UnemploymentInsuranceandDisabilityInsuranceintheGreatRecession AndreasI.Mueller * ColumbiaUniversityandIZA JesseRothstein UniversityofCalifornia,BerkeleyandNBER TillM.vonWachter UCLA,NBERandIZA September2013 Abstract: Disabilityinsurance(DI)applicationsandawardsarecountercyclical.Onepotential explanationisthatunemployedindividualswhoexhausttheirunemployment Insurance(UI)benefitsuseDIasaformofextendedbenefits.Weexploitthe haphazardpatternofuibenefitextensionsinthegreatrecessiontoidentifythe effectofuiexhaustionondiapplication,usingbothaggregatedataatthestatev monthandstatevweeklevelsandmicrodataonunemployedindividualsinthe CurrentPopulationSurvey.WefindnoindicationthatexpirationofUIbenefits causesdiapplications.ourestimatesaresufficientlyprecisetoruleouteffectsof meaningfulmagnitude. *WethankChrisHansman,EricJohnson,JeehwanKim,andAnaRoccaforexcellentresearch assistance.rothsteinisgratefultotherussellsagefoundationandthecenterforequitablegrowth atucberkeleyforfinancialsupport.muellerandvonwachter sresearchwassupportedbytheu.s. SocialSecurityAdministrationthroughgrant#1DRC12000002V01V00totheNationalBureauof EconomicResearchaspartoftheSSADisabilityResearchConsortium.Thefindingsandconclusions expressedaresolelythoseoftheauthorsanddonotrepresenttheviewsofssa,anyagencyofthe FederalGovernment,ortheNBER.EVmail:amueller@columbia.edu,rothstein@berkeley.edu, tvwachter@econ.ucla.edu.
1 I. Introduction+ As of theendof2012, 8.8 million adult Americans received Social Security DisabilityInsurance(SSDI)benefits.SSDIisasocialinsuranceprogramthatcollects mandatory premiums from workers and uses them to pay benefits to former workerswhohavebecomedisabled. 1 Figure1plotstheshareoftheworkingVage populationreceivingssdiovertime.itshowsthatthissharehasmorethandoubled since 1990. The rapid growth has prompted concerns about SSDI s sustainability, and recent projections indicate that the SSDI trust fund will be exhausted in 2016 (SocialSecurityAdministrationBoardofTrustees,2012). As Figure 1 indicates, the growth rate of SSDI rolls accelerated during the recessionsoftheearly1990sandearly2000s,andperhapsduringthe2007v2009 recession as well.figure 2 illustrates the number of applications for SSDI benefits and the number of new awards, both expressed as shares of the civilian nonv institutional population aged 20V64, along with the unemployment rate. Since the 1980s,SSDIapplicationsandawardshaverisenindownturns,thenfallenbeginning ayearortwoaftertheunemploymentpeak(black,daniel,andsanders2002;autor andduggan2003;dugganandimberman2009;coeetal.2012).ssdiapplications percapita,forexample,roseata6.7%annualratebetween1989and1994,fellata 4.6%annualrateduringtheexpansionyears1994through1999,thenroseagainat an10.5%annualratebetween1999and2004.dugganandimberman(2009)find that between 1984 and 2003 a one percentage point increase in the national 1Another program, SSI, provides benefits to disabled adults and children based on financial need, regardlessofworkhistory.ssicaseloadshavealsogrownrapidly.
2 unemployment rate was associated with an increase of roughly 8V9 percent in the number of applications filed for SSDI benefits. They conclude that nearly one quarter of the rise in male SSDI participation between 1984 and 2003 can be attributed to the recessions of the early 1990s and early 2000s. 2 The cyclical patternisnotablyweakerafter2004(vonwachter2010).applicationsdeclinedat onlya0.3%annualratebetween2004and2007,thengrewata6.5%rate farfrom proportionaltothemagnitudeofthegreatrecession between2007and2011. 3 Neither the older strongly countercyclical pattern nor its dampening in the lastdecade arewellunderstood.oneexplanation for countercyclical application ratesthatwouldbegenerallyconsistentwiththepurposesofthessdiprogramis thatemployers willingnesstohire(andmakeaccommodationsfor)individualswith moderately workvlimiting disabilities may vary with the tightness of the labor market.ssdieligibilityisrestrictedtoindividualswithfunctionalimpairmentsthat preventthemfromperformingtheirpreviousjobsorfromadjustingtoothertypes ofwork.theworker sage,education,andexperienceareconsideredinassessinghis orhersuitabilityforalternativeemployment;asthejobsavailabletoaworkerwith a given profile likely depend on economic conditions, there may well be workers who meet the medical eligibility criteria in bad economic times who would not be consideredtobesufficientlydisabledwerethelabormarkettighter. 4 2Othercontributingfactorsincludeanagingpopulation,increasedfemalelaborforceparticipation (which increases women s eligibility for SSDI benefits), more generous benefits, rising income inequality,andchangesinthedisabilitydeterminationprocess(dugganandimberman,2009). 3The slow decline after the 2001 recession is consistent with other evidence that the subsequent expansionwasrelativelytepid. 4Inprinciple,medicaleligibilitydoesnotdependontheavailabilityofpositions,butitseemslikely that workers qualifications are in practice judged relative to labor demand. JoffeVWalt (2013)
3 Other potential explanations for the cyclical sensitivity of SSDI applications attribute it to moral hazard. Consider a worker with a moderate health problem e.g., back pain that makes work unpleasant but not impossible. In principle, this worker should not be eligible for SSDI. But if he applies, a generous medical examiner might award him benefits (JoffeVWalt2013). His decision to applywill dependinpartonthegenerosityofssdibenefitsrelativetothemarketwagethathe cancommand.ifarecessionreduceshismarketwage,hemaybetippedoverinto SSDIapplication(AutorandDuggan2003;Black,Daniel,andSanders2002). A related hypothesis is that workers use SSDI to insure employment losses rather than wage declines. Displaced workers can generally claim unemployment insurance (UI) benefits. But UI is timevlimited and recessions are associated with sharpincreasesinunemploymentduration.workerswhoexhausttheiruibenefits butwhoarestillunabletofindworkmayturntossdiforongoingincomesupport. SSDIrecipientstendtoremainontheprogram,andoutofthelabormarket, untilretirement(autorandduggan,2006).asaresult,anyuseofssdiasasource of extended unemployment benefits is extremely expensive. Indeed, a backvofvthev envelope calculation, discussed below, suggests that savings from avoided SSDI casescouldplausiblyfinancealargeshareofthecostsofextensionsofuibenefits. ButlittleisknownaboutthedegreetowhichSSDIisinfactusedinthisway. This paper uses data from the Great Recession and its aftermath to investigate the relationship between UI exhaustion and SSDI applications. Our describesadoctorwho believesheneeds[toknowindividuals educationalattainment]indisability casesbecausepeoplewhohaveonlyahighschooleducationaren tgoingtobeabletogetasitvdown job.
4 analysistakesadvantageofagreatdealofvariabilityofuibenefitdurationsduring thedownturn.potential benefits reached as high as 99 weeks in 2009, remained highforseveralyears,thendeclinedsubstantiallyin2012. 5 AteachpointinthisperiodtherewassubstantialcrossVsectionalvariation,, duetovagariesofstatelawandtodiscontinuoustriggersinfederalprograms.this meantthatworkerslaidoffatroughlythesametimewereeligibleforverydifferent UIdurationsdependingonthelocationandexacttimingofthelayoff,andthusthat UI exhaustion rates varied substantially over time and across states. We use this variation to identify the effect of UI exhaustion on SSDI usage, using timevseries analyses, statevbyvmonth panels, event studies of weekly SSDI applications surrounding UI extensions, and microdata on unemployed workers to isolate differentcomponentsofthevariationinexhaustiontiming. SeveralrecentpapershaveexploredUIVDIinteractions.LindnerandNichols (2012)usevariationinbenefitamountsandeligibilitycriteriatoidentifythecausal effectofuiparticipationondiapplicationdecisions.themostrelevantpapertothe current project is Rutledge (2012). With both aggregate statevmonth application data and microdata from the Survey of Income and Program Participation (SIPP), RutledgeexaminestheeffectofUIbenefitdurationextensionsonSSDIapplication decisions and allowance rates. He focuses on the effect of a UI extension on the behaviorofthosewhowerealreadyclaiminguiwhenthespellwasannounced. 5ManymodelsshowthatUIshouldbemoregenerousduringrecessions(e.g.,Landais,Michaillat,and Saez2010),asmoralhazardcostsarerelativelylowandconsumptionsmoothingbenefitshighwhen unemploymentiselevated.afulldiscussionofoptimaluidesignisbeyondthescopeofthispaper.
