Energy Consumption and Inequality in the U.S.: Who are the Energy Burdened?

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Energy Consumption and Inequality in the U.S.: Who are the Energy Burdened? Octavio M. Aguilar and Cristina Fuentes-Albero 2025-026 Please cite this paper as: Aguilar, Octavio M., and Cristina Fuentes-Albero (2025). “Energy Consumption and Inequality in the U.S.: Who are the Energy Burdened?,” Finance and Economics Discussion Series 2025-026. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2025.026. 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.

Energy Consumption and Inequality in the U.S.: Who are the Energy Burdened?* Octavio M. Aguilar‡ Cristina Fuentes-Albero‡ April 1, 2025 Abstract Usingabroaddefinitionofenergyconsumptionthatincludesbothresidentialenergyuse andgasolinefortransport,weidentify20%ofhouseholdsinthePSIDasenergyburdened(EB) basedonatwice-the-median,income-basedthreshold. Logitanalysisshowsthatbeingnonwhite, being single with dependents, receiving public assistance, having no post-secondary education,andbeingunemployedincreasetheprobabilityofbeingEB.Wedocumentfourkey empiricalfacts:(1)EB/non-EBstatusispersistent;(2)EBhouseholdshavesignificantlyhigher marginal propensities to consume and marginal propensities to consume energy compared to non-EB households; (3) EB households experience lower expected energy consumption growthdespitehavinghigherexpectedincomegrowthrelativetonon-EBhouseholds;and(4) EBhouseholdsfacemorevolatileenergyconsumptionandincomethannon-EBhouseholds. Lastly, we show that both consumption inequality and energy consumption inequality have risen more moderately than income inequality over the 1999 to 2021 period. Inequality in residential energy consumption increased until 2009, then declined, whereas inequality in gasoline consumption for transport has risen steadily, reaching a level 50% higher in 2021 thanin1999. JELCLASSIFICATION: E21,I32. KEYWORDS: ENERGYCONSUMPTION,ENERGYBURDEN,INEQUALITY *WethankMatthiasPaustian,AndreasTischbirek,JavierFerri,andseminarparticipantsattheFederal ReserveBoard,theconference"MacroeconomicImplicationsofDecarbonizationPoliciesandActions",the 2024 Southern Economic Association, and the 2024 Workshop of the G7 CCMWG modeling experts for veryusefulcommentsandsuggestions. Theviewsexpressedinthispaperaresolelytheresponsibilityof theauthorsandshouldnotbeinterpretedasreflectingtheviewsoftheBoardofGovernorsoftheFederal ReserveSystemorofanyoneelseassociatedwiththeFederalReserveSystem. Thecollectionofdataused inthisstudywaspartlysupportedbytheNationalInstitutesofHealthundergrantnumberR01HD069609 andR01AG040213,andtheNationalScienceFoundationunderawardnumbersSES1157698and1623684. ‡Address: Board of Governors of the Federal Reserve System, 20th Street and Constitution Avenue NW, Washington, D.C. 20551, USA Email addresses: octaviomaguilar@outlook.com (O.M. Aguilar), cristina.fuentes-albero@frb.gov(C.Fuentes-Albero).

1 INTRODUCTION The study of empirical patterns in consumption and income using micro-data is ubiquitous in the macroeconomic literature.1 However, less attention has been devoted to the analysisofenergyconsumptionexpendituresusingmicro-dataunderamacroapproach. In this paper, we document four new empirical facts about energy consumption and energy burden (EB) status in the U.S. using the Panel Study of Income Dynamics (PSID). In particular, we show that (i) EB status is persistent; (ii) EB households have significantly larger marginal propensities to consume for both energy and all goods; (iii) EB households have lower expected energy consumption growth despite having higher expected income growth; and (iv) EB households face more volatile energy consumption and income. SinceBoardman(1991),theliteratureonenergyconsumptionandenergypovertyhasdefined household energy consumption as expenditures on electricity, gas, and other fuels for domestic use (hereafter, residential energy).2 In our PSID sample, U.S. households spend, on average, approximately 6% of their disposable income on residential energy andanadditional5%onenergyfortransport. Therefore,wearguethatenergyconsumptionshouldincludenotonlythetraditionalresidentialenergycomponentbutalsoexpendituresonenergyfortransport. Figure 1 presents expenditure shares in overall consumption by income decile in the PSID, based on survey waves from 1999 to 2021. As shown by the orange bars, the share of residential energy expenditures in total consumption declines monotonically with income.3 However, the share of expenditures on energy for transport, shown in lavender, remains relatively stable across income deciles—except for the top decile. This suggests the prevalence and potential significance of gasoline-related energy expenditures in assessinghouseholds’energyvulnerabilitystatus.4 Using the broad definition of energy consumption described above, we classify households as energy burdened (EB) if the share of energy expenditures to disposable income 1For example, Aguiar, Bils, and Boar (2024), Lewis, Melcangi, and Pilossoph (2024), Attanasio, Hurst, andPistaferri(2015),Heathcoteetal.(2023),andMeyerandSullivan(2023). 2See,forexample,Hernandez(2016)orLegendreandRicci(2015). 3Using data from the Residential Energy Consumption Survey (RECS), Linn, Liang, and Qiu (2023) documentthattheratioofkilowatthours(kWh)ofelectricityconsumptionper$1,000ofincomedecreases withincome. 4Using CEX data from 1999 to 2013, Oni (2024) shows a negative relationship between household expendituresharesonenergy, whichincludesresidentialandcommuting-relatedexpenditures, andincome levels. 1

FIGURE 1: HOUSEHOLDS’ EXPENDITURE BY INCOME DECILE, 1999-2021 100 4.9 6.0 6.0 5.8 5.9 5.9 5.7 5.7 5.3 4.1 10.8 8.6 7.4 6.8 6.3 6.2 5.6 5.4 5.0 4 4 . . 6 2 80 17.3 16.2 14.9 13.6 11.4 9.1 7.9 6.2 20.4 60 13.9 15.0 16.4 17.6 19.4 22.2 23.4 26.4 31.5 12.4 2.9 2.8 2.4 2.3 2.4 2.2 2.0 2.0 1.5 3.6 40 13.2 18.1 19.9 20.6 21.5 22.2 23.0 23.3 23.4 23.3 3.9 4.7 5.4 6.0 6.2 6.6 6.4 6.7 6.3 5.8 20 29.6 26.7 25.6 24.8 24.2 23.4 22.9 21.9 21.2 19.0 0 01..21 01..44 01..52 01..85 01..86 01..96 12..00 12..34 1 3 . . 2 1 1 4 . . 5 6 )%( noitpmusnoC latoT fo erahS Childcare Education Food Health Vehicle expenses Home expenses Rental equivalence Rent payment Residential energy Gas for transport 1 2 3 4 5 6 7 8 9 10 NOTE:Disposableincomedecilesarebasedonthepooledsurveydataforallwavesbetween1999and2021. Residential refers to expenditures in electricity, gas and other fuel for domestic use and gas for transport referstoexpendituresinenergyfortransport. exceedstwicethemedianshareinthesample. Thatis,weuseatwice-the-medianincomebased indicator to conclude that 20% of U.S. households in our sample are classified as EB. The average energy burden—defined as the ratio of energy expenditures to disposable income—is 25% for EB households, compared to only 7% for non-energy-burdened (non-EB) households. Most EB households (81%) are concentrated in the bottom two quintilesoftheincomedistribution. WecharacterizeEBhouseholdsusinglogitregressionanalysisandfindthattheprobability of being energy burdened is inversely related to household income. Consistent with Wang et al. (2021)’s findings using the Residential Energy Consumption Survey (RECS), we find that, after controlling for socioeconomic characteristics, Black, Asian, and Hispanic households are significantly more likely to be energy burdened relative to White households. Additionally,ouranalysissuggeststhatbeingmarried,employed,orhaving post-secondaryeducationreducestheprobabilityofbeingenergyburdened. Conversely, households with children are more likely to be energy burdened. Finally, consistent with Best and Sinha (2021)’s findings for RECS data, we find that receiving government assistanceintheformofsubsidizedhousing,foodassistance,orheatingsubsidiesisassociated withahigherprobabilityofbeingenergyburdened. Inthispaper,weputforwardfourempiricalfactsaboutenergyconsumptionandenergyburdenedstatus inthePSID. First,EB statusispersistent. Indeed, anEB household hasa probability of roughly 50% of being EB in the next survey wave, while a non-EB house- 2

hold has a nearly 90% probability of remaining non-EB across waves. Logit regressions suggestthattheprobabilityofremainingEBintwoconsecutivewavesisinverselyrelated to household income. Households with an employed head or post-secondary education aremorelikelytotransitiontonon-EBstatusinthefollowingwave. However,homeowners, Black households, and those with a female or unmarried head of household have a higher probability of remaining EB. Second, EB households have significantly larger marginal propensities to consume (MPCs) and marginal propensities to consume energy (MPCEs)thannon-EBhouseholds. Third,despiteexperiencinghigherincomegrowth,EB households have lower expected energy consumption growth than non-EB households. Finally, EB households exhibit more volatile expected energy consumption and income thannon-EBhouseholds. We also examine trends in income, consumption, and energy consumption inequality in our sample. Consistent with Heathcote et al. (2023) and Meyer and Sullivan (2023), we find that the rise in income inequality (24%) is substantially larger than that in consumption inequality (14%). Moreover, the increase in energy consumption inequality closely mirrors that of overall consumption inequality. Notably, inequality in total energy consumption and residential energy consumption rose by nearly 30% on average between 1999 and 2009 but declined by 13% in the following decade. In contrast, inequality in transportenergyconsumptionsurgedby46%overthesameperiod. When comparing inequality trends within EB and within non-EB households using the 90:10 inequality ratio,5 we find that income inequality increased by 43% between 1999 and 2021 for EB households, whereas the rise for non-EB households was just 26%. Conversely,thegrowthinconsumptioninequalityforEBhouseholdswashalfthatofnon-EB households. Wealsofindthatresidentialenergyconsumptioninequalityhasbeendeclining for EB households throughout the sample period, whereas for non-EB households, this decline began only after 2009. Finally, we identify transport energy consumption as the category with the largest increase in inequality over the sample period, rising by 68% for EB households and 50% for non-EB households. These findings underscore the importance of incorporating transport energy consumption in studies of household energy consumption in the U.S., as focusing solely on residential energy would overlook a substantialcomponent. The remainder of the paper is structured as follows. Section 2 describes the PSID data, 5The90:10ratioforincomeisdefinedastheratiooftheincomeneededtorankamongthetop10%of households in the distribution (the 90th percentile) to the income at the threshold of the bottom 10% of households(the10thpercentile). 3

