The Vaccine Boost: Quantifying the Impact of the COVID-19 Vaccine Rollout on Measures of Activity

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) The Vaccine Boost: Quantifying the Impact of the COVID-19 Vaccine Rollout on Measures of Activity Maria D. Tito and Ashley Sexton 2022-035 Please cite this paper as: Tito, Maria D., and Ashley Sexton (2022). “The Vaccine Boost: Quantifying the Impact of the COVID-19 Vaccine Rollout on Measures of Activity,” Finance and Economics Discussion Series 2022-035. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2022.035. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) 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 Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

The Vaccine Boost: Quantifying the Impact of the COVID-19 Vaccine Rollout on Measures of Activity* MariaD.Tito† AshleySexton‡ April26,2022 Abstract Thispaperinvestigatestheimpactofvaccineadministrationonthreemaindimensions of activity: spending, mobility, and employment. Our investigation combines two parts. First, we exploit the variation in vaccine administration across states. In panelregressionsthatincludealargesetofcontrols,wefindthattherollouthasasignificantimpactonspending,whiletheresultsonmobilityandemploymentaremixed. Second,toaddressconcernsofendogeneity,welookattheimpactofvaccinelotteries on spending. Using a dynamic event design setting, we find that lotteries have significantlyboostedvaccinationratesaboutaweekafterannouncement, withaneffect that lasts over the next several days and increases new vaccinations between 3.5 and 5percent. Thisboostinvaccinationrates,inturn,translatesintoasignificantincrease inretailspending,whichislargerandsomewhatmorepersistentthanwhatwedocumentinourstate-levelpanelregressions. Alltold,ourfindingsimplythatthevaccine rolloutadded,onaverage,0.5percentagepointtoGDPgrowthin2021. Keywords: COVID-19VaccineRollout,EconomicActivity *Wewouldliketothanktheparticipantstothe2021ICEA“PublicPolicyLessons”conferenceandto the2022SGEconferencefortheirinsightfulsuggestions.Theviewsexpressedinthisarticlearethoseofthe authorsanddonotnecessarilyreflectthoseoftheFederalReserveSystem. †FederalReserveBoard.Contact:maria.d.tito@frb.gov. ‡CollegeBoard

Several measures of economic activity have shown improvement since the start of the COVID-19 vaccine rollout. However, over the same period, other factors—such as the course of the pandemic and the associated policy interventions—have influenced the response in activity; our analysis isolates the contribution of vaccine administration from otherconfounders. Withafocusonthreemaindimensionsofactivity—spending,mobility,andemployment— our investigation combines two parts. First, we exploit the variation in new vaccine administration across states. In panel regressions that include a large set of controls—such as day- and state-quarter fixed effects; new vaccine distribution; the trends in new cases, hospitalizations,anddeaths;temperatures;theemploymentrate;demographiccharacteristics; and the Oxford Stringency Index, a measure of policy responses to the COVID-19 outbreaks—wefindthattherolloutisassociatedwithasignificantincreaseinretailspending. Theimpactonspending,however,takessometimetomaterialize,roughlyconsistent with the timing of vaccination to achieve a high degree of effectiveness: In our analysis, the effect of new vaccine administration on retail spending is significant only around 20 daysafterreceivingtheshot,withanimpactthatpersistsoverthefollowing10days.1 The impact on restaurant spending occurs at an even longer lags (around 50 days) from the initialvaccination. Incontrast,theresultsformobilityandemploymentaremixed. Inasecondpartofourempiricalanalysis,werelyontheintroductionofvaccinelotteriestoinstrumentforvaccineuptakeand,thus,controlforallotherfactorsthatcouldhave influenced vaccinations or affected economic activity across states. These lotteries were specifically designed to boost vaccination rates and are an ideal instrument for our analysis; in fact, these state-level interventions were unexpected before the official announcement and offer variation across states and over time. Using a dynamic event design setting, wefindthatlotterieshavesignificantlyboostedvaccinationratesaboutaweekafter announcement,withaneffectthatlastsoverthenextseveraldaysandincreasesnewvaccinationsbetween3.5and5percentacrosslotteryadopterscomparedwithstateswithout 1Whilefullprotectionisachievedaround14daysafterreceivingtheseconddose,earlystudieshave indicatedahighdegreeofeffectivenessofpartialimmunization(ofaround80percentforbothModerna’s andPfizer’svaccines)14daysafterthefirstdosebutbeforetheseconddose.See,forexample,Thompson etal.[2021]. 1

lotteries–that is, either never adopters or not-yet adopters. This effect, in turn, translates intoasignificantboosttoretailspending,withanimpactthatislargerandsomewhatmore persistentcomparedwithwhatwefindinourstate-levelanalysis. Inparticular,lookingat magnitudes, we find that a one standard deviation increase in vaccinations explains 2.25 standard deviations increase in retail spending about 30 days post-vaccination and over thefollowingtwoweeks. Thiseffectisconsistentwithanincreaseinretailspendingatthe monthlyrateof0.27percent. Finally, to quantify the impact on overall economic activity, we map the effects we documented on retail spending to the effect on GDP. Our estimates imply that vaccine administration,throughretailsales,boostedGDPgrowth,onaverage,0.5percentagepoint (pp)in2021. WhileRobertsonetal.[2021a]estimatealotterycostpermarginalvaccination of 55 USD, our estimates of the impact of vaccinations attribute an increase in GDP of about400billionUSDor1500USDpervaccination,suggestinganimportantcontribution ofthoseinterventionsfortherecovery. Our work contributes to the vast literature on evaluating public policy interventions. Our paper is most closely related to Hansen and Mano [2021], who document the impact of vaccinations on economic activity at the county level through an instrumental variable strategythatreliesonlocalpharmacydensity. Whilewhattheyfindisroughlycomparable to our state-level estimates of vaccinationon economic activity, their empiricalstrategy is not robust to some endogenous factors—such as demand shocks, which affected the local reallocation of vaccine doses across counties, or urban density, which is significantly correlated with the shape of the recovery. In contrast, lotteries and other monetary incentives are likely to be exogenous to all other sources of state-level variation, with even trends in vaccinations before their introduction not significantly different between states thatadoptedalotteryandstatesthatdidnotorthosethatreliedonlesssignificantmonetaryincentives. 2