5 WeextendRutledge sanalysisinthreeimportantways.first,ourconceptual modelviewsuiextensionsasasourceofvariationinthetimetouiexhaustion ratherthanasadirectdeterminantofssdiapplications,consistentwitha behavioralmodelinwhichindividualsmakedecisionsbasedonthebenefits availabletothemwithoutregardtothelegallabelingofthosebenefits.second,our empiricalspecificationsarecloselytiedtothisconceptualmodel,andarethuseasily interpretableintermsofthedeterminantsoftheunderlyingapplicationdecision. ThiscontrastswithRutledge sspecifications,whicharenotcloselyalignedtoa behavioralmodelandfocusonlegallabeling isanextensionineffectornot? ratherthanontrueincentives.third,weintroducetwonewdatasourcesthathave notbeenusedpreviouslytostudyuivdiinteractions.wehaveobtainedaccessto microadministrativessadatathatweusetotabulateweeklyssdiapplicationsand thecorrespondingawardrates.wealsousematchedcurrentpopulationsurvey (CPS)samplestoexamineindividualVleveldeterminantsofDIreceipt. II. A+simple+model+of+UI4DI+interactions+ Autor and Duggan (2003) model the choice between work and SSDI applicationformarginallydisabledworkers.theyshowthatsomepartiallydisabled workerswillstayintheirexistingjobs,butifdisplacedwillprefertoexitthelabor forceinordertoreceivedibenefitsratherthantosearchforanewjobatalower wage. Autor and Duggan interpret the cyclicality of SSDI applications as an indicationthattherearemeaningfulnumbersofworkersofthistype. Autor and Duggan s (2003) model does not incorporate unemployment insurance.weextendtheirmodeltodoso,drawingonrothstein s(2011)modelof
6 UI and job search. In our model, a displaced worker can choose in each period whethertosearchforworkortoremainidle. 6 Onlysearchcanleadtoanewjob, whileadiapplicationcanbesubmittedonlywhenintheidlestate. SearcherspaysearchcostscUandhaveaprobabilityfoffindingemployment eachperiod.theycandrawonuptonperiodsofunemploymentbenefits,worthbui perperiod.bycontrast,workersoutofthelaborforcedonotpaysearchcostsbut haveprobability0offindingemploymentandcannotdrawuibenefits. Inaperiodthatanindividualisoutofthelaborforce,heorshemayapplyfor DI benefits by paying an application cost ca. The probability that an application is successfulisp.weassumethatdieligibilitydecisionsareperfectlycorrelatedover time,sothataworkerwhoisrejectedoncewillnotlaterreapply.aworkerwhose applicationissuccessfulcandrawapervperiodbenefitofbdiinanyfutureperiodin whichheorsheisoutofthelaborforce,untilsuchpointasheorsheisreemployed. Thisbasicsetupgivesrisetoadynamicdecisionproblemwithstatevariables n {0,1,,N},indexingthenumberofweeksofUIbenefitentitlementremaining, and A {0; V1; 1}, describing the worker s DI entitlement. A=0 indicates a worker who has not applied for DI benefits; A=V1 a worker who has applied but been rejected; and A=1 a worker who has been awarded benefits.lettingδ indicate the discountrate,u(y)theflowutilityassociatedwithpervperiodcashincomey, 7 andve thecontinuationvalueofanewjob,theutilityassociatedwithjobsearchis: 6AsUIbenefitsarepaidonlytoworkerswithsufficientworkhistorieswhoareinvoluntarily displaced,wefocusonworkerswhopreferworktossdiapplication,sowillnotvoluntarilyquit existingpositionsinordertoapplyfordibenefits. 7Wedonotmodelsavingorborrowing.
7 VU(n,A)=u(bUI)VcU+δ[fVE+(1Vf)max{VU(nV1,A),VI(nV1,A)}]forn>0and VU(0,A)=u(0)VcU+δ[fVE+(1Vf)max{VU(0,A),VI(0,A)], where VI represents the value of idleness. 8 This depends on the worker s DI applicationstatus.thosewhohave not yetappliedfordibenefits or who have appliedbutbeenrejectedreceive: VI(n,A)=u(0)+δmax{VU(n,A),VI(n,A)},forA {0;V1}andanyn 0. ThosewhohavebeenapprovedforDIbenefitsreceive: VI(n,1)=u(bDI)+δmax{VU(n,1),VI(n,1)}. Finally,theutilityofaworkerwhoappliesforDIbenefitsis: VA(n,0)=u(0) ca+δ[pmax{vu(n,1),vi(n,1)}+ (1Vp)max{VU(n,V1),VI(n,V1)}]. Figure 3 shows how the worker s policy choice varies with f and p, for a particularsetofotherparameters.first,workerswithhighjobvfindingprobabilities searchforworkuntiltheyfindjobs,evenbeyondtheexpirationoftheiruibenefits. Thisistheupperareainthefigure.Second,inthelowerleft,workerswithlowjobV findingprobabilitiesbutalsolowdiawardprobabilitiessearchforworkuntiltheir UIbenefitsareexhausted,thenexitthelaborforcewithoutapplyingforDI. 9 Third, workersinthelowerright region,withveryhighdiawardprobabilitiesbutvery low jobvfinding chances, simply apply for DI immediately after displacement, 8Becauseweassumethatparametersarestationary,itcanbeshownthatanyworkerwhochooses searchwitha=0andn 1willalsochoosesearchthefollowingperiod.ThemaxoperatorsintheV U expressionsarethusrelevantonlyforn=1. 9Withtheparametervaluesused,jobsearchisworthwhileforthedurationofUIbenefitsevenifthe jobvfindingprobabilityiszero,astheuibenefitislargerthanthesearchcost.ifb UIislowenough relativetoc U,however,apolicyofexitingthelaborforceimmediatelyafterjoblossbecomesoptimal forlowvf,lowvpworkers.
8 withouteverlookingforwork.finally, workers with somewhat lower DI award chances and/or somewhat higher jobvfinding probabilities search for work until theiruibenefitsareexhausted,thenapplyfordibenefits. ItisthislasttypeofworkerthatcouldproduceacausaleffectofUIbenefit durationsondiapplications,astheseworkerscanbedeterredfromapplyingfordi byauiextension.forsome,thisistemporary theywillstillbejoblessattheendof theextendedbenefits,andwillapplytodithen.butotherswillfindjobsduringthe extended search period, and thus may be permanently diverted from the DI program. This diversion could be substantial. To see this, suppose that {f, p} have a uniform distribution on [0, 0.1] X [0, 1] among displaced workers and that other parametersareasinfigure3.then17%ofworkers,and35%ofthosewhoexhaust 26periodsofUIbenefits,areoftheUIVbeforeVDItype.WhenUIbenefitslastfor26 weeks, UIVbeforeVDI workers comprise 83% of DI applicants and 79% of DI awardees.theaverageuivbeforevdidiapplicanthasapervperiodjobvfindingrateof 1.5%.Thus,somewouldfindjobsifgivenlongerUIbenefitdurationsduringwhich tosearch.withourparameters,a26vperiodextensionofuibenefits(toatotalof52 periods)wouldpermitjustunderonevthirdoftheuivbeforevdiworkerswhowould otherwise apply for DI to instead find new jobs before their benefits run out. This would reduce steadyvstate DI applications and awards by a bit over onevquarter, whileincreasinguipaymentsbyabout40%. An effect of this magnitude would be enormously important. Because individuals awarded DI benefits tend to draw them until retirement, the present
9 valueofasinglediawardisaround$300,000.bycomparison,weeklyuipayments average around $300. Thus, the parameters used in Figure 3 and a uniform distribution of {f, p} imply that DI savings from a 26Vweek UI extension would amounttooverthreetimestheonvbudgetcostofthatextension.inotherwords,a UI extension would be selfvfinancing even if the effect on steadyvstate DI awards wereonlyonevthirdaslargeasinthissimplesimulation. But the parameters used are just approximations, and the assumption of a uniform {f, p} distribution is entirely unsupported. It seems more likely, for example,thatfandparenegativelycorrelated.thiswouldincreasetheshareofuiv beforevdiworkers,thoughperhapsalsoreducetheiraveragejobvfindingrates.nonv uniformityofthetwomarginaldistributionscouldoffsetanysucheffect.theeffect ofuibenefitdurationondiapplicationsisthusanempiricalquestion. III. Data+and+DI+trends+ WerelyonthreedatasourcestomeasuretrendsinSSDIapplicationand receipt.first,weusepubliclyavailabletabulationsfromthesocialsecurity Administration(SSA)ofSSDI,SSI,orSSDI/SSIapplicationsatthestateVbyVmonth levelbetweenaugust2004anddecember2012. Second,weobtainedaccesstoSSA sdisabilityresearchfile,arestrictedv accessmicrodatafilecoveringtheyears2008v2010andcontainingobservationson individualssdiapplicationslinkedtoapplicationoutcomes.weusethesedatato constructastatevbyvweekpanelofapplicationcounts.wealsocalculateeventual awardratesforeachweeklyapplication cohort,usinginformationonawardsover theremaininghorizoninthesample.
