 BruneiDarus BRN 0 0 N.A. 0 0 1 2 3 1 0 0 1 0 0 1 0 0 1 Bhutan BTN 0 0 N.A. 0 0 1 1 1 2 0 0 2 0 0 1 0 0 1 Botswana BWA 0 0 N.A. 0 0 2 1 1 2 0 0 2 0 0 2 0 0 2 CentralAfri CAF 0 0 N.A. 0 0 1 0 0 2 0 0 2 0 0 2 0 0 1 Canada CAN 0 0 N.A. 3 3 2 3 3 2 0 0 1 0 0 1 1 1 1 Switzerland CHE 3 3 N.A. 1 2 1 2 2 1 0 0 1 0 0 1 0 0 2 Chile CHL 3 3 N.A. 0 0 2 2 3 2 0 0 2 0 0 2 1 1 2 China CHN 0 0 N.A. 2 3 2 2 2 2 1 1 2 0 0 2 0 0 2 Coted’Ivoir CIV 0 0 N.A. 0 0 1 1 1 1 0 0 2 0 0 2 0 0 1 Cameroon CMR 0 0 N.A. 0 0 1 1 1 1 0 0 2 0 0 2 0 0 1 Congo,Dem. COD 0 0 N.A. 0 0 2 1 1 2 0 0 2 0 0 2 0 0 1 Congo,Rep. COG 0 0 N.A. 0 0 1 1 1 1 0 0 2 0 0 1 0 0 2 Colombia COL 0 0 N.A. 0 0 2 3 2 2 0 0 1 0 0 1 1 2 1 Comoros COM 0 0 N.A. 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 CaboVerde CPV 0 0 N.A. 0 0 1 1 2 1 0 0 2 0 0 1 0 0 1 CostaRica CRI 0 0 N.A. 0 0 2 2 3 2 0 0 1 0 0 2 1 2 1 Cuba CUB 0 0 N.A. 0 0 1 3 2 2 0 0 1 0 0 2 1 1 1 CaymanIslan CYM 0 0 N.A. 0 0 N.A. 2 3 N.A. 0 0 N.A. 0 0 N.A. 1 3 N.A. Cyprus CYP 0 0 N.A. 0 0 1 1 3 1 0 0 1 0 0 1 0 0 2 CzechRepubl CZE 0 0 N.A. 0 0 2 3 2 2 0 0 1 0 0 1 0 0 1 Germany DEU 3 3 N.A. 1 2 1 2 3 2 2 2 1 0 0 2 0 0 2 Djibouti DJI 0 0 N.A. 0 0 N.A. 1 1 N.A. 0 0 N.A. 0 0 1 0 0 1 Dominica DMA 0 0 N.A. 0 0 1 1 3 2 0 0 1 0 0 1 1 3 1 Denmark DNK 3 3 N.A. 0 0 1 1 2 2 0 0 1 0 0 1 0 0 2 DominicanRe DOM 0 0 N.A. 0 0 2 3 2 2 0 0 2 0 0 2 1 1 2 S.10

DiseaseSeverityandHealthExpenditureResponseDummy(Cont.) 1968Flu SARS H1N1 MERS Ebola Zika CountryName Country Mortality Case/Pop Health Expendi- Mortality Case/Pop Health Expendi- Mortality Case/Pop Health Mortality Case/Pop Health Mortality Case/Pop Health Mortality Case/Pop Health Code Rate ture Rate ture Rate Expenditure Rate Expenditure Rate Expenditure Rate Expenditure Algeria DZA 0 0 N.A. 0 0 1 3 1 2 2 2 2 0 0 2 0 0 1 Ecuador ECU 0 0 N.A. 0 0 2 3 2 2 0 0 2 0 0 2 1 2 1 Egypt,Arab EGY 0 0 N.A. 0 0 1 2 2 2 1 1 2 0 0 2 0 0 2 Eritrea ERI 0 0 N.A. 0 0 2 0 0 2 0 0 N.A. 0 0 N.A. 0 0 N.A. Spain ESP 0 0 N.A. 1 1 2 2 2 1 0 0 1 1 2 1 0 0 1 Estonia EST 0 0 N.A. 0 0 2 3 2 1 0 0 1 0 0 2 0 0 2 Ethiopia ETH 0 0 N.A. 0 0 2 1 1 2 0 0 2 0 0 2 0 0 2 Finland FIN 1 1 N.A. 0 0 1 1 2 1 0 0 2 0 0 1 0 0 1 Fiji FJI 0 0 N.A. 0 0 1 1 3 1 0 0 1 0 0 2 0 0 1 France FRA 2 2 N.A. 3 2 1 3 1 1 2 1 1 0 0 1 0 0 1 FaroeIsland FRO 0 0 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. Micronesia, FSM 0 0 N.A. 0 0 2 1 3 2 0 0 1 0 0 1 0 0 2 Gabon GAB 0 0 N.A. 0 0 1 1 1 1 0 0 1 0 0 1 0 0 2 UnitedKingd GBR 3 3 N.A. 1 1 2 2 3 2 3 2 1 1 1 1 0 0 1 Georgia GEO 0 0 N.A. 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 Ghana GHA 0 0 N.A. 0 0 2 2 1 2 0 0 2 0 0 2 0 0 1 Gibraltar GIB 0 0 N.A. 0 0 N.A. 1 3 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. Guinea GIN 0 0 N.A. 0 0 1 0 0 1 0 0 1 3 3 2 0 0 2 Gambia,The GMB 0 0 N.A. 0 0 2 0 0 2 0 0 2 0 0 1 0 0 1 Guinea-Bissa GNB 0 0 N.A. 0 0 1 0 0 1 0 0 1 0 0 2 0 0 1 EquatorialG GNQ 0 0 N.A. 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 Greece GRC 2 2 N.A. 0 0 2 3 2 1 3 2 1 0 0 1 0 0 1 Grenada GRD 0 0 N.A. 0 0 1 1 2 1 0 0 1 0 0 1 1 3 2 Greenland GRL 0 0 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. Guatemala GTM 0 0 N.A. 0 0 2 2 2 1 0 0 1 0 0 2 1 2 2 Guam GUM 0 0 N.A. 0 0 N.A. 2 3 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. Guyana GUY 0 0 N.A. 0 0 1 1 2 2 0 0 2 0 0 1 1 2 1 HongKongSA HKG 1 1 N.A. 3 3 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. Honduras HND 0 0 N.A. 0 0 2 2 2 2 0 0 2 0 0 2 1 1 2 Croatia HRV 0 0 N.A. 0 0 2 2 3 1 0 0 1 0 0 1 0 0 1 Haiti HTI 0 0 N.A. 0 0 2 1 1 2 0 0 1 0 0 2 1 1 2 Hungary HUN 3 3 N.A. 0 0 2 3 2 1 0 0 1 0 0 1 0 0 2 Indonesia IDN 0 0 N.A. 1 1 1 2 1 1 0 0 1 0 0 2 0 0 