Tracking Labor Market Developments during the COVID-19 Pandemic: A Preliminary Assessment

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Tracking Labor Market Developments during the COVID-19 Pandemic: A Preliminary Assessment Tomaz Cajner, Leland D. Crane, Ryan A. Decker, Adrian Hamins-Puertolas, Christopher Kurz 2020-030 Please cite this paper as: Cajner, Tomaz, Leland D. Crane, Ryan A. Decker, Adrian Hamins-Puertolas, and Christopher Kurz (2020). “Tracking Labor Market Developments during the COVID- 19 Pandemic: A Preliminary Assessment,” Finance and Economics Discussion Series 2020-030. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2020.030. 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.

Tracking Labor Market Developments during the COVID-19 Pandemic: A Preliminary Assessment TomazCajner LelandD.Crane RyanA.Decker ∗ AdrianHamins-Puertolas ChristopherKurz Firstdraft: April15,2020 Abstract Manytraditionalofficialstatisticsarenotsuitableformeasuringhigh-frequencydevelopmentsthat evolveoverthecourseofweeks,notmonths. Inthispaper,wetrackthelabormarketeffectsofthe COVID-19pandemicwithweeklypayrollemploymentseriesbasedonmicrodatafromADP.These dataareavailableessentiallyinreal-time,andallowustotrackbothaggregateandindustryeffects. CumulativelossesinpaidemploymentthroughApril4arecurrentlyestimatedat18million;justduringthetwoweeksbetweenMarch14andMarch28theU.S.economylostabout13millionpaidjobs. For comparison, during the entire Great Recession less than 9 million private payroll employment jobswerelost.Inthecurrentcrisis,themostaffectedsectorisleisureandhospitality,whichhassofar lostorfurloughedabout30percentofemployment,orroughly4millionjobs. Keywords:labormarket,economicmeasurement,bigdata. JELClassification:J2,J11,C53,C55,C81. ∗All authors are at the Federal Reserve Board of Governors. We thank ADP for access to and help with the payroll microdata that underlie the work described by this paper. In particular, this work would not have been possible without the support of Matt Levin, Ahu Yildirmaz, and Sinem Buber. The analysis and conclusions set forthherearethoseoftheauthorsanddonotindicateconcurrencebyothermembersoftheresearchstafforthe BoardofGovernors.

1 Introduction TheCOVID-19pandemic,andtheassociatedsocialdistancingmeasures,havedeeplyaffected the U.S. labor market. The unprecedented speed of deterioration in labor market conditions calls for higher frequency indicators than are currently available in traditional statistical data suchastheBLSCurrentEmploymentStatistics(CES)andtheCurrentPopulationSurvey(CPS). AndwhiletheweeklyinitialclaimsforunemploymentinsurancecollectedbytheDepartment ofLaborprovideinformationtohelpunderstandlabormarketdevelopments,claimsonlyprovideapartialpicturebecausetheycaptureonlyjobdestructionandnotdecreasedhiring.1 In this paper, we build on our extensive previous work in which we constructed employment measures from the ADP payroll microdata; see Cajner et al. (2018), Cajner et al. (2019a), Cajneretal.(2019b),andCajneretal.(2020). Inparticular,weusetwomeasuresofemployment: i)activeemployment,whichcorrespondstothenumberofindividualsactiveinthepayrollsystemregardlessofwhethertheywerepaidornotinagivenpayperiod;andii)paidemployment, whichcorrespondstothenumberofindividualsissuedapaycheckinagivenpayperiod.2 The distinction between active and paid employment is especially useful in the context of labor market developments during the COVID-19 episode. More precisely, changes in active employmentshouldreflectonly(permanent)layoffs,butnot(temporary)furloughs. Ontheother hand,changesinpaidemploymentshouldcapturebothlayoffsandunpaidfurloughs. Inother words, our active employment series should be similar in concept to CPS employment, while our paid employment series should be analogous to CES employment.3 We construct both typesofemploymentseriesattheweeklyfrequency,whichallowsustotrackthedynamicsof employmentlossesoccurringsincetheonsetoftheCOVID-19crisis. Importantly, in previous work we found that ADP data are reasonably representative of 1Additionally,notallworkersthatloseajobapplyforunemploymentinsuranceandthesevereprocessingdelays ofinitialclaimsthatoccurredinthesecondhalfofMarch2020implysomefurtherchallengeswhenusinginitial claimsdatatoinfercurrentlabormarketconditions. 2Active employees include wage earners with no hours in the pay period, workers on unpaid leave, and the like. Paidemployeesincludeanywageorsalaryworkersissuedregularpaychecksduringthepayperiodaswell asbonuschecksandpayrollcorrections. 3Active employment may also be affected by delays in payroll system maintenance. For example, if workers are permanently separated, some might not be automatically removed from payroll systems. That said, in the previousworkcitedabovewefoundthatADPactiveemploymentisbetterthanpaidemploymentforforecasting CESemployment. 1