10 Third,weusetheAnnualSocialandEconomicSupplement(ASEC) supplementtothecurrentpopulationsurvey(cps),administeredinthespringof eachyear. 10 Respondentsareaskedabouttheirincomefromvarioussourcesinthe previouscalendaryear.thosewhoreportincomefromsocialsecurityareaskedto listreasonsforthis.wemeasuressdireceiptasthepresenceofpositivesocial Securityincomeforsomeonewhonames disability asoneofthereasons. Figure4showstrendsinthenumberofdisabledworkerSSDIrecipientsfrom thepublishedssadata,alongwithtwoseriescomputedfromthecpsasec.one seriescountsallindividualsaged16andoverwhoreportsocialsecuritydisability income.thesecondexcludesthoseoverage66(67after2009,reflectingan increaseinthefullretirementage),asdisabledindividualsabovethisagereceive retirementpaymentsratherthanssdi.theformerseriesmatchesthe administrativerecordsreasonablywell,thoughshowsasomewhatflattertrajectory. Thelatterisnotablylower,suggestingboththatmanyrecipientscontinuebelieving theyarereceivingdisabilitybenefitsevenaftertheyareformallyconvertedtothe retirementprogramandthatthecpssurveymissessometruessdirecipients. Intheanalysisbelow,weidentifyunemployedworkers,aged20V64,inthe basicmonthlycpssurveyandaskwhethertheexpirationoftheiruibenefitsearly incalendaryearyisassociatedwithahigherprobabilityofreceivingssdiincomein thatyear.thisismadepossiblebytherotatingpaneldesignofthecps,which meansthatjustunderhalfoftherespondentsinthey+1asecfilecanbematchedto 10TheASECisoftenknownasthe MarchCPS. ItborrowstheMarchsamplefromtheregular monthlycpssurvey,supplementingthiswithportionsofthefebruary,april,andnovember(ofthe previousyear)monthlycpssamples.
11 basiccpsinterviewsinfebruary,march,oraprilofyeary,orinnovemberofyear yv1.thecpsisanaddressvbasedsample,somatchesareonlypossiblefor individualswhodonotmovebetweensurveys.weareabletomatcharound95%of ASECrespondentstooneofthesurroundingmonthlysurveys.Mergesbetween yearvyandyearvy+1asecsaremoredifficult,withmatchratesaround75%. 11 InthebasicCPSsurvey,unemployedworkersareaskedthereasonfortheir unemployment(e.g.,layoffvs.voluntaryquit)andthenumberofweeksthatthey havebeenunemployed.weusetheformertoproxyforuieligibilityandthelatterto assigneachunemployedindividualtothedateofdisplacement.wethenusea databaseofstateuirules,discussedinsectioniv,toassignthedatethattheworker wouldhaveexhaustedhisuibenefitsifhewaseligibleforfullbenefitsandifhe drewbenefitscontinuouslyfromthedateofdisplacementuntilexhaustion. IV. UI+during+the+Great+Recession+and+its+aftermath+ A. Extended)UI)Programs) Workersdisplacedfromcoveredemploymentwithsufficientworkhistories are generally eligible for up to 26 weeks of regular unemployment insurance benefits.butattimesduringthelastfewyears,workerswhohaveexhaustedtheir regular benefits might have drawn as many as 53 additional weeks of Emergency Unemployment Compensation (EUC) and as many as 20 more weeks of Extended Benefits (EB),bringingthetotalashighas99weeks.There has been substantial 11Thisexcludesobservationsthatshouldnotmatchduetothestructureofthesurvey(e.g.,thosein theirsecondsamplerotationinyeary).about1%ofmonthlyvtovasecmatchesand6v8%ofasecvtov ASECmatchesshowdiscrepanciesinage,race,gender,oreducation.Discrepantobservationsare discarded.
12 variationinthismaximumovertimeandacrossstates,resultingfromdifferencesin statepolicies,fromchangingfederallaw,andfrom triggers thatconditionedboth EUCandEBbenefitsonstateeconomicconditions. TheEUCprogramwasfirstauthorizedinJune2008. 12 Itinitiallyprovided13 weeks of federallyvfinanced benefits to supplement the regular 26 weeks. At the time,therecessionwasexpectedtoberelativelybrief,andeucwassettoexpirein March2009.Asthedownturnprovedtobedeeperandlongerlastingthaninitially expected, EUC was gradually expanded. In November 2008, EUC benefits were extendedto33weeksinstateswithunemploymentratesabove6percentandto20 weekselsewhere.theywereextendedagaininnovember2009,to34weeksinlow unemploymentstatesand53weeksinhighunemploymentstates. EUC complemented a preexisting program, EB, which was designed to providesupplementalweeksofbenefitsintimesofeconomicdistress.stateschoose whethertoparticipateineband,iftheyparticipate,selectfromamenuofpossible triggersthatwillactivateebbenefits.activationprovides13weeksofebbenefits (ontopoftheregularandeuceligibility),or20weeksinstatesthathaveadopteda moregeneroustriggerandthathaveunemploymentratesabove8%.thefirststate tobecomeeligibleforebbenefitsinthegreatrecessionwasalaska,injune2008; fiveadditionalparticipatingstatesbecameeligiblebyjanuary2009. ThecostofEBbenefitsisordinarilysplitbetweenthestateandtheFederal government,buttheamericanrecoveryandreinvestmentactof2009(arra;also 12Itresembledother,similartemporaryprogramscreatedinpastrecessions.Thediscussionhere drawsonrothstein(2011)andfujita(2010).
13 knownastherecoveryact)providedforfullfederalfunding.afterthis,anumber of states passed legislation to adopt the program or to liberalize their triggers.by May2009,recipientsin27statescouldreceiveEBbenefits,and11oftheseoffered 20weeksofbenefits.Eligibilitycontinuedtoexpand,withbetween36and39states payingebbenefitsthroughmostoflate2009,2010,andearly2011. Both EUC and EB benefits were gradually rolled back starting in mid 2011. TheEBrollbackwaslargelyautomatic,duetorulesthatconditioneligibilityonnot just a high but also a rising unemployment rate. During the aftermath of the recession,unemploymentremainedhighbutslowlydeclined.thenumberofstates payingebbenefitsfellthroughthesecondhalfof2011andthefirsthalfof2012.by July2012,onlyIdahowasstillpayingbenefits;ittriggeredoffinearlyAugust. The major rollback of EUC came in February 2012, when legislation made severalchanges.first,eucdurationswerecutby6to14weeks,dependingonthe state unemployment rate(though states with rates between 7 and 8.5% or above 9% were unaffected). Second,further cuts were scheduled for September 2012. Third, additional weeks of EUC benefits were provided to highvunemployment states that did notqualifyfor (ordidnotparticipatein)the EB program. This provisionprovidedtenextraweeksinmarch,april,andmayof2012;noneinjune, July,andAugust;andfourextraweeksfromSeptemberonward. On top of the basic story of haphazard expansion and rollback, additional variation in EUC durations arose from the temporary nature of the program. The program was initially set to expire in March 2009.In February 2009, the ARRA
14 extendeditthroughdecemberofthatyear. 13 During2010,Congressthenextended itseveraltimesforonlyafewmonthseach:fromdecember2009tofebruary2010, then to April, to June, and to November 2010.Several of these extensions were retroactive, authorized only after the program had already expired. The first expirationlastedonlyafewdays,buttwootherslastedforabouttwoweekseach andinjuneandjuly2010theprogramwasallowedtoexpireforafullsevenweeks. AlongerVtermextensionfinallytookeffectinDecember2010. Figure 5 shows the average, minimum, and maximum number of weeks of benefitsavailableovertimethroughtherecession,combiningtheregular,euc,and EBprograms.ThisfigureismadefromadatabaseofUIavailabilityatthestateVbyV week level, constructed by Rothstein(2011) but updated here to the end of 2012. Maximum benefit durations reached 99 weeks from late 2009 through mid 2012, andtheaveragestatewasclosetothemaximumthroughmuchofthisperiod.states begantofallawayfromthemaximumduringearly2012. ThethreeexpirationsoftheEUCprogramin2010arequiteprominentinthe figure,asdurationsfalldramaticallyineach.however,thesharpdeclinesindicated likelyoverstatethechangesexperiencedbyindividualrecipients.eucbenefitsare dividedintotiers atitspeak,the53weeksofmaximumeucbenefitsweredivided intofourtiersof20,14,13,and6weeks,respectively.whentheprogramexpired recipients were permitted to continue to draw benefits until they exhausted their current tier but could not begin a new tier, while people who exhausted their 13ARRAalsomadeUIbenefitsmoregenerousinanumberofways,includingbyprovidinga $25/weeksupplementtoUIbenefitsandbyexemptingthefirst$2,400ofbenefitsfromincometaxes. Bothprovisionsweretemporary.