1 India IND 0 0 N.A. 1 1 1 3 2 2 0 0 2 0 0 1 0 0 2 Ireland IRL 0 0 N.A. 1 2 2 2 3 1 0 0 1 0 0 1 0 0 1 Iran,Islami IRN 0 0 N.A. 0 0 2 2 2 2 2 2 2 0 0 N.A. 0 0 N.A. Iraq IRQ 0 0 N.A. 0 0 N.A. 2 2 1 0 0 1 0 0 1 0 0 1 Iceland ISL 0 0 N.A. 0 0 2 2 3 2 0 0 1 0 0 2 0 0 2 Israel ISR 0 0 N.A. 0 0 1 3 3 1 0 0 2 0 0 2 0 0 1 Italy ITA 2 2 N.A. 1 1 1 2 2 1 1 1 1 1 1 1 0 0 1 Jamaica JAM 1 1 N.A. 0 0 1 3 2 1 0 0 1 0 0 1 1 2 2 Jordan JOR 0 0 N.A. 0 0 1 2 3 2 2 3 1 0 0 2 0 0 1 Japan JPN 3 3 N.A. 0 0 1 2 2 1 0 0 1 0 0 1 0 0 1 Kazakhstan KAZ 0 0 N.A. 0 0 2 0 1 2 0 0 2 0 0 1 0 0 2 Kenya KEN 0 0 N.A. 0 0 2 1 1 2 0 0 2 0 0 2 0 0 1 KyrgyzRepub KGZ 0 0 N.A. 0 0 2 0 0 2 0 0 2 0 0 2 0 0 1 Cambodia KHM 0 0 N.A. 0 0 1 3 1 2 0 0 1 0 0 1 0 0 2 Kiribati KIR 0 0 N.A. 0 0 1 1 2 1 0 0 1 0 0 2 0 0 2 St.Kittsan KNA 0 0 N.A. 0 0 1 3 2 1 0 0 1 0 0 2 1 3 2 Korea,Rep. KOR 0 0 N.A. 1 1 2 3 2 2 2 3 1 0 0 2 0 0 2 Kuwait KWT 0 0 N.A. 1 2 1 2 3 2 2 3 1 0 0 2 0 0 1 LaoPDR LAO 0 0 N.A. 0 0 2 2 2 2 0 0 2 0 0 1 0 0 1 Lebanon LBN 0 0 N.A. 0 0 1 2 3 2 1 2 2 0 0 2 0 0 2 S.11

DiseaseSeverityandHealthExpenditureResponseDummy(Cont.) 1968Flu SARS H1N1 MERS Ebola Zika CountryName Country Mortality Case/Pop Health Expendi- Mortality Case/Pop Health Expendi- Mortality Case/Pop Health Mortality Case/Pop Health Mortality Case/Pop Health Mortality Case/Pop Health Code Rate ture Rate ture Rate Expenditure Rate Expenditure Rate Expenditure Rate Expenditure Liberia LBR 0 0 N.A. 0 0 1 0 0 2 0 0 2 3 3 2 0 0 1 Libya LBY 0 0 N.A. 0 0 1 2 2 2 0 0 N.A. 0 0 N.A. 0 0 N.A. St.Lucia LCA 0 0 N.A. 0 0 1 2 3 1 0 0 1 0 0 1 1 2 1 Liechtenstei LIE 0 0 N.A. 0 0 N.A. 0 2 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. SriLanka LKA 0 0 N.A. 0 0 1 3 2 2 0 0 1 0 0 1 0 0 2 Lesotho LSO 0 0 N.A. 0 0 1 1 2 2 0 0 2 0 0 2 0 0 1 Lithuania LTU 0 0 N.A. 0 0 2 3 1 1 0 0 1 0 0 1 0 0 2 Luxembourg LUX 0 0 N.A. 0 0 1 2 3 2 0 0 2 0 0 1 0 0 1 Latvia LVA 0 0 N.A. 0 0 1 0 0 1 0 0 1 0 0 1 0 0 2 MacaoSAR,C MAC 0 0 N.A. 1 3 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. Morocco MAR 0 0 N.A. 0 0 1 2 2 2 0 0 1 0 0 1 0 0 2 Monaco MCO 0 0 N.A. 0 0 1 0 2 1 0 0 1 0 0 1 0 0 1 Moldova MDA 0 0 N.A. 0 0 2 0 0 2 0 0 2 0 0 2 0 0 1 Madagascar MDG 0 0 N.A. 0 0 1 2 2 1 0 0 1 0 0 2 0 0 2 Maldives MDV 0 0 N.A. 0 0 1 2 2 1 0 0 2 0 0 2 0 0 2 Mexico MEX 0 0 N.A. 0 0 2 2 3 1 0 0 2 0 0 1 0 0 1 MarshallIsl MHL 0 0 N.A. 0 0 1 2 3 1 0 0 2 0 0 1 0 0 2 NorthMacedo MKD 0 0 N.A. 0 0 1 3 1 1 0 0 1 0 0 1 0 0 2 Mali MLI 0 0 N.A. 0 0 1 1 1 1 0 0 2 3 2 1 0 0 1 Malta MLT 0 0 N.A. 0 0 1 2 3 1 0 0 2 0 0 2 0 0 2 Myanmar MMR 0 0 N.A. 0 0 2 1 1 2 0 0 2 0 0 2 0 0 2 Montenegro MNE 0 0 N.A. 0 0 N.A. 3 2 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. Mongolia MNG 0 0 N.A. 1 3 1 2 3 1 0 0 2 0 0 2 0 0 1 Mozambique MOZ 0 0 N.A. 0 0 2 2 1 2 0 0 2 0 0 2 0 0 2 Mauritania MRT 0 0 N.A. 0 0 2 0 0 1 0 0 1 0 0 2 0 0 1 Mauritius MUS 0 0 N.A. 0 0 1 3 2 1 0 0 1 0 0 2 0 0 1 Malawi MWI 0 0 N.A. 0 0 2 1 1 2 0 0 2 0 0 2 0 0 2 Malaysia MYS 0 0 N.A. 3 2 1 3 2 1 3 2 1 0 0 2 0 0 1 Namibia NAM 0 0 N.A. 0 0 2 2 2 1 0 0 2 0 0 2 0 0 1 NewCaledoni NCL 0 0 N.A. 0 0 N.A. 2 3 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. Niger NER 0 0 N.A. 0 0 1 0 0 2 0 0 1 0 0 1 0 0 1 Nigeria NGA 0 0 N.A. 0 0 2 3 1 1 0 0 2 2 2 1 0 0 1 Nicaragua NIC 0 0 N.A. 0 0 1 2 3 2 0 0 2 0 0 2 1 2 2 Netherlands NLD 3 3 N.A. 0 0 2 3 2 1 1 2 1 0 0 1 0 0 1 Norway NOR 3 3 N.A. 0 0 2 2 3 2 0 0 2 0 0 2 0 0 1 Nepal NPL 0 0 N.A. 0 0 1 2 1 2 0 0 2 0 0 2 0 0 2 Nauru NRU 0 0 N.A. 0 0 N.A. 1 3 1 0 0 2 0 0 2 0 0 2 NewZealand NZL 0 0 N.A. 1 2 1 2 3 2 0 0 1 0 0 2 0 0 2 Oman OMN 0 0 N.A. 0 0 1 2 3 1 2 3 1 0 0 2 0 0 1 Pakistan PAK 0 0 N.A. 0 0 1 0 0 1 0 0 1 0 0 2 0 0 