Personal Tax Changes and Financial Well-being: Evidence from the Tax Cuts and Jobs Act

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Personal Tax Changes and Financial Well-being: Evidence from the Tax Cuts and Jobs Act Christine L. Dobridge, Joanne Hsu, and Mike Zabek 2024-029 Please cite this paper as: Dobridge, Christine L., Joanne Hsu, and Mike Zabek (2024). “Personal Tax Changes and Financial Well-being: Evidence from the Tax Cuts and Jobs Act,” Finance and Economics DiscussionSeries2024-029. Washington: BoardofGovernorsoftheFederalReserveSystem, https://doi.org/10.17016/FEDS.2024.029. 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.

Personal Tax Changes and Financial Well-Being: Evidence from the Tax Cuts and Jobs Act Christine L. Dobridge* 1, Joanne Hsu†2, and Mike Zabek‡ 1 1Board of Governors of the Federal Reserve System 2University of Michigan May 14, 2024 Abstract Weestimatetheeffectsofpersonalincometaxdecreasesonfinancialwell-being,including qualitativesubjectiveassessmentsandquantitativemeasures. Aplausiblycausaldesignshows thattaxdecreasesintheTaxCutsandJobsActmadesurveyrespondentsmorelikelytosaythey were “living comfortably” financially, with null effects at lower levels of subjective financial well-being. Estimates from a similar design using credit bureau data show that people who had larger tax decreases were modestly more likely to open new accounts, and more likely to have higher consumer credit balances. Tax decreases had effects on credit scores that are indistinguishable from zero. Results suggest that larger tax decreases improve financial wellbeinginwaysnotfullyproxiedbytypicaladministrativedata. Keywords: Taxes;subjectivewell-being;householdfinances;credit;financialwell-being JELNumbers: H24,G50,I31 *20thStreetandConstitutionAvenueN.W.Washington,DC20551;Christine.L.Dobridge@frb.gov,BoardofGovernorsoftheFederalReserveSystem,20thStreetandConstitutionAvenue,NW,Washington,DC20551 †426ThompsonSt,AnnArbor,MI48104;jwhsu@umich.edu. ‡Correspondingauthor: 20thStreetandConstitutionAvenueN.W.Washington,DC20551;Mike.Zabek@frb.gov. WethankPrestonHarry,ZofshaMerchant,andShookaSaketforexcellentresearchassistance. Theprojectbenefited from useful comments by Elliot Anenberg, Neil Bhutta, Breno Braga, Lisa Dettling, Adrienne DiTommaso, Jeff Larrimore, Alicia Lloro, Andrew Whitten, David Splinter, and seminar participants at the Board of Governors of the Federal Reserve System, the Federal Reserve Bank of Richmond, the Federal Reserve Bank of Philadelphia, LMUMunich,theNationalTaxAssociationSpringSymposium,theNationalTaxAssociationAnnualMeetings,the Allied Social Science Associations Annual Meetings, and the Consumer Financial Protection Bureau. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the Federal Reserve research staff, the Board of Governors, the Federal Reserve System, or the University of Michigan. Any remainingerrorsoromissionsaretheauthors’responsibility. Declarationsofinterest: none.

I Introduction CongresshasenactedchangestoU.S.individualincometaxesunderalmosteveryPresidentialadministrationinthepost-warperiod,mostrecentlyreducingtaxesbyover$1trillioninthe2017Tax CutsandJobsAct(TCJA).Richliteratureshavestudiedeffectsofvariouspersonaltaxes,including many studies of indirect effects.1 Less is known, however, about the direct impact of tax changes on consumers’ financial well-being. While it is clear that after a consumer experiences a tax cut, increasedafter-taxincomeshouldimprovefinancialwell-being,theexactpathwaysandmagnitude oftheeffectsareunclear. Doesadditionalincomereducepeople’sstressaboutmanagingfinancial obligations? Do tax cuts help people to meet aspirational financial goals, like owning a home? Understanding the implications of tax cuts for financial well-being would inform policy debates around personal taxes, including the active debate about extending the TCJA’s personal tax cuts beyond2025.2 This paper estimates effects of the TCJA personal tax cuts on measures of household financial well-being—both qualitative self-assessments and quantitative financial outcomes—drawn from rich survey data on household finances as well as a detailed panel of administrative consumer creditoutcomes. Weuseadifference-in-differencesempiricalstrategywithacontinuoustreatment that leverages considerable variation in TCJA-induced personal tax rates. Comparing effects on subjectivefinancialwell-beingwithothermorewell-studiedoutcomesincludingcreditutilization, delinquencies, and home ownership shows how information about subjective financial well-being cangiveafullerpictureofhowtaxchangesaffectedpersonalfinances. While we expect tax decreases to have weakly positive effects on financial well being, their effects on many other financial outcomes can be theoretically ambiguous due to the potential of countervailing consumption and savings responses. Consumers might use their increased disposableincometospend,save,orpaydowndebt,thereforegeneratinga-prioriuncleareffectsoncredit 1For example, Keane (2011) surveys a literature on changes in labor supply in response to taxation and Saez, SlemrodandGiertz(2012)surveyaliteratureonadjustmentsinoveralltaxableincomes. 2PublicLaw115-97. ThepersonalincometaxchangesintheTCJAwereeffectivefrom2018through2025andas ofearly2024,therewereseveraldifferingproposalstoextendthem(seeTaxFoundation,2024). Thebeginningofthe TCJAextensionsdebateissummarizedinBuhl(2022). 1

outcomeslikenewloansortotalloanbalances. Someconsumersmayevenincreasetheirspending in anticipation of continued tax cuts or a further relaxation of liquidity constraints.3 One example is a consumer who purchases a durable good like an appliance or a car using credit that they intend to pay off over time. Consumers might also choose to accumulate savings or lower their credit balances by paying down existing debts. All else equal, we would expect delinquencies to weakly decrease because lower income taxes increase consumers’ disposable incomes and hence their ability to pay their bills on time.4 Furthermore, any theorized effects could be attenuated in ourdataifthesalienceofthetaxcutsislow. UsingrichpanelsurveydatafromtheFederalReserveBoard’sSurveyofHouseholdEconomics and Decisionmaking (SHED), we show that households that received larger tax cuts subsequently reported larger improvements in subjective financial well-being—a higher propensity to report “living comfortably” after the TCJA. An event-study analysis shows null effects and no pre-trend leading up to 2017 and a roughly stable positive effect in 2018 and 2019. In addition to living morecomfortablyfinancially,householdsthatreceivedlargertaxcutswerealsolesslikelytohave student debt and more likely to be homeowners after the tax cut. We do not detect effects on emergencysavingsorwhetherahouseholdwouldcoveranunexpected$400expensewithcashor anequivalent. To determine if these improvements are limited to subjective measures, we draw on large-scale administrative data from the Federal Reserve Bank of New York/Equifax Consumer Credit Panel (CCP) dataset and investigate more quantitative financial well-being outcomes (Bhutta, Skiba and Tobacman, 2015; Hu et al., 2018; Argys et al., 2020; Black et al., 2023). Our results show that consumers with larger tax cuts were modestly more likely to open new accounts and had higher credit balances following the enactment of the TCJA. Increases in credit usage indicate greater spending power after TCJA, which taken together with the survey results, suggests higher credit 3While it is unclear exactly how long lasting consumers expected the changes to be, theory suggests a larger consumptionresponsetothislong-livedtaxchangerelativetothepreviouslystudied,one-timetaxrebatesin2001and 2008(Agarwal,LiuandSouleles,2007;Sahm,ShapiroandSlemrod,2010;Gross,NotowidigdoandWang,2014). 4Since missing payments yields costly fees, additional interest accrual, and higher interest rates on loans (both forcurrentloans,inthecaseofpenaltyinterestratesforcreditcards,andforanynewloansvialowercreditscores), individualsfaceastrongincentivetousetheincometohelpkeeptheirloanscurrentandstayoutofdelinquency. 2

usage is consistent with improvements in subjective financial well-being. We also observe some modest decrease in delinquencies after TCJA, which would be consistent with an improvement in subjective well-being as well, though some pre-trend in delinquencies diminishes a causal interpretation. Finally, wefind a zero estimated effect on creditscores overall. The lack ofcredit score changes alongside other outcomes suggests that tax cuts improve subjective well-being without alteringtheoverallcreditriskprofilefromalender’sperspective. Our most direct contributions are to literatures on the effects of tax changes, as well as income shocks more broadly, on household finances. A sizeable body of work analyzing the effects of tax orfiscalpolicychangesinvolvinglump-sumpaymentshasshownthatconsumersreactbychanging consumption, which might be financed by additional debt, as well as by paying down existing debt. In a study of Singaporeans, consumers reacted to the announcement of a one-time, fiscalpolicy-related income shock by increasing their spending, primarily using credit cards (Agarwal and Qian, 2014). Agarwal, Liu and Souleles (2007) and Sahm, Shapiro and Slemrod (2010) find evidence that borrowers use tax rebates to pay down debts; Skiba (2014) shows that receiving a tax rebate reduces the short-term likelihood of taking out a payday loan. Several related studies focus on changes to Medicaid enrollments induced by fiscal policy, with Hu et al. (2018) and Milleretal.(2021)showingthatMedicaidexpansionshadnotableeffectsonfinancialwell-being, includingreducingunpaidcreditbalancesincollectionandover-limitcreditcardspending,though the expansions had little effect on credit scores or other credit delinquency measures. Argys et al. (2020)findthatsuddendisenrollmentfromMedicaidresultedinanotabledeclineinwell-beingas measuredbycreditscoresanddelinquencies. Effects of the TCJA on financial well-being are of particular interest because tax cuts from the TCJA are more representative of other policy-relevant gradual and long-lived income fluctuations than commonly studied tax changes like lumpy rebates or stimulus checks. Under the TCJA, households quickly saw small incremental changes in their disposable incomes as the Internal RevenueService(IRS)updatedwithholdingtablesaftertheTCJAwasenacted.5 Thechangeswere 5See the updated withholding tables, which are available on the IRS’s website at: https://www.irs.gov/pub/irsprior/p15–2018.pdf 3

sosmallthatmultiplepublicopinionpollsshowedthatfewpeoplebelievedtheyowedlessintaxes under the TCJA (Tax Policy Center, 2019). Still, Scharlemann and van Straelen (2022) show the TCJA had meaningful household financial effects, finding the legislation increased a household’s probability of mortgage refinancing conditional on refinancing incentives. Hotchkiss, Moore and Rios-Avila (2021) estimate a family utility model of the TCJA’s effects, concluding that welfare of all families increased after the TCJA but the largest effects were for high-income households, thosewithself-employmentincome,thosewithchildren,andthosethatrenttheirhome. Thesmall andlong-livedchangesindisposableincomesbroughtbytheTCJAmostresembletheincremental increases to income driven by many other policy-relevant phenomena. In terms of one of those policies, minimum wage hikes, Aaronson, Agarwal and French (2012) find greater spending on durables, often debt financed. Dettling and Hsu (2021) also show that higher minimum wages are associatedwithgreateraccesstolow-costcredit,aswellasreducedpaymentdelinquency. Another contribution of this work is showing effects on a qualitative measure of subjective financial well-being alongside more quantitative outcomes from survey and administrative data sources.6 Ouranalysisvalidatesthatincreasesinsubjectivefinancialwell-beingwereaccompanied byincreasesinthenumbersofcreditaccounts,increasedcreditbalances,decreaseddelinquencies, a lower share of people holding student loans, and higher homeownership rates among people exposed to larger tax decreases. These other measures validate that subjective financial well-being can be useful in cases where more detailed data are not available. Our results also illustrate how subjective financial well-being can provide additional information that can help categorize the extentthatconsumersexperiencechangespositively,muchlikesubjectivewell-beingmoregenerally (Lachowska, 2017). In our case, the increase in subjective financial well-being alongside a zero result for credit scores shows how focusing only on credit scores can miss a measurable improvement in how people experience financial well-being. Using subjective financial well-being as an 6Lachowska(2017)showsthat2008taxrebateshadsubstantialeffectsonhigh-frequencymeasuresofsubjective well-being, and van Praag, Frijters and Ferrer-i-Carbonell (2003) relate financial well-being to broader measures of subjectivewell-being.Diener,OishiandTay(2018)providearecentsummaryoftheextensiveliteratureinpsychology on subjective well-being, including a cross-country examination of associations between progressive taxation and subjectivewell-beingbyOishi,SchimmackandDiener(2012). 4

outcome of interest is particularly useful in cases where it is difficult to determine the extent that consumers experience a particular financial development as being harmful, such as the extent that havinguncollectedmedicaldebtaffectsfinancialwell-being(Brevoort,GrodzickiandHackmann, 2020) or the extent that banning payday loans helps or harms consumers’ financial well-being (Bhutta,SkibaandTobacman,2015).7 This work also informs policymakers considering potential effects of tax policy or other related fiscalpolicychangesonhouseholdfinancialwell-being. GiventhatmostoftheTCJApersonaltax rate reductions are slated to expire at the end of 2025, our analysis has a few implications for the extension debate, specifically that a personal tax increase with distributional properties similar to those of the TCJA tax decreases would likely have modest effects on consumer credit utilization and performance and meaningful negative effects on subjective well-being. To the extent that our results indicate the TCJA led to an increase in consumer credit balances, which we interpret as beingconsistentwithaconsumptionresponsetogreaterdisposableincome,anotherimplicationof thisworkisthatapersonaltaxincreasecouldleadtosomepull-backinhouseholdconsumption. The remainder of the paper proceeds as follows. We present background information on the TCJA in Section II, describe the datasets and the empirical strategy in Section III, discuss the resultsinSectionIVandconcludeinSectionV. II Policy Background The TCJA made considerable changes to the U.S. federal income tax code, including both personal and business tax provisions. The law was enacted rapidly—it was introduced on November 2, 2017, and became public law on December 22, 2017—and the personal income tax provisions applied to the tax year beginning January 1, 2018. Among the key changes to the personal tax code were (1) reducing income tax brackets (see Table A.1 for pre- and post-TCJA tax brackets), (2) increasing the standard deduction, (3) reducing the deductibility of mortgage interest from 7NandaandBanerjee(2021)reviewthestudiesofsubjectivefinancialwell-beinganddocumentasignificantuptick inrecentstudies,primarilywithinthefieldofmarketing. 5