15 regular benefits were not permitted to enter the EUC program. 14 This tended to smooth over the expirations, limiting the disruption produced. Butthedegreeof smoothing depended importantly on the exact date of job loss, as this determined theworker spositioninthetierstructureatthetimeofeucexpiration. Each eventual reauthorization provided for the retroactive payment of benefits to individuals who would have received EUC but for the temporary exhaustion. The longvterm unemployed are unlikely to have substantial liquid savings or easy access to credit (Gruber 1997), however, so many may have felt seriousfinancialcrunchesduringtheexpirations. B. Modeling)UI)Exhaustion) The complex history of EUC and EB created a great deal of variation in the durationofuibenefitsandthusinthetimingofuiexhaustion.unfortunately,while the Employment and Training Administration (ETA) compiles weekly counts of initialuiclaims,nocomparabledataseriesisavailableforexhaustions.wetaketwo approachestoapproximatingthenumberofexhaustions. OurfirstexhaustionseriesisconstructedfromstateVbyVmonthlevelETAdata onthenumbersoffirstpaymentsandfinalpaymentsineachprogramandeuctier. For each state in each month,we compute the number of final payments in any program or tier minus the number of first payments in the EUC tiers or EB.This closelyapproximatesexhaustion,buttherearethreesourcesofslippage.first,this 14100%federalfinancingofEBexpiredeachtimetheEUCprogramdid.Manystatesconditioned theirebparticipationoncontinuedfederalfunding,andcutoffebbenefitswithinaweekortwoof thejune2010expiration.ebbenefitslostduringthisperiodwereingeneralnotpaidretroactively.
16 method incorrectly counts as exhaustions individuals who found new jobs or abandonedtheirjobsearchesupontheexpirationofaparticulartierorprogrambut who had more benefits available on another tier or program. Second, when individualsreceivetheirfinalpaymentsfromoneprogramortierinthelastweekof a calendar month, the initial payment on the next program or tier appears in the next month s data. This creates excess volatility in measured exhaustions. Third, when EUC benefits were expanded when new tiers were introduced, when the program was retroactively reauthorized, or when a state triggered on to new benefits many people received first payments who had not received final payments in the previous week. We estimate negative numbers of exhaustions at thesetimes.thesemomentsarequiteusefulforidentificationofuieffects,however, as they represent periods when UI exhaustions were low or zero. We present analysesbelowthatzeroinondiapplicationdynamicssurroundinguiextensions. ThesolidlineinFigure6showstheestimatednumberofUIexhaustionseach month, using this method. Exhaustions were fairly stable, at around 210,000 per month, through early 2008. Measured exhaustions turned sharply negative in July and August of 2008, following the creation of EUC. They then became volatile, bouncing around a lower mean through the rest of 2008 and 2009 with two dips into negative terrain following EUC expansions in February 2009 and December 2009VJanuary 2010. Exhaustions spiked enormously during the temporary EUC expiration in June 2010, only to turn negative again in August 2010 after the programwasreauthorized.followingthisepisode,theserieshasbouncedarounda levelsimilartothatseenbeforetherecessionbuthigherthanthe2008v9average.
17 Although the spikes and negative values represent measurement problems, thebroadpatterns declinesinexhaustionsin2009v10followedbyanincreasein 2011V12 correspond to real dynamics. In2009V10, benefit durations were quite long,andmanyrecipientsfoundjobsorexitedthelaborforcebeforetheyexhausted benefits,whilethecohortsthatwereapproachingexhaustionwereprimarilythose thathadlosttheirjobsbeforetherecessionsowerenotparticularlylarge.in2011v 12, durations remained long, but the large 2009 cohorts were exhausting their benefits,offsettingtheeffectofextendeddurationsontheexhaustionrate. We simulate an alternative UI exhaustion measure to use as a check on the administrative data. Webegin with weekly data on initial claims for regular UI benefits by state. We then use our statevbyvweek database of UI availability to identify the week that each entering UI cohort would have exhausted its benefits, assumingeligibilityforfullbenefitsandcontinuousclaiming.next,weestimatethe probability that an individual entering unemployment in each week would have survivedinthatstatus(ratherthanbecomingreemployedorexitingthelaborforce) until the expiration of benefits. The survival probabilities are described in the appendix;theyarebasedonestimatedaverageuiexithazardsthatareallowedto vary smoothly over time and discretely with unemployment duration. By multiplyingthesizeoftheenteringcohortbythesurvivalprobability,weestimate the number of UI exhaustions produced by the cohort when its benefits end, then
18 aggregateacrossallcohortsthatexhaustedtheirbenefitsineachmonthtoconstruct anestimatedexhaustionseries. 15 TwoseriesobtainedviathismethodareplottedinFigure6,correspondingto differentdefinitionsof exhaustion. Thefirstseries,plottedasadottedline,judges an individual to have exhausted her benefits in the first week that she did not receiveanonvtimebenefitpayment,evenifshewaslaterpaidretroactivelyforthat week. This series mirrors the general trends in the administrative measure, but shows zero exhaustions rather than negative numbers in months following EUC introduction and expansions. It also, however, shows an enormous spike in June 2010,whenEUCwasallowedtoexpire.(Thisdatapointiscensoredinthegraphto controltheoverallscale;infact,theseriesshowsnearly2.5millionexhaustionsthat month.) It is unclear whether this accurately reflects the expirations that are relevant to SSDI application decisions.ifrecipientswereconfidentthatcongress wouldeventuallyreauthorizetheprogramretroactivetoitsexpiration,andifthey had access to sufficient credit to borrowagainsttheireventualbenefits,thisspike dramaticallyoverstatesthenumberoftrueexhaustions. Our second simulated exhaustion series, graphed as a dashed line, counts individuals to exhaust their benefits only when they receive their final payments under any program, ignoring temporary breaks that are repaid retroactively. This doesnotshowapronouncedspikeinjune2010butdoesabetterjobofmirroring 15ThereisanadditionaladjustmenttoaccountforthefactthatnotallclaimsforUIbenefitsleadto actualbenefitpayments.
19 the patterns in the administrative data in 2011. We use this as our preferred exhaustionseriesintheanalysesbelow. Oursimulatedfinalexhaustionseriesexplains9%ofthetimeseriesvariation in the administrative data measure (and 21% when JuneVAugust 2010 are excluded). There is substantial acrossvstate variation concealed behind the aggregate time series shown in Figure 6.New York, for example, saw essentially zero exhaustions in 2008 and 2009, while Virginia saw as many or more exhaustionseachmonthin2008asbeforetherecession.weexploitthisvariationin many of the estimates below. A natural concern is that the statevbyvmonth exhaustion measures may be particularly noisy at the statevbyvmonth level. However, they do seem to have substantial signal: The elasticity of the administrativedataexhaustionmeasurewithrespecttosimulatedfinalexhaustions, controlling for state and month effects, is 0.24, with a standard error of 0.03. 16 WhenweexcludetheJune August2010period,theelasticityrisesto0.28. V. Analyses+of+UI4DI+interactions+using+aggregate+data+ Inthissection,wepresenttimeVseries,stateVbyVmonthpaneldata,andstateV byvweek event studies of the relationship between UI exhaustions and DI applicationsaswellasawardrates.recallthatthemodelinsectioniisuggestedthat somemarginallydisableduirecipientsmightbeinducedtoapplyforssdibenefits by the impending or actual exhaustion of their UI benefits. This would imply a 16Asanalternativetomodelinglogexhaustions therearemanyzerosatthestatevbyvmonthlevel wenormalizemonthlyexhaustionsineachstatebytheaveragenumberofmonthlyexhaustionsin thestatein2005v2007.theelasticityreportedinthetextisbasedonthenormalizedseries,which weuseforallfurtheranalyses.