1 Panama PAN 0 0 N.A. 0 0 1 2 2 2 0 0 2 0 0 2 1 2 2 Peru PER 0 0 N.A. 0 0 1 2 3 2 0 0 2 0 0 1 1 1 2 Philippines PHL 0 0 N.A. 3 2 2 2 2 2 1 1 2 0 0 1 0 0 2 Palau PLW 0 0 N.A. 0 0 1 1 3 1 0 0 2 0 0 2 0 0 2 PapuaNewGu PNG 0 0 N.A. 0 0 1 1 1 1 0 0 2 0 0 2 0 0 1 Poland POL 0 0 N.A. 0 0 1 3 1 2 0 0 1 0 0 1 0 0 2 PuertoRico PRI 0 0 N.A. 0 0 N.A. 0 1 N.A. 0 0 N.A. 0 0 N.A. 3 3 N.A. Korea,Dem. PRK 0 0 N.A. 0 0 N.A. 1 1 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. Portugal PRT 2 2 N.A. 0 0 1 1 3 1 0 0 1 0 0 1 0 0 2 Paraguay PRY 0 0 N.A. 0 0 2 3 2 2 0 0 2 0 0 2 1 1 2 WestBankan PSE 0 0 N.A. 0 0 N.A. 2 3 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. FrenchPolyn PYF 0 0 N.A. 0 0 N.A. 2 3 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. Qatar QAT 0 0 N.A. 0 0 2 3 1 1 2 3 1 0 0 2 0 0 1 S.12

DiseaseSeverityandHealthExpenditureResponseDummy(Cont.) 1968Flu SARS H1N1 MERS Ebola Zika CountryName Country Mortality Case/Pop Health Expendi- Mortality Case/Pop Health Expendi- Mortality Case/Pop Health Mortality Case/Pop Health Mortality Case/Pop Health Mortality Case/Pop Health Code Rate ture Rate ture Rate Expenditure Rate Expenditure Rate Expenditure Rate Expenditure Romania ROU 1 1 N.A. 1 1 2 2 2 1 0 0 1 0 0 1 0 0 2 RussianFede RUS 0 0 N.A. 1 1 2 2 1 1 0 0 2 0 0 2 0 0 1 Rwanda RWA 0 0 N.A. 0 0 2 1 2 2 0 0 2 0 0 1 0 0 2 SaudiArabia SAU 0 0 N.A. 0 0 1 2 3 2 2 3 2 0 0 2 0 0 1 Sudan SDN 0 0 N.A. 0 0 2 3 1 2 0 0 2 0 0 2 0 0 1 Senegal SEN 0 0 N.A. 0 0 2 0 0 1 0 0 2 1 2 1 0 0 1 Singapore SGP 0 0 N.A. 3 3 1 2 3 1 0 0 1 0 0 1 0 0 2 SolomonIsla SLB 0 0 N.A. 0 0 2 3 1 1 0 0 1 0 0 2 0 0 1 SierraLeone SLE 0 0 N.A. 0 0 2 0 0 2 0 0 1 2 3 2 0 0 1 ElSalvador SLV 0 0 N.A. 0 0 1 3 2 1 0 0 1 0 0 1 1 1 1 SanMarino SMR 0 0 N.A. 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 Somalia SOM 0 0 N.A. 0 0 N.A. 1 1 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. Serbia SRB 2 2 N.A. 0 0 2 3 2 1 0 0 2 0 0 1 0 0 1 SouthSudan SSD 0 0 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. SaoTomeand STP 0 0 N.A. 0 0 2 3 3 2 0 0 2 0 0 1 0 0 2 Suriname SUR 0 0 N.A. 0 0 2 2 2 2 0 0 2 0 0 1 3 3 2 SlovakRepub SVK 0 0 N.A. 0 0 1 3 2 1 0 0 2 0 0 1 0 0 1 Slovenia SVN 0 0 N.A. 0 0 2 3 2 1 0 0 1 0 0 1 0 0 1 Sweden SWE 1 1 N.A. 1 3 1 2 2 1 0 0 1 0 0 2 0 0 1 Eswatini SWZ 0 0 N.A. 0 0 2 1 1 2 0 0 1 0 0 2 0 0 2 Seychelles SYC 0 0 N.A. 0 0 1 1 3 2 0 0 2 0 0 1 0 0 2 SyrianArab SYR 0 0 N.A. 0 0 2 3 2 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. TurksandCa TCA 0 0 N.A. 0 0 N.A. 0 3 N.A. 0 0 N.A. 0 0 N.A. 1 3 N.A. Chad TCD 0 0 N.A. 0 0 1 1 1 1 0 0 1 0 0 1 0 0 1 Togo TGO 0 0 N.A. 0 0 1 0 0 2 0 0 2 0 0 2 0 0 2 Thailand THA 0 0 N.A. 3 2 1 2 3 1 1 1 1 0 0 1 0 0 1 Tajikistan TJK 0 0 N.A. 0 0 2 0 1 2 0 0 2 0 0 2 0 0 2 Turkmenistan TKM 0 0 N.A. 0 0 2 0 0 1 0 0 2 0 0 2 0 0 2 Timor-Leste TLS 0 0 N.A. 0 0 1 0 1 1 0 0 1 0 0 1 0 0 1 Tonga TON 0 0 N.A. 0 0 1 3 2 1 0 0 2 0 0 2 0 0 2 Trinidadand TTO 0 0 N.A. 0 0 2 2 2 1 0 0 1 0 0 1 1 3 1 Tunisia TUN 0 0 N.A. 0 0 1 2 2 2 2 2 2 0 0 2 0 0 1 Turkey TUR 0 0 N.A. 0 0 2 3 1 1 3 1 1 0 0 2 0 0 2 Tuvalu TUV 0 0 N.A. 0 0 2 1 3 2 0 0 1 0 0 2 0 0 2 Tanzania TZA 0 0 N.A. 0 0 2 2 1 1 0 0 2 0 0 1 0 0 2 Uganda UGA 0 0 N.A. 0 0 2 1 1 2 0 0 2 0 0 1 0 0 1 Ukraine UKR 0 0 N.A. 0 0 2 0 0 2 0 0 2 0 0 1 0 0 2 Uruguay URY 0 0 N.A. 0 0 2 3 2 2 0 0 2 0 0 2 1 1 2 UnitedState USA 3 3 N.A. 1 2 2 3 2 2 1 1 2 2 1 2 1 1 2 Uzbekistan UZB 0 0 N.A. 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 St.Vincent VCT 0 0 N.A. 0 0 1 1 2 1 0 0 1 0 0 1 1 3 1 Venezuela,R VEN 0 0 N.A. 0 0 2 3 2 2 0 0 1 0 0 2 1 2 2 BritishVirg VGB 0 0 N.A. 0 0 N.A. 1 2 N.A. 0 0 N.A. 0 0 N.A. 1 3 N.A. VirginIslan VIR 0 0 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. 