U.S. businesses, with a wide range of coverage across business size, industry, and geography. Inthisrespect,ADPdataareuniqueamongprivatesectordatasourcescurrentlybeingusedto tracklabormarketsduringtheCOVID-19crisis. Moreover,ourmainADPindexesarebasedon ADP data aggregated with size-by-industry weights derived from the BLS Quarterly Census ofEmploymentandWages(QCEW),sothedataareabletoprovideagoodmeasureoftheU.S. privatesector. WealsoemphasizethatourADP-basedemploymentmeasuresaredistinctfrom those published jointly by ADP and Moody’s in the monthly National Employment Report (NER). Wefindthat, amongpaycheck-to-paycheckcontinuingbusinesses, activeemploymenthas cumulatively declined by a bit more than 6 million between February 15 and April 4; recall thatthisdeclinecaptureslaidoffworkerswhowereremovedfromthepayrollsystem. Onthe otherhand,paidemploymenthascumulativelydeclinedbyabout18millionduringthesame period. Most severe job losses were experienced in the second half of March; our estimates suggestthattheU.S.economylostabout13millionjobsbetweenMarch15andMarch28. For comparison, during the entire Great Recession (between December 2007 and February 2010) 8.8 million jobs were lost. Our estimates of employment losses are also comparable to those inotherrecentpapers: Brynjolfssonetal.(2020)estimatethatabout16millionAmericanslost jobs through April 5, while Coibion, Gorodnichenko and Weber (2020) estimate that about 20 millionjobswerelostbyApril8. Our estimate of employment losses is based on the sample of PACs that can be longitudinally matched over consecutive pay periods, which implies we do not account for lower business entry and higher business exit. In some preliminary work, we find that reduced entry and elevated exit could potentially increase cumulative employment losses, perhaps by as muchas5to10million.4 4Moreover,recentlyreleasedweeklydataonnewbusinessapplicationsfromtheCensusBureaushowasharp decline;seeHaltiwanger(2020). 2

2 Methodology for Constructing Weekly Employment Indexes TheweeklyemploymentdatausedinthispapercomefromthecompanyADP,whichprocesses payrollsforabout20percentoftotalU.S.privateemployment.5 Themicrodataareatthelevel of Payroll Account Controls (PAC), which often correspond to business establishments (but maysometimescorrespondtofirms)asdefinedbytheCensusBureauandtheBLS.EachPAC updates their record at the end of each pay period. The record consists of the date payroll was processed, employment information for the pay period, and many time-invariant PAC characteristics (such as an anonymized PAC identifier, NAICS industry code, zip code, etc.).6 PAC records include both the number of individuals employed (“active” employees) and the number of individuals issued a paycheck in a given pay period (“paid” employees). Active employees include wage earners with no hours in the pay period, workers on unpaid leave, and the like. Paid employees include any wage or salary workers issued regular paychecks duringthepayperiodaswellasthoseissuedbonuschecksandpayrollcorrections. ThedatabegininJuly1999,butareavailableattheweeklyfrequencyonlysinceJuly2009. As argued in Cajner et al. (2018), ADP payroll data appear to be quite representative of the U.S. economy, though the data somewhat overrepresent the manufacturing sector and large businesses,ascomparedtotheQCEWuniverseofestablishments. Weaddresstheseissuesby reweightingthedataasexplainedbelow. The process of transforming the raw data into usable aggregate series is complex, and we refertheinterestedreadertoCajneretal.(2018)fordetails. Inshort,wecalculatetheweighted averagegrowthofemploymentatPACsappearinginthedatafortwoconsecutiveweeks. We build a weekly time series of employment for each PAC, estimating employment at the PAC each Saturday. Technically, the employment concept is PAC employment for the pay period thatincludestheSaturdayinquestion,aswecannotobservewithin-payperiodchanges. Lacking any information on events within a pay period, we assume that businesses adjust their employment discretely at the beginning of each pay period and that employment is constant within the pay period. This assumption is consistent with the typical practice of human re- 5OtherpapersthatuseADPdataincludeCho(2018)andGrigsby,HurstandYildirmaz(2019). 6Whenaccessingthemicrodata,wefollowanumberofprocedurestoensureconfidentiality.Businessnamesare notpresentinthedataweaccess. 3