$1,000,000 to $750,000 of mortgage debt, (4) limiting the deductibility of state and local income andpropertytaxesto$10,000,(5)raisingthethresholdforthealternativeminimumtaxforhouseholds, (6) raising and expanding the child tax credit, and (7) allowing for the deductibility of qualified business income for pass-through corporations. The Joint Committee on Taxation (JCT) projected that these personal tax provisions would substantially reduce tax revenue—by $1,127 billionfrom2018to2027(JointCommitteeonTaxation,2017b). TheJCTalsoprojectedthebulk of the total reduction in tax revenue in the $100,000 to $200,000 and the $200,000 to $500,000 taxpayerincomecategories—by$51billion(approximately$1,700pertaxpayerunit)and$47billion (approximately $5,100 per taxpayer unit), in these income categories, respectively, in 2019. For taxpayers with income less than $50,000, the TCJA was projected to reduce tax revenue by about$14billion($150pertaxpayerunit)in2019(JointCommitteeonTaxation,2017a). III Data and Empirical Specification III.I Data sources To study how financial well-being changed following passage of the TCJA, we use two main data sources that include individual-level financial outcomes. In addition, we apply information that facilitates the use of a detailed microsimulation model of the U.S. tax code provided by the NationalBureauofEconomicResearch(NBER)—theTAXSIMmodel—tocomputeanestimated taxchangeforeachhouseholdorindividual. First, to understand subjective financial well-being, we turn to data from the Survey of HouseholdEconomicsandDecisionmaking(SHED),conductedbytheBoardofGovernorsoftheFederal ReserveSystem eachyearsince 2013. Dependingon theyear, 6,000to12,000 individualsanswer a broad range of questions related to household financial positions and well-being, including a rating of their overall financial well-being, as well as detailed questions about banking and credit, abilitytocopewithunexpectedexpenses,housingpositions,andretirementpreparation. Since respondents to the SHED are drawn from a broader online panel survey we can observe 6

a subset of them across multiple years.8 We focus our analysis on the period from 2015 to 2019. We begin the sample in 2015 (three years before the TCJA enactment) to provide for an ample pre-period and to focus on a period when SHED questionnaires are relatively stable. We end the sample in 2019 (two years after the enactment) to avoid capturing the effects of the COVID-19 pandemic in 2020.9 To keep our focus on tax decreases, we restrict our sample to people whom we observe in 2017, which is the year we use to calculate the size of people’s TCJA-induced tax decreasesbasedonpre-legislationcharacteristics. To gain further insight into how U.S. household finances changed following the TCJA, we use information reported in the Federal Reserve Bank of New York/Equifax Consumer Credit Panel (CCP). The CCP dataset is an individual-level, anonymized panel of consumer credit records, drawnattheendofeachquarterfromEquifax—–oneofthethreemajorcreditbureausintheUnited States. The data include detailed information drawn from credit reports, including loan balances, creditlimits,andpaymentstatus. Asidefromvariablesonageandgeographiclocation,thedataset is generally limited to information about credit status. The CCP does not contain information, for example,abouthouseholdincome,employmentstatusordemographiccharacteristicslikeraceand education. Individualswhodonotinteractwithconventionalcreditmarketswillnotappearinthis dataset at all. As such, assessing evidence from both administrative and survey data together give usafullerpictureoftheeffectsoftheTCJAtaxchanges. We use a 10 percent sample of the overall CCP and create a panel with characteristics similar to those of the SHED panel: limiting the period to 2015 to 2019, requiring an individual to be in the dataset in 2017 as well as at least one year following, and using end-of-year observations to smooth through fluctuations in financial positions over the course of the year. We also require an individualtohaveareportedEquifaxRiskScore(atypeofcreditscore)andtohaveanon-missing value for the number of new accounts established in a year, in order to ensure a person’s credit records are sufficiently populated to make inferences across time. More details on construction of 8The sample in each year is drawn from continuing members of the Ipsos KnowledgePanel—an online panel of individuals originally recruited primarily via address-based sampling, with some members being recruited prior to 2009viarandomdigitdialing. 9Arobustnessappendixshowsthatwehavesimilarresultswhenweincludeawiderarrayofyears. 7

boththeSHEDandCCPdatasamplesarepresentedintheDataAppendix. III.II Computing estimated tax changes Since the TCJA changed many aspects of the tax code simultaneously, the goal of this paper is to study how the overall changes in an individual’s tax liability affected financial well-being. Our primary variable of interest, therefore, is the decrease in an individual’s estimated average tax rate under the pre-TCJA tax code in 2017 compared with the post-TCJA tax code in 2018. For both the SHED and the CCP analyses, we estimate pre- and post-TCJA tax rates using characteristics measured in 2017, before the tax law change, inputted into the TAXSIM model (Feenberg and Coutts, 1993). This procedure gives us measures of tax decreases that are due to interactions between pre-determined individual characteristics and the specifics of the policy change. Given the substantial interactions between the federal and state tax codes, we calculate the change in the total household average federal-plus-state tax rate—the federal-plus-state personal income tax liability divided by total income—to capture the overall change in tax burden for households in eachofthesegroups. For the analysis using individual-level survey data from the SHED, the estimated tax rates are primarilybasedonpre-TCJAindividualcharacteristics—householdincome,maritalstatus,number ofchildren,whetherthepersonreportsthattheycareforanadult,monthlymortgagepayments,and state of residence. In addition to the individual-level variables, we include some limited censustract-level statistics on mortgage interest payments derived from the Equifax Credit Risk Insight Servicing and ICE, McDash®Data (CRISM) dataset and property taxes from the U.S. Census Bureau’sAmericanCommunitySurvey. IntheCCP,wedonotobserveinformationonindividuals’incomesorotherhousehold-leveltax inputs. Instead,weuseTAXSIMtocalculatehyperlocal,representativetaxratesatthecensustract level,separatelyformortgageholdersandnon-holdersaswellasforsingleandjointfilingstatus.10 10Census tracts are small, sub-county geographic areas that include between about 420 and 1,200 housing units, approximating a neighborhood; for more information, see the discription on the U.S. Census Bureau’s website at https://www.census.gov/history/www/programs/geography/tracts and block numbering areas.html. 8

We input pre-TCJA data on census-tract-level median household or worker incomes and property taxesfromtheAmericanCommunitySurveyandoncensus-tract-levelmedianmortgagepayments fromCRISM.Specifically,weassigndifferentrepresentativetaxratechangestoindividualsinthe CCP based on (1) the individual’s census tract of residence in 2017, (2) whether we observe a mortgage for an individual in 2017, and (3) household size in 2017 (used to assign single or joint filing status). In robustness checks, we also present results using alternative inputs for computing representative tax changes, including estimates of dependents and business income. We provide a detailed description of the tax change calculations and data sources for the SHED and CCP analysesintheDataAppendix. Figure 1, Panel A, shows the distribution of tax rate decreases in the SHED sample, and Panel B shows the same distribution for the CCP sample. Each distribution is centered around a 2 percentagepointdecreaseandexhibitsconsiderablevariation. BecausethetaxdecreasesintheSHED are based on individual-level responses, we capture a higher variance in tax decreases using the SHED, an advantage of having individual-level data on incomes. The SHED data, however, also show some signs of binning due to income categories being collected as a categorical variable.11 EstimatedtaxdecreasesintheCCPdataalsoshowconsiderablevariation,reflectiveofsubstantial dispersioninhyperlocalmeasuresofincome,propertytaxes,andmortgagepayments,inparticular. In the CCP sample we observe a mean tax decrease of 1.8 percentage points, a median of 1.9, and arangefrom0.93atthe10thpercentileto2.7atthe90thpercentile. PanelsCandDofFigure1presentboxplotsoftaxratechangeswithintheincomebinsofindividuals recorded in the SHED—for the SHED and CCP samples, respectively. For both samples, the tax rate reduction tends to rise with income (Tax Policy Center, 2017), but there is substantial variationinthetaxreductionwithineachdecile. 11Thevariablehad21distinctcategoriesin2017,soitdoesprovidesignificantvariation,despitebeingbinned. For simplicity,weusethemidpointofeachbininthetaxdecreasecalculation,omittingobservationsinthetop-mostbin, whichisunbounded. 9

III.III Outcomes of interest Our main outcome of interest, subjective financial well-being, is meant to encompass feelings of security, agency, and satisfaction around financial decisions. We measure financial well-being using the following survey question, which is available through all years of the SHED: “Overall, which one of the following best describes how well you are managing financially these days?” Respondents answer “(1) Finding it difficult to get by, (2) Just getting by, (3) Doing okay, or (4) Living comfortably.” We operationalize the survey responses with dummy variables for reporting a specific category or higher.12 As described by Consumer Financial Protection Bureau (2017), financial well-being is related to economic outcomes, financial behaviors, and psychological outlooksincombinationwithoneanother. Sotwopeoplewithidenticallevelsoflifetimeconsumption and wealth can have different levels of financial well-being without any mismeasurement. Since we focus on changes within individuals, we expect that our results will primarily be influenced by economic outcomes—most notably income. However, changes in subjective well-being also reflect interactions with various behavioral responses that could affect how much positive effect a givenchangeinincomewillhaveonfinancialwell-being.13 In addition to financial well-being, we include several financial outcomes from the SHED. In terms of possible future financial goals we include homeownership—an important vehicle for wealthaccumulation. Intermsofcurrentfinancialsecurity,weincludebothaninfluentialquestion asking if someone would pay an unexpected $400 expense solely with cash or its equivalent and a question asking if someone has an emergency fund to cover larger unexpected expenses. Somewhereinthemiddlebetweenfinancialgoalsandcurrentneeds,weincludethepresenceofstudent loans,whichhasalsobeencitedasabarriertowealthaccumulation. Table 1 reports summary statistics for the sample of SHED respondents, just before the TCJA 12Federal Reserve Board (2018) and Federal Reserve Board (2021) validate our question relative to a longer batteryofquestionsproposedandusedbytheConsumerFinancialProtectionBureau(2017)tomeasuretheconceptof financialwell-being. 13The full definition of financial well-being from Consumer Financial Protection Bureau (2017) that we seek to measureisbasedoninterviewswithexpertsonfinancialliteracy, financialplanners, andindividualconsumers. “Financialwell-beingisastateofbeingwhereinapersoncanfullymeetcurrentandongoingfinancialobligations,can feelsecureintheirfinancialfuture,andisabletomakechoicesthatallowthemtoenjoylife.” 10

wasenactedinlate2017. Mostpeoplereportrelativelyhighlevelsoffinancialwell-being,with34 percent giving the highest category, living comfortably, and 41 percent saying that they are doing okay. Additionally,70percentofadultsarehomeownersand13percenthavestudentloans.14 To understand whether the effect of larger tax changes on subjective financial well-being translates to meaningful differences in other, more commonly used financial metrics, we turn to the CCP data and analyze outcomes related to household credit utilization and performance. First, as a summary measure of household credit positions and credit risk from a lender’s perspective, we study the Equifax Risk Score, a type of credit score (similarly to Bhutta, Skiba and Tobacman (2015)). Credit scores are positively associated with the financial health of consumer; higher scores indicate lower delinquency risk, and lenders typically offer more credit at more favorable terms to those consumers. Next, as measures of credit utilization, we use the number of new accounts opened over the past 12 months and the natural log of total outstanding consumer credit balances. Increased credit usage could affect financial well-being, for example, if an increase in creditbalancesreflectsaconsumptionresponsetohigherdisposableincome(Dinerstein,Yannelis and Chen, 2023). Finally, as a measure of credit performance, we use the number of delinquent accounts over a time horizon from 60 days delinquent to in severely derogatory status.15 Table 2 reports summary statistics in 2017 for all variables used in the main analysis and in robustness checks with the CCP data. We observe that the mean credit score in the dataset is 703, with a slightly higher median level of 726. The mean number of new accounts is about 0.9 in 2017, with amedianofzero,andthemeannumberofdelinquenciesis0.3,alsowithamedianofzero. 14Thestatisticsaregenerallysimilartothosefortheweightedcross-sectionalsurveypresentedinFederalReserve Board(2018). However,theredoesappeartobepositiveselectionthatcouldbeduetothelackofweights(whichwe includeasarobustnesscheck)andtherequirementthatpeoplearerandomlyselectedintothesampleandthenagree tobeinterviewedinmultiplewaves. 15Thesevariableswereselectedtoavoidissueswithtimingandreporting. Forexample,newmortgagesmayonly appearoncreditreportsafteralag. Weselected60ormoredaysdelinquenttoavoidthefactthatnotalllendersreport 30 day delinquencies. More severe outcomes, like bankruptcies, are slow moving processes that can take years to resolve,buttypicallyareprecededbydelinquencies,whichwedoanalyze. 11