20 positive correlation betweenui exhaustions and SSDI applications. Insofar as the marginaldiapplicantsarelesslikelytobeawardedbenefits,itshouldalsoproduce anegativecorrelationbetweenuiexhaustionsandssdiacceptancerates. A. Time)Series)Analyses) We begin by overlaying our simulated final UI exhaustion series with the numberofmonthlyssdiapplications,infigure7.thereislittlesigninthisgraphof a positive relationship between UI exhaustions and DI applications. Though UI exhaustions fell to well under half of their usual rate through most of 2009, DI applications rose by about 20% in late 2008 and early 2009. 17 UI exhaustions returned to close to their prevcrisis level in late 2010; DI applications plateaued aroundthattimeandhaveremainedroughlystablesince. Table 1 presents timevseries analyses of the log of seasonallyvadjusted aggregate monthly DI applications. The first column includes only the simulated numberoffinaluiexhaustionsinthemonth,measuredasashareoftheiraverage level during calendar years 2005V2007. The coefficient is negative, the opposite of theexpectedsignifuiexhaustionsleadtodiapplications,butisinsignificantand small.column2addsaquadratictimetrend,whilecolumn3addsacontrolforthe unemployment rate. The unemployment rate coefficient is positive and quite precisely estimated, indicating that a one percentage point increase in unemployment is associated with a 3.9% increase in DI applications. The UI 17WeseasonallyadjusttheDIseriesusingstateVlevelregressionsoflogmonthlyapplicationson calendarmonthdummies,controllingforquadratictimetrends,anindicatorforobservationssince February2009,andthenumberofweeksinthemonth.Wethensumadjustedstateapplicationsto formanationalseries.
21 exhaustion coefficient becomes positive and marginally significant (t=2.01) when the unemployment rate is controlled, but is quite small: A doubling of UI exhaustionsisassociatedwithonlya1.5%increaseindiapplications. Column 4 adds several controls: the number of initial UI claims, seen as proxies for economic conditions; an indicator for JuneVAugust 2010 observations, whentheexpirationofeucmakesitdifficulttomeasureperceiveduiexhaustions; andanindicatorfortheperiodafterfebruary2009.thesehaveessentiallynoeffect onthecoefficientofinterest. Column5addstheaveragesofthreeleadsandthreelagsofUIexhaustions. EachofthesemightcapturetrueeffectsofUIexhaustionsonDIapplications,which need not be exactly contemporaneous. But there is little indication that the contemporaneousspecificationmissesanimportantpartoftheresponse neither the lag nor the lead is significant, the contemporaneous effect is basically unchanged,andthepointestimateofthecumulativeeffectisalmostexactlyzero. Columns6V8explorealternativemeasuresofUIexhaustions.Incolumn6we usethesimulatedseriesforinitialexhaustions(thedottedlinefromfigure6),while incolumn7weusetheexhaustionseriescomputedfromadministrativerecordson EUCandEBinitialandfinalpayments(thesolidlinefromFigure6).Neitherofthese series indicates any relationship between exhaustions and DI applications. Finally, in column 8 we replace the counts of exhaustions with an indicator for the four months in which our simulations suggest that there were zero UI exhaustions, immediately following the introduction of the EUC program in mid 2008 and its
22 expansion in late 2009.This specification indicates that DI applications fell about 1.9%inthesemonths,implyingsimilarresponsivenesstothatfoundincolumns3V5. Alltold,thespecificationsinTable1indicatethatanyeffectofUIexhaustions on DI applications is quite small and sensitive to the way that exhaustions are measured. By contrast, there is a robust and large relationship between the unemployment rate and DI applications that does not appear to reflect an associationbetweenoverallunemploymentanduiexhaustions. B. Panel)Data)Analyses) We nextturntopaneldataanalysesoflog monthly DI applications at the statelevel,intable2.theseallowustocontrolforotherfactorsthatinfluencethe time pattern of DI applications, identifying the exhaustion effect from differences across states in exhaustion trends.there is substantial variation in these trends, driveninpartbythetimingoflayoffsandinpartbyvariationinuiavailability. Column 1 begins with a simple specification that includes state and month fixed effects, the unemployment rate, and the statevlevel index of final UI exhaustions. The unemployment rate coefficient is positive and significant, though somewhat smaller than in Table 1. The UI exhaustion coefficient is almost exactly zero.moreover,itisextremelypreciselyestimated,withastandarderrorlessthan halfofthoseintable1,andwecanthusruleoutelasticitiesofdiapplicationswith respecttouiexhaustionslargerthan0.005. Columns2and3explorealternativecontrolsforeconomicconditions.These havelittleeffectontheresults.column4includeslagsandleadsoftheexhaustion index. These are both insignificant, and the point estimates indicate a cumulative
23 elasticityofdiapplicationswithrespecttoexhaustionsofonly0.018.incolumn5, weincludeeachofthethreeleadsandthreelagsoftheexhaustionseriesseparately. Pointestimates(notshown)indicateacumulativeelasticityof0.004,withnegative coefficients for the contemporaneous and immediate leads and lags and positive coefficientsonthelongerleadsandlags.thisistheoppositeofthepatternthatone wouldexpectfromacausaleffectofanticipatedorrecentpastuiexhaustion. Column6excludestheJuneVAugust2010observations,whenUIexhaustions aredifficulttodefineprecisely.thishaslittleeffect. TherearetwosourcesofvariationinoursimulatedUIexhaustionmeasure: VariationinthesizeofenteringUIcohorts(i.e.,inthenumberofnewclaimants)and variationinthedurationofuibenefits.wehavealsocreatedalternativesimulations thatholdthe cohort size constant, so that benefit durations are the only source of variation. When we use these measures as instruments for the original measures, results are quite similar to those seen in Table 2, and the upper bounds of the confidenceintervalsareifanythingsmaller. Finally,columns7V9ofTable2exploreouralternativeUIexhaustionseries. They indicate slightly more positive effects, though they still rule out elasticities largerthan0.006.moreover,column9indicatesthatdiapplicationsriseinmonths whennewuiextensionstakeeffect,andtheconfidenceintervalrulesoutdeclines largerthan0.4%.wereturntothisinvestigationbelow. The published data cannot be used to examine award rates, as awards are reported for the month of final adjudication rather than for the month of initial application.asanalternative,weusethessamicrodatatoexaminetheacceptance
24 rate for SSDI applications filed in each state in each month in 2008, 2009, and 2010. 18 Table3presentsresultsparalleltothoseinColumns1,4,7and8ofTable2. <RESULTS(SUPPRESSED(PENDING(DISCLOSURE(REVIEW.>( ThepaneldataanalysesinTable2offernosignthatDIapplicationsrespond to UI exhaustions. We can always rule out elasticities larger than 0.02, and most specificationsruleoutelasticitiesonevquarterthissize. Atthispoint,itisworthconsideringhowlargeaneffectwouldneedtobeto bequantitativelyimportant.onewaytoapproachthisistocomparetheempirical estimatestotheelasticitiesimpliedbythetoymodelinsectionii.inthatmodel,a doubling of UI durations reduced steadyvstate UI exhaustions by about half and steadyvstate DI applications by a quarter. (The shortvrun effects would be much larger.)theestimatesintable2,then iftheycanbeinterpretedascausal imply much,muchsmalleruiexhaustioneffects. AnotherapproachistocomparethecostofUIextensionstotheresultingDI savings.asnotedearlier,thepresentvalueofadiawardisaround$300,000,while UIbenefitscostaround$300perweek.Thus,ifextendingUIbenefitsbyoneweek divertsevenoneinonethousandrecipientsfromgoingondi,thedisavingswould paytheentirecostoftheuiextension. However,thefirstVordereffectofaUIextensionislikelytobetomerelydelay DI applications rather than to permanently displace them. Rothstein (2011) estimatesthatthelongvtermunemployedhadmonthlyjobvfindingratesaround10 percent through 2009 and 2010.If we suppose that marginal DI applicants have
25 similar jobvfinding rates to this and if we assume a DI award rate of onevthird, roughlymatchingtherecentaverage,thenafourvweekuiextensionwouldbefully financedthroughdisavingsifitdeterred120diapplicationsper1,000potentialui exhaustees.thisalmostcertainlyunderstatestheneededamountofdeterrence,as marginaldiapplicantsareprobablylessemployablethantheaveragelongvtermui recipientandlikelyhavelowerawardratesthanaveragediapplicants. Recall that the estimates in Table 2 always ruled out elasticities of DI applicationswithrespecttouiexhaustionslargerthan0.02.diapplicationsareof thesameroughorderofmagnitudeasuiexhaustions,sothisimpliesareductionof not more than 20 DI applications per 1000 UI exhaustees whose benefits are extended,wellbelowthebreakveven point.moreover,thisisbasedontheupper limitoftheconfidenceintervals;pointestimatesimplyzeroorevennegativeeffects. C. Event)Analyses)) Wenextuseouradministrativemicrodatatoconducteventstudiesof weeklydiapplicationsintheperiodsimmediatelysurroundingextensionsofui benefits.thesehaveseveralpotentialadvantagesovertheanalysesabove.first, theydonotrequireustorelyonourimperfectuiexhaustionmeasures;wecanbe confidentthatuiexhaustionsdeclineddrasticallyfollowingnewbenefitextensions. Second,theeventstudyframeworkallowsustomoreflexiblyexaminethetime patternofanyapplicationresponsestouiextensions.third,theonlyleverbywhich