source departments, according to which job start dates often coincide with the beginning of pay period. The restriction to “continuers” allows us to abstract from changes in the size of ADP’s client base. Growth rates are weighted by PAC employment and further weighted for representativeness by size and industry (treating PACs as establishments for weighting purposes). We use QCEW employment counts by establishment size and two-digit NAICS as the target population. Formally, let w be the ratio of QCEW employment in a size-industry cell j,t j to ADP employment in cell j in week t, let C(j) be the set of ADP businesses in cell j, let e i,t be the employment of the i’th business, and let g = ei,t −ei,t−1 be the weekly growth rate of i,t ei,t−1 businessi.7 Aggregategrowthisestimatedas: ∑J w ∑ e g j=1 j,t−1 i∈C(j) i,t−1 i,t g = . (1) t ∑J w ∑ e j=1 j,t−1 i∈C(j) i,t−1 Cumulating the weekly growth rates yields a weekly index level for employment. We benchmark the data annually to QCEW employment levels and use a forward benchmarking projection akin to the CES birth and death model. While we believe benchmarking is important, since the QCEW (when available) represent the most complete and accurate estimate of employment,therawADPdataalignwellwithofficialsourcesevenbeforebenchmarking. The toppanelofFigure1comparesthemonthlychangeinemploymentintheunbenchmarkedADP- FRB series to the (QCEW-benchmarked) CES series through February 2020. The series track eachotherclosely,indicatingthatbotharepickingupthesameunderlyingsignal(i.e.,trueU.S. payrollgrowth.) ThebottompanelofFigure1showsthemorerecentevolutionofmonthlyestimates,includingthesharpdeclinesinMarch2020.8 NotethattheADP-FRBmonthlyseriesis calculatedasthechangeinthemonthlyaveragelevelofemployment,whereastheCEScalculates differences between pay periods including the 12th of the month. As such, the monthly ADP-FRB series is more influenced by the last half of March, which helps explain the larger declineweseeinthatseries. Returningtotheweeklydata,thelaststepistoseasonallyadjusttheseriesusingthemethods of Cleveland and Scott (2007), which combine a fixed coefficient regression with locally 7Forweighting,weuseMarchQCEWemploymentvaluesforeachyear. 8Inthebottompanelweuseourpreferred,benchmarkedversionoftheADP-FRBseries. 4

400 200 0 -200 -400 -600 -800 -1000 Jan2006 Jan2008 Jan2010 Jan2012 Jan2014 Jan2016 Jan2018 Jan2020 sboJ fo sdnasuohT CES ADP-FRB active employment, not benchmarked 500 0 -500 -1000 -1500 -2000 -2500 Jan2019 Mar2019 May2019 Jul2019 Sep2019 Nov2019 Jan2020 Mar2020 sboJ fo sdnasuohT CES ADP-FRB active employment, benchmarked Source:CES,ADP,authors’calculations. Figure1: MonthlyChangeinPrivatePayrollEmployment weightedregressionsontrigonometricfunctions. NotethattheClevelandandScottapproach wasemployedbytheBLStoseasonallyadjusttheweeklyunemploymentclaimsrelease.9 Sincetheprimaryfocusofthispaperisinweeklydata,itisworthnotingthedistributionof payfrequenciesintheADPdata. AsofMarch2017,22percentofPACswereissuingpaychecks weekly, 46 percent biweekly, 21 percent semi-monthly, and 11 percent monthly (in terms of employment, these shares are 23 percent, 55 percent, 18 percent, and 4 percent, respectively). 9Fortheweeklyseasonaladjustment,wespecificallycontrolforholidayweeks,includingThanksgiving,MemorialDay,LaborDay,NewYears,Christmas,andJuly4th.Wealsoaccountforstrongemploymentweeksleadingup toholidaysandtheseasonalemploymentrelatedtoChristmas. SpecialthankstoCharlieGilbertforhisassistance withseasonaladjustment. 5

ThesefractionsarenotfarfromwhattheBLSreports.10 3 Measuring Employment Losses during COVID-19 Pandemic Figure2summarizesourmainaggregateresultsbyshowingcumulativeemploymentchanges since February 15 2020, weekly employment changes, and initial claims for unemployment insurance, which serve as a comparison point. As the BLS March employment report indicated, significant payroll employment losses had already occurred in the first half of March (theBLSCESemploymentnumbersrefertothepayperiodincludingMarch12),andourADP- FRB employment numbers are consistent with an early March decline. Employment losses then sharply accelerated in the second half of March; our ADP-FRB paid employment numbersuggeststhatanadditional13millionjobswerelostbetweenMarch15andMarch28. For comparison, during the entire Great Recession (between December 2007 and February 2010) 8.8millionjobswerelost. OurestimateofjoblossesbetweenMarch15andMarch28ishigherthanthesumofinitial unemployment insurance claims filed during those two weeks. One likely explanation is that ourestimatesofemploymentlossesalsoincludereducedhiring,whileinitialclaimspredominantlycapturejobdestruction(workersfiredorfurloughed). Additionally,initialclaimslagjob losses,someworkersneverfileaclaim,andunemploymentofficeswerefacingsubstantialprocessingdelaysinthesecondhalfofMarch. Wecurrentlyseesomestabilizationofemployment losses in the week ending April 4, although it is possible that the estimate for that week will getreviseddownwardwhenmoredatabecomeavailable(revisionpropertiesaredescribedin thenextsubsection). 10SeeBLS(2019)"LengthofpayperiodsintheCurrentEmploymentStatisticssurvey." 6

Cumulative Job Loss since 15 Feb 2020 Weekly, SA Millions of Jobs 2 2 0 0 -2 -2 -4 -4 -6 -6 -8 -8 -10 -10 -12 -12 -14 -14 ADP-FRB Active Employment -16 ADP-FRB Paid Employment -16 -18 BLS CES -18 -20 -20 15 22 29 7 14 21 28 4 Feb. Mar. Apr. ADP FRB First Differences Weekly, SA Millions of Jobs 2 2 0 0 -2 -2 -4 -4 -6 -6 ADP-FRB Active Employment -8 ADP-FRB Paid Employment -8 -10 -10 15 22 29 7 14 21 28 4 Feb. Mar. Apr. Average Weekly Initial Claims Weekly, SA Millions of Claims 7 7 UI Claims 6 6 5 5 4 4 3 3 2 2 1 1 0 0 15 22 29 7 14 21 28 4 Feb. Mar. Apr. Note:ADP-FRBfiguresrefertothepayperiodsincludingeachSaturday,UIclaimsrefertotheweekendingoneach Saturday. Source:ADP,authors’calculations. Figure2: LaborMarketDevelopmentssinceFebruary152020 7