Section 3 discusses the classification of energy-burdened households, Section 4 characterizes EB households, Section 5 puts forward empirical regularities for EB status and energy consumption, Section 6 explores the evolution of inequality, and Section 7 concludes. 2 DATA We use data from the Panel Study of Income Dynamics (PSID), a longitudinal survey of U.S. households. The PSID began in 1968 with a sample of approximately 5,000 households, of which around 3,000 were representative of the U.S. population (the Survey Research Center, SRC, sample or core sample), while about 2,000 were low-income families (the Survey of Economic Opportunity, SEO, sample) (PSID, 2024). The core sample comprised 60% of the original sample, while the SEO sample accounted for the remaining 40%. The original families and their split-offs have been followed continuously, with annual waves until 1996 and biannual waves since 1997. In this paper, we use data from 1999 to 2021. We begin our sample in 1999 for two reasons. First, the PSID underwent a redesign in the late 1990s, introducing a new consumption module that, since 1999, has collected 70% of consumption expenditures, covering categories such as food, housing, transportation, education, and child care. Second, the wealth module has been included ineverywavesince1999,providingdetailedinformationonassetholdings. We exclude households with top-coded data in the following variables: non-housing wealth, mortgage, home value, rental payment, health insurance, any component of our consumption measure, and any component of our income definition. Following Kaplan, Violante, and Weidner (2014), we exclude households with missing information on race, education, or state of residence. We also exclude households with missing or faulty information regarding homeownership status. Following Andrés et al. (2022), we remove households with contradictory information on homeownership—specifically, those reporting not owning a house while also reporting positive net equity. Additionally, we exclude households with after-tax income and/or annual consumption below $2,000, as in Aguiar, Bils, and Boar (2024). Similar to Kaplan, Violante, and Weidner (2014), we excludehouseholdswhoseincomeincreasesbymorethan500percentordecreasesbymore than 100 percent. We also remove households with consumption below $5 and those reporting zero energy consumption. Finally, we exclude households with extreme average propensities to consume (APCs), keeping only those with APCs less than or equal to 2. Afterapplyingtheserestrictions,ourpooledsamplecontains58,303observations. 4

In our analysis, in addition to household-level characteristics, we use four key variables: income,consumption,andtwodefinitionsofenergyconsumption. SimilartoFisheretal. (2019), we express all monetary values in constant 2019 dollars. Following Aguiar, Bils, and Boar (2024), we apply family longitudinal weights. Our definition of income corresponds to family disposable income and includes salaries and other compensation, as well as private and government transfers.6 From this total, we subtract rent payments, property taxes, mortgage interest, and home insurance. We use the NBER TAXSIM (version 35) calculator to compute after-tax income.7 Similarly to Aguiar, Bils, and Boar (2024), we define consumption expenditures as food at home and food away from home, utilities, gasoline, public transportation, childcare, health expenditures, education, vehicles spending for purchases, repairs, insurance and parking, and spending on shelter, whichincludesrentalpaymentsforrentersand,forhomeowners,6%oftherespondent’s valuationoftheirhome. Weusetwodefinitionsforenergyconsumption: (i)residentialenergyconsumption,which includes expenditures in electricity, gas and other fuel for home use;8 and (ii) overall energy consumption, which encompasses residential energy consumption and expenditure in gasoline for transport. In the U.S., households devote 6% of their income, on average, toresidentialenergyexpendituresandabout5%toenergyfortransportexpenditures. In addition,accordingtotheAmericanCommunitySurveyandtheU.S.BureauoftheCensus, in 2020, 85% of U.S. workers commute by private vehicle with 75% of all commuters drivingaloneintheirprivatecar. Only5%ofworkersusepublictransporttocommuteto work and 3% walked to work.9 Given the prevalence of private motor vehicles in daily commutes in addition to the geographical and urban/rural planning characteristics of the U.S., we argue that expenditure on gas for transport is a basic energy expenditure 6Thus, family income accounts for, for both the household head and other adults in the household: salaries; dividends; rentpaymentsreceived; workers’compensation; trustfundincome; financialsupport fromrelativesandnon-relatives;childsupportreceived;alimonyreceived;SupplementalSecurityIncome (SSI);TemporaryAssistanceforNeedyFamilies(TANF)andotherwelfare;pensions/annuities;lumpsum payments(e.g.,inheritances,itemizeddeductions);andfinancialsupportgiventoothers. 7The NBER TAXSIM calculator and background information are available at https://www.nber.org/research/data/taxsim. 8ManyEuropean-basedstudiesonenergypovertyincludeonlyutilities(electricityandgasforheating and cooking) in the definition of energy consumption given that the first definition of fuel poverty dates backtoBoardman(1991)andincludedspaceheating,waterheating,lights,appliancesandcookinginthe definitionofspendingonenergyservices. 9The data comes from U.S. Bureau of the Census, Journey to Work: 2000, Tables 1 and 2, 1990-2000, March2004(www.census.gov/population/www/socdemo/journey.html).andtheU.S.BureauoftheCensus, 2015-2019 American Community Survey Five-Year Estimates, “Explore Census Data,” Beta version. Data also available at U.S. Department of Energy (DOE), Oak Ridge National Lab (2022) Transportation EnergyDataBookEdition40. 5

for American households. Moreover, given that the size of the share of income devoted to energy for transport is as large as that for residential energy, we argue that these energy expenditures should be taken into account when talking about households’ energy consumption. In the U.S., other surveys provide household-level information on energy consumption, such as the U.S. Energy Information Administration’s Residential Energy Consumption Survey (RECS) and the U.S. Census Bureau’s Household Pulse Survey. However, these surveys rely on cross-sectional research designs, making them less suitable for tracking householdsovertimeand,therefore,assessingthepersistenceofenergyinsecurity. Additionally, household income data in these surveys is less granular than in the PSID. While the PSID records actual reported income, RECS and the Household Pulse Survey only categorizeincomeintobins. 3 ENERGY POVERTY, INSECURITY, AND VULNERABILITY: DEFINING ENERGY-BURDENED HOUSEHOLDS In the literature, energy poverty is often referred to as fuel poverty, energy insecurity, or energy vulnerability, despite these terms having slightly different meanings or connotations. Moreover, while energy poverty is generally defined as a household’s inability to meet its energy needs, its specific meaning varies depending on a country’s level of development. In developing economies, energy poverty refers to a lack of access to basic energyservicesnecessaryforfundamentalneedssuchascookingorlighting. Incontrast, indevelopedeconomies,itreferstotheunaffordabilityofenergyservicesforheatingand cooling, in addition to lighting and cooking. In Section 3.1, we review definitions of energy poverty, insecurity, and vulnerability in developed economies and propose the use of the term energy-burdened households. Then, in Section 3.2, we classify households as energyburdenedornotenergy-burdenedusingavarietyofindicatorsfromtheliterature. Weproposeasourbaselineindicatoratwice-the-median(2M)income-basedmeasurethat incorporates a broader definition of energy expenditures, including both residential energy costs and gas for transport. Using this baseline indicator, we classify 20% of U.S. householdsinourdatasampleasenergy-burdened. 6

3.1 DEFINITIONS For developed economies, the concept of energy poverty was put forward by Isherwood and Hancock (1979) in the U.K. following the 1970s energy crisis. They defined "householdswithhighfuelexpenditureasthosespendingmorethantwicethemedian(i.e. 12%) onfuel,light,andpower". ItwasBoardman(1991)whointroducedthefirstformaldefinitionofenergypoverty: "ahomewouldbeenergypoorifitsexpenditureinenergyservices exceeded10%ofitstotalincome",whichwasusedbytheEnglishHousingConditionSurvey (EHCS) to measure "affordable warmth" in the 1990s and to define fuel poverty, and henceenergypoverty,intheUKFuelPovertyStrategyof2001. Indeed,the10%threshold isubiquitousintheempiricalresearchasthemostcommon,ifnotpreferred,measurefor energypoverty. In the U.S., however, the term most commonly used is energy insecurity. The U.S. Energy Information Administration (EIA) defines energy insecurity as the inability to adequately meet household energy needs, where household energy needs refer to domestic or residential energy services (Hernandez, 2016). The EIA provides information on energy insecurity in the Residential Energy Consumption Survey (RECS). In the RECS survey, energy insecurity is related to five energy insecurity issues: reducing or forgoing food or medicine to pay energy costs; leaving home at unhealthy temperature; receiving disconnect or delivery stop notice; unable to use heating equipment; and unable to use air-conditioning equipment. Therefore, energy insecurity is a mix of household energy expenditures, the physical conditions of housing units, and energy-related behaviors.10 The U.S. Census Bureau’s Household Pulse Survey definition of energy insecurity includes three conditions: (1) a difficulty paying energy bills, (2) reduced or forewent basic necessities like food and medicine to pay an energy bill, or (3) kept home at an unsafe temperaturebecauseofenergycostconcerns. Inthe2020RECSsurvey(thelastoneavailable),27%ofhouseholdsreportedfacingsometypeofenergyinsecurityandaround20% of respondents gave up basic necessities to pay their energy bill; while about 30% of respondentsdidsointhe2023U.S.CensusHouseholdPulseSurvey. The term energy vulnerability lacks a clear definition. For example, Legendre and Ricci (2015) define fuel- or energy-vulnerable households as those for whom domestic energy expenditures are the primary factor driving them into poverty—that is, households pushed into poverty due to their domestic energy costs. Thus, energy vulnerability re- 10Steele and Bergstrom (2021) estimate and compare alternative empirical approaches to generate an energy insecurity index using data from the 2015 Residential Energy Consumption Survey (RECS). They concludethat,in2015,between9and22%ofU.S.householdswereenergyinsecure. 7

lates to exposure to energy shocks. However, as highlighted by Middlemiss and Gillard (2015), energy vulnerability is also linked to a household’s inability to access the energy servicesnecessaryforanadequatestandardofliving. Specifically,theyarguethatenergy vulnerability encompasses both the likelihood of experiencing energy poverty and the capacitytoadapttochangesinenergypoverty. Despite the distinctions described above, the terms energy poverty, energy insecurity, and energy vulnerability are often used interchangeably in the literature. In this paper, we propose using the term energy-burdened households and define a household as energy burdenedifitsenergyexpendituresasashareofdisposableincome—itsenergyburden— exceedtwicethemedianshareinthesample. 3.2 INDICATORS In the literature, energy poverty is measured using two main approaches: income-based indicators and expenditure-based indicators. Income-based indicators assess energy expenditures as a share of disposable income—commonly referred to as energy burden— while expenditure-based indicators measure energy expenditures as a share of total consumption expenses, capturing relative energy consumption. Among income-based indicators, the most commonly used are the twice-the-median (2M) indicator, the Minimum Income Standard (MIS) indicator, and the Low Income High Cost (LIHC) indicator. In contrast, expenditure-based indicators primarily rely on variations of the 2M approach. These indicators are typically defined in the literature using residential energy expenditures as the measure of energy consumption. However, as previously noted, the share of disposable income spent on transport-related energy is substantial. Therefore, energy consumption should account not only for utility expenditures but also for gas for transport. Accordingly,asshowninTable1,wecomputetheshareofenergy-burdenedhouseholds using both residential energy expenditures alone and total energy consumption acrossallindicatorsunderanalysis. Under 2M income-based indicators, a household is energy burdened if the percentage of disposable income spent on energy consumption to maintain energy services exceeds a given threshold. Therefore, these indicators provide a threshold to classify households as energy burdened. Since Boardman (1991), most of the literature uses a 10% threshold. Indeed,the10%indicatorwastheofficialenergypovertyindicatorintheUKfrom2001to 2013. Wearguethat,while,inorigin,the10%thresholdcoincidedwithtwice-the-median 8