26 UIexhaustionmightbemanipulatedistheextensionofUIbenefits,soreducedVform eventstudiesofuiextensionsaredirectlyinformativeaboutpolicyeffects. 19 Inimplementingtheeventstudy,wefacetwochallenges.First,wecannot identifytheindividualsatriskofuiexhaustionindidata.thus,asabove,we examinetheeffectofuiextensionsonaggregatediapplications.second,many statessawrepeateduiextensionsoverrelativelyshortperiodsin2008and2009, whichmakesitdifficulttodistinguishlongvruneffectsofoneextensionfromshortv runeffectsofthenext.thus,whileafullassessmentoftheimpactofuiextensions wouldconsiderthecumulatedneteffect,startingfromthedatethattheextensionis firstanticipatedandextendinguntilwellafterthelastcohortaffectedbythe extensionexhaustsitsuibenefits,wefocusonshortervrunimpactsandon extensionsthatdonotcloselyoverlap. WedefineeventdatesastheweeksonwhichUIextensionscameintoeffect, asreportedin TriggerNotices publishedbytheu.s.departmentoflabor.we estimatespecificationsoftheform: "#("## " ) = + + + " + " + ", where"## " representsthelogofthenumberofssdiapplicationsfiledinstates inweekt. and aretimeandstatefixedeffects,respectively. measuresthe differencefromthenationalweeklytrendkweeksafter(or k weeksbefore,when k<0)theeventdate,andnisthenumberofweekstheextensionwasinplace.xst 19Onecaninterprettheeventstudyestimatesasthe reducedforms correspondingto2sls estimatorsinwhichuibenefitextensionsareusedasinstrumentsforuiexhaustion.wediscussed 2SLSestimateslikethisabove.
27 containspolynomialsofdegreethreeforthestatevlevelunemploymentrateaswell asthestatevlevelinsuredunemploymentrate.notethatthestatevlevel unemploymentrateisonlyavailableatthemonthlyfrequency. <RESULTS(SUPPRESSED(PENDING(DISCLOSURE(REVIEW>( VII.+Analysis+of+UI4DI+interactions+using+Current+Population+Survey+microdata+ Alloftheaboveanalysesareecological,aimingtotietrendsinaggregateDI applications and awards to trends in UI exhaustion. As a final exercise we turn to ourmergedcpsmicrodatatoexaminetheindividualvlevelrelationshipbetweenui exhaustionanddireceipt. Table4presentssummarystatisticsforthemergedCPSsample,poolingdata for calendar years 2005 2011 (with ASEC observations from the 2006V2012 surveys).werestrictthesamplethroughouttoindividualsaged20v64(inthebase monthsurvey),andwethosewhoseunemploymentspellsstartedin2003orearlier. Thefirstcolumnpresentsstatisticsforthefullsample(N=240,163).75%are employedattheinitialmonthlysurvey(generallyinmarchofyeary,thoughsome arefromfebruaryorapril,orfromnovemberofyearyv1),5%areunemployed,and 21%areoutofthelaborforce.Thesubsampleofunemployedworkersisdescribed in column 2, while column 3 summarizes job losers. Unfortunately, we cannot measureuireceiptdirectly.however,thereasonforunemploymentappearstobe an adequate proxy: Of those who were unemployed at the initial monthly survey and said that they had been involuntarily displaced, 40% report on the following March sasecsurveyhavingpositiveuiincomefortheyear.thiscomparestoonly
28 9%ofthosewhosaythattheyhadvoluntarilylefttheirpreviousjoborwerenew entrantsorreentrantstothelaborforce. Column 4 presents statistics for UIVeligible workers who would have exhausted their benefits before the end of the year in which they were initially observed,hadtheyremainedunemployedforthatlong.(ofcourse,notallworkers reached that point some presumably were reemployed before exhausting their benefits.butwecannotmeasurethesetransitions.)alluirecipientsin2005v2007 are in this category, as the base surveys were completed by April and UI benefits lasted only 26 weeks in those years.inlateryears,onlyworkerswhohadalready beenunemployedforsometimebytheinitialsurveywereatriskofexhaustingtheir UIbenefitswithinthecalendaryear. Of the UI eligible workers in the base month survey, 57% would have exhausted their benefits by the end of the calendar year, and 37% would have exhausted them by the midpoint of the year. The potential exhaustees closely resemble the overall pool of UI eligible workers in their demographic characteristics.theaveragetimeofexpirationisinearlymarchofyeary. The final rows of the table show the share of individuals who report in the yearvy+1asecsurveyhavingreceivedssdiincomeduringyeary.thisis3.2%for thefullsample,butoverhalfofthesealsoreportedhavingssdiincomeinyearyv1. We exclude them from our analysis of the effect of UI expiration. Only1.4% of individualswhodidnothavessdiincomeinyearyv1haditinyeary.unemployed workersandparticularlyjoblosershavebelowvaveragessdireceiptrates.among UIrecipients,thosewhowouldhaveexhaustedtheirbenefitsbeforetheendofyear
29 yhavesomewhathigherssdirecipiencyratesthandothosewhosebenefitswould havecontinuedbeyondtheendoftheyear. Table5presentsouranalysisofUIexpirationandDIreceiptinthematched CPSVASECsample.Weestimatespecificationsoftheform: DIisy=logit(URsyβ+LFisyγ+Xisyδ+Disyθ+κs+πy), wherediisyisanindicatorforreceiptofssdiincomebyindividualiinstatesinyear y(asreportedonthey+1asecsurvey);ursyistheunemploymentrateinstatesin yeary;lfisyisavectorofmeasuresoftheindividual slaborforcestatus,including dummies for unemployment and NILF(employment is the excluded category) and measuresofunemploymentduration;andxisyisameasureofuiexhaustionbefore theendofyeart.disyisavectorofdemographiccontrols dummiesforages40v49, 50V54,55V59,and60V64,with20V39theexcludedcategory;alinearagecontrol;and agenderindicator.κsandπyarefixedeffectsforstatesandyears,respectively. Column 1 presents a specification that includes only the demographic controls, state and year FEs, and the state unemployment rate. The latter enters with a negative coefficient, though it is insignificant and the implied effect is very small. Column 2 adds indicators for four labor force statuses at the base survey: Unemployedduetojobloss,unemployedduetovoluntaryquitortolabormarket entry or reentry, nonvparticipation in the labor force due to disability, and nonv participationforotherreasons.(theexcludedcategoryisemployment.)thosewho are not employed at the base survey have substantially higher probabilities of receivingssdithanarethebasevsurveyemployed.thoseoutofthelaborforcehave
30 higher probabilities than the unemployed, particularly so for those who attribute theirnonvparticipationtodisability. Logit coefficients can be difficult to interpret, particularly when positive outcomesarerare.wethususethecoefficientstopredicthowmuchlowerthessdi receipt rate would be if these individuals had instead been employed. These estimates, reported in the bottom rows of Table 5, indicate that unemployment accounts for twovthirds of the observed 0.97 percentage point rate of DI receipt amongtheunemployed. Column 4 adds the unemployment duration(measured as of the end of the calendaryear,assumingthattheinitialunemploymentspelllastsuntilthen),alone andinteractedwithuieligibility.thosewhoareemployedoroutofthelaborforce atthebasesurvey are assigned durations of 0.The longervterm unemployed are morelikelytoreceivessdiincomethanarethoseunemployedforshorterperiods, particularlyamongjobvlosers. Column5addstwoUIexpirationmeasures.Thefirstisacontinuousmeasure oftime(inyears)fromthedateofuiexpirationtotheendofthefocalyear.those whosebenefitscontinuedbeyondtheendofthefocalyeararecodedaszeros.the second measure is an indicator for an expiration before June 30. Because DI applicationstakeseveralmonthstoprocess,anindividualwhoseuibenefitsexpired lateintheyearandwhoappliedforssdiimmediatelythereafterwouldbeunlikely to receive DI income in that calendar year.those whose applications were filed early in the year, however, should have reasonable probabilities of receiving DI incomebytheendoftheyear.thus,ifuiexpirationsleadinrelativelyshortorderto
31 DI applications and if some of those applications are successful, both variables shouldbepositivelyassociatedwithdireceipt. 20 The estimates do not indicate this. Expiration before June 30 has a positive coefficient while the continuous time since UI expiration is negative, but they are not individually or jointly (p=0.84) significant, and both are quite small. The coefficientsimplythatdelayingalluiexhaustionsbeyondtheendofthefocalyear would reduce DI receipt among the unemployed by only 0.01 percentage points. This result is unaffected by the addition of quadratic and cubic terms in the state unemploymentrate,incolumn6. Column7restrictsthesampletotheunemployedandcolumn8tojoblosers. These dramatically reduce the sample size, reducing precision. Point estimates indicatesomewhatlargerexhaustioneffects inthefinalcolumn,theysuggestthat expiration of UI benefits raises the probability of DI receipt among the UIVeligible unemployedby0.32percentagepoints. Recall that Figure 6 indicated that approximately 250,000 individuals exhausttheiruibenefitseachmonth.theestimateincolumn8oftable5indicates that this induces about 800 DI awards over the next 6V12 months. This should be inflatedbyperhaps50%toaccountforawardsmadeonappeal,foratotalof1,200 eventualinducedawards. 21 Thisisabout1.4%oftheaveragenumberofawardsper month in recent years. This figure is strikingly consistent with the application 20ImplicitinthisparameterizationisanassumptionthattheprobabilityofDIreceiptriseswiththe timeelapsedsincethediapplication,perhapsparticularlyquicklyaround6monthsaftertheinitial application. 21BenítezVSilvaetal.(1999)estimatethat46%ofapplicantsareawardedbenefitsinthefirststageof reviewandthatthisrisesto73%afterappeals.firstvstageawardsaremadein5months,onaverage, butawardsmadeonappealtakeanaverageof15months.