III.IV Empirical specification To study how personal income tax changes affect financial well-being, we use a difference-indifferences regression specification with the continuous treatment variable of people’s predicted tax decreases. This strategy allows us to exploit heterogeneity in the size of tax decreases from 2017to2018toseeifindividualswhohadlargertaxdecreasesalsoexperiencelargerchangesina financialoutcome.16 For the SHED, we use individual panel survey data to estimate linear probability models of a series of dichotomous outcomes (Y ) for an individual i in year t. Our coefficient of interest, it β, is interpreted as the effect of a 1 percentage point decrease in someone’s average tax rate, calculatedusingchangesinthetaxcodefrom2017to2018usingpre-TCJA-determineddatafrom 2017 in TAXSIM. Because the tax decreases went into effect in early 2018, we estimate effects by comparing outcomes in 2018 and 2019 with the same outcomes before and during 2017. In termsofcontrols,α representsanindividualfixedeffectthatcapturesalltime-invariantdifferences i between individuals. Additionally, we include α , a state-by-year fixed effect that controls for st time-varying differences across states—including the local economic cycle—and a rich series of individual-levelcontrolsX ,includinghouseholdincomes,acubicinage,ruralstatus,andcurrent it employment status. Results are unweighted and standard errors are clustered by state using the respondent’s state when we first observe them. The SHED analysis regression specification is as follows: Y =βTaxdecrease1(t >2017)+α +α +γX +ε . (1) it i i st it it For the CCP analysis, we use a similar specification with some adaptation for the different strengthsoftheadministrativecreditbureaudata. Asbefore,ourcoefficientofinterestβ measures the effect of a 1 percentage point decrease in a household’s average tax rate, beginning in early 16Callaway, Goodman-Bacon and Sant’Anna (2024) show that the coefficient of interest in these models can be interpretedasthecausaleffectofaonepercentagepointdecreaseinaveragetaxratesunderastrongparalleltrends assumption. Thisstrongparalleltrendsassumptionrequirestraditionalparalleltrendsbytreatmentstatusaswellas someadditionalrestrictionsontreatmenteffectheterogeneitybythesizeofthetaxcutinourcontext. Notealsothat sincethetaxchangesalloccurredin2017,ourestimationstrategyisunaffectedbyvariationintreatmenttimingasin Rothetal.(2023). 12

2018, on an outcomeY . Also as before, we use individual fixed effects (α) to control for timeit i invariant individual-level differences like education level, gender, race, or ethnicity. In contrast to the SHED specification, the high number of observations in the CCP dataset allows us to includemuchmoredetailedcounty-by-yearfixedeffectsα inthisspecification. Thesefixedeffects ct control for highly localized, time-varying shocks to household financial conditions, such as the potential for changes in local-area hiring conditions due to the TCJA corporate tax rate changes. SincetheCCPdatahavefewerindividual-leveldemographiccharacteristics,however,weareonly able to control for age bins at an individual level. Standard errors are clustered at the county level to allow for arbitrary correlation of errors within a local geography. The CCP analysis regression specificationisasfollows: Y =βTaxdecrease1(t >2017)+α +α +γ Agebin +ε . (2) it i i ct 1 i it Inbothspecifications,variablesforthetreatmentgroupandapost-perioddummyaresubsumed by the individual and the state-by-year or county-by-year fixed effects (compared with a standard difference-in-differencesspecificationasinBertrand,DufloandMullainathan(2004)). Wediscuss robustnesstothespecificationsinSectionIV. IV Results IV.I SHED results In our analysis of the SHED data, we find evidence that tax reductions led to increases in the share of people who were living comfortably financially. As seen in column (1) of Table 3, a 1 percentagepointlargertaxreductionincreasesthelikelihoodthatsomeonewillsaytheyareliving comfortably by 1.5 percentage points. Columns (2) and (3) show insignificant, though somewhat noisy,estimatesforthelikelihoodthatsomeonewasmorelikelytobegettingbyorbetter,ordoing okay or better, as a result of the TCJA tax decreases. Together, these results indicate that larger 13

tax reductions improve subjective financial well-being at the top of the well-being distribution, whichisconsistentwithpreviousanalysesshowingthattheTCJAdisproportionatelybenefitedthe alreadywell-off(TaxPolicyCenter,2017;Hotchkiss,MooreandRios-Avila,2021). Our estimated effect size of 1.5 percentage points suggests that tax decreases can have impacts on financial well-being that are comparable to effects of business cycles and other year-over-year changes. Multiplying our 1.5 percentage point effect size by the 1.9 percentage point average tax reduction due to the TCJA implies an increase of 2.9 percentage points in the likelihood that someonewaslivingcomfortablyinresponsetoanaverage-sizedtaxdecrease.17 Thishypothesized 2.9 percentage point increase was about three times the 1 percentage point increase in the share of people reporting they were living comfortably from 2017 to 2018 (from 0.33 to 0.34) and was similar to the total increase from 2017 to 2019 (from 0.33 to 0.36). We also cannot rule out the possibilitythattheTCJAalsohadmeaningfuleffectsatlowerthresholdsduetoourrelativelywide standarderrors.18 We now turn to factors that may affect one’s sense of financial well-being. First, we analyze student loan holdings, which have been cited as a barrier to wealth accumulation (Mezza et al., 2020). Column (4) in Table 3 shows that a 1 percentage point decline in taxes decreases the likelihood that someone has student loans by 0.6 percentage points, which is significant at the one percent level. The results are consistent with a net paydown of debt as in Sahm, Shapiro and Slemrod (2010), which showed survey evidence that half of consumers used the 2008 tax rebates topaydowndebt. We also find some evidence of increased homeownership, an important vehicle for wealth accumulation—particularly since, for many families, a home is their largest asset. Column (5) shows that a 1 percentage point tax rate reduction led to a 0.6 percentage point increase in the 17This 2.9 percentage point increase would be the average effect of the TCJA under the strong assumptions that treatmenteffectsarehomogeneousandtherearenogeneralequilibriumeffects,inadditiontoalessrestrictiveparallel trendsassumption. 18Cross-sectionaldifferencesinthefractionofsurveyrespondentsreportinglivingcomfortablyaremuchlargerthan year-to-yearaggregatedifferences,sotheeffectsofthetaxreductionaremuchsmallerincomparison. Forexample, thesharelivingcomfortablyin2017goesfrom0.15forfamiliesearninglessthan$40,000peryearto0.62forfamilies earningmorethan$100,000. 14

likelihoodthatsomeoneownstheirhome,whichissignificantatthe10percentlevel.19 We find insignificant and small effects for measures of savings and liquidity. In column (6), we report statistically insignificant and economically modest negative effects on the likelihood that someone would handle an unexpected $400 expense with cash or its equivalent.20 We similarly find,asshownincolumn(7),insignificantlypositiveeffectsonthelikelihoodthatsomeonehasan emergencyfundofthreemonthsofexpenses. Event studies in Panel A of Figure 2 show no evidence of a pre-trend in the share of people living comfortably before the tax changes and a stable effect after the tax change. The dark line in Panel A of Figure 2 is roughly flat from 2015 to 2017 and close to zero. The estimated effect jumps up to around 1.5 percentage points in 2018 and stays there in 2019. The other two levels of financial well-being, shown in Panels B and C, are also undetectably different from zero before thetaxcut. Aswiththeregressionresults,weseenodetectableeffectofthetaxdecreasesatthese othertwo,lowerlevelsoffinancialwell-beingaftertheimplementationoftheTCJA.21 Overall,weseeevidencefromsurveyresponsesthattheTCJAledpeopletofeelmorecomfortable financially. Some people may have also felt more comfortable because tax decreases helped them pay down student loan obligations but we see no detectable positive effects on emergency savings or whether a household would cover an unexpected $400 expense with cash or an equivalent. Another factor that lends credibility to the result is that it is identified off of changes in people’s subjective well-being over time, not cross-sectional differences. So the result cannot be due to permanent differences in people’s dispositions or similarly unchanging differences in how 19The relationship between larger tax cuts and increasing homeownership may seem puzzling, since the TCJA removed some of the incentives for home-ownership by raising the standard deduction and imposing caps on both mortgageinterestandstate-leveltaxes,includingpropertytaxes. However,ourresultsareconsistentwiththechannel of increased after-tax incomes offsetting some of these effects, similar to findings in analyses from the Tax Policy Center(McClelland,MuccioloandSayed,2022). 20Thismeasureisbasedonaskingpeopletogivethepotentiallymultiplewaysthattheycouldpayfor“anemergency expensethatcosts$400”basedontheircurrentfinancialsituation. Thevariableisoneifthepersonwouldcoverthe expense exclusively using cash, savings, or a credit card paid off at the next statement. The variable is zero if the personsaidtheywouldpayforatleastpartofthe$400expensebyborrowingorsellingsomething,oriftheysaidthey wouldnothavebeenabletocovertheexpense. 21Thefigureplotstheβ coefficientestimatesand95percentconfidenceintervalsfromthefollowingevent-study k specification: Y it =∑2 k= 01 2 9 015,k̸=2017 β k ·I{k =t}·Taxdecrease i +α i +α st +γX it +ε it . Results are unweighted and standarderrorsareclusteredbystateusingtherespondent’sstatewhenwefirstobservethem. Eventstudiesforother SHEDoutcomesarepresentedinAppendixFigureA.1 15

peopleratetheirsubjectivewell-beingforafixedsetofcircumstances. However,itisstillrelevant tounderstandifthesedifferencesarereflectedinother,morequantitativemeasures,andweturnto thatquestionusinglarge-scaleadministrativecreditbureaudatafromtheCCPinthenextsection. IV.II CCP results Table 4 presents results from the CCP analysis, showing the effects of larger tax decreases on the Equifax Risk Score, the number of new accounts opened, the total balances of consumer credit accounts, and the number of delinquencies. We observe no statistically significant relationship between the size of the tax decreases and credit scores in column (1). In contrast, we observe a modest and statistically significant increase in new accounts in column (2): a 1 percentage point tax rate reduction led to an estimated increase in the number of new accounts by 0.03 across the sample. An average 1.76 percentage point tax cut implies a 0.05 increase in the number of new accounts for the average individual—an effect of small magnitude compared to about 0.9 new accountsopenedonaveragebyanindividualin2017. Lookingattotalconsumercreditbalances(column3),wefindtotalbalancesincreasefollowing the TCJA, with the interpretation that a 1 percentage point tax rate reduction increases consumer credit balances by 0.9 percent. This result is qualitatively consistent with observing an increase in new credit accounts and is consistent with greater spending power for consumers after the TCJA. Next, we turn to consumer delinquency effects, a straightforward measure of individuals missing debt payments and experiencing financial strain. We observe modest but statistically significant effectsfordelinquencies(column4). Wefindthata1percentagepointreductioninthepersonaltax rate is associated with a decline in the number of delinquencies of 0.006. The effect size of 0.006 impliesthattheaverage1.76percentagepointtaxcutwoulddecreasedelinquenciesby0.01,which isquitesmallcomparedtotheoverallaveragenumberofdelinquenciesof0.3in2017. Noincrease in delinquencies also points to the post-TCJA rise in credit usage reflecting an improvement in financial well-being overall. Finally of note, the null result on credit scores is consistent with the combinationofresultsondelinquenciesandaccountbalances—lowerdelinquencieswouldtendto 16