1 Data Ouranalysiscombinesseveralsourcesofdailystate-leveldata. Toquantifytheprogressin vaccineadministrationacrossstates,wedrawontheU.SCentersforDiseaseControland Prevention’sCOVID-19vaccinationsbyjurisdictiondata. Thespendingindicatorsinourinvestigationarebasedonsector-leveldatafromFiserv andSafeGraph. Inparticular, Fiserv, oneofthelargestcardintermediariesinthecountry, tracks consumer spending using card transaction data; while each observation in the initial data corresponds to a single card swipe—such as debit, credit, or gift card—data are then aggregated to the state (and national) level, the database we have access to.2 Safe- Graph aggregates data on customers’ visits to select businesses using cellphone GPS signals. Inadditiontostatevariation, bothdatasourcesprovideinformationonspendingat the sector-level; in our analysis, we will focus on the retail and restaurant sectors, which commonlycharacterizeretailspending. Toquantifymobilitypatterns,werelyontwoindicators: theAppledrivingindexand theINRIXindexofpassengerdistancetraveled. MeasuresofemploymentconditionscomefromHomebase,aproviderofclock-in/clockouttrackingsoftwarefocusedonsmallbusinesses.3 Finally,wecomplementourmainindicatorswithadditionalinformationfromtheNew York Times COVID-19 cases and deaths database, the Health and Human Services’ data on hospitalization, the Oxford stringency index, the National Oceanic and Atmospheric Administration’s National Climatic Data, and employment and demographics data from theBureauofLaborStatistics’CurrentPopulationSurvey(CPS). Because of data availability, our sample covers the period between February 10, 2021 andMarch19,2022. 2Formoredetails,seeAladangadyetal.[2019]. 3FormoredetailsonHomebase,seeCraneetal.[2020]. 3

2 Empirical Analysis 2.1 State-LevelAnalysis In the first part of our exploration, we leverage the state-level variation in the vaccine rollout to evaluate its effects on various economic outcomes. In particular, our baseline specificationrelatesthenumberofnewpeopleinstate s thatreceivedthefirstdoseofthe vaccineatdayt−j,NewAdm. ,withcurrentindicatorsofactivity, s,t−j y = β +β NewAdm. +γX +d +d +d +ε , j = 0,...,60 (1) st 0 1,j s,t−j st sq s t st where y denotes measures of spending, mobility, or employment. With a specification st thatinvestigatestheimpactofvaccinationsoverdifferentlagsfromreceivingthefirstdose of a COVID vaccine—and lags ranging from the contemporaneous impact up to 60 days after—weaimtodescribethehigh-frequencybehavioralresponseofindividualsaftervaccinations, an objective akin to constructing empirical impulse response functions. The evolution of individual behavior along various economic measures is, then, captured by β j = 0,...,60,thecoefficientsofinterest. Thetimehorizonofourevaluation—awindow 1,j of about two months after the first dose—tends to largely capture the full high-frequency impactofvaccinationsonvariousindicators.4 To separately identify the interaction between vaccinations and economic indicators fromotherconfounders,ourspecificationcontrolsforseveralfactors. First,werelyonnew vaccine distribution to capture supply shocks in vaccine availability. Second, we include newcases,newhospitalizations,andnewdeathstoaccountfortheeffectofthepandemic on vaccine administration and economic activity. Third, we add the Oxford stringency index, a composite measure of non-pharmaceutical interventions that records policy responses to the course of the disease. Fourth, we control for heating and cooling degree days because of the interactions among health, behavioral outcomes, and weather variables. Fifth, we draw on CPS data to include labor market—the employment rate—and demographics—share of male population, share of white non-hispanic population, share 4Extendingthewindowsforlongertimehorizonsdoesnotbringforthnovelresults,anditismorelikely tocapturealsotheimpactofotherfactors. 4

ofpopulationwithhighschooldegreeorless,shareofyoung(aged24orless)andprimeage(25-64)population—characteristics,which,accordingtoRobertsonetal.[2021b],have beenimportantcorrelatesofvaccinehesitance;wealsorelyonstate-quarterdummies,d , sq tocaptureotherrelevantstate-levelfeaturesvaryingacrossstatesandquarters. Finally,we include day- and state-fixed effects to absorb, respectively, common shocks across states andtime-invariantstate-levelcharacteristics—suchasdifferentattitudestowardsvaccinations,geographiccharacteristics,andsimilarfactors. Results Figures 1-3 illustrate the impact of new vaccinations on our main indicators of economic activities over the following 60 days after receiving the first dose. In those figures, each marker denotes the coefficient from model (1) for a specific lag j, while the shaded areas represent the 95 percent confidence interval. For example, looking at the left panel of figure 1, the impact of new vaccinations on retail spending is negative and significant until around15daysafterreceivingthefirstdoseofanyvaccine. Afterward,thecoefficientestimategraduallyincreases,peakingaround30dayspost-vaccination—orbeforeindividuals achieve full immunization. The magnitude of the coefficient at 30 days implies that, everythingelseequal,a1percentincreaseinthedependentvariableisassociatedwitha11/2 percent increase in retail spending around 30 days after receiving the first dose. The impact of vaccination on retail spending remains significant for around 10 days and, then, becomesinsignificantthroughtheendoftheperiodinouranalysis. Figure1: Vaccinations: ImpactonSpending 5