0 0 N.A. 1 3 N.A. Vietnam VNM 0 0 N.A. 2 3 2 3 1 2 0 0 2 0 0 1 0 0 2 Vanuatu VUT 0 0 N.A. 0 0 1 1 1 1 0 0 1 0 0 1 0 0 1 Samoa WSM 0 0 N.A. 0 0 1 2 3 2 0 0 1 0 0 1 0 0 2 Yemen,Rep. YEM 0 0 N.A. 0 0 2 2 2 1 3 2 2 0 0 1 0 0 N.A. SouthAfrica ZAF 3 3 N.A. 3 1 2 2 3 2 0 0 2 0 0 2 0 0 2 Zambia ZMB 0 0 N.A. 0 0 2 1 1 2 0 0 2 0 0 1 0 0 2 Zimbabwe ZWE 0 0 N.A. 0 0 N.A. 1 1 N.A. 0 0 1 0 0 2 0 0 2 PanelB:CorrelationbetweenDiseaseSeverityandHealthExpenditureAdjustment 1968Flu SARS H1N1 MERS Ebola Zika MortalityRate Case/Pop MortalityRate Case/Pop MortalityRate Case/Pop MortalityRate Case/Pop MortalityRate Case/Pop MortalityRate Case/Pop HealthSpendingAdjustment N.A. N.A. -0.0003 -0.1219 -0.0893 -0.0502 -0.119 -0.0282 0.1036 0.6779 -0.0128 -0.1313 Significance N.A. N.A. 0.9986 0.5529 0.2706 0.5297 0.5626 0.8911 0.7757 0.0312 0.9425 0.459 Obs N.A. N.A. 26 26 154 159 26 26 10 10 34 34 NOTE: PanelAdepictstheseveritydummyandhealthexpendituresadjustmentdummy,bycountryandwithineachdiseaseepisode. Fortheformer,weuseeithermortalityrateorcase-topopulationrate.0meansunaffected.Forthe1968Flu,1,2and3meansisolated,regionalandwidespread.Forthehealthexpendituresadjustmentdummy,wedividecountriesintothreegroups basedonthechangeinhealthexpenditureinthecrisisonsetyear,normalizedbythepreviousyear’sGDP.PanelBreportsthecross-countrycorrelationbetweenhealthspendingadjustmentand theseveritymeasure(mortalityrateorcasesrate)foreachepisodeinaffectedcountries. S.13

TableS.8TheEffectofHealthCrisesonGDPGrowth: TradeLinkages(SeverityofCrises) GDPgrowthrate% (1) (2) (3) (4) (5) (6) SamplePeriod: 1988-2018 Shock -2.22** -1.98** (1.03) (0.97) MortalityRate -2.07** -2.40* (0.86) (1.22) Cases/Pop -2.50*** -1.54*** (0.62) (0.55) ShocktoTradePartner -0.52** -1.11 -1.04 (0.23) (0.71) (0.65) TradeWeightedbyIndirectShock -1.00** (0.38) TradeWeightedbyMortalityRates -0.10 (0.07) TradeWeightedbyCases/Pop -0.14*** (0.02) Trade/GDP 0.19 0.17 0.24 0.32 0.23 0.21 (0.33) (0.33) (0.35) (0.38) (0.35) (0.34) DomesticCredit/GDP -0.73 -0.73 -0.76 -0.76 -0.76 -0.73 (0.46) (0.46) (0.49) (0.49) (0.48) (0.46) Log(Population) 0.12** 0.11** 0.11** 0.12** 0.11** 0.12** (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) Log(GDPpercapita) -0.20** -0.21** -0.23** -0.22** -0.22** -0.19* (0.09) (0.09) (0.10) (0.10) (0.09) (0.10) Recession -0.56 -0.57 -0.85* -0.83* -0.79* -0.47 (0.38) (0.38) (0.42) (0.44) (0.39) (0.32) BankingCrisis -1.54*** -1.54*** -1.45*** -1.44*** -1.46*** -1.52*** (0.37) (0.36) (0.41) (0.43) (0.40) (0.40) Constant 4.76*** 4.99*** 5.08*** 4.64*** 5.02*** 4.51*** (0.46) (0.51) (0.59) (0.50) (0.56) (0.45) Observations 4502 4502 4502 4502 4502 4502 WithinR2 0.065 0.066 0.051 0.045 0.055 0.061 DecadeFE Yes Yes Yes Yes Yes Yes CountryFE Yes Yes Yes Yes Yes Yes NOTE: ThedependentvariableistherealannualGDPgrowthrate. Shockdummyequalsoneforcountryiat onsetyeart,andzerootherwise.Shocktotradepartnerequalsto1ifoneofthecountry’stradingpartnerishit byahealthcrisis,and0otherwise. Theweighttradenetworkincolumn(2)isconstructedbymultiplyingthe shocktoacountry’stradingpartnerdummybytheshareofbilateraltradebetweenthesetwocountriesinthe country’stotaltrade(Tradeweightedbyindirectshock).Theweighttradenetworkincolumncolumn(4)and (6)isconstructedbymultiplyingthetradingpartner’sexpostmortalityrateorcasesnumberperpopulation by the trade share (trade weighted by morality rate and cases to population). Standard errors are corrected usingDriscollandKraay(1998)andreportedinparentheses. ∗,∗∗ and∗∗∗ indicatestatisticalsignificanceat the10%,5%,and1%level,respectively. S.14