raise credit scores, while higher account balances and new accounts would tend to depress credit scores.22 To evaluate the parallel trends assumption and show how the effects vary over time, Figure 3 presentsevent-studygraphsshowingresultsoftheeffectsofthetaxdecreasein2017onoutcomes presented in Table 4: new accounts (Panel A), consumer credit balances (Panel B), delinquencies (PanelC),andEquifaxRiskScore(PanelD).23 Thepre-trendssuggestthatthereareparalleltrends in the number of new accounts and consumer credit balances and that we can interpret effects of the tax cuts on credit usage as causal. For delinquencies, however, we observe what appears to be some decline for those who eventually have larger tax decreases starting in 2017. While it is possiblethatindividualsexpectingtaxreformtopassin2017wereabletomakebudgetarychanges to stave off delinquency, due to this pre-trend, we cannot conclude that the tax cuts had a causal effect on account delinquencies.24 For Equifax Risk Score, however, we observe a decline in the years leading up to TCJA passage in 2017 and then an increase in 2018 and 2019, suggesting that thepre-trendworksagainstfindingaresultfromtheTCJAtaxdecreases. On the whole, we interpret our evidence on the effect of personal tax decreases on quantitative measures of household financial well-being as consistent with the results on subjective measures. However,therelationshipbetweenincreasedcreditusageandfinancialwell-beingisnotclearcut. On one hand, these results could reflect a consumption response to greater disposable income, as in Dinerstein, Yannelis and Chen (2023). On the other hand, the results could also be generated by a greater reliance on borrowing to cover fixed expenses after a negative income shock, as 22We present results for the natural log of consumer credit balances, which exclude accounts with zero balance. Resultsforthenaturallogofbalancesplusonearequalitativelysimilar,withverysimilarcoefficientsandstatistical significance. Additionally,wefindsimilareffectsfordelinquenciesoverothertimehorizons,suchas30to120days pastdue,or60to120dayspastdue. 23Thefigureplotstheβ coefficientsandthe95percentconfidenceintervalfromthefollowingevent-studyspecifik cation: Y it =∑2 k= 01 2 9 015,k̸=2017 β k ·I{k=t}·Taxdecrease i +α i +α ct +γ 1 Agebin i +ε it . Standarderrorsareclusteredat thecountylevel. 24FindingsfromtheFederalReserveBankofNewYork’sSurveyofConsumerExpectationssuggeststhatfollowing the2016election,therewasanotableincreaseinindividuals’perceivedlikelihoodofnear-termincomeandpayrolltax declines,providingsomesupportiveevidencethatanticipationeffectscouldhavebeenatplayin2017. Thefraction of survey respondents expecting a decrease in the average income tax rate over the next 12 months rose from 11.5 percentinAugust2016to27.7percentinDecember2016. Thefractionexpectingadecreaseinthetopmarginalrate rosefrom15.2percentto41.2percentovertheperiod(FederalReserveBankofNewYork,2024). 17

in Dodini, Larrimore and Tranfaglia (2022). The implications of reduced delinquencies are more clear;sincedelinquenciesaresymptomsofdamagingfinancialevents,alowerprobabilityofdelinquency would be expected to reflect higher levels of financial well-being. Looking at our credit usage result in the context of the results on delinquencies and the SHED outcomes suggests that a positive, consumption response is more natural since we observe that the increased borrowing after a positive income shock is accompanied by improved financial well-being in terms of other variables—subjective financial well-being, home ownership, and student loans—and we do not observe a deterioration in delinquencies. Additionally bolstering this interpretation of increased borrowing due to a consumption response, we find that credit scores were essentially unchanged, so the tax decreases did not affect individuals’ credit risk profiles from a lender’s perspective on average. IV.III Robustness Robustness exercises for the SHED analysis show that the effects on financial well-being are of similarmagnitudeswithdifferentcontrolsandsamplerestrictions. AppendixTableA.3showsthat the estimated effect of a 1 percentage point tax decrease on the likelihood that someone is living comfortably is quite stable across specifications and samples. An exception is that effect sizes are larger when we omit the individual fixed effect and when controls are lagged twice, so they are pre-determined with respect to the effect of the tax change. The larger effect in the twice-lagged control sample appears due to the smaller sample size caused by this restriction, however, and not the inclusion of the twice-lagged controls themselves. Results are similar for a specification thatusescontemporaneouscontrols,butrestrictsthesampletoindividualsforwhomtwice-lagged controlsareavailable. WediscusstheseresultsinmoredetailinOnlineAppendixSectionA. FortheCCPanalysis,wepresentrobustnessresultsforalternativespecificationsandalternative methods for calculating the tax decrease in Appendix Table A.4 and describe them in detail in Online Appendix Section A. We find that the results for our preferred specification are similar in magnitudeandsignificancetothoseforalternativessuchasincludingcountyandyearfixedeffects 18

andincludingeconomiccontrolvariablesliketheunemploymentrateandemploymentgrowthwith countyandyearfixedeffects. Wealsofindourresultsarerobusttoalternativewaysofcalculating the tax decrease, including a method for incorporating dependents into the calculation, a method forincorporatingbusinessincome,andanalternativemethodforcalculatingpropertytaxliabilities. V Conclusion This paper provides plausibly causal evidence that larger personal income tax decreases after the TCJA led to greater improvements in subjective well-being in the two years after their enactment, with effects concentrated at the highest level of well-being: individuals reporting living comfortably. Improvements in subjective financial well-being are also accompanied by a decrease in the likelihoodofhavingstudentloansbutbylittledetectableeffectonmeasuresofhouseholdsavings. Largertaxdecreasesalsohavemodesteffectsonconsumercreditoutcomes,whicharegenerally consistentwiththeresultsonsubjectivewell-being. TCJAtaxdecreasesareassociatedwithasmall increase in the number of new accounts and an increase in consumer credit balances. On net, we observe no change in individuals’ credit scores and can not interpret the post-TCJA delinquencies declineascausal. Overall, our results are consistent with tax decreases improving people’s financial well-being whetherornottheynotablyimproveotherquantitativemeasureslikecashonhandorcreditscores. As policymakers weigh the costs and benefits of individual income tax changes in future policy debates—particularly with the TCJA individual tax cuts set to expire in 2025—this research providesinsightintooneimportantchannelofhowtaxpolicyaffectshouseholdwell-being. Still,since the tax decreases in the TCJA were concentrated among households earning higher incomes, our resultsleaveopenthepossibilitythattaxdecreasesthataffectlowerincomehouseholdscouldhave differenteffects. Householdslivingonlowerincomesmayusetheirtaxdecreasesdifferently—for example, in ways that may leave their levels of financial well-being unchanged, or move them alongdifferentmarginsintermsofborrowingandsaving. Additionalresearchcouldshedlighton 19

thesedifferencesandfurtherclarifythemechanismsbehindourresults. References Aaronson,Daniel,SumitAgarwal,andEricFrench.2012.“Thespendinganddebtresponseto minimumwagehikes.”AmericanEconomicReview,102(7):3111–39. Agarwal, Sumit, and Wenlan Qian. 2014. “Consumption and debt response to unanticipated incomeshocks: EvidencefromanaturalexperimentinSingapore.”AmericanEconomicReview, 104(12):4205–30. Agarwal,Sumit,ChunlinLiu,andNicholasS.Souleles.2007.“Thereactionofconsumerspendinganddebttotaxrebates—evidencefromconsumercreditdata.”JournalofPoliticalEconomy, 115(6):986–1019. Argys,LauraM.,AndrewI.Friedson,M.MelindaPitts,andD.SebastianTello-Trillo.2020. “Losing public health insurance: TennCare reform and personal financial distress.” Journal of PublicEconomics,187:104202. Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. 2004. “How much should we trust differences-in-differences estimates?” The Quarterly Journal of Economics, 119(1): 249– 275. Bhutta, Neil, and Lisa Dettling. 2018. “Money in the bank? Assessing families’ liquid savings usingtheSurveyofConsumerFinances.”BoardofGovernorsoftheFederalReserveSystem. Bhutta,Neil,PaigeMartaSkiba,andJeremyTobacman.2015.“Paydayloanchoicesandconsequences.”JournalofMoney,CreditandBanking,47(2-3):223–260. Black,SandraE,JeffreyTDenning,LisaJDettling,SarenaGoodman,andLesleyJTurner. 2023. “Taking it to the limit: Effects of increased student loan availability on attainment, earnings,andfinancialwell-being.”AmericanEconomicReview,113(12):3357–3400. Brevoort, Kenneth, Daniel Grodzicki, and Martin B. Hackmann. 2020. “The credit consequencesofunpaidmedicalbills.”JournalofPublicEconomics,187:104203. Buhl,John.2022.“Thetaxdebateisjustbeginning.”TaxVox: FederalBudgetandEconomy. Callaway, Brantly, Andrew Goodman-Bacon, and Pedro H. C Sant’Anna. 2024. “Differencein-differences with a Continuous Treatment.” National Bureau of Economic Research Working Paper32117. Consumer Financial Protection Bureau. 2017. “CFPB financial well-being scale: Scale developmenttechnicalreport.”ConsumerFinancialProtectionBureau. Deaton, Angus. 2008. “Income, health, and well-being around the world: Evidence from the GallupWorldPoll.”JournalofEconomicPerspectives,22(2):53–72. 20

Dettling, Lisa J., and Joanne W. Hsu. 2021. “Minimum wages and consumer credit: Effects on accessandborrowing.”TheReviewofFinancialStudies,34(5):2549–2579. Diener,Ed,ShigehiroOishi,andLouisTay.2018.“Advancesinsubjectivewell-beingresearch.” NatureHumanBehaviour,2(4):253–260. Dinerstein, Michael, Constantine Yannelis, and Ching-Tse Chen. 2023. “Debt moratoria: Evidence from student loan forbearance.” National Bureau of Economic Research Working Paper 31247. Dodini,Samuel,JeffLarrimore,andAnnaTranfaglia.2022.“FinancialrepercussionsofSNAP workrequirements.”BoardofGovernorsoftheFederalReserveSystem. Federal Reserve Bank of New York. 2024. “Survey of Consumer Expectations ublic Policy Survey.” Federal Reserve Board. 2016. “Report on the economic well-being of U.S. households in 2015.” BoardofGovernorsoftheFederalReserveSystem. Federal Reserve Board. 2018. “Report on the economic well-being of U.S. households in 2017.” BoardofGovernorsoftheFederalReserveSystem. Federal Reserve Board. 2019. “Report on the economic well-being of U.S. households in 2018.” BoardofGovernorsoftheFederalReserveSystem. Federal Reserve Board. 2020. “Report on the economic well-being of U.S. Households in 2019, featuring supplemental data from April 2020.” Board of Governors of the Federal Reserve System. Federal Reserve Board. 2021. “Economic well-being of U.S. households in 2020.” Board of GovernorsoftheFederalReserveSystem. Feenberg,Daniel,andElisabethCoutts.1993.“AnintroductiontotheTAXSIMmodel.”Journal ofPolicyAnalysisandManagement,12(1):189–194. Goodman, Lucas, Katherine Lim, Bruce Sacerdote, and Andrew Whitten. 2021. “How do business owners respond to a tax cut? Examining the 199A deduction for pass-through firms.” NationalBureauofEconomicResearch. Gross, Tal, Matthew J. Notowidigdo, and Jialan Wang. 2014. “Liquidity constraints and consumerbankruptcy: Evidencefromtaxrebates.”ReviewofEconomicsandStatistics,96(3):431– 443. Hotchkiss,JulieL.,RobertE.Moore,andFernandoRios-Avila.2021.“Impactofthe2017Tax CutsandJobsActonLaborSupplyandWelfareofMarriedHouseholds.”FederalReserveBank ofAtlantaFRBAtlantaWorkingPaper2021-18. Hu, Luojia, Robert Kaestner, Bhashkar Mazumder, Sarah Miller, and Ashley Wong. 2018. “TheeffectoftheAffordableCareActMedicaidexpansionsonfinancialwellbeing.”Journalof PublicEconomics,163:99–112. 21

Ipsos Group S.A. n.d.. “KnowledgePanel: A methodological overview.” Ipsos Group S.A. Available at https://www.ipsos.com/sites/default/files/ ipsosknowledgepanelmethodology.pdf. Joint Committee on Taxation. 2017a. “Distributional effects of the conference agreement for H.R.1,the‘TaxCutsandJobsAct’.” JointCommitteeonTaxation.2017b.“Estimatedbudgeteffectsoftheconferenceagreementfor H.R.1,the‘TaxCutsandJobsAct’.” Keane, Michael P. 2011. “Labor supply and taxes: A survey.” Journal of Economic Literature, 49(4):961–1075. Lachowska, Marta. 2017. “The effect of income on subjective well-being.” Journal of Human Resources,52(2):374–417. Larrimore, Jeff, Maximilian Schmeiser, and Sebastian Devlin-Foltz. 2015. “Should You Trust Things You Hear Online? Comparing SHED and Census Bureau Survey Results.” Board of GovernorsoftheFederalReserveSystem. Lee,Donghoon,andWilbertVanderKlaauw.2010.“AnintroductiontotheFRBNYConsumer CreditPanel.”FRBofNewYorkStaffReport,1(479). McClelland, Robert, Livia Mucciolo, and Safia Sayed. 2022. “New evidence on the effect of the TCJA on the housing market.” Tax Policy Center research report. Available at https://www.taxpolicycenter.org/publications/ new-evidence-effect-tcja-housing-market. Mezza,Alvaro,DanielRingo,ShaneSherlund,andKamilaSommer.2020.“Studentloansand homeownership.”JournalofLaborEconomics,38(1):215–260. Miller, Sarah, Luojia Hu, Robert Kaestner, Bhashkar Mazumder, and Ashley Wong. 2021. “The ACA Medicaid expansion in Michigan and financial health.” Journal of Policy Analysis andManagement,40(2):348–375. Nanda, Ambika Prasad, and Ranjan Banerjee. 2021. “Consumer’s subjective financial wellbeing: A systematic review and research agenda.” International Journal of Consumer Studies, 45(4):750–776. Oishi, Shigehiro, Ulrich Schimmack, and Ed Diener. 2012. “Progressive taxation and the subjectivewell-beingofnations.”PsychologicalScience,23(1):86–92.PMID:22157676. Roth, Jonathan, Pedro H.C. Sant’Anna, Alyssa Bilinski, and John Poe. 2023. “What’s trending in difference-in-differences? A synthesis of the recent econometrics literature.” Journal of Econometrics. Saez,Emmanuel,JoelSlemrod,andSethHGiertz.2012.“Theelasticityoftaxableincomewith respecttomarginaltaxrates: Acriticalreview.”JournalofEconomicLiterature,50(1):3–50. 22