3.1 EvolutionofReal-TimeEstimates Whencalculatingweeklyemploymentestimates,themostrecentobservationincorporatesdata fromPACswithweeklypayfrequencyandapproximatelyhalfofthePACswithbiweeklypay frequency. As additional PACs report payroll, our employment estimates for the more recent weeksrevise. Ourrealtimeestimationprocedureattemptstoaccountforthenot-yet-available datafromnonreportingPACsbyusingthehistoricalrelationshipbetweentheraw“firstprint” estimate and the final estimate, obtained when data from all PACs have arrived. While this procedure works reasonably well during normal times, the current sharp movements in employmentrepresentadditionalchallengesforthereal-timeestimationofemployment,sincethe forecastingpartoftheestimationmightworklesswellthanusual.11 Figure 3 shows the evolution of real-time estimates for active and paid employment since thebeginningofMarch. Thesedataindicatethatinitialemploymentestimateshavebeengenerally somewhat too high, revising down when more data became available. That said, after the second read, revisions appear to be relatively small. Thus, it seems likely that the current estimate of employment during the week including April 4 will get revised (most likely downward),butestimatesbeforethatweekarelikelytoremainlittlerevised. 11Indeed, our real-time adjustment to the most recent week of data in hand has performed worse during the COVID-19crisisthanatanypointsince2005. However, errorsinthereal-timeforecastsdonotpersist, sincethe adjustmentisreplacedbyreportingPACs. 8

ADP-FRB Active Employment Cumulative Job Loss since 15 Feb 2020 Weekly, SA Millions of Jobs 1 1 0 0 -1 -1 -2 -2 -3 -3 -4 -4 -5 -5 March 7 March 28 -6 -6 March 14 April 4 -7 March 21 -7 -8 -8 15 22 29 7 14 21 28 4 Feb. Mar. Apr. ADP-FRB Paid Employment Cumulative Job Loss since 15 Feb 2020 Weekly, SA Millions of Jobs 2 2 0 0 -2 -2 -4 -4 -6 -6 -8 -8 -10 -10 -12 -12 -14 March 7 March 28 -14 -16 March 14 April 4 -16 March 21 -18 -18 -20 -20 15 22 29 7 14 21 28 4 Feb. Mar. Apr. Note:ADP-FRBfiguresrefertothepayperiodsincludingeachSaturday. Source:ADP,authors’calculations. Figure3: Real-TimeEstimatesofPayrollEmploymentsinceFebruary152020 3.2 SectorLevelData Since ADP microdata also contain information about NAICS industry of PACs, we construct estimates of active and paid employment at the “supersector” level.12 These estimates are presentedinTable1andFigures4-7.13 In absolute terms, the most notable cumulative drops in paid employment were in leisure and hospitality (almost 4 million); trade, transportation, and utilities (almost 2.5 million); and professional and business services (1.3 million). Large paid employment losses were also 12OurADPsupersectordefinitionscorrespondtothebroadsupersectordivisionsreportedinCESreleases. 13Note that supersector estimates do not necessarily add up to topline estimates, since the topline estimate is constructedindependentlybyusingaggregateddataonly. 9

Percentchangeinemployment Abilitytotelework Activeemp Paidemp (share,inpercent) Miningandlogging -2.9 -2.9 ND Construction -2.2 -9.3 17.2 Manufacturing -2.1 -5.5 30.3 Trade,transportation,andutilities -4.1 -9.4 15.8 InformationServices -2.6 -4.7 53.3 FinancialServices -1.3 -1.9 57.4 ProfessionalandBusinessServices -2.4 -6.6 53.4 EducationandHealthServices +0.7 -4.5 25.9 LeisureandHospitality -10.5 -30.5 8.8 OtherServices -2.7 -9.9 27.7 Table1: CumulativeEmploymentChangesandAbilitytoTelework recorded in education and health (1 million); manufacturing (0.7 million); and construction (0.65 million). In relative terms, the largest employment losses were in leisure and hospitality (30.5 percent), other services (9.9 percent), trade, transportation, and utilities (9.4 percent), and construction (9.3 percent). Table 1 also reportsgeneral ability to telework as estimated by the BLS.14 It appears that the largest employment losses occurred in sectors with low general abilitytotelework. 14See,https://www.bls.gov/news.release/pdf/flex2.pdf.Forsimilarestimates,seealsoDingelandNeiman (2020). 10