S.3 Consumption and Investment We first estimate how the consumption and investment components of GDP were affected bypasthealthcrises. Therearemanyreasonswhyahealthcrisismightlowerconsumption andinvestment.S32 Forexample,withanincreaseinuncertaintyintheeconomy(seeBaker et al. (2020)), people might increase precautionary savings and thus reduce consumption andinvestmentplans. Theseeffectswillbeevenstrongerifpeopleexpectanegativeimpact of health crises on future income. The decline in spending could further strengthen the negativeimpactofcrisesontheproductionsideandslowdowntherecoveryphase. Figure S.5 reports the impulse response functions for the growth rates of private consumption expenditure and fixed investment. Private consumption growth in affected countries is 2.8% less than for unaffected countries in the onset year, with a 0.1% bounce-back oneyearlater. Perhapsnotsurprisingly,thedropinfixedinvestmentgrowthismuchlarger: 8.3%relativedeclineinaffectedcountriesintheonsetyear,withanegative1.0%oneyear later and a bounce-back only two years later. The sharp and persistent drop in investment andalargerbounce-backtwoyearslaterisconsistentwiththeobservedgreatervolatilityin investment, in this case likely due to the heightened uncertainty accompanying the health shockandrecession(Bakeretal.(2016)). Thedynamicsofconsumptionandinvestmentbehaviorduringthehealthcriseshelpus understandtheoutputdynamics. Whentheoutbreakoccurs,thenegativeshockelicitscuts in both consumption and investment expenditures. The effect on consumption is relatively short-lived — when output starts to recover in the first year, consumption resumes. For investment, it takes one more year to recover from the negative shock. Furthermore, the bounce-back in investment is not sufficient to offset the negative impact the health crisis causes. Asaresult,thehealthcrisiscanhaveapersistenteffectonoutput. S32Malmendier and Shen (2020) show that personal experiences from negative economic shocks “scar” consumer behavior in the long run. The authors do not directly address health crises per se, but instead showthathouseholdswhohavelivedthroughtimesofhighunemploymentspendsignificantlylessonfood and total consumption, after controlling for income, wealth, employment, demographics, and the current unemployment rate. Their model of experience-based learning is suggestive of a channel through which a shock like COVID could have persistent effects. Carroll et al. (2020) also study the negative impact of COVIDonconsumptionspending. S.15

FigureS.5TheEffectofHealthCrisesonConsumptionandInvestment PanelA:PrivateConsumptionGrowth PanelB:FixedInvestmentGrowth tnecreP 5 2 1- 4- 7- 01- 0 1 2 3 4 5 Years tnecreP 5 2 1- 4- 7- 01- 0 1 2 3 4 5 Years NOTE: Impulse response functions (IRF) are estimated based on the local projection method as in Jorda` (2005): git+H =αH i + ∑4 s=1 βH s git−s+∑4 s=0 δH s Dit−s+Xit+εit,withH=0,1,···,5,wheregitistheannualrealgrowthrateofprivateconsumptioninPanelA andfixedinvestmentinPanelBforcountryiatyeart,Ditisadummyvariableindicatingadiseaseeventhittingcountryiinyeart,with Xit includingcountry-levelcontrolssuchasTrade/GDP,DomesticCredit/GDP,populationandlogGDPpercapita. Wealsoincludea decadedummy,USrecessiondummy,abankingcrisisdummyandcountryfixedeffects. StandarderrorsarecorrectedusingDriscoll andKraay(1998).Onestandarderrorbandsareshown. S.4 Recovery in GDP growth: A Higher-frequency Look Our analysis using annual data and a large sample of countries suggests that bounce-back occurs in the year after the health shock. It is interesting to investigate by how much and howquicklybounce-backoccursusinghigherfrequencydata. Wehaveavailablequarterly GDP