Sahm, Claudia R., Matthew D. Shapiro, and Joel Slemrod. 2010. “Household response to the 2008 tax rebate: Survey evidence and aggregate implications.” Tax Policy and the Economy, 24(1):69–110. Scharlemann, Therese C., and Eileen van Straelen. 2022. “Fiscal policy and geographic variationintherefinancingchannel.”AvailableatSSRN. Sherter,Alain.2019.“Nearly40%ofAmericanscan’tcoverasurprise$400expense.”CBSNews. Skiba, Paige Marta. 2014. “Tax rebates and the cycle of payday borrowing.” American Law and EconomicsReview,16(2):550–576. TaxFoundation.2024.“WhereDotheCandidatesStandonTaxes?” Tax Policy Center. 2017. “Distributional analysis of the conference agreement for the Tax Cuts andJobsAct.”Urban-BrookingsTaxPolicyCenter. Tax Policy Center. 2019. “A last look at the 2019 filing season.” Urban-Brookings Tax Policy Center. van Praag, B.M.S, P. Frijters, and A. Ferrer-i-Carbonell. 2003. “The anatomy of subjective well-being.”JournalofEconomicBehaviorandOrganization,51(1):29–49. 23

VI Figures and Tables Figure 1: Distribution of TCJA personal tax reductions and relationship between tax reductions andincomelevelsintheSHEDandCCPanalysissamples 15 10 5 0 tnecreP 15 10 5 0 -4 -2 0 2 4 6 8 Tax decrease (pp) PanelA:SHEDtaxdecreases tnecreP -4 -2 0 2 4 6 8 Tax decrease (pp) PanelB:CCPtaxdecreases 6 4 2 0 -2 )pp( esaerced xaT 6 4 2 0 -2 0-10 10-20 20-30 30-40 40-50 50-75 75-10 1 0 00-12 1 5 25-15 1 0 50-250 Household income (thousands) PanelC:SHEDtaxdecreasesandincomes )pp( esaerced xaT 0-10 10-20 20-30 30-40 40-50 50-75 75-10 1 0 00-12 1 5 25-15 1 0 50-200 Household income (thousands) PanelD:CCPtaxdecreasesandincomes ThisfigurepresentshistogramsofthedistributionoftheTCJApersonaltaxreductionsinpercentagepoints(pp)forthe SHEDanalysissample(PanelA)andfortheCCPanalysissample(PanelB),aswellasboxplotsoftherelationship betweenthetaxreductionsandsampleincomefortheSHEDandCCPsamples(PanelCandPanelD).Theboxplots fortheSHEDanalysissampleshowdatafortheincomebinsreportedtothesurveyin2017.TheboxplotsfortheCCP analysisshowdataforthecorrespondingbinsofcensus-tractmedianincome(drawnfromtheU.S.CensusBureau’s American Community Survey) assigned to each individual in the sample in 2017. Sample selection is described in Section III.I, and a detailed data description for all variables is provided in the Data Appendix. Data sources are Federal Reserve Board, SHED; Federal Reserve Bank of New York/Equifax Consumer Credit Panel; U.S. Census Bureau,AmericanCommunitySurvey;CRISM;NBER,TAXSIM;andauthors’calculations. 24

Figure2: Eventstudiesofeffectsonfinancialwell-being .04 .02 0 -.02 -.04 )pp( tceffe detamitsE .04 .02 0 -.02 -.04 2015 2016 2017 2018 2019 Year PanelA:Livingcomfortably )pp( tceffe detamitsE 2015 2016 2017 2018 2019 Year PanelB:Doingokay .04 .02 0 -.02 -.04 )pp( tceffe detamitsE 2015 2016 2017 2018 2019 Year PanelC:Gettingby This figure presents event-study effects of the TCJA tax decrease on various categories of households’ subjective financial well-being: whether a household reports they are “Living comfortably” (Panel A), “Doing okay” or better (Panel B), or “Getting by” or better (Panel C). The black lines in the figures indicate coefficient estimates, and the graylinesindicate95percentconfidenceintervalsobtainedbyestimatingaspecificationsimilartoEquation(1),but interacting the TCJA tax decrease per individual with dummy variables for the years 2015, 2016, 2018, and 2019 (asdescribedinfootnote21). Intheyearsbeforethetaxchange,peoplewhowouldexperiencelargertaxreductions followingtheTCJAhadnosignificantdifferencesintheirprobabilityofsayingtheywerelivingcomfortably. Inthe yearsfollowingthetaxchange,thereportedlikelihoodoflivingcomfortablyincreased. Sampleselectionisdescribed inSection III.I,outcome variablesare describedinSection III.III,and adetailed datadescriptionfor allvariables is providedintheDataAppendix. DatasourcesaretheFederalReserveBoard,SHED;CRISM;NBER,TAXSIM;and authors’calculations. 25

Figure3: Eventstudiesofeffectsonconsumercreditoutcomes .04 .03 .02 .01 0 -.01 tceffe detamitsE .03 .02 .01 0 -.01 2015 2016 2017 2018 2019 Year PanelA:#Newaccounts tceffe detamitsE 2015 2016 2017 2018 2019 Year PanelB:ln(Consumercreditbalances) .01 .005 0 -.005 tceffe detamitsE 1 .8 .6 .4 .2 0 2015 2016 2017 2018 2019 Year PanelC:#Delinquencies tceffe detamitsE 2015 2016 2017 2018 2019 Year PanelD:EquifaxRiskScore Thisfigurepresentsevent-studyestimatesoftheTCJAtaxdecreaseonconsumercreditoutcomes: thenumberofnew consumer credit accounts (Panel A), the natural log of consumer credit balances (Panel B), the number of accounts 60daysdelinquenttoinseverelyderogatorystatus(PanelC),andtheEquifaxRiskScore(PanelD).Theblacklines in the figures indicate coefficient estimates and the gray lines indicate 95 percent confidence intervals obtained by estimatinga specificationsimilarto Equation(2), but interactingthe TCJAtaxdecrease perindividualwith dummy variablesfortheyears2015,2016,2018,and2019(asdescribedinfootnote23). Weobservepatternsthatsuggestthe paralleltrendsassumptionholdsparticularlyforthenumberofnewaccountsandforconsumercreditbalances.Sample selectionisdescribedinSectionIII.I,outcomevariablesaredescribedinSectionIII.III,andadetaileddatadescription for all variables is provided in the Data Appendix. Data sources are Federal Reserve Bank of New York/Equifax ConsumerCreditPanel;U.S.CensusBureau,AmericanCommunitySurvey;CRISM;NBER,TAXSIM;andauthors’ calculations. 26

NONCONFIDENTIAL // EXTERNAL Table1#: Summarystatistics: SHEDanalysis Number of Mean Median Standard observations deviation Variable of interest Tax reduction (percent) 5,830 1.94 1.93 1.10 Outcome variables Living comfortably 5,830 0.34 0.00 0.47 Doing okay 5,830 0.41 0.00 0.49 Just getting by 5,830 0.19 0.00 0.39 Struggling to get by 5,830 0.07 0.00 0.25 Student loans 5,830 0.13 0.00 0.34 Home owner 5,830 0.70 1.00 0.46 Would cover $400 with cash or equivalent 5,830 0.62 1.00 0.49 Has emergency fund 5,830 0.54 1.00 0.50 Individual control variables Household income less than $25,000 5,830 0.19 0.00 0.39 Household income $25,000 to $49,999 5,830 0.23 0.00 0.42 Household income $50,000 to $99,999 5,830 0.30 0.00 0.46 Household income $100,000 to $199,999 5,830 0.25 0.00 0.44 Household income $200,000 to $249,999 5,830 0.03 0.00 0.18 Less than high school 5,830 0.03 0.00 0.16 High school or GED 5,830 0.25 0.00 0.43 Some college 5,830 0.34 0.00 0.47 College or more 5,830 0.39 0.00 0.49 Working 5,830 0.56 1.00 0.50 Nonmetropolitan 5,830 0.14 0.00 0.34 White 5,830 0.73 1.00 0.44 Black 5,830 0.10 0.00 0.30 Latino or Latina 5,830 0.11 0.00 0.31 Woman 5,830 0.47 0.00 0.50 Age 5,830 53.42 56 16.92 This table presents summary statistics for variables used in the analysis of the effect of the TCJA tax decrease on variables from the SHED. Column (1) presents the number of observations, column (2) presents the mean value, column (3) presents the median, and column (4) presents the standard deviation. The sample shown in the table consistsofvaluesfromthe2017surveyforpeopleincludedinthebaselineSHEDanalysisdatasample(asincluded in the specification shown in Table 3). Sample selection is described in Section III.I, and a detailed description for all variables is provided in the Data Appendix. Data sources are the Federal Reserve Board, SHED; and authors’ calculations. 27

Table2: Summarystatistics: CCPanalysis Number of Standard observations Mean Median deviation Variable of interest Tax reduction (percent) 1,128,407 1.76 1.85 0.73 Outcome variables Equifax Risk Score 1,128,407 703 726 105 # New accounts 1,128,407 0.88 0.00 1.27 # Delinquencies 1,128,407 0.30 0.00 1.10 ln(Consumer credit balances) 943,664 9.00 9.44 1.90 Control variables State level: Real GDP growth (percent) 1,128,407 2.60 2.34 1.62 County level: Unemployment rate (percent) 1,128,407 3.98 3.80 1.21 Average weekly wage growth (percent) 1,128,407 3.50 3.16 2.70 Employment growth (percent) 1,128,407 1.43 1.32 1.63 Zip code level: Ordinary dividends per return (thousands) 1,128,407 1.67 0.75 4.01 Net capital gains per return (thousands) 1,128,407 4.81 1.69 16.15 Individual level: Age 1,126,782 50 50 19 This table presents summary statistics for variables used in the analysis of the effect of the TCJA tax decrease on consumer credit outcomes. Column (1) presents the number of observations, column (2) presents the mean value, column(3)presentsthemedian,andcolumn(4)presentsthestandarddeviation.Thesampleshowninthetableconsists ofvaluesfrom2017forallindividualsincludedinthebaselineCCPdatasample(asincludedinthespecificationshown inTable4). Ageistabulatedforindividualswithanon-missingagevalueinthesampleandthelogofconsumercredit balances is tabulated for individuals with a non-zero balance. Individuals with a missing age are included with a separatedummyvariableinthebaselinespecifications. Othervariablesaretabulatedforallindividualsincludedinthe sample.SampleselectionisdescribedinSectionIII.I,andadetaileddescriptionforallvariablesisprovidedintheData Appendix.DatasourcesareFederalReserveBankofNewYork/EquifaxConsumerCreditPanel;U.S.CensusBureau, AmericanCommunitySurvey; CRISM;BureauofLaborStatistics,LocalAreaUnemploymentStatistics; Bureauof EconomicAnalysis, GrossDomesticProductbyState; IRSIndividualIncomeTaxStatistics; NBER,TAXSIM;and authors’calculations. 28

NONCONFIDENTIAL // EXTERNAL Ta# ble3: EffectsoftaxreductionsonSHEDvariables Living Doing okay Getting by Student Homeowner Would Emergency comfortably loans handle $400 fund (1) (2) (3) (4) (5) (6) (7) Dependent variable mean (2017) 0.34 0.74 0.93 0.13 0.70 0.62 0.54 Tax reduction (percentage points) 0.015*** -0.0017 -0.0025 -0.0062*** 0.0057* -0.0027 0.0028 (0.0051) (0.0048) (0.0040) (0.0020) (0.0029) (0.0054) (0.0042) Observations 15,854 15,854 15,854 15,854 15,854 15,854 15,854 Number of people 5,953 5,953 5,953 5,953 5,953 5,953 5,953 Individual fixed effects X X X X X X X Year-by-state fixed effects X X X X X X X Individual level controls X X X X X X X Thistablepresentsestimatesoftheeffectsofa1percentagepointreductioninanindividual’staxrateduetotheTCJAonvariousmeasuresofsubjectivefinancial well-being(columns1to3)aswellasotherfinancialoutcomes,includinghavingastudentloanoutstanding(column4),homeownershipstatus(column5),abilityto handleanunexpected$400expensewithcashoritsequivalent(column6),andhavinga3-monthemergencyfund(column7).Theresultssuggestthattaxreductions increasedthelikelihoodthatsomeonesaidtheywerelivingcomfortablyfinancially,increasedthelikelihoodofowningahome,anddecreasedthelikelihoodpeople hadstudentloans. ResultsareestimatedfromEquation(1). Thisspecificationincludesindividualfixedeffects,state-by-yearfixedeffects,andcontrolsforacubic inage,ruralstatus,employmentstatus,andcategoriesofhouseholdincome. Standarderrorsareclusteredbystateandarereportedinparentheses. ***,**,and* indicatelevelsof1percent,5percent,and10percentsignificance,respectively. TheregressionsexcludepeopleinthetopbinofSHEDhouseholdincomeswhereit isnotpossibletocomputeamidpointofincomeforthepurposesoftheTAXSIMcalculation. SampleselectionisdescribedinSectionIII.I,outcomevariablesare describedinSectionIII.III,andadetaileddatadescriptionforallvariablesisprovidedintheDataAppendix. DatasourcesaretheFederalReserveBoardSHED; CRISM;NBER,TAXSIM;andauthors’calculations. 29

