key: cord-0867035-zmewkev1 authors: Burlinson, Andrew; Giulietti, Monica; Law, Cherry; Liu, Hui-Hsuan title: Fuel poverty and financial distress date: 2021-07-22 journal: Energy Econ DOI: 10.1016/j.eneco.2021.105464 sha: 8950a810afd2f1699bb770480a1d0c06835c1f0f doc_id: 867035 cord_uid: zmewkev1 Governments and advocacy groups have drawn attention to the precarious position of those members of society who are unable to attain an adequate level of energy services, i.e. the fuel poor. Concerns have also arisen about the ability of fuel poor individuals to adapt to the hardship recently brought about by the COVID-19 pandemic. This paper contributes to the literature by exploring empirically the link between fuel poverty and financial distress prior to and during the first wave the COVID-19 pandemic. The analysis is based on the most recent longitudinal, nationally representative survey of the United Kingdom, Understanding Society (UKHLS, Wave 10, January 2018–February 2020). After correcting for the effects of potential endogeneity in the variables of interest, our results identify a statistically robust relationship between fuel poverty indicators and self-reported measures of current financial distress, with stronger effects for subjective indicators. The fuel poverty indicators however exert only a limited influence on an individual's expectation of their future financial situation. Our analysis of the first wave of the COVID-19 pandemic also confirms that fuel poverty contributed to financial distress. Our main findings are robust to a suite of specification and sensitivity checks. Our results lead to recommend assessing measures which target fuel poverty on the basis of their potential indirect effect on financial distress. Fuel poverty is considered a distinct form of poverty, not least because tackling it has the potential to garner a "win-win-win" for policymakers through improvements in economic hardship, mental and physical health, and energy/carbon savings (Boardman, 1991; Green and Gilbertson, 2008; Hills, 2011) . Broadly, over the last three decades, fuel poverty has been defined as the household's inability to achieve thermal comfort to levels commensurate with a healthy standard of living at a reasonable cost (Boardman, 1991; Hills, 2012) . 1 The incidence of fuel poverty depends on three central driversincome, energy efficiency and energy prices (Moore, 2012; Thomson et al., 2017) . Recent estimates show fuel poverty affects over 20% of households in the United States (US) and China, 10% of households in Australia and France and close to 10% of households in Japan (Legendre and Ricci, 2014; Okushima, 2017; Zhang et al., 2019; Awaworyi Churchill et al., 2020; Wang et al., 2021) . 2 The prevalence of fuel poverty in Great Britain (GB) -the focus of the present papervaries by nation with 10% of households identified as fuel poor in England, 25% in Scotland and 12% in Wales (BEIS, 2021a; Hinson and Bolton, 2020) . A near consensus has formed around the body of evidence documenting the deleterious impact that fuel poverty exerts on the health of households, including higher rates of mortality and higher risk of cardiovascular, inflammatory and mental health conditions (see e.g. Crossley and Zilio, 2018; Marmot Review Team, 2011; Public Health England, 2014; Thomson et al., 2001) . Whilst financial distress is a potential mediator between fuel poverty and health outcomes (Hills, 2011; Marmot Review Team, 2011) , the fuel poverty and financial distress nexus is hitherto underexplored, especially in the economics literature. 3 A better understanding of fuel poverty-induced financial distress is paramount in order to evaluate the full impact of policy interventions affecting energy consumption and expenditure. With rising energy prices and stagnant real income in GB (BEIS, 2020a), low-income households face difficult trade-offs between energy and other necessities, diminishing savings and/or incurring debt in order to maintain optimal levels of thermal comfort (Harrington et al., 2005; Hills, 2011; Anderson et al., 2012; Grey et al., 2017; Munyanyi et al., 2021) . In 2018, just over 1.1 million gas and 1.3 million electricity consumers were in arrears or repaying fuel debt in GB (Ofgem, 2019) . In 2017, the total amount of debt and arrears accruing by gas and electricity consumers has been shown to reach around £1.1 billion in total for GB (Citizens Advice Bureau, 2018) . What is more, financial distress has also manifested itself in countries characterised by lower energy prices and a wider adoption of cooling technologies. For example, according to the most recent US Residential Energy Consumption Survey, at least 7 million households' forgone necessities to pay energy bills, 6 million (7 million) households are unable to cool (heat) their homes due to financial constraints and 2 million households received disconnection notices every month (EIA, 2018) . Fuel poverty may not only impose financial constraints, but also further impacts the mental health and well-being of households (Ofgem, 2019 (Ofgem, , 2021 . Indeed, Hills (2011:89) 's review of early evidence on the measurement, causes and impacts of fuel poverty suggests that a "chain of causation could potentially be from income (not exclusively low income), to debt, to poor mental health". Crucially, this 'financial security' chain represents one of two key pathways that could explain the causal mechanism between fuel poverty indicators and well-being and self-reported health outcomes established in recent economics literature. Most recently, using the Household Income and Labour Dynamics in Australia longitudinal survey, Awaworyi Churchill et al. (2020) unveil the negative relationship between fuel poverty