Mining and Logging Mining and Logging Cumulative Job Loss since 15 Feb 2020 ADP FRB First Differences Weekly, SA Millions of Jobs Weekly, SA Millions of Jobs 0.02 0.02 0.02 0.02 0.01 0.01 0.01 0.01 0.00 0.00 0.00 0.00 -0.01 -0.01 -0.01 -0.01 -0.02 -0.02 -0.02 -0.02 -0.03 -0.03 -0.03 -0.03 ADP-FRB Active Employment ADP-FRB Active Employment ADP-FRB Paid Employment ADP-FRB Paid Employment -0.04 -0.04 -0.04 -0.04 -0.05 -0.05 -0.05 -0.05 15 22 29 7 14 21 28 4 15 22 29 7 14 21 28 4 Feb. Mar. Apr. Feb. Mar. Apr. Construction Construction Cumulative Job Loss since 15 Feb 2020 ADP FRB First Differences Weekly, SA Millions of Jobs Weekly, SA Millions of Jobs 0.0 0.0 0.0 0.0 -0.1 -0.1 -0.1 -0.1 -0.2 -0.2 -0.2 -0.2 -0.3 -0.3 -0.3 -0.3 -0.4 -0.4 -0.4 -0.4 -0.5 -0.5 -0.5 -0.5 ADP-FRB Active Employment ADP-FRB Active Employment -0.6 ADP-FRB Paid Employment -0.6 -0.6 ADP-FRB Paid Employment -0.6 -0.7 -0.7 -0.7 -0.7 -0.8 -0.8 -0.8 -0.8 15 22 29 7 14 21 28 4 15 22 29 7 14 21 28 4 Feb. Mar. Apr. Feb. Mar. Apr. Manufacturing Manufacturing Cumulative Job Loss since 15 Feb 2020 ADP FRB First Differences Weekly, SA Millions of Jobs Weekly, SA Millions of Jobs 0.0 0.0 0.0 0.0 -0.1 -0.1 -0.1 -0.1 -0.2 -0.2 -0.2 -0.2 -0.3 -0.3 -0.3 -0.3 -0.4 -0.4 -0.4 -0.4 -0.5 -0.5 -0.5 -0.5 ADP-FRB Active Employment ADP-FRB Active Employment -0.6 ADP-FRB Paid Employment -0.6 -0.6 ADP-FRB Paid Employment -0.6 -0.7 -0.7 -0.7 -0.7 -0.8 -0.8 -0.8 -0.8 15 22 29 7 14 21 28 4 15 22 29 7 14 21 28 4 Feb. Mar. Apr. Feb. Mar. Apr. Note:Themostrecentweekofthesectorseriesisnotcorrectedforincompletereportingofbiweekly,semimonthly, andmonthlypayers. Source:ADP,authors’calculations. Figure4: SectorPayrollEmploymentsinceFebruary152020 11

Trade, Trans. and Util Trade, Trans. and Util Cumulative Job Loss since 15 Feb 2020 ADP FRB First Differences Weekly, SA Millions of Jobs Weekly, SA Millions of Jobs 0.50 0.50 0.2 0.2 0.25 0.25 0.00 0.00 -0.0 -0.0 -0.25 -0.25 -0.2 -0.2 -0.50 -0.50 -0.75 -0.75 -0.4 -0.4 -1.00 -1.00 -1.25 -1.25 -0.6 -0.6 -1.50 -1.50 -0.8 -0.8 -1.75 ADP-FRB Active Employment -1.75 ADP-FRB Active Employment ADP-FRB Paid Employment ADP-FRB Paid Employment -2.00 -2.00 -1.0 -1.0 -2.25 -2.25 -2.50 -2.50 -1.2 -1.2 15 22 29 7 14 21 28 4 15 22 29 7 14 21 28 4 Feb. Mar. Apr. Feb. Mar. Apr. Information Services Information Services Cumulative Job Loss since 15 Feb 2020 ADP FRB First Differences Weekly, SA Millions of Jobs Weekly, SA Millions of Jobs 0.10 0.10 0.10 0.10 0.05 0.05 0.05 0.05 -0.00 -0.00 -0.00 -0.00 -0.05 -0.05 -0.10 -0.10 -0.05 -0.05 -0.15 -0.15 -0.10 -0.10 -0.20 ADP-FRB Active Employment -0.20 ADP-FRB Active Employment ADP-FRB Paid Employment ADP-FRB Paid Employment -0.15 -0.15 -0.25 -0.25 -0.30 -0.30 -0.20 -0.20 15 22 29 7 14 21 28 4 15 22 29 7 14 21 28 4 Feb. Mar. Apr. Feb. Mar. Apr. Financial Services Financial Services Cumulative Job Loss since 15 Feb 2020 ADP FRB First Differences Weekly, SA Millions of Jobs Weekly, SA Millions of Jobs 0.20 0.20 0.05 0.05 0.15 0.15 0.00 0.00 0.10 0.10 -0.05 -0.05 0.05 0.05 -0.10 -0.10 -0.00 -0.00 -0.15 -0.15 -0.05 -0.05 -0.20 -0.20 ADP-FRB Active Employment -0.10 ADP-FRB Active Employment -0.10 -0.25 ADP-FRB Paid Employment -0.25 ADP-FRB Paid Employment -0.30 -0.30 -0.15 -0.15 -0.35 -0.35 -0.20 -0.20 15 22 29 7 14 21 28 4 15 22 29 7 14 21 28 4 Feb. Mar. Apr. Feb. Mar. Apr. Note:Themostrecentweekofthesectorseriesisnotcorrectedforincompletereportingofbiweekly,semimonthly, andmonthlypayers. Source:ADP,authors’calculations. Figure5: SectorPayrollEmploymentsinceFebruary152020(continued) 12