Table4: Effectsoftaxreductionsonconsumercreditoutcomes Equifax Risk # New credit ln(Consumer Score accounts credit balances) # Delinquencies (1) (2) (3) (4) Dependent variable mean (2017) 703 0.881 9.841 0.298 Tax reduction (percentage points) -0.0746 0.0305*** 0.0089*** -0.0055*** (0.0692) (0.0022) (0.0034) (0.0015) Observations 5,513,155 5,513,155 4,592,701 5,513,155 Number of people 1,160,652 1,160,652 1,008,534 1,160,652 Individual fixed effects X X X X County-by-year fixed effects X X X X Age bin control X X X X Thistablepresentsestimatesoftheeffectsofa1percentagepointreductionintheaveragehyperlocalpersonaltaxrate duetotheTCJAontheEquifaxRiskScore(column1),thenumberofnewcreditaccounts(column2),thenaturallog oftotalconsumercreditbalances(column3), andthenumberofaccountsthatare60daysdelinquenttoinseverely derogatory status. Results suggest the tax reductions led to an increase in the number of new credit accounts and consumercreditbalances,ledtoadecreaseindelinquencies,andhadnostatisticallysignificanteffectoncreditscores. Results are estimated using Equation (2). This specification includes county-by-year fixed effects, individual fixed effects,andacontrolforanindividual’sagebracket. Standarderrorsareclusteredatthecountylevelandarereported inparentheses. ***,**,and*indicatelevelsof1percent,5percent,and10percentsignificance,respectively. Sample selectionisdescribedinSectionIII.I,outcomevariablesaredescribedinSectionIII.III,andadetaileddatadescription for all variables is provided in the Data Appendix. Data sources are Federal Reserve Bank of New York/Equifax ConsumerCreditPanel;U.S.CensusBureau,AmericanCommunitySurvey;CRISM;NBER,TAXSIM;andauthors’ calculations. 30

Appendices A Robustness Appendix SHED Robustness Analyses TableA.3presentsarobustnessanalysisofhowtheestimatedeffectofTCJAtaxreductionsonthe likelihood someone reports that they are living comfortably financially varies as we include various controls and sample restrictions. The first row reports the estimated coefficient and clustered standard error for the specification and sample indicated. The first five columns show coefficients as we progressively add more fine-grained controls, leading up to our preferred specification in column (5). Columns (6) and onward involve changing the sample by re-weighting, adding additional observations, or restricting to a sub-sample, including the sub-samples necessary to include differentsetsofcontrols. The first five columns of Table A.3 show that the effect size decreases by more than half when we include individual fixed effects but is relatively stable as we include other, more fine-grained controls. Column (1) shows the estimated effect including only the tax reduction variable without any additional controls. When we do not include any controls, our estimate is that a 1 percentage point tax reduction leads to a 3.0 percentage points higher likelihood, which is double the magnitude of our preferred estimate in column (5). However, the coefficient decreases to 1.1 percentage points in column (2) when we include both individual and year fixed effects. The coefficient remains stable when we include state-by-year fixed effects in column (3) as well as an age cubic, an indicatorfor living inametropolitan area, andanindicator forwhethertheperson wasworkingin column(4). Thecoefficientisslightlylarger,butnotdetectablyso,whenwemovetoourpreferred specificationincludinganumberofbinsforhouseholdincomesincolumn(5). Columns (6) through (10) of Table A.3 show that changes in the weighting and sample of observations included have small effects on the point estimates and do not change the qualitative results. Column (6) shows estimates using the baseline specification (Equation 1) in column (5) 31

when we include the yearly cross-sectional survey weights. The weights are not included in the baselinespecificationbecausetheyarecalculatedfortheentirecross-sectionineachyear,whereas we use a sub-sample of the cross-section for this analysis. It is reassuring, however, that our resultsaresimilarwhenweincludethem,thoughwithlargerstandarderrors. Column(7)showsthat we also find similar results when we include people in the top-most bin, where we cannot determine a midpoint. We also find very similar results in column (8), where we exclude people in the bottom-mostincomebin,wherethepercentagepointtaxreductionmaybeparticularlysensitiveto mis-measurement of income, since income is in the denominator of the tax reduction variable.25 Column(9)showsthattheresultsarealsosimilarwhenweincludemoreyearsofSHEDdata,covering from 2013 to 2020. Finally, column (10) shows that the results are similar when we include onlyrespondentsthatwecanobservebeforeandafterthetaxchange. Column (11) shows results when we include twice-lagged controls as opposed to the contemporaneous controls in our specification, and column (12) shows results using contemporaneous controls, but restricting to the sample with data available to include twice-lagged controls in the specification. Theestimatedeffectsareverysimilaracrosscolumns(11)and(12). Inourpreferred specification,weincludecontemporaneouscontrolsbecausethesparsenatureoftheSHED’spanel componentseverelyrestrictsoursamplewhenwerestricttopeoplewhowerealsointerviewedtwo years previously. We include twice-lagged controls to check that our main results still hold when weconditiononlyonvariablesthatarepre-determinedbeforethetaxcut,thusalleviatingconcerns that the so-called “bad controls” problem is biasing our results (i.e., that the TCJA tax reduction may be affecting the control variables directly, leading to bias). Column (10) shows that when we include these twice-lagged controls, our sample is less than one-third the size of the baseline sample and our estimated coefficient is much larger—4.1 percentage points as opposed to the baselinevalueof 1.5percentagepoints. Thecoefficientin column(10)is also muchlessprecisely estimated, given the much smaller sample—the standard error is roughly three times larger than for our preferred specification. To check whether the sample change is causing the differences or 25We code household income at $500,000 when the top-most bin indicates a household income of more than $250,000. Forthebottom-mostbin,wetakethemidpointof$2,500betweenzeroandthelowestbinof$5,000. 32

the controls change is causing the differences, column (11) estimates our preferred specification withthecontemporaneouscontrolsusingthesub-sampleofobservationsforwhichwehavetwicelagged controls available. Estimating the preferred specification on this sub-sample results in a very similar estimate of the tax reduction effect to when we use twice-lagged controls, showing that the differences in the estimate are driven by the sample restriction and not the pre-treatment controlvariables. Overall,TableA.3showsthattheestimatedeffectoftaxreductionsonthelikelihoodthatsomeone reports living comfortably financially is positive and significant across a range of specificationsandsub-samples. Theeffectsizeismuchlargerwhenweomitindividualfixedeffects. When we restrict to the roughly one-third of respondents for which we have enough observations to use (twice) lagged controls, the point estimate is larger, but also with a loss of precision. The larger point estimates appear to have more to do with the sample restriction imposed by requiring twice-lagged data,however, because coefficientsare quite similarwhen we usethe standard setof contemporaneouscontrolswiththissmallersample. CCP Robustness Analyses TableA.4presentsarobustnessanalysisofhowtheestimatedeffectofTCJAtaxreductionsonconsumercredit-relatedoutcomesvariesunderalternativespecificationsandunderalternativemethods forcalculatingtheTCJAtaxreduction. We examine robustness for the three primary consumer credit outcomes of interest shown in rows of the table: the Equifax Risk Score, the number of new credit accounts, and the number of delinquencies reported that are 60 days past due to in severely derogatory status. The baseline specification results from Table 4 (estimated using Equation (2)) are presented in column (1) for comparison. ThedetailedconstructionofallvariablesisdescribedintheDataAppendix. In the baseline specification, we include a control for an individual’s age bin as well as countyby-year and individual fixed effects. Columns (2) to (4) show that the results are generally similar using different variations of fixed effects and control variables. In the specification presented in 33

column(2),whichincludescounty-by-yearandindividualfixedeffects,wealsoincludezip-codelevel deciles of total realized capital gains and ordinary dividends received in 2017 as additional control variables. These variables control for wealth levels at a hyperlocal level, which helps alleviate concerns that wealth effects could be driving our results. For example, one may have the concern that unobserved increases in equity market wealth related to the TCJA tax cuts could be causingadowntrendindelinquenciesifhouseholdsusewealthgainstostaycurrentondebtservice payments. Inthespecificationsshownincolumns(3)and(4),wepresentresultsforincludingless granular fixed effects. Column (3) presents results that include county and year fixed effects, and column (4) presents results including county and year fixed effects as well as a number of county-level economic controls (end-of-year unemployment rate, employment growth, and total wage growth), the state-level real GDP growth of the previous four quarters, and zip-code-level deciles of realized gains and ordinary dividends per return received in 2017. We observe that the results are little changed across these alternative specifications, suggesting that neither localized annualshocksnordifferencesinlocalizedeconomicconditionsareakeydriverofourresults. Columns (5) to (7) show that the results are generally similar when employing alternative inputs for calculating the hyperlocal, representative TCJA tax reduction using the TAXSIM model, suggesting that our results are not especially sensitive to measurement of these inputs. All regressions in these columns are estimated using the baseline specification, Equation (2). Column (5) shows results using average county-level house prices in 2017 and state property tax rates to estimate property taxes paid, instead of the Census Bureau’s estimates of median census-tract-level property taxes, and results are similar to those in the baseline. Column (6) shows results including an estimate of business income for individuals in the median income group at the zip-code level, as derived from the IRS Individual Income Tax Statistics report in 2017. The TCJA also madesubstantialchangestothetreatmentofbusinesstaxincomeforpass-throughcompanieslike S corporations and partnerships, in which business income is taxed under the personal tax code (Goodman et al., 2021). Including the business income estimates—which are most often zero for the median income group—we observe essentially identical results as in the baseline. This out- 34

come suggests that the TCJA’s changes to the tax treatment of business income for pass-through companies is not driving our results. Finally, one of the drawbacks of the CCP dataset is that several relevant inputs into an estimated tax rate calculation are unobservable, like the number of dependents. Therefore, in column (7), we present results for including a probability-weighted estimate of the tax cut incorporating dependents in a household based on an individual’s age and assumed tax filing status, as described in detail in the Data Appendix. Though we observe some statistically significant, negative results for the Equifax Risk Score in this specification, the effect is economically modest. In addition, the results for the number of new accounts and for the number of delinquencies are little changed compared with the baseline, suggesting measurement error due to the inability to observe dependents in the CCP dataset is not meaningfully affecting these results. 35

B Appendix figures and tables FigureA.1: EventstudiesofeffectsonotherSHEDvariables .04 .02 0 -.02 -.04 )pp( tceffe detamitsE .04 .02 0 -.02 -.04 2015 2016 2017 2018 2019 Year PanelA:Studentloans )pp( tceffe detamitsE 2015 2016 2017 2018 2019 Year PanelB:Homeowner .04 .02 0 -.02 -.04 )pp( tceffe detamitsE .04 .02 0 -.02 -.04 2015 2016 2017 2018 2019 Year PanelC:Wouldpaya$400expensesolelywithcash oranequivalent )pp( tceffe detamitsE 2015 2016 2017 2018 2019 Year PanelD:Hasathree-monthemergencyfund This figure presents event-study effects of the TCJA tax decrease on various SHED outcome variables: whether a householdreportshavingastudentloanoutstanding(PanelA),owningahome(PanelB),iftheywouldpaya$400 expense with cash or a cash equivalent (Panel C), and having a three-month emergency fund (Panel D). The black linesinthefiguresindicatecoefficientestimates,andthegraylinesindicate95percentconfidenceintervalsobtained by estimating a specification similar to Equation (1), but interacting the TCJA tax decrease with dummy variables fortheyears2015,2016,2018,and2019(asdetailedinfootnote21). SampleselectionisdescribedinSectionIII.I, outcome variables are described in Section III.III, and a detailed data description for all variables is provided in the DataAppendix. DatasourcesFederalReserveBoard,SHED;CRISM;NBER,TAXSIM;andauthors’calculations. 36

TableA.1: Personalincometaxbracketspre-andpost-TCJA Pre-TCJA Post-TCJA Income threshold Income threshold Marginal tax rate Single filers Joint filers Marginal tax rate Single filers Joint filers 10% $0 $0 10% $0 $0 15% $9,325 $18,650 12% $9,525 $19,050 25% $37,950 $75,900 22% $38,700 $77,400 28% $91,900 $153,100 24% $82,500 $165,000 33% $191,650 $233,350 32% $157,000 $315,000 35% $416,700 $416,700 35% $200,000 $400,000 39.60% $418,400 $470,700 37% $500,000 $600,000 ThistableshowspersonalincomemarginaltaxratebracketsforsinglefilersandjointfilersbeforeandaftertheenactmentoftheTCJA.Therateshownisapplied totaxableincomeabovetheamountsgiveninthefollowingcolumnuptotheincomelevelforthenexthighesttaxbracket. DatasourceistheInternalRevenue Service. 37