and subjective well-being. Similarly, Kahouli (2020)'s and Awaworyi Churchill and Smyth (2021) 's findings further reveal that fuel poverty adversely impacts self-assessed health in France and general health in Australia, respectively. However, the intermediate mechanisms linking fuel poverty to health outcomes requires further investigation. First, alleviating fuel poverty through the "living conditions" pathway could impact health psychosocially (e.g. anxiety and depression), in all age groups, and/or physiologically, particularly the elderly and infants, via higher levels of thermal comfort (Harrington et al., 2005; Green and Gilbertson, 2008; Hills, 2011; Marmot Review Team, 2011; Gilbertson et al., 2012: 12; Ormandy and Ezratty, 2012) . Second, tackling the deleterious impacts of fuel poverty through the "financial security" pathway could improve health via lower levels of financial stress (Hills, 2011; Gilbertson et al., 2012: 12) : Fuel poverty could also damage mental health as a result of stress arising from financial worry - (Harrington et al., 2005: 263) However, the second pathway, despite its potential importance, remains underexplored in the relevant literature. 4 This is somewhat surprising since the quasi-experimental evaluation of the United Kingdom's flagship fuel poverty initiative (The Warm Front Scheme) concluded: The alleviation of fuel poverty and the reduction of stress associated with greater financial security emerge as the most likely route to health, both mental and physical. - (Gilbertson et al., 2012 : 132) Gilbertson et al. (2012) analysed cross-sectional surveys of 2685 lowincome households living in five urban areas of England, collecting preintervention (2001/2) and post-intervention (2002/3) data. Their analysis indicates that the "financial security" pathway is the principal route connecting the alleviation of fuel poverty to self-reported health (i. e. fuel povertystresshealth), whilst the "living conditions" pathway serves as the secondary route (i.e indoor temperaturethermal comfort health) (Green and Gilbertson, 2008; Gilbertson et al., 2012) . The present paper draws upon a nationally representative survey of the UK, Understanding Society: the UK Household Longitudinal Study (UKHLS), in order to empirically examine this intermediate link between fuel poverty and financial distress. Considering the growing policy attention on the precarious position of the fuel poor and on the increased deprivation caused by the current pandemic, we employ three UKHLS' COVID-19 web surveys to examine whether fuel poverty contributes to financial distress during the first wave of the Coronavirus pandemic. The empirical analysis focuses only on Great Britain (GB) which includes England, Scotland, and Wales due to the different energy market and regulatory arrangements in Northern Ireland. In this paper, we suggest that fuel poverty increases the probability of falling behind on bills and finding one's current financial situation difficult to deal with, prior to and during the COVID-19 pandemic. Our findings are not only robust across a series of specification checks, but also rely on methodologies which address potential endogeneity concerns including instrumental variable estimation and Oster (2019)'s bounding approach. However we find less pronounced evidence to suggest that fuel poverty affects the surveyed individuals' expectations about their financial future. We contribute to the existing literature in three key ways. First, to the best of our knowledge, this is the first paper to quantitatively investigate fuel poverty as a determinant of financial distress using representative surveys. Using more recent data, the present paper complements quasiexperimental (Gilbertson et al., 2012) and qualitative (Harrington et al., 2005: 263; Grey et al., 2017) analyses of energy efficiency interventions in low-income households/communities, by testing the external validity of the key intermediate link (fuel povertystress) in the "financial security" pathway. Establishing determinants of financial distress is crucial due to its long-term consequences for income and health inequalities, particularly for low-income households who are more exposed during periods of economic and financial crises (Arber et al., 2014; Olafsson, 2016) . Most recently, for example, sharp falls in income are expected as a result of the COVID-19 pandemic, potentially sharpening the trade-offs between expenditure on necessities, savings and debt. Indeed, "what they [households] normally spend their money on will matter for how well they can weather this storm" (IFS, 2020: 2). Necessity goods, such as gas and electricity (Meier et al., 2013) , will form a rising proportion of disposable income for households unable to flexibly adjust their spending in response to a fall in income (IFS, 2020). We therefore add to the growing literature seeking to uncover the determinants of financial distress. Over the last decade, studies have investigated financial distress through the lens of the difficulties associated with student loan debt (Elliott and Lewis, 2015; Bricker and Thompson, 2016) , medical insurance (Dobkin et al., 2018; Hu et al., 2018; Mazumder and Miller, 2016) , and mortgage repayment (Gathergood, 2012) . Notable recent contributions in the economics and finance literature explore the channels through which cognitive and noncognitive abilities affect measures of financial distress (Xu et al., 2015; Parise and Peijnenburg, 2019 ). Yet the role of energy, and thus fuel poverty, in determining household financial distress has so far been overlooked. 