TABLE 1: SHARE OF ENERGY-BURDENED HOUSEHOLDS (IN %) Residential OverallEnergy Income-basedindicators 2Mthreshold 23 20 10%threshold(residential) 14 – 20%threshold(overall) – 10 MIS 10 11 LIHC1 29 30 LIHC2 11 10 Expenditure-basedindicators 2Mthreshold 19 13 10%threshold(within-home) 21 – 20%threshold(overall) – 15 Note: Fortheincome-basedapproach,the2Mthresholdforresidentialenergyburdenis7.6%,whilethe2M thresholdforoverallenergyburdenis15.5%. Theshareofhouseholdsthatfallbelowpovertybecauseof energyexpensesis2%forresidentialenergyconsumptionand3%foroverallenergyconsumption. Forthe expenditure-based approach, the 2M threshold for residential energy relative consumption is 11%, while the2Mthresholdforoverallenergyrelativeconsumptionis23%. energy burden in the data,11 differences across countries and over time make the 10% inadequate to measure the extent of energy poverty in an economy. We propose to come back to the origins and compute the threshold associated with the 2M indicator using PSIDdatafrom1999to2021. Using pooled data, the 2M threshold for residential energy consumption is 7.6%, while for overall energy consumption is 15.5%. Thus, a 10% approach would underestimate theshareofenergy-burdenedhouseholdswhenusingonlyresidentialenergyandwould overestimate the share when considering overall energy consumption. We suggest that when considering the 10% approach with overall energy consumption, we should use a 20%thresholdinsteadofa10%threshold. AsshowninTable1,while14%ofhouseholds are energy-burdened under a 10% threshold using only residential energy expenditures, 23%ofhouseholdsareenergy-burdenedusingthe2Mindicator. Whenconsideringoverall energy expenditures, only 10% of households are energy burdened under the 20% 11Boardman(1991)chosethisvalueforthethresholdbecauseitwastwicethemedianenergyexpenditure intheUKin1988. 9

threshold, while 20% are classified as being energy burdened under the 2M threshold rule. Figure 2 illustrates the overlap in our sample between the twice-the-median (2M) indicator based on residential energy consumption and the 2M indicator based on total energy consumption. Ofthe13,590householdsclassifiedasenergyburdenedundertheresidentialenergy-based2Mindicator,66%arealsoclassifiedasenergyburdenedunderthetotal energy-based 2M indicator. Conversely, of the 11,889 households identified as energy burdened under the total energy-based 2M indicator, 76% are also classified as energy burdenedundertheresidentialenergy-based2Mindicator. FIGURE 2: ENERGY BURDEN FOR RESIDENTIAL AND OVERALL ENERGY Residential 2M Overall Energy 2M NOTE: Author’scalculationsfromthePSID.Figurereportscounts. The Minimum Income Standard (MIS) approach proposed by Moore (2012) defines a household as energy poor if she does not have enough income to pay for energy costs, after covering housing and other needs. This indicator identifies households that would be above the poverty threshold but fall below it because of their energy expenditure. We computetheMISindicatorusingthefederalpovertythresholdspublishedbytheU.S.Department of Health and Human Services. As shown in Table 1, 11% of households in our sample are energy burdened using the MIS approach with overall energy consumption and10%withresidentialconsumption. UndertheLowIncomeHighCost(LIHC)approachputforwardbyHills(2012),ahouseholdisenergypoorifherincomeisbelowcertainpovertythresholdandherenergycosts are higher than an energy expenditure threshold. For the LIHC 1 measure, the income threshold is 60% of the median equivalent income net of housing and energy costs and theenergythresholdasthemedianequivalentenergyexpenditure.12 Theshareofenergy- 12Equivalent income is calculated as income over the square root of family size. When considering 10

burdened households in the U.S. is about 30% for both residential energy expenditures and overall energy consumption.13 The LIHC 1 indicator has been the official metric for energypovertyintheUKsince2013and,asindicatedbyBurlinson,Giulietti,andBattisti (2018), it is gaining traction in European-based studies. The LIHC 2 measure is based on theindicatorproposedbyRomero,Linares,andLópez(2018). UndertheLIHC2measure, a household is energy poor if her energy expenditure exceeds the median energy expenditure and her income net of energy expenditures exceeds 60% of the median household income (not equivalent income) net of the mean energy expenditure. In this case, about 10% of U.S. households are energy burdened both when considering residential energy consumptionandwhenusingtotalenergyconsumption. All the measures above look at energy consumption vis-a-vis income. In a study for Canada,Greenetal.(2016)suggest,“toidentifyhouseholdsasbeinginenergypovertyif energy accounts for at least 10% of their total expenditures". Therefore, we also consider 2MindicatorsusingrelativeenergyconsumptionasshowninthebottompanelofTable1. Underastandard2Mapproach,13%ofhouseholdareenergyburdenedwhenusingtotal energy consumption while 19% of households are classified as energy burdened when using residential energy consumption. Following the 10% threshold suggested by Green et al. (2016), the share of energy-burdened households increases to 21% when using residential energy consumption. The share of energy-burdened households increases to 15% whenconsideringtotalenergyconsumptionanda20%threshold. For the remainder of the paper, we use the 2M indicator based on total energy consumption as our baseline measure, as it is our preferred metric for assessing household energy burdeninthePSID. 4 WHO ARE THE ENERGY-BURDENED HOUSEHOLDS? In this section, we first study the defining characteristics of the EB households identified using our baseline indicator, that is, the 2M indicator for total energy consumption. We then assess whether EB households are just traditional hand-to-mouth consumers. Finally, we study the role of these characteristics in increasing the probability of being only residential energy expenditures, the income threshold is 60% of the median equivalent income net ofhousingandresidentialenergycostsandtheresidentialenergythresholdasthemedianequivalentenergyexpenditure. Whenconsideringtotalenergyexpenditures,theincomethresholdis60%ofthemedian equivalentincomenetofhousingandoverallenergycostsandtheoverallenergythresholdasthemedian equivalentenergyexpenditure. 13Using the OECD weights to compute equivalent incomes, the share of energy-burdened households usingtotalenergyconsumptionis31%. 11

energy-burdenedusingamultinomiallogitapproach. 4.1 HOUSEHOLD CHARACTERISTICS Table 2 reports the share of EB households in each decile of the disposable income distribution. There is a negative relationship between the share of EB households and income. For example, while 67% of households in the first decile of the income distribution are energy burdened, less than 1% of households in the top decile are energy burdened. Additionally,wereportthedistributionofEBhouseholdsacrossincomequintilesinTable3. We conclude that 84% of EB households are concentrated in the first two quintiles of the incomedistribution. TABLE 2: SHARE OF ENERGY-BURDENED HOUSEHOLDS IN EACH DECILE (IN%) Decile 1 2 3 4 5 6 7 8 9 10 Share(%) 67 46 32 22 16 11 6 4 2 0.5 TABLE 3: ENERGY-BURDENED HOUSEHOLDS ACROSS INCOME QUINTILES (IN%) Quintile 1 2 3 4 5 Share(%) 55 26 13 5 1 In Table 4, we provide the averages for income, consumption, and energy consumption acrossthetwodifferenttypesofhouseholds. Notably,EBhouseholdsgenerateanincome that amounts to just 37% of non-EB households’ income but their total consumption is 76% of non-EB households’ consumption. However, energy consumption by EB householdsis49%higherthatofnon-EBhouseholds,whichisdrivenbybothhigherconsumptiononresidentialenergyaswellasgasfortransport. In Table 5 we report the average propensity to consume (APC), defined as consumptionto-income ratios, for both types of households. The APC for energy is also known in the literature as energy burden. Both Table 4 and Table 5 show how aggregate values mask the consumption behavior of EB households. For example, as shown in Table 5, the APC for overall consumption for all households is 0.84, while the same APC for EB households is 1.51. In addition, the average APC for EB households across all measures of energy consumption is nearly triple the amount of all households. The average EB householdspends25%ofherincomeontotalenergyconsumption,15%ofherincomein 12

TABLE 4: SAMPLE STATISTICS: MEANS AllHouseholds EnergyBurdened Non-EnergyBurdened Income $66,843 $28,044 $74,916 Consumption $43,571 $34,405 $45,478 Energyconsumption $4,430 $6,097 $4,083 Residentialconsumption $2,278 $3,040 $2,119 Transportconsumption $2,152 $3,057 $1,964 TABLE 5: AVERAGE PROPENSITY TO CONSUME AllHouseholds EnergyBurdened Non-EnergyBurdened Consumption 0.84 1.51 0.70 Energyconsumption 0.10 0.25 0.07 Residentialconsumption 0.05 0.15 0.04 Transportconsumption 0.04 0.11 0.03 utilities, and 11% in gas for transport. However, non-EB households spend only about 7%oftheirincomeinenergyexpenditures. WefurtherourcharacterizationofthedifferencesbetweenEBandnon-EBhouseholdsby lookingattheirdemographicandeconomiccharacteristics,aspresentedinTable6. Relativetonon-EBhouseholds,amongEBhouseholds,thereisahighershareofunemployed (10% compared to 5%), renters (46% compared to 41%), dwellings in mobile homes (10% compared to 4%) and government housing (8% compared to 4%). Additionally, a larger share of EB households receive heating subsidies (12% compared to 3%) and participate inwelfareprogramssuchasfreeschoollunchorSNAP(35%comparedto13%). Lastly,regardingdemographicandgeographiccharacteristics,EBhouseholdshaveahighershare of unmarried head of households (60% compared to 40%), higher share of Black households (50% compared to 29%), and higher share of households located in the South (53% comparedto40%). Weusethisbasiccharacterizationofthedemographic,socioeconomic, and geographic characteristics of EB and non-EB households as motivation to estimate whetherthesefactorscontributetobeingenergyburdenedintheSection4.3. 13