Table 1 summarizes the cumulative impact on retail spending, aggregating across differentlags. Inparticular,weestimatetheeffectofvaccinationsonspendingusingaspecificationthatincludesalllags 60 ∑ y = β + β NewAdm. +γX +d +d +d +ε (2) st 0 1,j s,t−j st sq s t st j=0 and we report point estimates associated with linear combinations of coefficients over specifiedperiods. Column(1)referstoretailspending,andwecontinuetoreporttheelasticityofspendingtochangesinvaccinations—thatis,thecoefficientincolumn(1)implies that an increase of new vaccine administration of 1 percent is associated with 4 percent higherspendingforabout10days(fromthe21stthroughthe30thdaypost-vaccination).5 Quantifyingtheimpactofvaccinationsintermsofstandarddeviationsoftheexplanatory variable—for comparability across various measures—implies that a one standard deviationincreaseinvaccinationsisassociatedwitha39percentofastandarddeviationhigher spendingatretailbusinesses21dayspost-vaccinationandoverthenext10days. Columns(2)-(4)focusonspendingatretailestablishmentsthatwereclassifiedasnonessential during the more acute phase of the pandemic—and thus, more likely to be subject to mandated closures during that time. While the direct indicators of spending at clothing and at sports and hobby stores—based on the Fiserv data and shown in columns (2) and (3)—appear to have not been significantly affected by the vaccine rollout, we find a significant impact on visits at those types of establishments (column (4)). The effect that we reportincolumn(4),however,isbasedonalinearcombinationofthecoefficientsbetween 15and25dayspost-vaccination;thepointestimatewouldgraduallybecomeinsignificant ifweweretoextendthewindowoftheanalysismuchfurther. Thepatternoflittlechanges inspendingandasignificantlyincreaseinvisitssuggeststhatasubstitutioninthetypeof spending; this interpretation is confirmed by the results in table A1, where the indicator forvisitsatnonessentialestablishmentsbecomesinsignificantaftercontrollingforchange intotalspendingatnon-storeretailers. 5Thecoefficientestimateisrobusttothechoiceofsimilarwindowsaroundthe30thdaypostvaccination. 6

Table1: EffectsoftheVaccineRolloutonRetailSpending (1) (2) (3) (4) (5) (6) Retail Nonessential Restaurants VARIABLES Spending Clothing Sports&Hobby Visits Spending Visits NewAdm. 4.394*** -0.181 0.006 6.361*** 0.017** 5.260*** (1.516) (3.550) (0.033) (3.002) (0.008) (2.204) OtherControls1 y y y y y y State-QuarterFE y y y y y y DayFE y y y y y y StateFE y y y y y y Obs. 2,676 2,510 1,701 1,119 2,676 1,119 R-squared 0.553 0.487 0.681 0.765 0.793 0.680 NumberofStates 51 49 36 51 51 51 Source:Fiserv,Inc.,SafeGraph,CDC,CPS,andNOAA. 1Othercontrolsincludenewvaccinedistribution;newcases,hospitalizations,anddeaths;heating andcoolingdegreedays;theemploymentrate;demographicscharacteristics;andtheOxford stringencyindex. RetailSales:Percentagechangeinretailsalesspendingrelativeto2019. Grocery,Spending:Percentagechangeingroceryspending(NAICS445)relativeto2019. Clothing:Percentagechangeinspendingatapparelstores(NAICS448)relativeto2019. Sports&Hobby:Percentagechangeinspendingatsportinggoods,hobby,book,andmusic stores(NAICS451)relativeto2019. Nonessential,Visits:Percentagechangeinvisitstononessentialretailstoresrelativeto2019. Restaurants,Spending:Percentagechangeinrestaurantspending(NAICS722)relativeto2019. Restaurants,Visits:Percentagechangeinvisitstorestaurantsrelativeto2019. NewAdm.:Log-numberof7-daymovingaverageofnewdailyvaccineadministration,cumulatedeffectaftervaccination. Legend:∗∗∗significantat1%,∗∗at5%,∗at10%. Notes:StateFEregressions.Pointestimatesforthemainexplanatoryvariablearebasedonlinearcombinationsofcoefficients:incolumns(1)-(3),wereportthelinearcombinationfromthe 21st-daythroughthe30th-daylag;incolumn(5),wereportthelinearcombinationfromthe51stdaythroughthe60thday;incolumns(4)and(6),wereportthelinearcombinationfromthe15thdaythroughthe25th-daylag.Robuststandarderrors,clusteredatthestatelevel,arereportedin parenthesis. 7