32 elasticitiesobtainedfromtheaggregateanalysesabove,andagainindicatesthatany effectsofuiexhaustionondiuptakearequitesmallrelativetotheoverallflow. VIII.++ Conclusion+ ThispaperhasusedtheunevenextensionofUIbenefitsduringandafterthe Great Recession to isolate variation in UI exhaustion that is not confounded by variation in economic conditions more broadly. Using a variety of analytical strategies,wehaveexaminedtherelationshipbetweenuiexhaustionanduptakeof DIbenefits.Noneoftheanalysespresentedhereindicateameaningfulrelationship. Although we cannot rule out small effects, all of the analyses indicate that the elasticityofdiapplicationswithrespecttouiexhaustionis0.02orsmaller,fartoo small to account for the cyclical pattern of DI application or to contribute meaningfullytothecostvbenefitanalysisofuiextensions. There are a number of caveats to this result.most importantly, we must make assumptions about the timing of DI applications and awards induced by UI exhaustion. For the aggregate analyses, we must assume that any induced applicationsoccurwithinthreemonths(beforeorafter)thedateofuiexhaustion, while our CPS analysis can detect only induced applications thatleadtoreceiptof paymentswithinthesamecalendaryearasanearlieruiexhaustion.theremaybe effectsatlongerlags UIexhausteesmaywaitsixmonthsormorebeforeapplying for SSDI, or awards made to exhaustees might be disproportionately likely to requireanappealofaninitialrejection.thesepossibilitiesmeanthatacausallink betweenuiexhaustionanddicannotbeconclusivelyruledout.
33 Nevertheless,theanalysisherecounselsagainstthelikelihoodofsuchalink. It rather tends to support alternative explanations for the countercyclicality of DI applications. For example, the cyclical pattern may simply reflect variation in the potentialreemploymentwagesofdisplacedworkers(davisandvonwachter2011) or changes in the employment opportunities of the marginally disabled that influence SSA s evaluation of the applicant s employability. These alternative explanationsmayhavequitedifferentpolicyimplicationsthanwouldalinktoui.it is not clear, for example, that more stringent functional capacity reviews would reducerecessionvinduceddiclaimsiftheseclaimsreflectexaminers judgmentsthat theapplicantsaretrulynotemployableintheextantlabormarket.
34 References+ Autor,DavidH.andMarkG.Duggan.2003.Theriseinthedisabilityrollsandthe declineinunemployment.quarterly(journal(of(economics118,no.1:157 206. Autor,DavidH.andMarkG.Duggan.2006.ThegrowthintheSocialSecurity disabilityrolls:afiscalcrisisunfolding.journal(of(economic(perspectives20, no.3:71 96. Black,Dan,KermitDaniel,andSethSanders.2002.Theimpactofeconomic conditionsonparticipationindisabilityprograms:evidencefromthecoal boomandbust.american(economic(review92,no.1:27 50. Coe,NormaB.,KellyHaverstick,AliciaH.Munnell,andAnthonyWebb.2011.What explainsstatevariationinssdiapplicationrates?workingpaperno.2011v 23,CenterforRetirementResearchatBostonCollege. Davis,StevenJ.andTillvonWachter.2011.Recessionsandthecostsofjobloss. Brookings(Papers(on(Economic(Activity,Fall:1V61. Duggan,MarkandScottA.Imberman.2009.Whyarethedisabilityrolls skyrocketing?thecontributionofpopulationcharacteristics,economic conditions,andprogramgenerosity.inhealth(at(older(ages:(the(causes(and( consequences(of(declining(disability(among(the(elderly,ed.davidm.cutlerand DavidA.Wise.Chicago:UniversityofChicagoPress. Fujita,Shigeru.2010.Economiceffectsoftheunemploymentinsurancebenefit. Business(Review((Federal(Reserve(Bank(of(Philadelphia),FourthQuarter. JoffeVWalt,Chana.2013.UnfitforWork:TheStartlingRiseofDisabilityinAmerica. PlanetMoneyforThis(American(Life.http://apps.npr.org/unfitVforVwork/ (accessedmarch22,2013). Landais,Camille,PascalMichaillat,andEmmanuelSaez.2010.Optimal unemploymentinsuranceoverthebusinesscycle.workingpaperno.16526, NationalBureauofEconomicResearch,Cambridge,MA. Lindner,StephanandAustinNichols.2012.Theimpactoftemporaryassistance programsondisabilityrollsandrevemployment.workingpaperno.2012v2, CenterforRetirementResearchatBostonCollege. Rothstein,Jesse.2011.UnemploymentinsuranceandjobsearchintheGreat Recession.Brookings(Papers(on(Economic(Activity,Fall:143 214. Rutledge,MatthewS.2012.TheimpactofUnemploymentInsuranceextensionson DisabilityInsuranceapplicationandallowancerates.WorkingPaperno. 2011V17,CenterforRetirementResearchatBostonCollege(revisedApril 2012). SocialSecurityAdministrationBoardofTrustees.2012.The(2012(Annual(Report(of( the(board(of(trustees(of(the(federal(oldwage(and(survivors(insurance(and( Federal(Disability(Insurance(Trust(Funds.TechnicalReportGPO73V947. Washington,DC:U.S.GovernmentPrintingOffice. vonwachter,till.2010.theeffectoflocalemploymentchangesontheincidence, timing,anddurationofapplicationstosocialsecuritydisabilityinsurance. WorkingPaperno.NB10V13,NationalBureauofEconomicResearchPapers onretirementresearchcenterprojects,cambridge,ma.