Prof. & Bus. Services Prof. & Bus. Services Cumulative Job Loss since 15 Feb 2020 ADP FRB First Differences Weekly, SA Millions of Jobs Weekly, SA Millions of Jobs 0.3 0.3 0.2 0.2 0.0 0.0 0.1 0.1 -0.2 -0.2 0.0 0.0 -0.4 -0.4 -0.1 -0.1 -0.2 -0.2 -0.6 -0.6 -0.3 -0.3 -0.8 -0.8 -0.4 -0.4 -1.0 ADP-FRB Active Employment -1.0 -0.5 ADP-FRB Active Employment -0.5 ADP-FRB Paid Employment -0.6 ADP-FRB Paid Employment -0.6 -1.2 -1.2 -0.7 -0.7 -1.4 -1.4 -0.8 -0.8 15 22 29 7 14 21 28 4 15 22 29 7 14 21 28 4 Feb. Mar. Apr. Feb. Mar. Apr. Edu. & Health Edu. & Health Cumulative Job Loss since 15 Feb 2020 ADP FRB First Differences Weekly, SA Millions of Jobs Weekly, SA Millions of Jobs 0.3 0.3 0.2 0.2 0.2 0.2 0.1 0.1 0.0 0.0 0.0 0.0 -0.2 -0.2 -0.1 -0.1 -0.4 -0.4 -0.2 -0.2 -0.6 -0.6 -0.3 -0.3 -0.4 -0.4 -0.8 -0.8 ADP-FRB Active Employment -0.5 ADP-FRB Active Employment -0.5 -1.0 ADP-FRB Paid Employment -1.0 -0.6 ADP-FRB Paid Employment -0.6 -1.2 -1.2 -0.7 -0.7 -1.4 -1.4 -0.8 -0.8 15 22 29 7 14 21 28 4 15 22 29 7 14 21 28 4 Feb. Mar. Apr. Feb. Mar. Apr. Leisure + Hosp. Leisure + Hosp. Cumulative Job Loss since 15 Feb 2020 ADP FRB First Differences Weekly, SA Millions of Jobs Weekly, SA Millions of Jobs 0.0 0.0 0.00 0.00 -0.5 -0.5 -0.25 -0.25 -1.0 -1.0 -0.50 -0.50 -1.5 -1.5 -2.0 -2.0 -0.75 -0.75 -2.5 -2.5 -1.00 -1.00 -3.0 -3.0 -1.25 -1.25 -3.5 -3.5 ADP-FRB Active Employment -1.50 ADP-FRB Active Employment -1.50 -4.0 ADP-FRB Paid Employment -4.0 ADP-FRB Paid Employment -4.5 -4.5 -1.75 -1.75 -5.0 -5.0 -2.00 -2.00 15 22 29 7 14 21 28 4 15 22 29 7 14 21 28 4 Feb. Mar. Apr. Feb. Mar. Apr. Note:Themostrecentweekofthesectorseriesisnotcorrectedforincompletereportingofbiweekly,semimonthly, andmonthlypayers. Source:ADP,authors’calculations. Figure6: SectorPayrollEmploymentsinceFebruary152020(continued) 13

Other Services Other Services Cumulative Job Loss since 15 Feb 2020 ADP FRB First Differences Weekly, SA Millions of Jobs Weekly, SA Millions of Jobs 0.1 0.1 0.0 0.0 -0.0 -0.0 -0.1 -0.1 -0.2 -0.2 -0.1 -0.1 -0.3 -0.3 -0.2 -0.2 -0.4 -0.4 -0.5 -0.5 -0.3 -0.3 ADP-FRB Active Employment ADP-FRB Active Employment -0.6 ADP-FRB Paid Employment -0.6 ADP-FRB Paid Employment -0.4 -0.4 -0.7 -0.7 -0.8 -0.8 -0.5 -0.5 15 22 29 7 14 21 28 4 15 22 29 7 14 21 28 4 Feb. Mar. Apr. Feb. Mar. Apr. Note:Themostrecentweekofthesectorseriesisnotcorrectedforincompletereportingofbiweekly,semimonthly, andmonthlypayers. Source:ADP,authors’calculations. Figure7: SectorPayrollEmploymentsinceFebruary152020(continued) 14