NONCONFIDENTIAL // EXTERNAL TableA#.2: SHEDanalysissampleconstruction 2015 2016 2017 2018 2019 Total Cross sectional observations 5,642 6,610 12,447 11,316 12,173 48,188 Reason excluded Only observed once 2,695 3,885 6,024 4,371 4,041 21,016 Insufficient tax info 1,408 504 230 3,340 4,897 10,379 Item nonresponse 11 80 134 57 32 314 Household income topcoded 40 54 229 150 152 625 Baseline sample 1,488 2,087 5,830 3,398 3,051 15,854 Thistableshowsthenumberofobservationsintheoverall,cross-sectionalSHEDineachyearalongwiththenumber thatareexcludedfromtheanalysissampleforvariousreasonsbrokenoutbyeachyear. Thefirstrowgivesthetotal number of observations in the dataset for that year. The next rows give reasons why an observation is not in the analysissample. “Onlyobservedonce”meansthatpeopleare notobservedinanyoftheother yearsanalyzed—the most common reason that people are not included in the analysis sample. “Insufficient tax info” means that we did nothaveenoughinformationtocalculatetaxratechangesforthatpersoneitherbecausetheywerenotinterviewedin 2017orbecauseofitemnonresponsein2017. “Itemnonresponse”meansthattheobservationwasexcludedbecause an outcome or control variable was refused or otherwise unavailable due to item nonresponse. “Household income topcoded”meansthattheobservationwasremovedbecauseitincludedahouseholdincomeinthetopmost,unbounded bin. Thelastrowgivesthebaselinesample. NotethatthesizeoftheSHEDwasnearlydoubledfrom2017forward. DatasourcesaretheFederalReserveBoard,SHED;CRISM;NBER,TAXSIM;andauthors’calculations. 38

TableA.3: Robustnessoftheeffectonlivingcomfortably(SHED) Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Tax reduction (percentage points) 0.025*** 0.011** 0.011** 0.012** 0.015*** 0.013* 0.015*** 0.015** 0.012*** 0.015*** 0.041*** 0.040*** (0.0034) (0.0042) (0.0049) (0.0050) (0.0051) (0.0078) (0.0049) (0.0058) (0.0044) (0.0051) (0.013) (0.013) Observations 15,854 15,854 15,854 15,854 15,854 15,854 18,550 15,399 18,785 13,056 5,171 5,171 Number of people 5,953 5,953 5,953 5,953 5,953 7,735 5,773 5,953 4,890 4,079 4,079 Individual fixed effects X X X X X X X X X X X Year fixed effects X State-by-year fixed effects X X X X X X X X X X Age cubic, metro, and working X X X X X X X X X Household income bins X X X X X X X X Cross-sectional yearly weights X Includes top-coded incomes X Excludes bottom incomes X Data from 2013 to 2020 X Both pre and post TCJA X Twice-lagged controls X Contemporaneous controls with twice-lagged control sample X ThistablepresentsarobustnessanalysisoftheeffectoftheTCJAtaxreductiononthelikelihoodofahouseholdreportingtheywerelivingcomfortablyfinancially acrossanumberofalternativespecifications,comparedwiththebaselinespecification(Equation1). BaselineresultsarepresentedinTable3andshownincolumn (5). AlternativespecificationsaredescribedindetailintheSHEDrobustnessappendixandincludethecontrolslistedinrowsofthetable. Unlessotherwisenoted, controlsarefromthesameyearastheobservationandtheregressionsareunweighted. Unlessotherwisespecified,theregressionsexcludebothpeoplewhoreport having no family incomes and the top bin of family incomes where it is impossible to compute a midpoint. Across specifications, tax reductions increased the likelihoodthatsomeonesaidtheywerelivingcomfortably. Effectsarelargerwithoutindividualfixedeffectsandwithinthesampleofobservationswherewehave twice-laggedcontrolsavailable,bothwhenusingthetwice-laggedcontrolsandwhenusingcontemporaneouscontrols. Standarderrorsareclusteredbystateand arereportedinparentheses. ***,**,and*indicatelevelsof1percent,5percent,and10percentsignificance,respectively. SampleselectionisdescribedinSection III.I,andadetaileddescriptionforallvariablesisprovidedintheDataAppendix. DatasourcesareFederalReserveBoard,SHED;CRISM;NBER,TAXSIM;and authors’calculations. 39

TableA.4: Robustnessoftheeffectonconsumercreditoutcomes(CCP) Alternative specification Alternative tax cut estimate County-by-year County and year House-price- Including Including fixed effects and County and year fixed effects and based property estimate of estimate of Baseline controls fixed effects controls tax estimate business income dependents (1) (2) (3) (4) (5) (6) (7) Outcome Variables Equifax Risk Score -0.0746 -0.0762 0.0259 0.0401 0.0077 -0.0746 -0.2804*** [0.0692] [0.0692] [0.0835] [0.0801] [0.0710] [0.0692] [0.0728] # New accounts 0.0305*** 0.0305*** 0.0285*** 0.0285*** 0.0284*** 0.0305*** 0.0321*** [0.0022] [0.0019] [0.0022] [0.0019] [0.0021] [0.0022] [0.0023] # Delinquencies -0.0055*** -0.0055*** -0.0053*** -0.0048*** -0.0064*** -0.0055*** -0.0059*** [0.0015] [0.0015] [0.0016] [0.0016] [0.0016] [0.0015] [0.0017] County-by-year fixed effects X X X X X Age control X X X X X X X Geographic controls X X Year fixed effects X X County fixed effects X X This table presents a robustness analysis of the effect of the TCJA tax reduction on consumer credit variables across a number of alternative specifications, as compared with the baseline specification (Equation 2), and alternative methods for calculating the representative TCJA tax reduction. Results are presented for thethreedependentvariablesshowninrows: theEquifaxRiskScore, thenumberofnewaccountsandthenumberofaccountsthatare60daysdelinquenttoin severelyderogatorystatus. BaselineresultsarepresentedinTable4andincolumn(1). Alternativespecificationsandmethodsforcalculatingthetaxreductionare describedindetailintheCCProbustnessappendix. Wefindthattheresultsforourpreferredspecificationaresimilarinmagnitudeandsignificancetothosefor almostallofthealternativesshown. Standarderrorsareclusteredatthecountylevelandarereportedinparentheses. ***,**,and*indicatelevelsof1percent,5 percent,and10percentsignificance,respectively. SampleselectionisdescribedinSectionIII.I,andadetaileddescriptionforallvariablesisprovidedintheData Appendix. Data sources are Federal Reserve Bank of New York/Equifax Consumer Credit Panel; U.S. Census Bureau, American Community Survey; CRISM; Bureau of Labor Statistics, Local Area Unemployment Statistics; Bureau of Economic Analysis, Gross Domestic Product by State ; IRS Individual Income Tax Statistics;NBER,TAXSIM;andauthors’calculations. 40

C Data Appendix Survey of Household Economics and Decisionmaking The Survey of Household Economics and Decisionmaking (SHED) has been conducted annually in the fourth quarter of the year since 2013. The survey was designed by Federal Reserve Board staff and fielded by Ipsos, a private firm focused on consumer research. Survey respondents are members of Ipsos’s KnowledgePanel, and are selected into that panel using address-based sampling. The surveys are administered online, and potential respondents who do not have internet accessareprovidedwithadevicewithaninternetconnectionthatwillallowthemtotakesurveys. Ipsos Group S.A. (n.d.) provides more information on the KnowledgePanel’s construction and methodology. ResponseratesfortheSHEDarerelativelyhighforanonlinepanel. Thelowestresponserateis after initial contact, which Ipsos estimates at 13 percent in 2018. From here, 64 percent complete aninitialsurveyand54percentofthoserespondentscompletedtheSHED.Sincedrop-offsineach stage compound, that fact implies a relatively low response rate moving through each stage—4.3 percent. While the specific figures here apply to 2018 and come from Federal Reserve Board (2019), they are similar in other years. Previous analyses of statistics drawn from the SHED haveshownthattheyoftencorrespondtostatisticsfromcomparablequestionsdrawnfromsurveys with substantially higher response rates from first contact. Larrimore, Schmeiser and Devlin- Foltz(2015)andFederalReserveBoard(2016)showsthatseveralstatisticsfromtheSHEDmatch thoseintheCensusBureau’sCurrentPopulationSurvey,AmericanCommunitySurvey(ACS),and SurveyofIncomeandProgramParticipation. Thisresultsuggeststhatthevariousmeasuresusedto retainarepresentativeonlinepanelsamplelimitthelevelofbiasduetounobservabledifferences. Ipsos and the Federal Reserve take several steps to improve the representativeness of the online panel. From2014to2017,thesurveyexplicitlyincludedanoversampleofpeoplewithhousehold incomesbelow$40,000peryearalongsideanadditionalsampleofpeoplewhohadcompletedthe survey in the previous year. From 2018 onward, Ipsos gave larger incentive payments as well as morefollowupemailstopeoplewhobelongedtotargetgroupswithlowerresponserates—adults aged 18 to 20, adults with less than a high school degree, and adults who are either Hispanic or non-White.26 In addition to benefiting from efforts to improve response rates, the survey includes post-stratificationweightsdesignedtomakethesamplesinvariousyearsnationallyrepresentative. We include these weights as a robustness check and find similar results. We do not include these weightsinourbaselineanalysis,however,sincetheydonotapplytoouranalysissampleofpeople whoareinterviewedinmultipleyears. Sampleconstruction Our SHED analysis sample is a panel of respondents who are interviewed across multiple years. Wecanincludepeopleacrossmultipleyearsbecauseoftwofeaturesofthesampledesign. Thefirst reason we can observe people after the tax decrease is that the main sample in each year is drawn fromcontinuingrespondentsintheKnowledgePanel. SincetheSHEDhasarelativelylargesample 26Moredetailsonthespecificresponseratesandsurveyframesineachyearareavailableinthesurveyreportsfor eachyearontheFederalReserveBoard’swebsite. See,forexample,FederalReserveBoard(2016,2018,2019,2020, 2021). 41

thatincludesasubstantialshareoftheKnowledgePanelineachyear,alargenumberofrespondents who are randomly drawn from the KnowledgePanel in a current year were also randomly drawn in one or more previous years. The second reason is that the sampling frame included an explicit sampleofpeoplewhorespondedinthepreviousyearsin2014through2017.27 We can link 56 percent of observations from the SHED waves of 2015 to 2019 to the same personinanotheryearoftheSHED.Howeverthenumberdecreasesto35percentwhenwerestrict tohavinginformationin2017thatweusetoestimatetherespondent’staxchange. TableA.2gives thenumberofobservationsineachyear’ssurveyalongwithabreakdownofcategoriesofresponses we need to exclude from the analysis. Linking is done based on a respondent’s being a member of the repeat panel in 2016 or 2017, which implies that the unique identifier of the person in that previous year was the same as in that year. Datasets in 2018 and 2019 also include a variable for the unique identifier for the person in previous years if they were a respondent. It may be possible tolinkmoreobservationswithadditionaldata. Notethatmostobservationsarenotlinkedbecause theyarenotsampled,notbecauseofnon-response. TableA.2alsopresentstheotherreasonswhypeopleareexcludedfromtheanalysissample,includingtop-codedhouseholdincomesanditemnon-responseinvariablesweuseintheregression but not the tax calculation. The final data sample includes 15,854 observations, or 33 percent of thecross-sectionalobservationsfrom2015to2019. SHEDoutcomes Financialwell-being. Financialwell-beingiscodedasresponsestothefollowingquestion(number B2): “Overall, which one of the following best describes how well you are managing financially these days?” Respondents can answer “living comfortably,” “doing okay,” “just getting by,” or “finding it difficult to get by.” Since the responses are ordinal, we present regression results in termsoftheshareofpeoplewhoaredoingatleastaswellasagivencategory. Forexample,inthe variablefordoingokayorbetter,weincludepeoplewhosaytheyarelivingcomfortablyalongside peoplewhosaytheyaredoingokay. The financial well-being question is a major focus of the SHED release each year (Federal ReserveBoard,2020)becauseitgivesinsightsintohowpeoplesubjectivelyfeelabouttheirfinances. Methodologically it resembles attempts to examine overall well-being—for example overall life satisfaction in the Gallup World Poll (Deaton, 2008)—in that it asks people to say where they standinahierarchyofcategoriesoffinancialwell-being. Thequestionalsogivessimilarresultsto more involved measures. In 2017 and 2020, the SHED also included a financial well-being scale developed by the Consumer Financial Protection Bureau (CFPB; Consumer Financial Protection Bureau,2017),andanswerstoeachmeasureshowedsimilartrends(FederalReserveBoard,2018, 2021). Other measures of personal finances. In addition to overall financial well-being, the SHED asks extensive questions about specific parts of people’s finances. In keeping with the more subjectivemeasuresofoverallfinancialwell-being,wealsoincludeseveralsomewhatsubjectiveconcepts that are similar to questions used in the CFPB’s financial well-being scale. These items 27Notethatwhileeachfeaturemakesitmorelikelythatpeoplewillbeinterviewedinconsecutiveyears,wedonot imposethisrestriction. Sosomepeoplearere-interviewedafterseveralyears. 42