TABLE 6: HOUSEHOLD CHARACTERISTICS (SHARES IN %) EnergyBurdened Non-EnergyBurdened Unemployed 10 5 Debtholders Mortgagedebt 32 41 Non-mortgagedebt 48 57 Renters 46 41 Dwellings Apartments 18 22 Mobilehomes 10 4 Governmenthousing 8 4 Typeofheating Gas 51 54 Electricity 39 37 Oil 4.80 4.80 Wood 1.30 1.10 Coal 0.10 0.10 Solar 0.00 0.20 Propane 2.40 1.80 Kerosene 0.60 0.00 Other 5.0 4.70 Receivegov’theatingsubsidy 12 3 Welfareprograms(freeschoollunchorSNAP) 35 13 Married 40 60 Race White 44 65 Black 50 29 Asian 1 1 Other 5 5 Region Northeast 11 14 Northcentral 24 26 South 53 40 West 12 20 4.2 ARE EB HOUSEHOLDS JUST HAND-TO-MOUTH HOUSEHOLDS? SinceEBhouseholdsareprimarilyconcentratedinthelowestincomequintiles,asshown inTable3,onemightquestionwhethertheyaresimplyhand-to-mouth(HTM)consumers. Toaddressthisissue,weidentifythesharesofhouseholdswhoarenotHtM(non-HTM), poorHTMhouseholds(P-HTM),andwealthyHTMhouseholds(W-HTM)inourdataset following Aguiar, Bils, and Boar (2024). In our setup, P-HTM households are defined as 14

households with net worth smaller than two months labor earnings as in Zeldes (1989) and W-HTM as those that are not P-HTM and have negative liquid wealth with absolute valueexceeding16.5%oftheirannualincome.14 TABLE 7: HAND-TO-MOUTH STATUS OF EB HOUSEHOLDS EnergyBurdened Non-EnergyBurdened P-HTM 32% 27% W-HTM 37% 24% Not-HtM 31% 49% AsshowninTable7,69%ofEBhouseholdsareclassifiedasHTM,with32%beingP-HTM and 37% W-HTM. This implies that a substantial share—31%—of EB households are not HTM. Among non-EB households, we also observe significant shares of P-HTM (27%) and W-HTM (24%), while 49% are classified as not-HTM. Table 8 presents the distribution of EB and non-EB households across the P-HTM, W-HTM, and not-HTM categories. Our findings indicate that while HTM households are more likely to be EB than non- HTM households, EB households exist across all HTM classifications in non-negligible proportions. Therefore, we argue that EB households represent a distinct and meaningful category within the PSID, warranting further empirical and theoretical investigation. TABLE 8: EB STATUS OF HTM HOUSEHOLDS P-HTM W-HTM Not-HTM Energy-Burdened 23% 28% 14% NotEnergy-Burdened 77% 72% 86% 4.3 REGRESSION ANALYSIS For our regression analysis, we are interested in how energy burden co-varies across different household characteristics. Following an empirical approach similar to Best and 14This criteria for liquid wealth was put forward by Kaplan, Violante, and Weidner (2014). Conversely toKaplan,Violante,andWeidner(2014),Aguiar,Bils,andBoar(2024)imposethatP-HTMandW-HTMare mutuallyexclusivehouseholdtypes. 15

Sinha(2021),Mohr(2018),orRomero,Linares,andLópez(2018),weestimatethefollowinglogitmodelfrom1999-2021usingthefullsampleofhouseholds: (cid:16) p (cid:17) ln = x ′ β+ε . (1) 1− p i i In equation 1, the outcome variable is the log of the odds of being energy burdened accordingtoour2Mapproachontotalenergyconsumption,and pistheprobabilityofbeing energy burdened. The vector of explanatory variables, x, includes socioeconomic status, home ownership status, type of dwelling, type of heating used, general demographics, andgeographiclocation.15 The results of our regression analysis are shown in Table 9, and are reported as average marginaleffects. Reportingthe marginaleffectsallow ustointerpretthe resultsasdifferencesinprobabilities,whichismoreinformativethananoddsratio. Eachcolumnaddsa set of covariates. We will discuss the results from the full specification in column 5. First, theestimatesprovideevidencethattheprobabilityofbeingenergyburdenedisinversely relatedtohouseholdincome. Inparticular,ourresultsshowthatbeinginthebottomtwo income quintiles is associated with a higher probability of energy burden relative to incomequintilesthree,four,andfive,whichsuggeststhatincomedirectlyalleviatesenergy burden. Whenexaminingownershipstatus,wefindthathomeownersexhibitahigherprobability of being energy burdened compared to renters, and that house size is negatively related to energy burden. Focusing on specific dwelling types, we find that, relative to living in an apartment, one-family houses and mobile homes face a particularly high probability of being energy burdened. In fact, households in mobile homes are 2.2 to 2.4 times more likely to experience energy burden, relative to one- and two-family style houses, respectively. Hence, the nature and size of housing are critical factors in determining energy burden. In terms of household characteristics, being a race other than White is associated with higher probability of being energy burdened, being the highest for Black households. Corroborating our findings is Wang et al. (2021), who conclude that Black households are more vulnerable than White and Asian households in the 2015 wave of RECS. In our case,wealsoconcludethatHispanichouseholdshaveasignificantlyhigherprobabilityof 15Wealso estimatea probitmodel anda linearprobabilitymodel (LPM).LPMs arewidely usedin empiricalwork,forexample,seeTito(2024),andChenetal.(2017). TheLPMservesasarobustnesscheckby includingyearandstatebyyearfixedeffects. WereporttheresultsinAppendixTableA1andconcludeour resultsarerobust. 16

TABLE 9: HOW ENERGY BURDEN COVARIES ACROSS DIFFERENT HOUSEHOLD PROFILES Variables (1) (2) (3) (4) (5) Socioeconomic Bottomtwoincomequintiles 0.128∗∗∗ 0.126∗∗∗ 0.111∗∗∗ 0.098∗∗∗ 0.097∗∗∗ Ownershipstatus Homeowner −0.029∗∗∗ 0.005 0.016∗∗∗ 0.018∗∗∗ Housesize6+ −0.019∗∗∗ −0.015∗∗∗ −0.010∗∗∗ −0.013∗∗∗ Typeofdwelling Onefamilyhouse 0.087∗∗∗ 0.093∗∗∗ 0.088∗∗∗ 0.089∗∗∗ Twofamilyhouse 0.079∗∗∗ 0.075∗∗∗ 0.065∗∗∗ 0.065∗∗∗ Mobilehome 0.191∗∗∗ 0.185∗∗∗ 0.142∗∗∗ 0.143∗∗∗ Rowhome 0.005 0.003 0.008 0.006 Heating Gas −0.001 0.005 0.011∗∗ 0.010∗∗ Oil 0.021∗∗∗ 0.056∗∗∗ 0.059∗∗∗ 0.057∗∗∗ Other(propane,wood,kerosene) 0.037∗∗∗ 0.079∗∗∗ 0.071∗∗∗ 0.066∗∗∗ Householdcharacteristics Race Black 0.115∗∗∗ 0.073∗∗∗ 0.074∗∗∗ Asian 0.004 0.024 0.029∗ Other 0.035∗∗∗ 0.031∗∗∗ 0.044∗∗∗ Hispanic 0.042∗∗∗ 0.047∗∗∗ 0.038∗∗∗ Married −0.033∗∗∗ −0.044∗∗∗ −0.045∗∗∗ Female 0.027∗∗∗ 0.015∗∗ 0.015∗∗ Kids 0.025∗∗∗ 0.018∗∗∗ 0.019∗∗∗ Head65+ 0.014∗∗∗ −0.050∗∗∗ −0.041∗∗∗ Othersocioeconomic Employed −0.072∗∗∗ −0.065∗∗∗ Postsecondaryeducation −0.068∗∗∗ −0.069∗∗∗ Subsidizedhousing 0.029∗∗∗ 0.029∗∗∗ Heatingsubsidy 0.133∗∗∗ 0.123∗∗∗ Behindonmortgage 0.139∗∗∗ 0.109∗∗∗ Geographiclocation Northeast −0.039∗∗∗ −0.038∗∗∗ Northcentral −0.029∗∗∗ −0.028∗∗∗ West −0.067∗∗∗ −0.067∗∗∗ Yeardummies? N N N N Y N 57,248 57,248 57,248 57,248 57,248 Pseudo R2 0.084 0.098 0.124 0.155 0.172 NOTE:Asterisksindicatethelevelofsignificanceoftheparameters,*p <.10;**p <.05;and***p <.01. being energy burdened than non-Hispanic households. Additionally, we find that head of households over the age of sixty-five have a lower probability of being energy bur- 17

dened,whileFemaleheadofhouseholdsandhavingchildrenisassociatedwithahigher probabilityofenergyburden. Wefindthattheheadofhouseholdbeingmarried,employed,andhavingpost-secondary educationreducetheprobabilityofbeingenergyburdened. Interestingly,householdsthat receive government-subsidized housing and heating subsidies exhibit a higher probability of energy burden. This relationship is also found by Best and Sinha (2021), who use RECS data and focus on residential energy consumption to conclude that the reception of assistance is positively correlated with fuel poverty. Lastly, geographic location plays a role as well; compared to the South, all other regions face lower probabilities of being energyburdened. Insummary,householdswithlowincomesandlowlevelsofeducation,particularlythose livinginsingle-familyormobilehomesintheSouth,andthosewithchildren,arethemost vulnerabletobeingenergyburdened. AsshowninAppendixTableA1ourresultsarerobust to alternative models, such as LPM and Probit. Moreover, as shown in Appendix Table A2, our logit results are robust when we re-estimate equation 1 on a sample excluding households identified as hand-to-mouth by either measure. This confirms that our findings are not confounded by hand-to-mouth status, indicating that our results are explainedsolelybyEBstatus. 5 ENERGY CONSUMPTION: EMPIRICAL FACTS While there is a vast literature documenting empirical facts on consumption and income using U.S. micro-data, the study of empirical regularities of energy consumption is in its infancy. In this section, we put forward four empirical facts for energy consumption and energy-burdened status in the PSID: (i) households tend to remain energy burdened over time; (ii) EB households have significantly larger marginal propensities to consume energy; (iii) EB households have lower energy consumption growth than non-EB households despite having higher income growth; and (iv) EB households have more volatile energyconsumptionandincomethannon-EBhouseholds. FACT 1: EB/NON-EB STATUS IS PERSISTENT To study the persistence of the EB status, we compute two-year transition rates between EB status in the PSID using the pooled sample and report them in Table 10. Transition probabilitiesarecomputedconsideringhouseholdsthatwereincludedintwoconsecutive waves,whichmeansthat,forexample,somehouseholdsinour1999samplearedropped 18