Figure2: Vaccinations: ImpactonMobility The right panel of figure 1 focuses on spending in the restaurant sector (NAICS 722), which has been more sensitive to the evolution of the pandemic given its high-contactintensity nature. After being significantly negative in the first few days post-vaccination, the coefficient estimates in the figure show a slight upward trend and become significant only around 50 days post-vaccination and through the end of our evaluation window. Relatedly, thecumulativeimpactonrestaurantspendingshownincolumn(5)oftable(1) ispositiveovertheperiodbetweenthe51stand60th-daypost-vaccination,butitisrather small, implying an increase of 0.1 percent of a standard deviation per standard deviation increaseinnewvaccineadministration. LookingattheSafeGraphmeasureforthesector, we continue to document a significant impact on visits at restaurant establishments over 10 days after the 15th day post-vaccination; this effect, however, continues to be related to some substitution patterns from online spending, and, in fact, it is not robust to the inclusionofthechangeinspendingatnon-storeretailers(column(2),tableA1). Turningtomobility, ourresultslookmoremixed. Whiletheleftpaneloffigure2suggests a boost in mobility according to the INRIX index of passenger distance traveled, even upon receiving the first dose of a COVID vaccine, the point estimates for the Apple mobility index are, at first, negative and significant, but become insignificant around 20 dayspost-vaccination. Asimilardichotomyappearsinthecumulativeeffects,reportedin columns(1)and(2)oftable2. Finally,employmentindicatorsappearlargelyunaffectedbythevaccinerollout. Interestingly, the point estimates for either hours worked (left panel, figure 3) or the number 8

Table2: EffectsoftheVaccineRolloutonMobilityandEmployment (1) (2) (3) (4) Mobility Employment Variable INRIX Apple HoursWorked BusinessOpen NewAdm. 3.057*** -15.706*** 1.445 0.119 (1.330) (6.746) (1.203) (0.612) OtherControls1 y y y y State-QuarterFE y y y y DayFE y y y y StateFE y y y y Observations 942 2,676 2,676 2,676 R-squared 0.552 0.879 0.800 0.837 NumberofStates 51 51 51 51 Source:INRIX,Apple,HouseholdPulseSurvey,Homebase,CDC,andNOAA. 1Othercontrolsincludenewvaccinedistribution;newcases,hospitalizations, anddeaths;heatingandcoolingdegreedays;theemploymentrate;demographicscharacteristics;andtheOxfordstringencyindex. INRIX:Percentagechangeinthe7-daymovingaverageofpassengerdistance traveled. Apple:Percentagechangeinthe7-daymovingaverageofthedrivingindex. Hoursworked:Percentagechangeinthenumberoftotalhoursworkedrelativeto2019insmallbusinessestablishments. BusinessOpen:Percentagechangeinthenumberofopenbusinessesrelative to2019insmallbusinessestablishments. NewAdm.:Log-numberofthe7-daymovingaverageofnewdailyvaccine administration,cumulatedeffect. Legend:∗∗∗significantat1%,∗∗at5%,∗at10%. Notes:StateFEregressions.Pointestimatesforthemainexplanatoryvariable arebasedonlinearcombinationsofcoefficients:incolumn(1),wereportthe linearcombinationfromthe1st-daythroughthe14thday;incolumn(2),wereportthelinearcombinationthroughthe60thday;incolumns(3)-(4),wereport thelinearcombinationfromthe5th-daythroughthe14thday.Allspecificationsincludestateanddayfixedeffects,newvaccinedistribution;newcases, hospitalizations,anddeaths;heatingandcoolingdegreedays;andtheOx- 9 fordstringencyindex.Robuststandarderrors,clusteredatthestatelevel,are reportedinparenthesis.

Figure3: Vaccinations: ImpactonEmployment ofopenbusinesses(rightpanel)arenegative,althoughlargelyinsignificant—withtheexceptionofafewdaysforthenumberofopenbusinessestowardstheendofourwindow. The negative relationship between the vaccine rollout and measures of employment in the first few days after vaccination is suggestive of the use of sick leave due to possible side effects after receiving the vaccine, but the fact that the negative sign of the coefficient persists throughout our evaluation window is somewhat puzzling. Looking over a windowbetweenthe5thandthe14thdaypost-vaccination,thecumulativeeffectsofvaccination on hours worked or the number of businesses open—shown in columns (4) and (5),respectively—areactuallyinsignificant,andtheyshowamoreintuitivepositivesign. 2.2 InstrumentingVaccineUptake: TheImpactofLotteries While our main specification includes a fairly exhaustive set of controls, it is not fully robust to concerns of further omitted variables or endogeneity. In this section, we explore an instrumental variable strategy that addresses those concerns. In particular, we rely on the implementation of vaccine lotteries. Between May 10 and July 1, 2021, 19 states announcedlotteriestoboostvaccinationrates. Participationtothelotteriesrequiredhaving received or receiving one shot of the vaccine; while, in some instances, individuals were notrequiredtotakeanyadditionalsteps,moststatessetupwebportalsforthesubmission of the vaccination record. Table A2 summarizes announcements and last extraction dates bystate. Evidence on the impact of lotteries or other monetary incentives on vaccine uptake 10