includethefollowing: • Would handle $400: Would cover a $400 expense with cash or a near equivalent (Sherter, 2019). Thismeasureiscommonlyemployedtoassessthefinancialsecurityoffamilies.28 Wealsoincludearguablylesssubjectivemeasuresabouttherespondent: • Emergencyfund: Hadanemergencyfundthatwouldcovertheirexpensesforthreemonths • Studentloans: Hadastudentloan • Homeowner: Ownedtheirhome Administrative measures from consumer credit reports in the CCP analysis also provide less subjectivemeasuresofpeople’sfinances. Computingtaxchanges The SHED has the advantage of providing many of the main inputs necessary to identify average tax rates, and we use them to compute implied changes in taxes using the NBER TAXSIM tax microsimulation model (Feenberg and Coutts, 1993). Specifically, we compute tax rates based on informationobtainedinthelatestyearavailablebeforethetaxlawwasimplementedsoastoavoid any possible behavioral effects of the tax law changes. For our main specification, we compare average tax rates in 2017, before the law was implemented, with average tax rates in 2018, after the law was implemented. Since we are comparing someone with the same characteristics before and after, the differences are due to changes in the tax law from 2017 to 2018, which we interpret astheeffectoftheTCJA.29 The SHED itself, as well as the other surveys given to the Ipsos KnowledgePanel, provide the bulkofourinputsintoTAXSIMthatidentifythesizeoftaxdecreases. Theyincludethefollowing: • Total family income, TAXSIM variable pwages: We use income as reported in a “panel variable” that asked all members of the Ipsos KnowledgePanel to report their household income. The variable is named ppincimp in the dataset. We use the midpoint of each income category. We generally exclude the top category, which is “$250,000 or higher” in 2017. Where it is included we arbitrarily use $500,000. Since we are unable to identify differenttypesofincome,weassumethatallofthefamily’sincomeiswageincome. • Filing status, TAXSIM variable mstat: We use an Ipsos KnowledgePanel variable giving maritalstatus,ppmarit. • Numberofdependents,TAXSIMvariablesdepx,dep13,dep17,anddep18: Weusethe number of individuals reported as living in the household of various ages under 18. The number of people in the household of various ages is drawn from “panel variables” asked of all members of the Ipsos KnowledgePanel, not the equivalent questions in the SHED 28AcomparisonoftheimplicationsofthisquestionwithmeasuresobtainedfromtheSurveyofConsumerFinances isinBhuttaandDettling(2018)andBox3ofFederalReserveBoard(2020). 29Coding was done through a submission to TAXSIM version 32, available at: https://users.nber.org/taxsim/taxsim32/. 43

itself. We include an additional adult dependent if the person reports living with extended family, parents, or a friend and report that they do so to care for that individual or group of individualsinthemainsurvey. • Mortgage interest payments, TAXSIM variable mortgage: To calculate an estimate of annual mortgage interest payments, we use question M4, which asks people who own their homes with a mortgage to report the range of their total monthly mortgage payment. To calculate the interest share of the total mortgage payment, we use the Equifax Credit Risk InsightServicingandICE,McDash®Data(CRISM)datasetfor2017:Q4tocalculatethemedianinterestpaymentshareofmortgagepaymentsbyzipcode. Wemapzipcodestocensus tracts using a crosswalk provided by the Department of Housing and Urban Development. We exclude zip codes with fewer than 10 observations. For missing tracts, we use median county-leveldata,andweusemedianstate-leveldataifcounty-leveldataaremissing.. • State of residence, TAXSIM variable state: We use the individual’s reported state of residenceaccordingtotheIpsosKnowledgePanel. • Property tax value estimation, TAXSIM variable proptax: For property tax estimates, we use census-tract-level data from the U.S. Census Bureau’s American Community Survey in 2017 on median real estate taxes paid with a mortgage (variable name HD01 VD03) and medianrealestatetaxespaidwithoutamortgage(variablenameHD01 VD04). Formissing tracts, we use median county-level data, and we use median state-level data if county-level dataaremissing. From TAXSIM output, we calculate an individual’s average tax rate under 2017 tax law as the sumoffederalandstateincometaxliabilitydividedbytheindividual’sincome(usingtheTAXSIM variable names: (fiitax + siitax)/pwages). We calculate an individual’s average tax rate under the 2018 tax code in the equivalent way. Then, for each individual, we define the TCJA tax rate reductionas-1*(2018taxrate–2017taxrate). Federal Reserve Bank of New York/Equifax Consumer Credit Panel (CCP) TheCCPisanindividual-level,anonymizedpaneldatasetofconsumercreditrecords,drawnatthe endofeachquarterfromEquifax—oneofthethreemajorcreditbureausintheUnitedStates. The primaryCCPdatasampleconsistsofa5percentrandomsampleofallU.S.individualswithSocial Security numbers and credit records, and each quarter, the panel is updated as new individuals establish credit records. Once an individual establishes a credit history and enters the sample, they remain in the sample continuously, whether or not they have credit activity in a particular quarter, until death. The data include detailed information drawn from credit reports, including loan balances, credit limits, and payment status. Aside from variables on age and geographic location, the dataset is generally limited to information about credit status. The CCP does not contain information, for example, about household income, employment status, or demographic characteristics like race and education level. Additional information about the dataset, including samplingandmethodology,isavailableinLeeandVanderKlaauw(2010). 44

Sampleconstruction To draw our sample of individuals’ credit histories, we use a 10 percent sample of the overall CCP dataset and include the years 2015 to 2019—three years before, to two years after, the TCJA enactment. We use end-of-year observations to smooth through fluctuations in financial positions over the course of the year and for consistency with the annual nature and timing of the SHED, which results in a total of 6,598,370 observations in the dataset. To be included in the sample, an individual is required to be in the dataset in 2017—the year before the TCJA implementation—as wellasin2018or2019,tohaveareportedEquifaxRiskScore(atypeofcreditscore),andtohave a reported value for the number of new accounts established in a year. These requirements limit thesampleto5,513,155observations. CCPoutcomes WestudyfiveoutcomesfromtheCCP: • EquifaxRiskScore: Anindividual’screditscore • #Newaccounts: Thenumberofaccountsopenedwithin12months • # Delinquencies: The sum of the number of accounts that have been reported 60, 90, and 120dayspastdueplusthenumberofaccountsthathavebeenreportedinseverelyderogatory status • ln(Consumer credit balances): The natural log of total outstanding account balances minus mortgageaccountbalances Computingtaxchanges TheCCPdatasetdoesnotincludeinformationonhouseholdincome,taxfilingstatus,orotherdata necessary to identify an individual’s average tax rate change following the TCJA. Therefore, we calculatehyperlocal(i.e.,census-tract-level),representativechangesineffectivepersonaltaxrates followingtheTCJAtostudyhowthetaxchangesaffectedhouseholdfinances. Using2017dataas inputs into the NBER’s TAXSIM model of tax liabilities, we calculate the representative average tax rate (federal plus state taxes) under 2017 tax law and under 2018 tax law. As in the SHED analysis,wedefinetheTCJAtaxratereductionas-1*(2018taxrate–2017taxrate). For each census tract observed in the CCP dataset in 2017, we calculate four representative tax rate changes: for single filers with a mortgage, for single filers without a mortgage, for joint filers withamortgage,andforjointfilerswithoutamortgage. Wethenassigntherelevanttaxratechange to an individual in the CCP based on the census tract in which they lived in 2017:Q4, whether we attribute joint or single filing status to them in 2017:Q4, and whether we observe them holding an outstandingmortgageloanin2017:Q4. AllcalculationsweredoneusingTAXSIMversion32. Using the TAXSIM inputs described in the next paragraph, we calculate the representative average federal plus state tax rates under 2017 tax law and under 2018 tax law. We calculate the totalaveragetaxratechangegiventhesubstantialinteractionsbetweenfederalandstatetaxcodes. The tax rate for each year is calculated as (Federal individual income tax liability (fiitax) + State individualincometaxliability(siitax))/(Wageandsalaryincomeofprimarytaxpayer(pwages)). 45

WedescribetheinputsintotheTAXSIMmodelindetail: • Filing status, TAXSIM variable mstat: To assign single or joint tax filing status to individuals in the CCP, we use information from a supplemental CCP data sample of individuals wholiveatthesameaddressaspeopleincludedintheprimarydatasample. Forhouseholds of more than one individual with a credit record in the CCP but fewer than five individuals, weassignaperson“marriedfilingjointly”filingstatusiftheagedifferencebetweenanytwo members of the household is smaller than 15 years. For households of more than four individuals with a credit record, we assign “single” filing status, as these individuals are more likelytoliveinmulti-unitbuildings. • Income measures, TAXSIM variable pwages: As measures of income by census tract, we use data from the U.S. Census Bureau’s American Community Survey from 2017. As our measure of income for joint filers, we use married-couple family median income (variable name HC03 EST VC13). As our measure of income for single filers, we use median earningsforworkers(variablenameHC01 VC124). • Mortgage interest calculation, TAXSIM variable mortgage: Using the CRISM dataset for 2017:Q4, we calculate an estimate of annual mortgage interest payments for each individual by multiplying the interest rate and mortgage balance, taking the median by zip code, and mapping zip codes to census tracts using a crosswalk provided by the Department of HousingandUrbanDevelopment.30 Weexcludezipcodeswithfewerthan10observations. For missing tracts, we use median county-level data, and we use median state-level data if county-leveldataaremissing. • Property tax value estimation, TAXSIM variable proptax: For property tax estimates, we use census-tract-level data from the U.S. Census Bureau’s American Community Survey in 2017 on median real estate taxes paid with a mortgage (variable name HD01 VD03) and medianrealestatetaxespaidwithoutamortgage(variablenameHD01 VD04). Formissing tracts, we use median county-level data, and we use median state-level data if county-level dataaremissing. • State of residence, TAXSIM variable state: We use the individual’s state of residence reportedintheCCPasof2017:Q4. Computingtaxchanges(alternativemethodsforrobustness) We also use several alternative methods to compute the hyperlocal, representative TCJA income taxchangestoassesstherobustnessofourprimarymeasure: • Alternative property tax calculation: To calculate an alternative estimate of the census-tract median property tax, we multiply the state property tax rate sourced from Tax-Rates.org by thecensus-tracthomevaluemedianfromthe“SelectedHousingCharacteristics”tableinthe U.S. Census Bureau’s American Community Survey. All other TAXSIM inputs remain the same. 30The crosswalks are available on the Department of Housing and Urban Development’s website at https://www.huduser.gov/portal/datasets/usps crosswalk.html. 46

• Incorporatingbusinessincomedataintothetaxchangeestimate,TAXSIMvariablepbusinc: ToincorporateanestimateofbusinessincomeintotheTCJAtaxratechangecalculation,we calculate median business income per return in a zip code for an individual in the median income group in the zip code using data from the IRS Individual Income Tax Statistics in 2017. AllotherTAXSIMinputsremainthesame. • Incorporatingdependentsintothetaxchangeestimate: WecalculateanestimateoftheTCJA tax rate change including dependents using data on the number of children and the fraction ofhouseholdswithchildrenfromtheCensusBureau’s2017report“America’sFamiliesand Living Arrangements.” While households may have non-child dependents, we focus on measuringthenumberofchildrenasdependentsforthisanalysis. Since we cannot observe whether someone has dependents, we calculate an expected tax decreasebycomputingaweightedaverageofexpectedtaxreductionsconditionalonhaving and not having dependents. To construct this variable, denoted Taxdecreased , we begin jafm bycomputinganexpectedtaxdecreaseconditionalonhavingdependents,TaxdecreaseD=2. jfm Therelevant“withdependents”taxdecreaseisbasedonthecensustractinwhichtheperson livedin2017:Q4(j),whetherweattributejointorsinglefilingstatustothemin2017:Q4(f), and whether we observe them holding an outstanding mortgage loan in 2017:Q4 (m). Since theaveragenumberofchildrenperparentisthesameformarriedandsinglefilersaswellas forindividualswithandwithoutamortgage,weuniformlyassignthe2017nationalaverage of two children as dependents (D = 2; TAXSIM variable depx). The weights applied to this variable are denoted by κ and are based on the fraction of individuals with children af by age, for single households and married households from the Census Bureau. The other component, TaxdecreaseD=0, is the estimate of the tax decrease conditional on not having jfm dependents (D=0), based on the person’s census tract, filing status, and mortgage status.31 Inequationform: Taxdecreased =κ ∗TaxdecreaseD=2+(1−κ )∗TaxdecreaseD=0 jafm af jfm af jfm Datasourcesanddefinitionsforcontrolvariables • StateGDP:Four-quarterpercentchangeinstategrossdomesticproduct,fromtheBureauof EconomicAnalysis • Unemployment rate: County-level unemployment rate for December from the Bureau of LaborStatisticsLocalAreaUnemploymentStatisticsreport • Employment growth: Twelve-month percent change in total county-level employment from theBureauofLaborStatisticsLocalAreaUnemploymentStatisticsreport • Wage growth: Four-quarter percent change in the county-level average weekly wage from theBureauofLaborStatisticsQuarterlyCensusofEmploymentandWagesreport 31Note that this latter tax decrease, TaxdecreaseD=0, is the same as the tax decrease used in the main text, jfm Taxdecrease. i 47

• Ordinary dividends: Total ordinary dividends per zip code in 2017, from the IRS Individual IncomeTaxStatistics • Realized capital gains: Total realized capital gains per zip code in 2017 from the IRS IndividualIncomeTaxStatistics. • Age: We calculate age by subtracting birth year in the CCP dataset from sample year, and we generate 13 age bins for the following age groups: 18 to 24, 25 to 29, 30 to 34, 35 to 39, 40 to 44, 45 to 49, 50 to 54, 55 to 59, 60 to 64, 65 to 69, 70 to 74, 75 to 79, and 80 or older. Wealsoincludeanagebinforobservationswithoutabirthyear 48