when computing the transition probabilities because they are not included in the 2001 TABLE 10: TRANSITION RATES EnergyBurdened Non-EnergyBurdened t+2 t+2 A.TotalEnergy EnergyBurdened 0.49 0.51 t Non-EnergyBurdened 0.12 0.88 t B.ResidentialEnergy EnergyBurdened 0.59 0.41 t Non-EnergyBurdened 0.11 0.89 t C.GasforTransport EnergyBurdened 0.48 0.52 t Non-EnergyBurdened 0.17 0.83 t NOTE: Weusea2MapproachtoclassifyhouseholdsasEBforeachtypeofenergyconsumption. wave. As shown in Panel A of Table 10, if a household is EB in year t using the 2M approach for total energy consumption, the probability of remaining EB in year t+2 is 49%. If a household is non-EB in year t, the probability of becoming energy burdened in year t+2 is only 12%, that is, the probability of remaining non-EB is 88%. Therefore, we arguethatEB/non-EBstatusispersistentacrosssurveywaves. InPanelBandPanelCof Table10,wereportthetransitionrateswhenEBstatusisdeterminedusinga2Mapproach forresidentialenergyconsumptionandgasfortransport,respectively. Theprobabilityof remainingEBwhenusingresidentialenergyconsumptionisalmost60%,whileitremains around 50% when using gas for transport. Notably, the persistence of being non-energy burdenedisaround85%irrespectiveofthedefinitionofenergyconsumption. Determinants of persistence in EB status: Next, we study which household characteristics are associated with persistent EB status by estimating a logit model similar to equation 1. Specifically, the dependent variable is the log of the odds of being energy burdened—by total energy, residential energy, or gas for transport—in two consecutive survey waves (e.g., 1999 and 2001).16 In particular, p is the probability of being EB in two consecutive periods and (1− p) is the probability of being EB in the first wave and 16In some cases, a household may appear in both categories. For example, if a household is energy burdenedin1999and2001butnotin2003,itwillbecountedas1forthe1999to2001periodandas0for the2001to2003period. 19

non-EBinthefollowingone. Thevectorofexplanatoryvariablesincludesasetofsocioeconomic,demographic,andgeographiccharacteristics. TABLE 11: CHARACTERISTICS OF PERSISTENT EB STATUS Variables TotalEnergy Residential Transport Socioeconomic Bottomtwoincomequintiles 0.034∗∗∗ 0.022∗∗∗ 0.032∗∗∗ Incomegrowth −0.005 −0.011 −0.011 Energyconsumptiondecline 0.041∗∗ 0.015 0.023 Ownershipstatus Homeowner 0.097∗∗∗ 0.085∗∗∗ 0.036∗∗∗ Householdcharacteristics Race Black 0.054∗∗∗ 0.096∗∗∗ −0.016 Asian 0.035 0.120∗ −0.088 Other 0.046 0.027 0.046∗ Hispanic −0.009 0.037 0.030 Married −0.065∗∗∗ −0.057∗∗∗ −0.035∗∗ Female 0.054∗∗∗ 0.081∗∗∗ −0.004 Kids −0.014 −0.033∗∗∗ 0.005 Head65+ −0.022 0.003 −0.064∗∗∗ Othersocioeconomic Employed −0.079∗∗∗ −0.104∗∗∗ −0.032 Postsecondaryeducation −0.094∗∗∗ −0.115∗∗∗ −0.061∗∗∗ Subsidizedhousing 0.019 0.034 0.014 Heatingsubsidy 0.075∗∗∗ 0.099∗∗∗ 0.012 Behindonmortgage −0.052 −0.053 0.022 Geographiclocation Northeast −0.024 0.001 −0.035∗ Northcentral −0.010 −0.026∗ −0.026∗ West −0.092∗∗∗ −0.115∗∗∗ −0.045∗∗∗ ✓ ✓ ✓ Yeardummies? N 6,610 7,479 8,621 Pseudo R2 0.060 0.086 0.041 NOTE: The dependent variable is equal to 1 if the household is energy burdened—bytotalenergy,residentialenergy,orgasfortransport—intwoconsecutiveperiods. Thedependentvariableisequaltozeroifthehouseholdtransitions from energy burden in period one to not-energy burdened in period two. Asterisks indicate the level of significance of the parameters, * p < .10; ** p <.05;and*** p <.01. Table 11 reports the results across three categories of EB status: total energy, residential energy, and transport energy. As before, the results are presented as average marginal effects, allowing us to interpret them as differences in probabilities. Across all specifications, households in the bottom two income quintiles are significantly more likely to 20

experience persistence in EB status; that is, the probability of remaining EB is inversely related to household income. Interestingly, income growth does not have a significant effectonEBstatuspersistence. Inotherwords,anincreaseinhouseholdincomebetween waves does not significantly alter the likelihood of remaining EB in the second wave. Similarly,declinesinresidentialenergyconsumptionorgasfortransportbetweenwaves do not significantly affect a household’s EB status persistence, as shown in the last two columns of Table 11. However, when classifying households as EB based on total energy consumption, a decline in total energy consumption is positively associated with persistentEBstatus. Homeownership increases the probability of EB persistence across all types of energy consumption,withastrongereffectforresidentialenergythanforgasfortransport. This pattern suggests that homeowners face challenges in adjusting their residential energy use, likely due to inefficient home infrastructure and higher maintenance costs, which contribute to persistent EB status. Demographically, relative to white households, Black households are more likely to experience persistence in total and residential EB status. Female-headed households also exhibit a higher probability of remaining EB for total andresidentialenergyconsumption,whereasmarriedhouseholds,householdswithchildren, and those with an elderly head (65+) are less likely to experience persistence in EB status. The strongest factors mitigating EB status persistence are employment and education. Being employed significantly reduces the probability of remaining EB in both total and residentialenergy,whilepostsecondaryeducationfurtherdecreasestheriskacrossallcategories. Finally,geographicdifferencesshowthat,relativetotheSouth,livingintheWest consistently reduces the probability of persistent EB status. These findings suggest that persistent energy burden status is shaped by a combination of economic, demographic, andgeographicfactors,withdisparitieslinkedtohomeownership,locationintheincome distribution,andrace. FACT 2: EB HOUSEHOLDS HAVE SIGNIFICANTLY LARGER MPC AND MPCES The marginal propensity to consume (MPC) out of transitory income shocks is defined as the fraction of a small, unanticipated, one-time increase in disposable income that a household spends within a given time period. In this paper, we introduce the concept of the marginal propensity to consume energy (MPCE), which we define similarly as the fraction of a small, unanticipated, one-time increase in disposable income that a household spends on energy goods and services. We exploit the panel dimension of the PSID 21

datatoestimateMPCsandMPCEsforEBandnon-EBhouseholdsinourdataset. Following the literature (see, for example, Kaplan, Violante, and Weidner, 2014, Auclert, 2019, and Commault, 2022), we use a semi-structural approach to estimate MPCs and MPCEs based on Blundell, Pistaferri, and Preston (2008). In the first step, we regress the logofthemeasureofconsumptionofinterest(allgoods,totalenergy,residential,ortransport)andthelogofdisposableincomeonobservablehouseholds’characteristicsandyear dummies. Following Commault (2022), these households’ characteristics include year of birth, family size, number of children, existence of outside dependent children, education, race, employment status, presence of an additional income recipient that is not the head of the household or spouse, and region. In these regressions, we also include interactiontermsbetweenyeardummiesandeducation,race,employmentstatus,andregion. We then compute the first-difference of the residuals of log consumption and log income ∆ ∆ denoted by c and y , respectively. Following Auclert (2019), we estimate the passit it through coefficient of log income on log consumption, ψ = cov(∆c it ,∆y i,t+2 ) . Specifically, i cov(∆y it ,∆y i,t+2 ) the Blundell, Pistaferri, and Preston (2008) estimator is implemented by an instrumental variable regression of ∆ c on ∆ y using ∆ y as an instrument.17 We then recover it it i,t+2 the estimates of the MPC by multiplying the estimated pass-through coefficient, ψ , by i the corresponding mean consumption-income ratio (i.e., the average propensity to consume). InTable12,wereporttheestimatedpass-throughcoefficientsoftransitoryincomeshocks to the various consumption measures for EB and non-EB households. All estimated pass-through coefficients are statistically significant, with those for EB households being larger than those for non-EB households. Consequently, the implied MPCs and MPCEs are substantially higher for EB households, indicating that they are more responsive to temporary, unexpected changes in income than non-EB households.18 Table A3 in the Appendix confirms that these results are robust to alternative classifications of EB and non-EB households. Recent contributions to the literature, such as Commault (2022) and Crawley (2020), have revisited the estimation procedure of Blundell, Pistaferri, and Preston(2008),demonstratingthattheMPCestimatesbasedonBlundell,Pistaferri,andPreston (2008) serve as a lower bound for MPCs. Therefore, we argue that our estimates for MPCs and MPCEs, reported in Table 12, may be at the lower end of the possible range. However, the magnitude of our MPC estimates aligns with recent estimates using the 17ThePSIDsurveyisbiannualalthoughbothincomeandconsumptionarereportedatannualfrequency. 18InAppendixFigureA.3,weplottheMPCforourtotalconsumptionmeasurebyyear. Duetothesmall samplesizeineachyear, theestimatesforEBhouseholdsarequitevolatile, rangingfrom0.16to0.49. As such,wefinditmoreappropriatetoestimateMPCsusingapooledsampleapproach. 22

TABLE 12: MARGINAL PROPENSITIES. 2M APPROACH TotalCons. TotalEnergy Residential Transport PanelA:Energy-BurdenedHouseholds ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ψ 0.161 0.324 0.130 0.313 i (0.027) (0.039) (0.042) (0.056) MPC 0.243 0.082 0.020 0.038 PanelB:Non-BurdenedHouseholds ∗∗∗ ∗∗∗ ∗∗ ∗∗∗ ψ 0.131 0.177 0.080 0.154 i (0.024) (0.031) (0.032) (0.044) MPC 0.092 0.012 0.003 0.005 NOTE: ψ i stands for the pass-through coefficient or short-run elasticity of consumption with respect to transitory income shocks and MPC refers to marginal propensity to consume. Standard errors are clustered at the household level and are reported in parenthesis,* p <.10;** p <.05;and*** p <.01. PSIDbyCho,Morley,andSingh(2024).19 FACT 3: EB HOUSEHOLDS HAVE LOWER ENERGY CONSUMPTION GROWTH THAN NON-EB HOUSEHOLDS DESPITE HAVING HIGHER INCOME GROWTH Table 13 reports the average growth rates of income, consumption, energy consumption, and its components in the sample. The entries in Table 13 are computed as follows: we calculate the growth rate between wave t and wave t+2 for households classified as EB (non-EB) in wave t, divide it by 2 to annualize the rate, and then take the average across EB (non-EB) households.20 The average income growth for EB households in the sample is substantially higher than for non-EB households. However, while the average growth rates of consumption, energy consumption, and its components for EB households are negative,theseratesarepositivefornon-EBhouseholds. 19Fisher et al. (2019) estimate MPCs using an Euler equation approach with PSID data from 1999-2013, findinganestimateof8%. 20Table A4 in Appendix B.1 reports the average growth rates for each wave, where, for example, 2001 23