is mixed. In particular, the closest papers to this part of our analysis, Dave et al. [2021] and Robertson et al. [2021a], document contrasting results.6 Looking across various state lotteries, Dave et al. [2021] argue that lotteries had no impact on vaccine administration, while Robertson et al. [2021a] suggest that 10 of the 12 statewide lotteries they studied generated a positive and statistically significant impact on vaccine uptake. Those papers differintermsoftheleveloftheanalysis(state-levelforDaveetal.[2021]vscounty-level forRobertsonetal.[2021a])andwhatcontrolsareusedinthestudy,butbothrelyondata throughearlyJuly. Whileweperformouranalysisatthestate-level,weextendthelottery datatoencompassallextractions;wealsoadoptamoreexhaustivespecification.7 Beforedescribingourstrategyfortheidentificationoflotteryeffects,figure4compares the trends in vaccination rates between states that have announced a lottery at any time in our sample and states that never did. Interestingly, future adopters of vaccine lotteries displayhigherlevelsofvaccineuptakesincemid-March2021. Whilegenerallythetrends invaccinationarenottoodissimilar,thefigurehighlightsacoupleofinstanceswheredeviationsinthetrendsbetweenthetwogroupsofstatesaremorevisible(shadedblueareas); thosedeviationsoccurinearlyMay,aroundthetimeoftheearliestlotteryannouncements, and around July 1st, when several extractions for various lotteries occurred. This quick comparison, however, is very rudimentary and does not take into account the different timingoftheannouncementsordifferencesinotherstate-levelfactors. To precisely identify the impact of lotteries on vaccinations, we rely on a dynamic difference-in-differenceestimationthatcontraststhevariationinvaccinationratesinstates thathaveimplementedthelotterywiththatofstatesthathavenot—butmayatsomepoint inthefuture—implementedit. Inparticular,ourbaselinespecificationisthefollowing 60 ∑ NewAdm. = δ + δ PostAnnounc. +ξX +d +d +d +(cid:101) (3) st 0 1,j s,t−j st sm s t st j=−15 6Severalotherpapershavelookedattheimpactoflotteriesorothermonetaryincentivesonvaccinations.Inparticular,Brehmetal.[2021]andMallowetal.[2021]documentapositiveimpactonvaccinations fortheOhiolottery,incontrastwiththefindingsofLangetal.[2021].OtherevidencepointstoapositiveimpactofguaranteedmonetaryincentivesinSweden(Campos-Mercadeetal.[2021])andintheU.S.(Daietal. [2021]),butnotforthevaccinehesitantpopulation(Changetal.[2021]). 7OursamplecoversdatathroughMarch2022,wellafterthelastextractiononAugust26intheKentucky lottery. 11

Figure4: ComparisonofVaccineAdministrationTrendsbetweenStateswithandwithout VaccineLotteries wherePostAnnounc. collectstheleadsandlagsrelativetothelotteryannouncement.8 s,t−j δ , j ∈ {−15,...,60}areourcoefficientofinterestsandidentifythedynamictreatmentef- 1,j fectsundertheassumptionofconditionalparalleltrendsoflotteryadoptersrelativetothe not-yetadoptersandtheneveradoptersgroups;thus,underthoseparalleltrendassumptions, the coefficients could be interpreted as a weighed average of the average treatment effect on the treated. Importantly, in our setting, treatment effects are heterogeneous in time since lotteries are adopted in different time periods across different states. As noted byGoodman-Bacon[2021],thissettingmayleadtoestimatesthatarebiasedawayfroma weighed average of the average treatment effect on the treated. This problem, however, is addressed in our panel event study design with a dynamic specification that includes two-wayfixedeffects. Whileparalleltrendassumptionsmightbelesslikelytoholdforpartofthecomparison groups,ourspecificationincludesaratherexhaustivesetofcontrols. Inparticular,wecontinuetocontrolforthesameexplanatoryvariablesasinequation(1);inaddition,equation 8Allperiodsbeyondsomespecifiedvaluesareaccumulatedintofinallagandleadpointstoavoidunbalancedleadsandlags. 12

(3)includesaregressorthatcapturesthetimetoextractionandadummyforthepresence of other types of incentives.9 Furthermore, we adopt a more stringent fixed effect specification relative to model (1), with state-month fixed effects, d , to capture differences sm in the monthly timing of lottery adoption across states as well as cross-state variation in economic conditions or other monthly factors that might have influenced the decision to announcealottery. While equation 3 represents the first stage of our instrumental variable strategy, we willthenusetheestimatesfromthatmodeltopredictnewvaccineadministrationanduse thispredictioninourbaselinemodels(1)and(2). Results Figure 5 summarizes the first-stage result. The coefficient estimates of the leads and lags aroundthelotteryannouncementsarerelativetotheperiodbeforeannouncement,coded as −1 and identified by the black vertical bar. The chart highlights no significant effects on new vaccinations in the 15 days before announcements, consistent with the parallel trend assumptions required for our estimation. However, we also find that there is no immediateincreaseinvaccinationsafterannouncement;thisfindingisconsistentwiththe factthateligibilityinseverallotteriesrequiredregisteringthroughanonlineportalbefore scheduledextractiondates. Onthe8thdaypost-announcement,thecoefficientestimateis significantaswellasinafewotherinstancesoverthefollowingtwoweeks. Table 3 summarizes the cumulative impact on new vaccine administration in the preand post-announcement period. Column (1) reports that the cumulative effect in the pre-period is effectively zero, confirming that the conditional parallel trend assumption is satisfied in the 15 days before announcement. After excluding the first 7 days postannouncement, we find a significant effect of lotteries on new vaccinations through the 45thdaypost-announcement(columns(2)and(3)). Whenextendingthewindowthrough day60(column(4)),wefindasimilarpointestimate,althoughlesstightlyestimated. Looking at magnitudes and using the coefficient from column (2), the effect implies that states thatannouncedlotteriesexperienceda3.5percentincreaseinvaccinationsaweekafteran- 9InformationonincentivesiscollectedfromapublicationoftheNationalGovernorsAssociationavailablehere. 13