3.3 TheDistributionofEmploymentChangeamongSmallBusinesses Small businesses (typically defined as those with fewer than 500 employees) have been a particularfocusofpolicymakers,buttheyareespeciallydifficulttostudyduetodatalimitations. WenextusetheADPmicrodatatocharacterizethedistributionofemploymentchangesamong small businesses as the U.S. entered the COVID-19 crisis. A particular advantage of the ADP data, as we note elsewhere, is the ability to distinguish between “active” employees (i.e., employees existing on payroll records, regardless of whether they received pay in a given pay period) and “paid” employees (i.e., employees who received pay in a given pay period). The difference between these measures may shed some light on the permanence of employment contractions—completeremovalofemployeesfrompayrollrecordsmayreflectpermanentjob destruction,whileadeclineinpaidemploymentcouldreflectbothpermanentandtemporary layoffs. ADP dataafford a richcharacterization ofthe distribution ofemployment changesamong smallbusinessunits,bothintermsofactiveandpaidemployees.15 Figure8reportsthedistribution of cumulative employment changes from February 15 through March 28 among ADP businesses operating in both periods16. For each initial employment size class (defined based on active employment), we report percentiles of employment change. The top panel corresponds to changes in active employment, while the bottom panel refers to changes in paid employment. Thefigurerevealsstrikingpatterns. We focus first on active employment, which shows markedly smaller declines than paid employment. Among the smallest initial size class (1-9 employees), every decile shows zero cumulativechangeinactiveemployment;thismaybedueto“integerproblems”(i.e.,forvery smallbusinessunits, evenasingleemployeelayoffreflectsasubstantialchangeinproductive capacity), or it may reflect selection issues in which businesses of this size that come under stress must exit entirely rather than shedding employment (in which case they do not appear 15HereweagainemphasizethatADPpayrollunitsmaynotmapdirectlytoeitherfirmorestablishmentconcepts usedinofficialdata;inthissection;wetreatpayrollunitsas“smallbusinesses”withtheusualcaveatthatsomeof themmayactuallybesmallestablishmentsoflargerfirms. 16Forthisfigure,weuserawdatawithoutseasonaladjustmentforsimplicity. Moreover,thecalculationsforthis figureareperformeddirectlyonthemicrodata,withoutadditionalweightingandindexmachineryusedelsewhere inthispaper. 15

Cumulative Growth Rate by Initial Business Size ADP-FRB Active Employment Percentile by Size Growth Rate 10 10 5 5 0 0 -5 -5 -10 Initial Business Size -10 1 - 9 10 - 19 20 - 49 50 - 99 100 - 249 250 - 499 -15 -15 -20 -20 10 20 30 40 50 60 70 80 90 Percentile Growth Rate Deciles by Initial Business Size ADP-FRB Paid Employment Percentile by Size Growth Rate 20 20 10 10 0 0 -10 -10 -20 -20 -30 -30 -40 -40 -50 -50 -60 Initial Business Size -60 -70 1 - 9 10 - 19 20 - 49 -70 -80 50 - 99 100 - 249 250 - 499 -80 -90 -90 -100 -100 10 20 30 40 50 60 70 80 90 Percentile Source:ADP,authors’calculations. Figure8: DistributionofCumulativeEmploymentChangesfromFebruary15toMarch28 inthesecalculations). Within the next initial size class (10-19 employees), however, the 10th percentile business saw an active employment decline of more than 10 percent over this period, though the 90th percentilebusinesssawgainsofalmost10percent. Roughlyspeaking,asidefromthesmallest (1-9) size category, absolute employment changes in either direction were smallest among the largest businesses; the interdecile range for the largest size class (250-499) is about 12 percentage points, while the range for the 10-19 class was more than 19 percentage points. It is also strikingthatmanybusinessessawsubstantialemploymentgainsoverthisperiod,eventhough the figure is limited to businesses traditionally thought of as “small” (i.e., fewer than 500 em- 16

ployees). Themedianbusinessineverysizeclasssawzerochangeinactiveemployment,with positivegainsamongmanyabove-medianbusinesses. We next turn to the bottom panel of Figure 8, which illustrates the employment change distribution in terms of paid employment. Here was see substantially larger moves. The 10th percentile business within every size class saw declines of at least 30 percent, with the largest class shown (250-499) seeing a decline of almost 80 percent. Even the smallest business size class(1-9)sawsubstantialdeclinesamongmanybusinesses. Thispanelalsorevealsextremely widedispersioninoutcomes;theinterdecilerangeofpaidemploymentchangesvariesfrom50 percentage points among the smallest businesses to over 90 percentage points for the 250-499 sizeclass. Ofcourse,“integerproblems”maybeplayingaroleamongthesmallestbusinesses. Interestingly, Bartik et al. (2020) (using rich Homebase data on small business hourly employment and hours) find that declines in hours worked between January and late March are driven almost entirely by business shutdowns (i.e., zero hours worked in a week) and hours reductions among retained employees, with minimal contribution from layoffs of hourly employees.17 Figure 8 is not markedly inconsistent with this while adding considerable color. Themediansmallbusinessshowslittlemovementineitheractiveorpaidemploymentbetween mid-February and late March, suggesting minimal layoffs among surviving businesses. But underlying this median result is considerable distributional variation: a bit less than half of survivingbusinessesexperiencednontrivialemploymentdeclines,andafewbusinessesexperienced dramatic declines approaching 80 percent. On the other hand, some businesses have experienced sizable employment gains. That being said, some tension remains between the resultsofBartiketal.(2020)andourresultsfromADPdatainthissectionandelsewhereinthe paper. Generallyspeaking,inothersectionswehavedocumentedsubstantialdeclinesinboth paidandactiveemploymentamongcontinuingbusinesses,andFigure8showsthatthehalfof businesses showing declines in active employment saw declines that are larger in magnitude thanthegainsamongtheroughlyhalfofbusinessesthatsawgains;andforpaidemployment, declines are seen up to the 60th percentile of small businesses. In other words, ADP data do suggest substantial layoffs among continuing businesses. A possible reason for this discrep- 17Homebasedatatrack“local”typesofbusinesseswithconcentrationinretailandleisureandhospitalityindustries;seeBartiketal.(2020)fordetails. 17