TABLE 13: AVERAGE ANNUAL GROWTH RATE (IN %) CONDITIONAL ON EB STATUS IN THE FIRST WAVE OF THE GROWTH RATE Burdened Non-Burdened Income 11.33 -0.83 Consumption -1.41 1.23 Energyconsumption -9.47 2.53 Residentialconsumption -5.87 2.11 Transportconsumption -12.07 2.51 Note: Growth rates are defined as the log-difference between the variable of interest in wave t+2 and in wavet,dividedbytwotoannualizethegrowthrate. We further explore these differences in growth rates between EB and non-EB households usingconsumptionandincomegrowthregressions. Specifically,weexaminewhetherbeing energy burdened predicts consumption and income growth, thereby characterizing the role of EB status in these dynamics. To do so, we propose the following consumption (income) growth equation, which closely follows the specification introduced by Aguiar, Bils, and Boar (2024) in their study on the role of hand-to-mouth status in aggregate consumptiongrowth: ∆ lnxi = β EB +ϕ ′ D +θ ′ X +εi , (2) j,t+2 i j,t i t i j,t j,t+2 Here, i represents income, overall consumption, total energy consumption, residential energyconsumption,andgasfortransportenergyconsumption. Thegrowthrateofvariable x isdefinedasthelog-differencebetweenitslevelinwave t+2andwave t,divided bytwo toannualizethe growthrate. In equation2,EB equals1 ifhousehold j isenergy j,t burdened in wave t, D is a vector of year dummies, and X is a vector of household t j,t characteristics. The household characteristics include a quadratic term for age and dummies for changes in marital status and family size. We estimate the growth regressions bothwithandwithouthouseholdfixedeffects. Panel A in Table 14 reports the results for the growth regressions. For each variable, the first column reports the β coefficient in the regression without household fixed effects i and the second column reports the estimates for the regression in which we control for householdfixedeffects. Inourcase,regardlessofwhetherwecontrolornotforhousehold fixedeffects,thesignofthecoefficientsisthesameforeachvariablewhilethemagnitude is larger when we control for household fixed effects. Our regression results allow us to conclude that EB households have a significantly higher rate of income growth than non-EBhouseholds, referstotheannualgrowthratebetweenwave1999andwave2001. 24

sdlohesuoHdenedruBygrenErofsnoissergeRytilitaloVdnahtworG :41 ELBAT tropsnarT laitnediseR snoCygrenE noitpmusnoC emocnI )2( )1( )2( )1( )2( )1( )2( )1( )2( )1( rossergeR htworG.A ∗∗∗ 512.0−∗∗∗ 721.0−∗∗∗ 321.0−∗∗∗ 470.0−∗∗∗ 781.0−∗∗∗ 901.0−∗∗∗ 430.0−∗∗∗ 620.0− ∗∗∗ 132.0 ∗∗∗ 821.0 denedruB-ygrenE )410.0( )800.0( )110.0( )600.0( )900.0( )500.0( )800.0( )400.0( )010.0( )700.0( ✓ ✓ ✓ ✓ ✓ oN oN oN oN oN ?stceffEdexiF 22.0 80.0 91.0 30.0 52.0 01.0 81.0 20.0 72.0 60.0 2R ytilitaloV.B ∗∗∗ 010.0 ∗∗∗ 310.0 ∗∗∗ 400.0 ∗∗∗ 120.0 ∗∗∗ 110.0 ∗∗∗ 210.0 200.0 100.0− ∗∗∗ 420.0 ∗∗∗ 640.0 denedruB-ygrenE )700.0( )600.0( )600.0( )500.0( )500.0( )400.0( )400.0( )300.0( )600.0( )500.0( ✓ ✓ ✓ ✓ ✓ oN oN oN oN oN ?stceffEdexiF 34.0 10.0 94.0 20.0 24.0 10.0 14.0 20.0 54.0 30.0 2R ;50. < p **;01. < p *,sretemarapehtfoecnacfiingisfolevelehtetacidnisksiretsA .leveldlohesuohehttaderetsulcerasrorredradnatS :ETON .047,71sieziselpmasehT .10.< p ***dna 25

while EB households face a significantly lower rate of consumption growth than non-EB householdsforallcategoriesunderstudy. FACT 4: EB HOUSEHOLDS HAVE MORE VOLATILE ENERGY CONSUMPTION AND INCOME THAN NON-EB HOUSEHOLDS We also follow Aguiar, Bils, and Boar (2024) in specifying the regression for the volatility ofconsumptionandincomegrowthasfollows: (cid:12) (cid:16) (cid:17) (cid:12) (cid:12)∆ ln xi (cid:12) = γ EB +ω ′ D +ψ ′ X +εi , (3) (cid:12) j,t+2 (cid:12) i j,t i t i j,t j,t+2 vol (cid:12) (cid:16) (cid:17) (cid:12) where the volatility measure, (cid:12)∆ ln xi (cid:12), is defined as the absolute value of the (cid:12) j,t+2 (cid:12) vol growthrateofvariablexminusthegrowthpredictedbytheregressionsinequation2. The results of these volatility regressions are reported in Panel B of Table 14. When comparing the estimates with and without fixed effects, we find that the estimated coefficients are smaller when controlling for household fixed effects. EB households experience greater volatility in future income and energy consumption (and its components) growth than non-EB households. However, the coefficient for total consumption growth volatility is not significantly different from zero. Thus, we conclude that EB households do not face greater volatility in future total consumption growth compared to non-EB households. 6 TRENDS IN INEQUALITY: INCOME, CONSUMPTION, AND ENERGY CONSUMPTION The study of income and consumption inequality trends has been a central focus in the literature. While there is broad consensus on the rise in income inequality over the past several decades, the trajectory of consumption inequality remains debated. On one side, Aguiar and Bils (2015) and Attanasio, Hurst, and Pistaferri (2015) find that consumption inequality has closely tracked income inequality since the 1980s. On the other, Krueger and Perri (2006), Heathcote et al. (2023), and Meyer and Sullivan (2023) document a more moderate increase in consumption inequality relative to income dispersion since the 1960s. In this paper, we extend this discussion by not only documenting trends in income and consumption inequality in our dataset but also examining the evolution of inequality in energy consumption and its components. To begin, in Figure 3, we present theGinicoefficientandthevarianceofthelogofenergyconsumptionanditscomponents. 26

FIGURE 3: ENERGY INEQUALITY OVER TIME, 1999-2021 0.44 0.43 0.42 0.41 0.40 0.39 0.38 iniG tropsnarT rof saG 0.35 0.34 0.33 0.32 0.31 0.30 0.29 iniG laitnediseR dna latoT 0.95 0.90 0.85 0.80 0.75 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 2019 2021 Year Total Energy Residential Energy Gas for Transport A.Gini ecnairaV fo goL 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 2019 2021 Year Total Energy Residential Energy Gas for Transport B.LogofVariance AsshowninFigure3,whileboththeGiniandvariancefortransportenergyconsumption exhibit an upward trend, residential and total energy consumption inequality increased until 2009 before declining. Although the Gini coefficient and the variance of logs are widelyusedintheliterature,theyarenotwithoutcriticism. AsnotedbyAttanasio,Hurst, andPistaferri(2015)andHeathcoteetal.(2023),thesemeasurescapturedispersionacross the entire population, potentially masking divergent trends across different segments of the distribution. Additionally, self-reported survey data often suffer from measurement issues, particularly at the lower end of the income distribution and the upper end of the consumptiondistribution. To mitigate these concerns, Meyer and Sullivan (2023) advocate for using percentile ratios,whicharelesssensitivetomeasurementerrorsintheextremetailsofthedistribution whenanalyzinginequalitypatterns. Inparticular,weusethe90:10,50:10,and90:50ratios where the 90:10 ratio describes inequality between the top and the bottom of the distribution, the 50:10 ratio describes inequality between the middle and the bottom of the distribution and the 90:50 ratio describes inequality between the top and the middle of thedistribution. Table15presentsincome,consumption,andenergyconsumptioninequalityforallhouseholdsbetween1999and2021. Asshowninthefirstcolumn,incomedispersion,measured by the 90:10 ratio, was substantially higher in 1999 than consumption and energy consumption dispersion. While income inequality remained relatively stable between 1999 and2009,itrosesignificantlyaftertheGreatRecession,increasingbyapproximately24% between 1999 and 2021. In contrast, consumption inequality grew by about 14% over the same period, with most of the increase occurring between 1999 and 2009. Thus, our re- 27

TABLE 15: CHANGES IN INEQUALITY ACROSS ALL HOUSEHOLDS (%) PERCENTAGE CHANGES InitialLevel 1999-2009 2009-2021 1999-2021 90:10ratio Income 8.29 -0.56 24.38 23.68 Consumption 4.69 8.18 4.94 13.53 Energy 4.53 29.04 -10.81 15.09 Residential 4.00 26.92 -15.96 6.67 Transport 5.83 2.86 45.83 50.00 50:10ratio Income 3.08 0.29 13.72 14.06 Consumption 2.23 3.36 2.55 6.00 Energy 2.40 20.83 -13.64 4.35 Residential 2.20 13.29 -11.73 0.00 Transport 2.50 -4.00 25.00 20.00 90:50ratio Income 2.69 -0.85 9.37 8.44 Consumption 2.10 4.66 2.33 7.11 Energy 1.87 6.80 3.28 10.29 Residential 1.82 12.04 -4.79 6.67 Transport 2.33 7.14 16.67 25.00 sults indicate that consumption inequality rose less than income inequality during this period.21 The surge in inequality in the lower half of the income distribution, as reflected in the 50:10ratio,whilesmallerthanfortheoveralldistribution,suggeststhathouseholdsatthe bottom have lost ground relative to the median, particularly since 2009. Meanwhile, the gap between the median and the lower end of the consumption distribution has steadily widened at a rate of 3% per decade, though the overall increase remains less than half of the rise in income inequality. In the upper half of the income distribution (measured by the90:50ratio),incomeandconsumptioninequalityhavegrownatsimilarrates. Notably, theincreaseinincomeinequalityinthelowerhalfofthedistributionisnearly70%larger than in the upper half, whereas the rise in consumption inequality in the lower half is about20%smallerthanintheupperhalf. We now turn to the evolution of inequality in energy consumption and its components. 21Using income data from the Current Population Survey and consumption data from the Consumer Expenditure Interview, Meyer and Sullivan (2023) also find that the rise in income inequality, measured by the 90:10 ratio, outpaced the rise in consumption inequality. They report a 25% increase in income inequalitysincethe1960s,comparedtoa9.5%riseinconsumptioninequality. 28