Figure5: LotteryAnnouncementsandImpactonNewVaccinations nouncement and over the next 20 days relative to those that did not introduce or had not yet made an announcement. As an alternative quantification, the effect of lotteries translatesintoanincreaseinvaccinationsofalmost2.5standarddeviationsinlotteryadopters whencomparingtoneverornot-yetadopters. FiguresA1-A3summarizethesecond-stageimpactofvaccinationsonourmeasuresof activityusingourimpulseresponseframework. Theresultsaregenerallyconsistentwith what we have documented in the previous section. Importantly, our estimate of the impactonretailspendingcontinuestobesignificant,althoughtheeffectisshiftedbyaweek, likely reflecting the slight delay in the impact of lotteries on vaccination after announcement. In our second stage analysis, three other main differences emerge. First, we do notfindapositiveandsignificantimpactofvaccinationsonrestaurantspending. Second, mobility indicators display a more similar behavior, with both INRIX and Apple indexes showing, atfirst, nosignificanteffectsand, lateron, pointingtoanegativeimpact. Third, the significance of the negative point estimates for measures of employment has largely disappeared. Table 4 looks at the cumulative impact of vaccinations on our main measures of eco- 14

Table3: CumulativeImpactofLotteriesonNewVaccinationsaroundAnnouncement (1) (2) (3) (4) Variable NewVaccineAdm. Before After 2-15days 8-30days 8-45days 8-60days CumImpact 0.501 3.565** 5.121** 3.545 (1.123) (1.624) (2.326) (2.686) OtherControls1 y y y y State-MonthFE y y y y DayFE y y y y StateFE y y y y Obs. 14,026 14,026 14,026 14,026 R-squared 0.580 0.580 0.580 0.580 NumberofStates 51 51 51 51 Source:CDCandNOAA. 1Othercontrolsincludenewvaccinedistribution;newcases,hospitalizations,anddeaths;heatingandcoolingdegreedays;theemployment rate;demographicscharacteristics;theOxfordstringencyindex;the timetoextraction;andadummyforthepresenceofotherincentives. NewVaccineAdm.:Log-numberofthe7-daymovingaverageofnew dailyvaccineadministration. Legend:∗∗∗significantat1%,∗∗at5%,∗at10%. Notes:Dynamicdifference-in-differenceregressions,cumulatedeffects before(column(1))andafter(columns(2)-(4))lotteryannouncment. Inthepost-announcementperiod,weexcludethefirst7daystoaccountforlearningabouteligibilityconditionsRobuststandarderrors, clusteredatthestatelevel,arereportedinparenthesis. 15

nomic activity in our instrumental variable setting. The cumulative estimates are evaluated over the 15 days after day 30—or, between day 31 and day 45 post-vaccination—a windowthatwechosebasedontheimpulseresponsefunctionresults.10 AsalreadyhighlightedbyfiguresA1-A3,ourresultsarerobustonlyforretailspending. Inparticular,our estimates imply that an increase in vaccinations raises retail spending by 23.8 percent— or 2.25 standard deviations per standard deviation—after 30 days from receiving the first doseofthevaccineandoverthefollowingtwoweeks. Inotherwords,oureffectsuggesta dailyboosttoretailsalesofabout1.6percentperdayfor15daysperpercentageincrease invaccinations–whichtranslatesintoamonthlyrateof0.27percent. 3 Implications for GDP growth Ouranalysissuggeststhatretailspendinghasreceivedasignificantboostfromtheprogress in the vaccine rollout. But what do those effects ultimately tell us about aggregate economicactivity? Wehavedrawnadirectinference—summarizedintable5—basedontwo features: (1)therelationbetweenourmainspendingindicatorandtheretailsalescomponent of personal consumption expenditures (PCE) and (2) its contribution to GDP. First, weestimatethegrowthrateofCensus’retailsales,theofficialsourceofGDPdatareleased bytheBureauofEconomicAnalysis(BEA),usingourFiservindicator. OurFiservspending measure is highly correlated with the Census’s data on retail sales, and our estimate suggests an average growth of 0.87 percent per month in 2021. As a result, our estimates predict that retail sales grew at almost 10 percent at an annual rate in 2021 (line 1, table 5)—vs. 101/2 percent using the BEA GDP Data. Second, we calculate the contribution of ourvaccineeffectstoretailsalesandtoGDP.Basedontheaveragegrowthofnewvaccine administration since the beginning of the year, we estimate that the vaccine uptake explainsabout15percentoftheaverageincreaseinretailsalesand,asaresult,accountsfor about0.5percentagepointofGDPgrowthoverthesametimehorizon(line3). Theimpact ofvaccinationsonGDPwecalculated,however,islikelyalowerboundasitfocusesona 10Theversionofmodel(2)usedforthisanalysisincludesonlythoselagsbeyondallothercontrols,asthe inclusionofadditionallagssignificantlyrestrictsthesampleforestimation.Resultsare,however,robustto theuseofamodelwithupto30lagsaroundthesamewindow. 16