data from OECD, though only for 47 countries. See Table S.3 for details. Figure S.6 displaysthequarterlyGDPgrowthdistributionofaffectedandunaffectedcountriessideby side. Weplotthesedistributionsoverthreedifferentintervalsofthreeconsecutivequarters: (1)fromfivequartersbeforetotwoquartersbeforeonset,(2)centeredintheonsetquarter, and(3)fromthreequarterstosixquartersaftertheonsetquarter. Wechooseathreequarter windowbecausetheofficialdeclarationofahealthcrisisbyWHOtendstobeconservative (slow). This consideration does not affect identification in our annual sample nearly as muchasitcouldaffectthequarterlyidentification.S33 Theaverage,annualizedgrowthrateinthethreequarterwindowcenteredonthehealth crisisonsetis-0.4%foraffectedcountriesand2.8%forunaffectedcountries. Thisisinline with our estimates using annual data above. In quarters 2 to 5 before the health crisis, the S33Inaddition,notethatallcountriesinthequarterlysamplewereaffectedbyH1N1,alsounliketheannual sample. Thisweakensidentification. S.16

averagegrowthrateinaffectedcountriesisnotmuchdifferentthaninunaffectedcountries, norisitinquarters3to6afterthehealthshock. Thissuggeststhatthebounce-backofGDP growth is quick. Examining the magnitudes of these comparative responses, however, we see that bounce-back is not sufficient to restore the level of GDP within this time interval, consistentwiththeresultsfromtheannualsample. We also estimate panel regressions using quarterly GDP growth data. Table S.9 confirms that our main results hold in the quarterly data. Health crises shocks lower GDP growth in affected countries compared to unaffected countries, with an impact magnitude that is slightly larger than in the annual data. Furthermore, each individual health crisis contributes to this negative effect, with the exception of Ebola (see Table S.10). We also use the high, medium or low severity dummy to replace the shock dummy in Table S.11 or directly weight the health shock by the severity of each health crisis in Table S.12. We find that a more severe health crisis is associated with larger declines in GDP growth. Our last exercise is a placebo test of randomly picking a country-quarter to replace our quarterly shock dummy, as seen in Table S.13. The insignificant coefficient on the artificially constructedvariablesuggeststhatouridentificationisvalid. S.17

FigureS.6QuarterlyGDPGrowthDistribution ytisneD 51. 1. 50. 0 5- 0 5 01 (-5) to (-2) Quarters(Affected Countries) Mean = 2.63 ytisneD 51. 1. 50. 0 5- 0 5 01 Onset (-1) to (+1) Quarters(Affected Countries) Mean = -.44 ytisneD 51. 1. 50. 0 5- 0 5 01 (+3) to (+6) Quarters(Affected Countries) Mean = 3.36 ytisneD 51. 1. 50. 0 5- 0 5 01 (-5) to (-2) Quarters(Unaffected Countries) Mean = 2.91 ytisneD 2. 51. 1. 50. 0 5- 0 5 01 Onset (-1) to (+1) Quarters(Unaffected Countries) Mean = 2.83 ytisneD 2. 51. 1. 50. 0 5- 0 5 01 (+3) to (+6) Quarters(Unaffected Countries) Mean = 3.32 NOTE:Therealquarterlyyear-over-yearseasonallyadjustedGDPgrowthratedistributionfortheaffectedandunaffectedcountrygroups.0representsthequarterwhenWHOdeclaresahealth crisishitsacountry. S.18