Based on the 90:10 ratio reported in the first column of Table 15, we conclude that inequality in transport energy consumption exceeds that of residential and total energy consumptionbutremainsbelowthelevelofincomeinequality. TABLE 16: CHANGES IN INEQUALITY MEASURED BY 90:10 RATIO(%) PERCENTAGE CHANGES InitialLevel 1999-2009 2009-2021 1999-2021 EBhouseholds Income 6.34 9.01 31.35 43.18 Consumption 4.49 9.04 -1.83 7.05 Energy 4.09 1.64 11.23 13.06 Residential 4.50 -2.22 -11.36 -13.33 Transport 8.33 8.00 55.56 68.00 Non-EBhouseholds Income 6.44 -2.22 28.67 25.82 Consumption 4.55 9.97 5.07 15.55 Energy 4.55 17.97 -1.75 15.90 Residential 4.00 25.00 -15.07 6.16 Transport 5.00 33.33 12.50 50.00 Residential and total energy consumption inequality fluctuated over the sample period: between 1999 and 2009, inequality rose sharply by about 28%, only to decline by an average of 14% over the following decade. As a result, the overall increase in residential energy consumption inequality over the full sample is modest at 7%, while the rise in totalenergyconsumptioninequalitycloselymirrorsthatofoverallconsumption. The50:10 ratio suggests that since 2009, households in the lower part of the distribution have lost less ground relative to the median in terms of residential and total energy consumption. Moreover, the decline in residential energy consumption inequality since 2009 has been presentatbothendsofthedistribution. In contrast, the trajectory of transport energy consumption inequality is markedly different. While it remained relatively stable during the first decade of our sample, it surged by46%between1999and2021—nearlydoubletheincreaseinincomeinequalityoverthe same period. Overall, transport energy consumption inequality increased by 50%. Examining different segments of the income distribution, we find that the rise in transport energyinequalityissimilaratboththetopandbottom. Next, we conduct a similar analysis comparing EB and non-EB households. Figure A.1 in Appendix A, plots the distribution of income and consumption for EB and non-EB householdsfrom1999through2021. AsshowninAppendixAFigureA.1,theincomeand 29

consumptiondistributionsforEBhouseholdshasshifteddownwardandtotherightover time, whereas those for non-EB households have remained relatively stable. Likewise, Figure A.2 in Appendix A, illustrates that the distributions of total energy consumption anditscomponentshaveshiftedtotherightforEBhouseholds. TABLE 17: ENERGY-BURDENED HOUSEHOLDS: CHANGES IN INEQUALITY: 1999-2021(%) PERCENTAGE CHANGES InitialLevel 1999-2009 2009-2021 1999-2021 90:10ratio Income 6.34 9.01 31.35 43.18 Consumption 4.49 9.04 -1.83 7.05 Energy 4.09 1.64 11.23 13.06 Residential 4.50 -2.22 -11.36 -13.33 Transport 8.33 8.00 55.56 68.00 50:10ratio Income 2.64 8.38 18.81 28.77 Consumption 2.11 4.67 14.74 20.11 Energy 2.18 -7.13 3.64 -3.75 Residential 2.18 1.15 -9.09 -8.05 Transport 3.33 3.20 16.28 20.00 90:50ratio Income 2.40 0.58 10.55 11.19 Consumption 2.13 4.17 -14.44 -10.87 Energy 1.88 9.44 7.33 17.46 Residential 2.07 -3.33 -2.50 -5.75 Transport 2.50 4.65 33.78 40.00 Table 16 reports 90:10 ratios for EB and non-EB households. As in the overall sample, income inequality has risen more than overall consumption and total energy consumption inequality for both EB and non-EB households. However, the increase in income inequality is 65% larger for EB households than for non-EB households (43% vs. 26%), while the rise in consumption inequality for EB households is about half that of non-EB households (7% vs. 16%). Although total energy consumption inequality has risen similarly for both groups, the timing differs: for non-EB households, the increase occurred primarilybetween1999and2009,whereasforEBhouseholds,ittookplacebetween2009 and2021. Residential energy consumption inequality has declined for EB households throughout the whole the sample period, falling by 13% from 1999 to 2021. In contrast, for non- EB households, it increased by 25% in the first decade before declining by 15% in the latter part of the sample. Consistent with our previous findings, inequality in gas for 30

transport consumption has steadily increased for both groups. Over the entire sample period,gasfortransportconsumptioninequalityroseby68%forEBhouseholdsand50% fornon-EBhouseholds,significantlyoutpacingtheincreaseinincomeinequalityforboth groups. Lastly, in Table 17, we examine the evolution of inequality among EB households by reporting the 50:10 and 90:50 ratios. First, income inequality, measured by the 90:10 ratio, increased by 43% between 1999 and 2021, with the largest increase occurring between 2009and2021(31.35%). Notably, the rise in income inequality was more modest in the earlier period (1999-2009: 9.01%), highlighting an acceleration in the past decade. Comparing the 50:10 and 90:50 ratios, we conclude that income inequality mostly increased in the bottom part of the EB distribution, as the 50:10 ratio grew by 28.77%, compared to an 11.19% rise in the 90:50 ratio. This suggests that income inequality widened more between lower-and middleincome EB households than between middle- and higher-income EB households. Consumption inequality, however, exhibited a different pattern. The overall increase in the 90:10 ratio was relatively small (7.05%), reflecting offsetting trends: while inequality in the bottom half of the distribution increased significantly (50:10: +20.11%), it declined by a similar magnitude in the upper half (90:50: -10.87%). This pattern suggests that while lower-income EB households saw greater inequality in consumption, inequality among higher-income EB households narrowed, suggesting a reduction in consumption gaps at thetopofthedistribution. Intermsofthecomponentsofenergyconsumption,residential energyinequalitydeclinedoverthesample,particularlyintheupperhalfofthedistribution (90:50: -5.75%). In contrast, transport energy inequality increased dramatically at both ends of the distribution, with the 90:10 ratio rising by 68% and the 50:10 ratio by 20%. 7 CONCLUDING REMARKS Using U.S. micro-data from the PSID, we show that energy burden is persistent, disproportionately affects lower-income households, and is associated with higher marginal propensities to consume and marginal propensities to consume energy. Our study also shows that energy-burdened households face more volatile energy consumption and incomegrowththannon-burdenedhouseholds,despiteexperiencinghigherincomegrowth on average. Importantly, we broaden the definition of household energy consumption to includegasolinefortransport,demonstratingthattraditionalmeasuresofenergyburden, 31

whichfocussolelyonresidentialenergyexpenditures,fallshortinidentifyingtheoverall scopeofenergyburdeninthehouseholdsector. Byexaminingtheevolutionofinequality in income, consumption, and energy expenditures, we find that income inequality has increased at a faster rate than energy consumption inequality, with the burden of rising inequality being more pronounced among energy-burdened households. The empirical facts documented in this paper provide a guidance to discipline the calibration of theoretical macroeconomic models with some degree of household heterogeneity regarding energyburdenandenergyconsumption. 32

References Aguiar, M. and M. Bils. 2015. “Has Consumption Inequality Mirrored Income Inequality?” AmericanEconomicReview105:2725–2756. Aguiar,M.,M.Bils,andC.Boar.2024.“WhoAretheHand-to-Mouth?” ReviewofEconomic StudiesForthcoming. Andrés, J., J. E. Boscá, J. Ferri, and C. Fuentes-Albero. 2022. “Households Balance Sheets andtheEffectofFiscalPolicy.” JournalofMoney,CreditandBanking54:737–778. Attanasio, O., E. Hurst, and L. Pistaferri. 2015. “The Evolution of Income, Consumption andLeisureInequalityintheUS,1980-2010.” InImprovingtheMeasurementofConsumer Expenditures,editedbyC.Carroll,T.Crossley,andJ.Sabelhaus.Univ.ChicagoPress. Auclert, A. 2019. “Monetary Policy and the Redistribution Channel.” American Economic Review109(6):2333–67. Best,R.andK.Sinha.2021. “FuelPovertyPolicy: GoBigorGoHomeInsulation.” Energy Economics97:105195. Blundell, R., L. Pistaferri, and I. Preston. 2008. “Consumption Inequality and Partial Insurance.” AmericanEconomicReview98(5):1887–1921. Boardman,B.1991. FuelPoverty: FromColdHomestoAffordableWarmth. BlehavenPress. Burlinson, A., M. Giulietti, and G. Battisti. 2018. “The Elephant in the Energy Room: EstablishingtheNexusBetweenHousingPovertyandFuelPoverty.” EnergyEconomics 72:135–144. Chen, J. J., V. Mueller, Y. Jia, and S. Kuo-Hsin Tseng. 2017. “Validating Migration Responses to Flooding Using Satellite and Vital Registration Data.” American Economic Review107(5):441–45. Cho, Y., J. Morley, and A. Singh. 2024. “Did Marginal Propensities to Consume Change withtheHousingBoomandBust?” JournalofAppliedEconometrics39:128–149. Commault, J. 2022. “Does Consumption Responds to Transitory Shocks? Reconciling NaturalExperimentsandSemistructuralMethods.” AmericanEconomicJournal: Macroeconomics14((2)):96–122. Crawley,E.2020. “InSearchofLostTimeAggregation.” EconomicLetters189:1089–1098. 33

Fisher, J., D. Johnson, T. Smeeding, and J. Thompson. 2019. “Estimating the Marginal PropensitytoConsumeUsingtheDistributionsofIncome,ConsumptionandWealth.” FederalReserveeBankofBoston,WPNo.19-4. Green, K. P., T. Jackson, I. Herzog, and M. Palacios. 2016. “Energy Costs and Canadian Households: HowMuchAreWeSpending?” FraserInstituteReport. Heathcote,Jonathan,FabrizioPerri,GiovanniL.Violante,andLichenZhang.2023. “More unequalwestand? InequalitydynamicsintheUnitedStates,1967–2021.” ReviewofEconomicDynamics50:235–266. URLhttps://www.sciencedirect.com/science/article/ pii/S1094202523000455. Hernandez, D. 2016. “Understanding ’energy insecurity’ and why it matters to health.” SocialScience&Medicine167:1–10. Hills,J.2012. “GettingtheMeasureofFuelPoverty.” FinalReportoftheFuelPovertyReview . Isherwood, R. M. and B.C. Hancock. 1979. “Household Expenditure on Fuel: DistributionalAspects.” EconomicAdviser’sOffice,DHSS:London,UK . Kaplan, G., G. Violante, and J. Weidner. 2014. “The Wealthy Hand-To-Mouth.” Brookings PapersonEconomicActivity45(1):77–153. Krueger,D.andF.Perri.2006. “DoesIncomeInequalityLeadtoConsumptionInequality? EvidenceandTheory.” ReviewofEconomicStudies73:163–193. Legendre, B. and O. Ricci. 2015. “Measuring fuel poverty in France: Which households are the most fuel vulnerable?” Energy Economics 49:620–628. URL https: //www.sciencedirect.com/science/article/pii/S0140988315000390. Lewis, D., D. Melcangi, and L. Pilossoph. 2024. “Latent Heterogeneity in the Marginal PropensitytoConsume.” Mimeo. Linn,J.,J.Liang,andY.Qiu.2023. “RisingU.S.IncomeInequalityandDecliningResidential Electricity Consumption: Is There a Link?” The Review of Economics and Statistics :1–46. Meyer,B.D.andJ.X.Sullivan.2023. “ConsumptionandIncomeInequalityintheUnited Statessincethe1960s.” JournalofPoliticalEconomy131(number2):247–284. 34