Table4: SecondStageEffectsoftheVaccineRolloutonActivity (1) (2) (3) (4) (5) (6) Spending Mobility Employment VARIABLES Retail Restaurant INRIX Apple Hours OpenBusinesses NewAdm. 23.770*** 3.535 -41.412 -6.213 -1.822 -0.950 (5.661) (6.077) (88.460) (7.070) (3.065) (1.745) OtherControls1 y y y y y y State-MonthFE y y y y y y DayFE y y y y y y StateFE y y y y y y Obs. 1,127 1,127 94 1,127 1,127 1,127 R-squared 0.810 0.805 0.949 0.976 0.979 0.986 NumberofStates 50 50 20 50 50 50 Source:Fiserv,Inc.,INRIX,Apple,Homebase,CDC,CPS,andNOAA. 1Othercontrolsincludenewvaccinedistribution;newcases,hospitalizations,anddeaths;heatingandcoolingdegreedays;theemploymentrate;demographicscharacteristics;theOxford stringencyindex;thetimetoextraction;andadummyforthepresenceofotherincentives. RetailSpending:Percentagechangeinretailsalesspendingrelativeto2019. RestaurantsSpending:Percentagechangeinrestaurantspending(NAICS722)relativeto2019. INRIX:Percentagechangeinthe7-daymovingaverageofpassengerdistancetraveled. Apple:Percentagechangeinthe7-daymovingaverageofthedrivingindex. Hoursworked:Percentagechangeinthenumberoftotalhoursworkedrelativeto2019in smallbusinessestablishments. BusinessOpen:Percentagechangeinthenumberofopenbusinessesrelativeto2019insmall businessestablishments. NewAdm.:Log-numberof7-daymovingaverageofnewdailyvaccineadministration,cumulatedeffectaftervaccination. Legend:∗∗∗significantat1%,∗∗at5%,∗at10%. Notes:Second-stageFEregressions.Pointestimatesforthemainexplanatoryvariablearebased onlinearcombinationsofcoefficients:incolumns(1)-(3),wereportthelinearcombinationfrom the21st-daythroughthe30th-daylag;incolumn(5),wereportthelinearcombinationfromthe 51st-daythroughthe60thday;incolumns(4)and(6),wereportthelinearcombinationfrom the15th-daythroughthe25th-daylag.Robuststandarderrors,clusteredatthestatelevel,are reportedinparenthesis. 17

singlechannel—althoughthemostimportant,accordingtoourestimates. Table5: Vaccinations: ImpactonGDPGrowth 2021Average 1.RetailSalesGrowth1 9.98% 2.RetailSalesContributiontoGDP 3.38% 3.VaccinationsImpact 0.54% Source:BEA,Census,andFiserv,Inc. 1RetailsalesgrowthpredictionbasedonFiservdata. Notes:EstimatesofvaccinerollouteffectsonGDP growth. Finally,weproposeacost-benefitanalysis,comparingourGDPimplicationswithcost estimates for the lotteries implemented. Our calculation suggest that the vaccine rollout added 400 billion to GDP in 2021—or about 1500 USD per vaccination. Robertson et al. [2021a] estimate that the cost of lotteries per marginal vaccination was 55 USD. Even accounting for the fact that the implementation of lotteries largely occurred between the middleofthesecondquarterandthethirdquarter,thebenefitperquarter—oraround375 USD—remains much higher than the cost of lotteries, pointing to the importance of the vaccinerolloutandofthestate-levellotteriesforeconomicactivity. 4 Conclusions Inthispaper,weanalyzedtheimpactofvaccineadministrationonthreemaindimensions of activity: spending, mobility, and employment. We complement an investigation using panel state-level data with an instrumental variable strategy that relies on the implementation of vaccine lotteries. Our results highlight the effect of new vaccinations on retail spending. In particular, relying on the results from our IV strategy, we find that lotteries have significantly boosted vaccination rates about a week after announcement, with an effect that lasted over the next several days and, overall, increased new vaccinations by at least 3.5 percent across lottery adopters compared to states without a lottery (either never adopters or not-yet adopters). This boost in vaccination rates, in turn, translates 18

intoasignificantincreaseinretailspending,witha1percentincreaseinnewvaccinations associated with a monthly growth of 0.27 percent in retail spending. All told, our findingsimplythatthevaccinerolloutadded, onaverage, about0.5percentagepointtoGDP growthin2021andthatthecostoflotterieswaswellbelowtheboosttoretailsales. 19

References AdityaAladangady,ShifrahAron-Dine,WendyDunn,LauraFeiveson,PaulLengermann, and Claudia Sahm. From Transactions Data to Economic Statistics: Constructing Realtime,High-frequency,GeographicMeasuresofConsumerSpending. FEDSWorkingPaper,2019-057,2019. Apple,Inc. MobilityTrendReport,2022. MargaretBrehm,PaulBrehm,andMartinSaavedra.TheOhioVaccineLotteryandStarting VaccinationRates. Technicalreport,2021. BureauofLaborStatistics. CurrentPopulationSurvey,2022. Pol Campos-Mercade, Armando N Meier, Florian H Schneider, Stephan Meier, Devin Pope,andErikWengström. MonetaryIncentivesIncreaseCOVID-19Vaccinations. Science,374(6569):879–882,2021. Tom Chang, Mireille Jacobson, Manisha Shah, Rajiv Pramanik, and Samir B Shah. Financial Incentives and Other Nudges do not Increase COVID-19 Vaccinations among the VaccineHesitant. Technicalreport,NationalBureauofEconomicResearch,2021. Leland Dod Crane, Ryan Decker, Aaron Flaaen, Adrian Hamins-Puertolas, and ChristopherJohannKurz. BusinessExitDuringtheCovid-19Pandemic: Non-TraditionalMeasuresinHistoricalContext. FEDSWorkingPaper,2020-089R1,2020. Hengchen Dai, Silvia Saccardo, Maria A Han, Lily Roh, Naveen Raja, Sitaram Vangala, Hardikkumar Modi, Shital Pandya, Michael Sloyan, and Daniel M Croymans. BehaviouralNudgesIncreaseCOVID-19Vaccinations. Nature,597(7876):404–409,2021. Dhaval Dave, Andrew I Friedson, Benjamin Hansen, and Joseph J Sabia. Association BetweenStatewideCOVID-19LotteryAnnouncementsandVaccinations. InJAMAHealth Forum,volume2,pagese213117–e213117.AmericanMedicalAssociation,2021. Fiserv, Inc. SpendTrend Transaction Data, 2022. URL https://www.firstdata.com/en_ us/insights/spendtrend.html. 20