"#$%&'()''*+'%&,"-"&./0'10'021%&'34',"5"6"1.'.3.".0/"/$/"3.16'-3-$61/"3.'1#&7' 89:;<='(>?9:89((' "#$%&'()*%+,-,%.$&,.+/01%1,&23/%14#*5%*&2.1&-#0&2/3%.%6,+,2*,%&72.12*% 8%2&0*%12(%+%83%*9:#6%2+;<%2*= >#0*+%&'>#+,2/>%+0*,$<?18,.,&$*2$,#.7@66,+%#6$;%?+$02*<72.1A0*%20#6B23#* >$2$,&$,+&
"#$%&'8)''@@*+'1--6",1/"3.0'1.7'1A1%70'10'021%&'34'-3-$61/"3.'1#&7'89:;<=' (>;B:89((' "#$%&'?--/,+2$,#.&2.1242*1&12$22--/<$#.%41,&23/%14#*5%*+2&%&= >#0*+%&'>#+,2/>%+0*,$<?18,.,&$*2$,#.7@66,+%#6$;%?+$02*<72.1A0*%20#6B23#* >$2$,&$,+&
"#$%&'C)''D3%E&%'-36","&0'FG'7&#%&&'34'7"01F"6"/G'H-I'1.7'J3F:4".7".#' -%3F1F"6"/G'H4I' ()*+,+-.-/01*21345-461,17*+ "# "$ "% "& "' "# "$ "% "& ' ()*+,+-.-/01/8,/19:1,;;.-<,=*41->1,;;)*?@5 A@,)<812*)@?@) A@,)<8G1/8@41@D-/ B;;.01,C@)1@D8,E>=461F: B;;.01-HH@5-,/@.0 ' "#$%&'>%%$%C$6#*1%&+*,-$,#.#68#1%/=@$;%*8#1%/-2*28%$%*&2*%DEF:GH:IJ H+#**%&-#.1,.K$#2-%*I-%*,#142K%.#*82/,L%1$#:2.12M#3$;2$/2&$&6#*%N%*JO 3P)FQ=RO3()FQ=SO+PFQ=TO+?F9OFQ=USO2.10H<JF<=
"#$%&'<)''@@*+'%&,"-"&./0='@@K'71/1'50)'LM@'K@NL' ''()*+,-./.,012*34.55.6027 " # $ %& ''<*3=6+>,+2*605?7 @A'*3;B,*%#C7 @A'*3;B,*%#D##E#F7 %88& %889 &&& &&9 &%& :,;+ "#$%&'>>?&%*,%&,.+/01%&#./<1,&23/%14#*5%*+2&%&=
"#$%&'B)''O.&P-63GP&./'".0$%1.,&'F&.&4"/'151"61F"6"/G'35&%'/2&'Q%&1/' R&,&00"3.' '(()*+,-+./+0(1(23*+45467407( " #" $" %" &"" "&841"" "&841""# "&841""$ "&841""% "&841"&" "&841"& 943( :5(;4<(+*343( =616>?> =4@6>?> "#$%&'V?N%*2K%&$2$%W&%*,%&*%-*%&%.$&2&,8-/%70.4%,K;$%12N%*2K%2+*#&&SQ &$2$%&-/0&$;%(,&$*,+$2.1X#/083,2=
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
"#$%&'(&&)*%&+%,)%+&"-"$.+)+&/0&-"1)/-"$&23&"44$)5"1)/-+ 6'7 687 697 6:7 6;7 6<7 6=7 6>7 "#$%&'(&)*+$,-."/#-&0"#1)*2& 345467 345478 45496 4549: 45498 ;,%."<%)&/=&>4463?&$@A5B 0454:>B 0454>:B 045448B 045448B 04544CB D*+$,-."/#-&"#1)*&0$@A5E& 3454>> <F)@5&7&;/#.+-B 0454>4B D*+$,-."/#-&"#1)*&0$@A5E& 45496 #)*.&7&;/#.+-B 04549:B (#"."$%&'(&)*+$,-."/#-&0"#1)*B 345449 04544:B '(&="#$%&<$G-&0"#1)*B 45449 045446B 90H/&)*+$,-."/#-&.+"-&;/#.+B 34549C 04544?B '#);<%/G;)#.&F$.)&0IJB 4547C 4547? 45479 4547K 4547K 4547? 045447B 045446B 045446B 045446B 045446B 04544KB %#0"#"."$%&'(&L%$";-B 3454>9 3454>6 3454>6 3454>6 04549CB 045498B 04549CB 045498B 90M,#)E&M,%GE&J,A,-.&>494B 454>7 454>6 454>> 454>4 045448B 045496B 045448B 045448B N/-.3JOOJ 45498 45479 45498 45498 45497 04549?B 045496B 04549?B 04549?B 04549?B P,$1F$."L&.";)&.F)#1 # G G G G G G G H 949 949 949 949 C6 949 949 949 H/.)-2&&I$;<%)&"#&$%%&L/%,;#-&"-&$&#$."/#$%&.";)&-)F")-&-<$##"#A&J,A,-.&>44:&./&Q)L);R)F&>49>& 0H/@);R)F&>44:&./&I)<.);R)F&>49>&"#&L/%,;#&6B5&&Q)<)#1)#.&@$F"$R%)&"-&%#0F)L)"<.-&=/F&II(E&IIQ(E& /F&II(SIIQ(&L%$";-BE&;)$-,F)1&$.&.+)&;/#.+%G&%)@)%&$#1&-)$-/#$%%G&$1T,-.)15&&H)U)G3V)-.&-.$#1$F1& )FF/F-E&$%%/U"#A&=/F&$,./L/FF)%$."/#-&$.&,<&./&=/,F&%$A-E&"#&<$F)#.+)-)-5
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
"#$%&'(&&)*++",-&./"/0./01.&23,&+"/14%5&67)89):6&."+;$%<&=>>?8=>@@ 9$$ AB%+;$3-%5& "/&#".%&.*,C%- AB%+;$3-%5& D&AE&%$0F0#$% :G4"*./&AE& #%23,%&%B5&32& 1"$%B5",&-%", H@I H=I HJI H'I "#$%&'( &$%&)* '%+'* (%),' -./01. *&2 #(2 ('2 (,2 34. #"5* ()5( #&5& #&5' 6&"5$7 6&"5"7 6&&5,7 6&&5,7 34.8*$9 ("2 "#2 "+2 ")2 :;<=0>?@A BCDE &$2 &'2 &'2 &,2 DE (&2 ()2 #&2 #$2 E@/.8=@11.4. "+2 ",2 "'2 ",2 F39 (&2 &+2 &,2 &'2 B0G@H8I@H=.8J>0><J :/K1@L.;,*2 $2 $2 $2 MA./K1@L.; *2 &$$2 &$$2 &$$2 @>8?A810G@H8I@H=.8NOB-P "&2 $2 $2 $2 MO8.1?4?G1. (2 ',2 &$$2 &$$2 MA./K1@L/.A>8;<H0>?@A%8L.0HJ &5(" &5(& &5#' N0J8@I8&"Q(&P 6$5*'7 6$5*#7 6$5'(7 MO8.RK?H.;8G.I@H.8&"Q(& *,5(2 &$$5$2 MO8.RK?H.;8G.I@H.8'Q($ ('5,2 '#5&2 C?/.8J?A=.8MO8.RK?H0>?@A%8L.0HJ8 $5," N0J8@I8&"Q(&P 6$5*&7 EESO8?A=@/.8?A8L.0H8L (5"2 &5#2 &5"2 &5#2 EESO8?A=@/.8?A8L.0H8L8?I8A@A.8?A8LT& &5#2 &5$2 $5)2 &5&2 @>.JU88E>0A;0H;8;.V?0>?@AJ8?A8JW<0H.8GH0=X.>J5883118J>0>?J>?=J8<J.8J0/K1?A48Y.?4Z>J8IH@/8>Z.8?A?>?018 /@A>Z1L8[\E8J0/K1.58
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`JWD%(.*&"?(&0&5( *?(%*;((4"&5*P*%,0?(%&5(1"%)&JW<((>)@&%(*&"+;&5(-(%"0,)9(10%(*+,*1&(% a*+$*%/l@jkkb)(-*%*&(./6n/80'-*%"+;&5((4"&5*p*%,"+(*85,$%*&"0+;%0$-&0 &5(*?(%*;((4"&5*P*%,10%&50)(<"&5,^JW@<(09&*"+'$.&"-."8*&"?(1*8&0%)10%$)( "+*,O$)&"+;c K 675()(.(*,$)&00$%*,O$)&(,5*P*%,)I CB[D "# "# ""# ""# "# "# ""# ""# "# "#" ""# ""# "# "#" ""# ""# "# "#" "# ""# "# "#" "# ""# #$%1"+*.)"'$.*&(,23(45*$)&"0+'(*)$%(10%)&*&()@<((>&")80'9"+() "+"&"*.8.*"')@-*/'(+&%*&()@*+,)$%?"?*.&0(45*$)&"0+6F0%'*../@"&")I CBdD " " "