TableS.9TheEffectofHealthCrisesonRealQuarterlyGDPGrowth QuarterlyGDPgrowthrate(YoY)% (1) (2) (3) (4) SamplePeriod: 1960-2018 1990-2018 AllEvents AllEvents AllEvents WithoutH1N1 Shock(Q) -3.73*** -3.80*** -2.32*** -0.98*** (1.23) (1.16) (0.52) (0.23) ConsensusForecast(Q) 1.37*** 1.35*** (0.22) (0.21) Trade/GDP 0.03 -0.03 0.57 0.48 (0.79) (0.80) (1.21) (1.16) DomesticCredit/GDP -1.81*** -1.94*** -1.20 -1.20 (0.56) (0.68) (1.35) (1.33) Log(Population) -0.25*** -0.31* -0.00 -0.01 (0.09) (0.17) (0.08) (0.08) Log(GDPpercapita) 0.59*** 0.71* 0.08 0.10 (0.18) (0.37) (0.23) (0.22) Recession -1.48** -1.85* -1.36** -1.29** (0.70) (1.06) (0.61) (0.63) BankingCrisis(Q) 0.29 0.52 -0.16 -0.26 (1.14) (1.25) (0.90) (0.90) Constant 3.38*** 3.48*** -1.59 -1.48 (0.81) (1.05) (1.67) (1.63) Observations 5218 3959 1240 1222 AdjustedR2 0.126 0.108 0.378 0.346 DecadeFE Yes Yes Yes Yes CountryFE Yes Yes Yes Yes NOTE:ThedependentvariableisrealquarterlyGDPgrowthrate,annualized.Thesampleperiodforcolumn(1)is1960-2018whilethe sampleperiodforcolumn(2)-(4)is1990-2018.Theshockdummyequalsoneforcountryihitbyahealthcrisisatonsetyeart,andzero otherwise.Incolumns(1)-(3),weincludeallsixhealthcriseswhilecolumn(4)excludesH1N1andthe1968Flu.Countryanddecade fixedeffectsareincluded. AllstandarderrorsarecorrectedusingDriscollandKraay(1998)andreportedinparentheses. ∗,∗∗and∗∗∗ indicatestatisticalsignificanceatthe10%,5%,and1%level,respectively. S.19

TableS.10TheEffectofHealthCrisisonRealQuarterlyGDPGrowth,byCrisis QuarterlyGDPgrowthrate(YoY)% (1) (2) (3) (4) SamplePeriod: 1960-2018 1990-2018 AllEvents AllEvents AllEvents WithoutH1N1 EBOLA 0.40 0.30 -0.21 -0.21 (0.35) (0.35) (0.26) (0.27) H1N1 -6.39*** -6.18*** -3.59*** (1.01) (1.24) (0.86) MERS -0.86*** -0.79*** -0.87*** -0.85*** (0.27) (0.27) (0.24) (0.23) SARS -1.34*** -1.55*** -1.45*** -1.46*** (0.39) (0.36) (0.28) (0.27) Zika -2.62*** -2.62*** -0.93*** -0.94*** (0.41) (0.40) (0.27) (0.27) Hkflu -0.77* (0.44) ConsensusForecast(Q) 1.34*** 1.35*** (0.22) (0.22) Trade/GDP 0.01 -0.06 0.53 0.48 (0.78) (0.79) (1.20) (1.16) DomesticCredit/GDP -1.76*** -1.90*** -1.22 -1.20 (0.56) (0.68) (1.34) (1.33) Log(Population) -0.25*** -0.32* -0.01 -0.01 (0.09) (0.17) (0.08) (0.08) Log(GDPpercapita) 0.60*** 0.72* 0.09 0.10 (0.18) (0.37) (0.23) (0.22) Recession -1.36** -1.69 -1.29** -1.31** (0.68) (1.06) (0.61) (0.63) BankingCrisis(Q) 0.21 0.42 -0.23 -0.26 (1.13) (1.25) (0.90) (0.90) Constant 3.36*** 3.42*** -1.47 -1.46 (0.83) (1.08) (1.67) (1.63) Observations 5218 3959 1240 1222 AdjustedR2 0.136 0.120 0.384 0.347 DecadeFE Yes Yes Yes Yes CountryFE Yes Yes Yes Yes NOTE: ThedependentvariableisrealquarterlyGDPgrowthrate,annualized. Thesampleperiodforcolumn(1)is1960-2018while thesampleperiodforcolumns(2)-(4)is1990-2018. Countryanddecadefixedeffectsareincluded. Allstandarderrorsarecorrected usingDriscollandKraay(1998)andreportedinparentheses.∗,∗∗and∗∗∗indicatestatisticalsignificanceatthe10%,5%,and1%level, respectively. S.20

TableS.11TheEffectofHealthCrisesonRealQuarterlyGDPGrowth,bySeverity QuarterlyGDPgrowthrate(YoY)% (1) (2) (3) (4) (5) (6) SamplePeriod: 1960-2018 1990-2018 1960-2018 1990-2018 HighMortalityRate -4.77*** -5.09*** -2.72*** (1.36) (1.25) (0.75) MediumMortalityRate -5.17*** -4.93*** -3.66*** (1.27) (1.31) (1.06) LowMortalityRate -2.45*** -2.60*** -1.24*** (0.88) (0.83) (0.27) HighCases/Pop -3.65*** -3.82*** -2.56*** (1.20) (1.23) (0.90) MediumCases/Pop -4.43*** -4.40*** -2.57*** (1.28) (1.19) (0.47) LowCases/Pop -3.02** -3.09*** -1.72*** (1.23) (1.11) (0.40) ConsensusForecast(Q) 1.36*** 1.37*** (0.22) (0.22) Trade/GDP 0.05 -0.02 0.56 0.03 -0.03 0.57 (0.80) (0.81) (1.21) (0.79) (0.80) (1.22) DomesticCredit/GDP -1.80*** -1.93*** -1.23 -1.81*** -1.93*** -1.19 (0.57) (0.68) (1.35) (0.56) (0.68) (1.35) Log(Population) -0.25*** -0.31* -0.00 -0.25*** -0.31* -0.00 (0.09) (0.17) (0.08) (0.09) (0.17) (0.08) Log(GDPpercapita) 0.59*** 0.71* 0.09 0.60*** 0.72* 0.08 (0.18) (0.37) (0.23) (0.18) (0.37) (0.23) Recession -1.45** -1.81* -1.33** -1.47** -1.85* -1.36** (0.69) (1.06) (0.60) (0.69) (1.06) (0.61) BankingCrisis(Q) 0.28 0.50 -0.18 0.29 0.52 -0.16 (1.13) (1.25) (0.89) (1.14) (1.25) (0.90) Constant 3.36*** 3.46*** -1.57 