Andrew Goodman-Bacon. Difference-in-Differences with Variation in Treatment Timing. JournalofEconometrics,225(2):254–277,2021. Thomas Hale, Noam Angrist, Rafael Goldszmidt, Beatriz Kira, Anna Petherick, Toby Phillips, Samuel Webster, Emily Cameron-Blake, Laura Hallas, Saptarshi Majumdar, and Helen Tatlow. COVID-19 Government Response Tracker, 2021. URL https: //doi.org/10.1038/s41562-021-01079-8. Niels-JakobHHansenandRuiCMano. COVID-19Vaccines: AShotinArmfortheEconomy. IMFWorkingPapers,2021(281),2021. HealthandHumanServices. HospitalUtilization,2022. David Lang, Lief Esbenshade, and Robb Willer. Did Ohio’s Vaccine Lottery Increase VaccinationRates? APre-Registered,SyntheticControlStudy. Technicalreport,2021. PeterJMallow,AlecEnis,MatthewWackler,andEdmondAHooker. COVID-19Financial Lottery Effect on Vaccine Hesitant Areas: Results from Ohio’s Vax-a-Million Program. TheAmericanJournalofEmergencyMedicine,2021. NationalOceanicandAtmosphericAdministration. NationalClimaticData,2022. Christopher Robertson, K Aleks Schaefer, and Daniel Scheitrum. Are Vaccine Lotteries WorththeMoney? EconomicsLetters,209:110097,2021a. ChristopherRobertson,DanielScheitrum,AleksSchaefer,TreyMalone,BrandonRMcFadden,KentDMesser,andPaulJFerraro. Payingamericanstotakethevaccine: Wouldit helporbackfire? JournalofLawandtheBiosciences,8(2):lsab027,2021b. SafeGraph. https://www.safegraph.com/,2022. MG Thompson, JL Burgess, AL Naleway, et al. Interim Estimates of Vaccine Effectiveness of BNT162b2 and mRNA-1273 COVID-19 Vaccines in Preventing SARS-CoV-2 InfectionAmongHealthCarePersonnel,FirstResponders,andOtherEssentialandFrontline Workers–Eight U.S. Locations, December 2020-March 2021. MMWR Morb Mortal WklyRep.,70:495–500,2021. 21

The New York Times. Coronavirus (Covid-19) Data in the United States. Retrieved on March23,2022,fromhttps://github.com/nytimes/covid-19-data,2021. U.S.CentersforDiseaseControlandPrevention. COVID-19VaccinationsbyJurisdiction, 2022. 22

A Additional Figures and Tables FigureA1: SecondStageEffectofVaccinations: ImpactonSpending FigureA2: SecondStageEffectofVaccinations: ImpactonMobility FigureA3: SecondStageEffectofVaccinations: ImpactonEmployment 23

TableA1: TheVaccineRolloutandVisits: ControllingforNon-StoreSpending (1) (2) Variables NonessentialVisits RestaurantVisits NewAdm. 3.156 4.223 (4.492) (3.985) Non-StoreSpending1 y y OtherControls2 y y State-QuarterFE y y DayFE y y StateFE y y Obs. 2,595 2,595 R-squared 0.218 0.209 NumberofStates 51 51 Source:SafeGraph,CDC,CPS,andNOAA. 1Changeinspendingatnon-storeretailers(NAICS454)relative to2019. 2Othercontrolsincludenewvaccinedistribution;newcases, hospitalizations,anddeaths;heatingandcoolingdegreedays; theemploymentrate;demographicscharacteristics;andtheOxfordstringencyindex. Nonessential,Visits:Percentagechangeinvisitstononessential retailstoresrelativeto2019. Restaurants,Visits:Percentagechangeinvisitstorestaurants relativeto2019. NewAdm.:Log-numberof7-daymovingaverageofnewdaily vaccineadministration,cumulatedeffect. Legend:∗∗∗significantat1%,∗∗at5%,∗at10%. Notes:StateFEregressions.Pointestimatesforthemainexplanatoryvariablearebasedonlinearcombinationsofcoefficientsfromthe15th-daythroughthe25th-daylag.Robuststandarderrors,clusteredatthestatelevel,arereportedinparenthesis. 24

TableA2: StateLotterySummary Announcement ExtractionDate State Date (Last) Arkansas May25 - California May27 July1 Colorado May25 July6 Delaware May25 June29 Illinois June17 August16 Kentucky1 June4 August26 Louisiana June17 July31 Maine June16 June30 Maryland May20 July3 Massachusetts June15 August19 Michigan July1 August3 Nevada June17 August26 NewMexico June1 August6 NewYork May20 June11 NorthCarolina June10 August1 Ohio May12 June20 Oregon May21 June27 Washington June3 July13 WestVirginia June1 August1 1OnMay10,Kentuckyofferedacouponforafreelotteryticket($225,000maximumcashawardtowinner) tothoseages18+whoreceivedaCOVID-19vaccine onlyat180KrogerandWalMartlocationsstatewide. Notes:Announcementdatesandlastextractiondates acrossstatesthatinstitutedlotteries. 25