Middlemiss, L. and R. Gillard. 2015. “Fuel poverty from the bottom-up: Characterising household energy vulnerability through the lived experience of the fuel poor.” Energy Research SocialScience6:146–154. Mohr, T. M. 2018. “Fuel Poverty in the US: Evidence using the 2009 Residential Energy ConsumptionSurvey.” EnergyEconomics74:360–369. Moore, R. 2012. “Definitions of Fuel Poverty: Implications for Policy.” Energy Policy 49:19–26. Oni,M.H.2024. “Commuting,HomeUtilities,andProduction: TheDistributionalEffects ofEnergyPriceShocks.” Mimeo.UniversityofArkansas. PSID. 2024. “Panel Study of Income Dynamics, public use dataset.” Produced and distributed by the Survey Research Center, Institute for Social Research, University of Michigan,AnnArbor,MI(2024). Romero,J.C.,P.Linares,andX.López.2018. “ThePolicyImplicationsofEnergyPoverty Indicators.” EnergyPolicy115:98–108. Steele, A. J. Harker and J. C. Bergstrom. 2021. “"Brr! It’s cold in here" measures of householdenergyinsecurityfortheUnitedStates.” EnergyResearch&SocialScience72. Tito, M. D. 2024. “Does the Ability to Work Remotely Alter Labor Force Attachment? An Analysis of Female Labor Force Participation.” FEDS Notes. Board of Governors of the FederalReserveSystem(U.S.). Wang, Q., M. Kwan, J. Fan, and J. Lin. 2021. “Racial Disparities in Energy Poverty in the UnitedStates.” RenewableandSustainableEnergyReviews137:110620. Zeldes,S.P.1989. “ConsumptionandLiquidityConstraints: AnEmpiricalInvestigation.” JournalofPoliticalEconomy97(2):305–46. 35

Online Appendix A Appendix: Figures FIGURE A.1: DISTRIBUTION OF INCOME AND CONSUMPTION 0.000025 0.000020 0.000015 0.000010 0.000005 0.000000 ytisneD 1999 2009 2021 0.000015 0.000010 0.000005 0.000000 0 50000 100000 150000 Income A.Income: EBHouseholds ytisneD 1999 2009 2021 0 50000 100000 150000 Income B.Income: Non-EBHouseholds 0.000025 0.000020 0.000015 0.000010 0.000005 0.000000 ytisneD 1999 2009 2021 0.000020 0.000015 0.000010 0.000005 0.000000 0 20000 40000 60000 80000 100000 Consumption C.Consumption: EBHouseholds ytisneD 1999 2009 2021 0 20000 40000 60000 80000 100000 Consumption D.Consumption: Non-EBHouseholds 36

FIGURE A.2: DISTRIBUTION OF THE COMPONENTS OF ENERGY CONSUMPTION 0.000034 0.000033 0.000032 0.000032 0.000031 ytisneD 1999 2009 2021 0.000034 0.000033 0.000032 0.000031 0.000030 0 2000 4000 6000 8000 10000 Total Energy Consumption A.TotalEnergy: EBHouseholds ytisneD 1999 2009 2021 0 2000 4000 6000 8000 10000 Total Energy Consumption B.TotalEnergy: Non-EBHouseholds 0.000034 0.000033 0.000033 0.000033 0.000033 ytisneD 1999 2009 2021 0.000034 0.000033 0.000033 0.000033 0.000033 0 1000 2000 3000 4000 5000 Within−Home Energy Consumption C.Residential: EBHouseholds ytisneD 1999 2009 2021 0 1000 2000 3000 4000 5000 Within−Home Energy Consumption D.Residential: Non-EBHouseholds 0.000034 0.000033 0.000032 0.000032 ytisneD 1999 2009 2021 0.000034 0.000033 0.000032 0.000032 0 2000 4000 6000 Gas for Transport Consumption E.GasforTransport: EBHouseholds ytisneD 1999 2009 2021 0 2000 4000 6000 Gas for Transport Consumption F.GasforTransport: Non-EBHouseholds 37

FIGURE A.3: MARGINAL PROPENSITIES TO CONSUME BY YEAR, 1999-2021 50 40 30 20 10 0 % ni CPM 2000 2005 2010 2015 2020 Energy Burdened Non−Energy Burdened NOTE: MPCsarecomputedusingthe2MapproachasdescribedinSection5. 38

B Appendix: Tables B.1 Robustness 39

TABLE A1: HOW ENERGY BURDEN COVARIES ACROSS DIFFERENT HOUSEHOLD PROFILES: PROBIT AND LPM ESTIMATES Variables Probit LPM Socioeconomic BottomTwoIncomeQuintiles 0.101∗∗∗ 0.113∗∗∗ Ownershipstatus Homeowner 0.016∗∗∗ 0.017∗∗∗ Housesize6+ −0.012∗∗∗ −0.010∗∗ Typeofdwelling Onefamilyhouse 0.088∗∗∗ 0.099∗∗∗ Twofamilyhouse 0.065∗∗∗ 0.078∗∗∗ MobileHome 0.144∗∗∗ 0.157∗∗∗ Rowhome 0.006 0.028∗∗∗ Heating Gas 0.009∗∗∗ 0.014∗∗∗ Oil 0.058∗∗∗ 0.056∗∗∗ Other(propane,wood,kerosene) 0.066∗∗∗ 0.062∗∗∗ Householdcharacteristics Race Black 0.073∗∗∗ 0.068∗∗∗ Asian 0.027∗ 0.036∗∗∗ Other 0.043∗∗∗ 0.033∗∗∗ Hispanic 0.037∗∗∗ 0.039∗∗∗ Married −0.045∗∗∗ −0.042∗∗∗ Female 0.014∗∗∗ 0.019∗∗∗ Kids 0.020∗∗∗ 0.018∗∗∗ Head65+ −0.040∗∗∗ −0.047∗∗∗ Othersocioeconomic Employed −0.065∗∗∗ −0.078∗∗∗ PostsecondaryEducation −0.069∗∗∗ −0.068∗∗∗ Subsidizedhousing 0.029∗∗∗ 0.027∗∗ Heatingsubsidy 0.126∗∗∗ 0.173∗∗∗ BehindonMortgage 0.111∗∗∗ 0.136∗∗∗ Geographiclocation Northeast −0.036∗∗∗ − Northcentral −0.028∗∗∗ − West −0.064∗∗∗ − Yeardummies? ✓ − State-yearFE? − ✓ YearFE? − ✓ N 57,248 57,361 R2 0.176 0.188 NOTE: Asterisksindicatethelevelofsignificanceoftheparameters,* p < .10;** p < .05;and*** p < .01. LPMclustersstandard errorsbyid-year. ProbitreportsthePseudoR2. 40

TABLE A2: HOW ENERGY BURDEN COVARIES ACROSS DIFFERENT HOUSEHOLD PROFILES WHO ARE NOT HTM Variables Logit Socioeconomic Bottomtwoincomequintiles 0.092∗∗∗ Ownershipstatus Homeowner 0.016∗∗∗ Housesize6+ −0.004 Typeofdwelling Onefamilyhouse 0.067∗∗∗ Twofamilyhouse 0.077∗∗∗ Mobilehome 0.110∗∗∗ Rowhome 0.001 Heating Gas −0.004 Oil 0.038∗∗∗ Other(propane,wood,kerosene) 0.036∗∗ Householdcharacteristics Race Black 0.050∗∗∗ Asian 0.030 Other 0.049∗∗∗ Hispanic 0.006 Married −0.041∗∗∗ Female −0.002 Kids 0.029∗∗∗ Head65+ −0.025∗∗∗ Othersocioeconomic Employed −0.045∗∗∗ Postsecondaryeducation −0.055∗∗∗ Subsidizedhousing 0.039∗∗ Heatingsubsidy 0.138∗∗∗ Behindonmortgage 0.096∗∗∗ Geographiclocation Northeast −0.005 Northcentral −0.018∗∗ West −0.042∗∗∗ ✓ Yeardummies? N 25,816 Pseudo R2 0.179 NOTE: Sub-sampleofhouseholdswhoareclassified asnothand-to-mouth. Asterisksindicatethelevelof significanceoftheparameters,* p < .10; ** p < .05; and*** p <.01. 41

TABLE A3: MARGINAL PROPENSITIES. 10% APPROACH VulnerableHouseholds Non-VulnerableHouseholds MPC 0.208 0.076 MPCetotal 0.052 0.011 MPCeresidential 0.010 0.004 MPCetransport 0.026 0.005 TABLE A4: AVERAGE ANNUAL GROWTH RATE (IN %) 2001 2003 2005 2007 2009 2011 2013 2015 2017 2019 2021 Income EB 17.25 11.96 15.40 12.34 10.50 8.43 8.68 10.10 11.48 13.10 14.56 Non-EB -0.43 -2.95 -1.56 -0.91 -0.54 -4.25 -1.75 -0.13 -0.32 0.91 1.63 Consumption EB 0.22 -1.06 1.12 -0.50 -2.90 -2.61 -2.74 -2.67 1.31 0.52 -2.91 Non-EB 2.97 1.55 2.50 2.19 -2.12 -0.05 0.69 0.44 2.52 1.12 1.80 EnergyCons EB -0.11 -13.03 -1.68 -4.24 -14.51 -3.00 -10.27 -13.56 -11.02 -7.45 -13.86 Non-EB 10.56 -1.08 9.17 8.42 -2.52 10.25 0.79 -3.46 -2.34 3.60 -1.92 Residential EB 1.83 -9.99 -2.88 -3.72 -4.45 -6.91 -8.25 -1.36 -8.70 -5.52 -11.65 Non-EB 6.49 -1.78 4.03 5.30 4.57 1.39 0.08 3.09 -0.38 3.31 -1.12 Transport EB -0.59 -12.88 1.70 -3.92 -24.59 1.60 -11.30 -24.98 -12.63 -9.26 -14.06 Non-EB 14.73 -1.23 13.61 10.88 -10.49 18.31 2.23 -10.97 -3.39 3.21 -4.13 Note: Growth rates are defined as the log-difference between the variable of interest in wave t+2 and in wavet,dividedbytwotoannualizethegrowthrate. 42