3.37*** 3.48*** -1.59 (0.81) (1.06) (1.67) (0.81) (1.05) (1.68) Observations 5218 3959 1240 5218 3959 1240 AdjustedR2 0.128 0.111 0.382 0.126 0.109 0.378 DecadeFE Yes Yes Yes Yes Yes Yes CountryFE Yes Yes Yes Yes Yes Yes NOTE: Thedependentvariableincolumn(1)-(6)isrealquarterlyGDPgrowthrate,annualized. Thesample period for columns (1) and (4) is 1960-2018 while the sample period for columns (2)-(3) and (5)-(6) is 1990-2018. Countryanddecadefixedeffectsareincluded. Allstandarderrorsareclusteredcorrectedusing Driscoll and Kraay (1998) and reported in parentheses. ∗, ∗∗ and ∗∗∗ indicate statistical significance at the 10%,5%,and1%level,respectively. S.21

TableS.12TheEffectofHealthCrisesonRealQuarterlyGDPGrowth: WeightedbySeverityofCrises QuarterlyGDPgrowthrate(YoY)% (1) (2) (3) (4) (5) (6) SamplePeriod: 1960-2018 1990-2018 1960-2018 1990-2018 MortalityRate -4.67* -4.65* -4.33** (2.68) (2.46) (1.66) Cases/Pop -8.36*** -8.18*** -2.29** (1.67) (2.01) (1.07) ConsensusForecast(Q) 1.41*** 1.40*** (0.24) (0.24) Trade/GDP 0.09 0.06 0.70 0.07 0.03 0.69 (0.83) (0.85) (1.30) (0.82) (0.84) (1.31) DomesticCredit/GDP -1.84*** -1.98*** -1.13 -1.81*** -1.95*** -1.15 (0.59) (0.71) (1.36) (0.58) (0.70) (1.36) Log(Population) -0.26*** -0.32* -0.01 -0.26*** -0.32* -0.01 (0.09) (0.18) (0.08) (0.09) (0.17) (0.08) Log(GDPpercapita) 0.60*** 0.71* 0.08 0.60*** 0.72* 0.09 (0.18) (0.37) (0.23) (0.18) (0.37) (0.23) Recession -1.55** -1.98 -1.43** -1.50* -1.90 -1.40** (0.78) (1.20) (0.67) (0.77) (1.18) (0.67) BankingCrisis(Q) 0.42 0.67 -0.04 0.38 0.62 -0.06 (1.18) (1.32) (0.96) (1.18) (1.31) (0.96) Constant 3.32*** 3.46*** -1.86 3.31*** 3.43*** -1.82 (0.83) (1.09) (1.80) (0.84) (1.10) (1.79) Observations 5214 3959 1240 5214 3959 1240 AdjustedR2 0.11 0.08 0.36 0.11 0.09 0.36 DecadeFE Yes Yes Yes Yes Yes Yes CountryFE Yes Yes Yes Yes Yes Yes NOTE: Thedependentvariableincolumn(1)-(6)isrealquarterlyGDPgrowthrate,annualized. Thesample period for columns (1) and (4) is 1960-2018 while the sample period for columns (2)-(3) and (5)-(6) is 1990-2018. Countryanddecadefixedeffectsareincluded. Allstandarderrorsareclusteredcorrectedusing Driscoll and Kraay (1998) and reported in parentheses. ∗, ∗∗ and ∗∗∗ indicate statistical significance at the 10%,5%,and1%level,respectively. S.22

TableS.13TheEffectofHealthCrisesonRealQuarterlyGDPGrowth: PlaceboTest QuarterlyGDPgrowthrate(YoY)% (1) (2) (3) (4) SamplePeriod: 1960-2018 1990-2018 AllEvents AllEvents AllEvents WithoutH1N1 Shock(Q) -0.27 -0.64 0.02 -0.07 (0.46) (0.53) (0.35) (0.32) ConsensusForecast(Q) 1.42*** 1.35*** (0.24) (0.21) Trade/GDP 0.10 0.06 0.69 0.49 (0.83) (0.86) (1.30) (1.16) DomesticCredit/GDP -1.85*** -1.99*** -1.15 -1.20 (0.60) (0.71) (1.37) (1.33) Log(Population) -0.26*** -0.32* -0.01 -0.01 (0.09) (0.18) (0.08) (0.08) Log(GDPpercapita) 0.60*** 0.72* 0.09 0.10 (0.18) (0.37) (0.24) (0.23) Recession -1.57* -2.00 -1.44** -1.28** (0.80) (1.22) (0.68) (0.64) BankingCrisis(Q) 0.45 0.71 -0.03 -0.26 (1.19) (1.33) (0.97) (0.90) Constant 3.33*** 3.47*** -1.87 -1.50 (0.84) (1.10) (1.81) (1.64) Observations 5218 3959 1240 1222 AdjustedR2 0.105 0.082 0.358 0.344 DecadeFE Yes Yes Yes Yes CountryFE Yes Yes Yes Yes NOTE: Thedependentvariableincolumn(1)-(4)isrealquarterlyGDPgrowthrate,annualized. Thesample period for column (1) is 1960-2018 while the sample period for columns (2)-(4) is 1990-2018. The shock variable is randomly generated. Country and decade fixed effects are included. All standard errors are clustered corrected using Driscoll and Kraay (1998) and reported in parentheses. ∗, ∗∗ and ∗∗∗ indicate statisticalsignificanceatthe10%,5%,and1%level,respectively. S.23