ancyisthatBartiketal.(2020)observebusinessshutdownswhenzerohourlyemployeesclock in; it is likely that some businesses are staying “open” with at least a few salaried employees, sotheycountascontinuingbusinessesinADPdatadespiteregisteringnohourlyemployment. Taken together, the various insights from Figure 8 reveal striking heterogeneity in the experiences of small businesses as the U.S. entered the COVID-19 crisis. The relatively muted moves in active employment (relative to paid employment) might, one would hope, suggest that many businesses do not see their layoffs as permanent (though it may also reflect recordkeeping inattentiveness), while the extreme declines in paid employment among more than half of small businesses leave ample room for concern about the state of the small business economy. 4 Conclusion Inthecomingmonths,theBLS’smonthlylaborreportswillcaptureboththedepthandbreadth of the current employment losses. Unfortunately, the speed of the recent declines makes it necessary to turn to nontraditional data, which can provide some higher-frequency insight abouteventsastheyunfold. Theresponsivenessofpolicymakersoverthepastfewweekshas been, in part, engendered by an understanding of how rapidly the COVID-19 pandemic has afflicted the economy. A broad spectrum of timely and high-frequency economic information hasfacilitatedthisunderstanding. Lookingforward,inadditiontothemonthlyandweeklyemploymentindexes,wearecurrentlygeneratingasetofdiffusionindexes,measuresofentryandexit,sizeandindustryclass measures, andanemploymentindexsolelycomprisedofindividualswithweeklypaychecks. Thiscorpusoflabordatawillnotonlyfurtherourunderstandingoftherapiddeteriorationof theemploymentsituation,butwillprovideevidenceoftimingandmagnitudeoftheeventual reboundinthelabormarket. References Bartik,AlexanderW.,MarianneBertrand,FengLin,JesseRothstein,andMattUnrath.2020. 18

“Labor Market Impacts of COVID-19 on Businesses: Update with Homebase Data Through April8.”mimeo. Brynjolfsson, Erik, John Horton, Adam Ozimek, Daniel Rock, Garima Sharma, and Hong YiTuYe.2020.“COVID-19andRemoteWork: AnEarlyLookatUSData.”mimeo. Cajner,Tomaz,LelandCrane,RyanA.Decker,AdrianHamins-Puertolas,ChristopherKurz, and Tyler Radler. 2018. “Using Payroll Processor Microdata to Measure Aggregate Labor Market Activity.” Board of Governors of the Federal Reserve System (U.S.) FEDS Working Paper2018-005. Cajner, Tomaz, Leland D. Crane, Ryan A. Decker, Adrian Hamins-Puertolas, and Christopher Kurz. 2019a. “Tracking the Labor Market with "Big Data".” Board of Governors of the FederalReserveSystem(U.S.)FEDSNotes2019-09-20. Cajner, Tomaz, Leland D. Crane, Ryan A. Decker, Adrian Hamins-Puertolas, and Christopher Kurz. 2019b. “Weekly Payroll Employment Data for the United States.” Board of GovernorsoftheFederalReserveSystem(U.S.)mimeo. Cajner, Tomaz, Leland D. Crane, Ryan A. Decker, Adrian Hamins-Puertolas, and Christopher Kurz. 2020. “Improving the Accuracy of Economic Measurement with Multiple Data Sources: TheCaseofPayrollEmploymentData.”BigDatafor21stCenturyEconomicStatistics. UniversityofChicagoPress. Cho, David. 2018. “The Labor Market Effects of Demand Shocks: Firm-Level Evidence from theRecoveryAct.”mimeo. Cleveland, William P., and Stuart Scott. 2007. “Seasonal Adjustment of Weekly Time Series withApplicationtoUnemploymentInsuranceClaimsandSteelProduction.”JournalofOfficialStatistics,23(2):209–221. Coibion, Olivier, Yuriy Gorodnichenko, and Michael Weber. 2020. “Labor Markets during theCovid-19Crisis: APreliminaryView.”mimeo. 19

Dingel, Jonathan I, and Brent Neiman. 2020. “How Many Jobs Can be Done at Home?” NationalBureauofEconomicResearchWorkingPaper26948. Grigsby,John,ErikHurst,andAhuYildirmaz.2019.“AggregateNominalWageAdjustments: NewEvidencefromAdministrativePayrollData.”NBERWorkingPaper25628. Haltiwanger,John.2020.“ApplicationsforNewBusinessesContractSharplyinRecentWeeks: AFirstLookattheWeeklyBusinessFormationStatistics.”UniversityofMarylandmimeo. 20