3 At the time of writing, searching the keywords "fuel poverty" or "energy poverty" and "financial distress" in the Scopus Database retrieves no documents. Several papers are retrieved when replacing the latter term with "household finance", the most relevant, of which, explores household selfdisconnection from energy supply (Rocha et al., 2019) 4 For detailed reviews of studies evaluating the first 'thermal comfort' pathway, particularly those using randomised or quasi-randomised control household energy efficiency interventions, see e.g. Liddell and Morris, 2010; McAndrew et al., 2021 . Perhaps most closely related to the present paper is Dorsey-Palmateer (2020)'s study of financially-constrained households in the US. Using a sub-sample from the 2017 American Housing Survey, the author examines the association between several indicators of financial distress (e.g. utility notices/disconnections, missed rent payments) and monthly combined utility costs (including energy and other utilities), monthly housing costs and monthly income. The author finds utility payments to be associated with a greater (dollar-for-dollar) impact on financial distress than monthly income and housing costs. Contrary to this approach, the present paper models the relationship between energyspecific covariates (fuel poverty) and financial distress, using a nationally representative survey. An important methodological difference from Dorsey-Palmateer (2020) is our deployment of methods to alleviate potential endogeneity concerns. In order to address endogeneity concerns, we propose a novel set of instrumental variables which complement those currently implemented in the literature in order to formalise the empirical relationship between fuel poverty and financial distress. It is important to note that due to practical challenges Green and Gilbertson and colleagues (2008; are unable to precisely target fuel poor households and subsequently rely on proxies for measurement of fuel poverty and financial distress. The authors ask households whether they "had difficulties paying their fuel bills". In addition, the authors use a four-point scale of general stress from no stress (1) to high stress (4) levels. In essence, the positive association between the two sets of variables is interpreted as the stress effects of fuel-induced financial pressure. The present paper, in contrast, uses commonly implemented indicators of fuel poverty (both objective and subjective) and self-reported measures of financial distress (quasiobjective and subjective). We therefore model a more proximal relationship between fuel poverty, unpaid bills and perceptions about financial distress now and finances in the future. To alleviate endogeneity concerns, we rely upon regional variation in energy prices (Awaworyi Churchill et al., 2020; Kahouli, 2020; Awaworyi Churchill and Smyth, 2021; Munyanyi et al., 2021) , and further add to the literature by introducing a robust set of instruments. We exploit the between-and within-region variation of nonlinear pricing in GB's retail energy market using annual regional-level gas and electricity retail unit prices (£/kWh), fixed charges (£/year), and the fixed charge to unit price ratioall of which are further disaggregated by payment methods (i.e. credit, direct debit and prepayment). This approach provides additional withinvariation compared to regional-level energy consumer price indices (Awaworyi Churchill et al., 2020; Awaworyi Churchill and Smyth, 2021; Munyanyi et al., 2021) and appears more robust than the sole use of unit prices (Kahouli, 2020) . Finally, we investigate the financial vulnerability of the fuel poor during the UK's first wave of the COVID-19 pandemic, April 2020-July 2020. COVID-19 has impacted the welfare of people worldwide, particularly the poorest, and has further exposed existing inequalities within and across countries (Fuchs-Schündeln et al., 2020; The Economist, 2020; Wildman, 2021) . Indeed, governments and advocacy groups have drawn attention to the precarious position of the fuel poor and their ability to adjust to income shortfalls prior to and during the pandemic (Citizens Advice Bureau, 2020; National Energy Action, 2020a; Scottish Government, 2020; The End Fuel Poverty Coalition, 2020) . The remainder of the paper is structured as follows: Section 2 describes the data and presents our empirical methodology; Section 3 presents our results, before discussing policy implications and drawing conclusions in Section 4. Our data are obtained from a longitudinal, nationally representative survey of the UK, Understanding Society: the UK Household Longitudinal Study (UKHLS) (University of Essex, 2020). We utilise the most recent General Population Sample, a random sample of the general UK population, Wave 10 (January 2018-May 2020) -referred to hereafter as the 'main survey'. We focus specifically on GB as the instrumental variables are confined to England, Scotland, and Wales. 5 As part of the main survey, a set of financial distress measures and fuel poverty indicators are collected alongside economic and socio-demographic characteristics. The final sample consists of 23,210 individuals with valid/non-missing values for our outcomes, key variables of interest and controls within the main survey. 6, 7 Drawing upon relevant literature, we use three dichotomous selfreported measures of financial distress (Table 1) . 8 BEHINDBILLS equals 1 when respondents report being behind on some or all bills, and 0 otherwise (Parise and Peijnenburg, 2019) . We set FINNOW equal to 1 if individuals found their current financial situation difficult or very difficult, and 0 otherwise. Whereas FINFUT equals 1 if the individual believes their financial situation will be worse off a year from now, and 0 otherwise. The latter two measures capture the individual's current The identification strategy hinges upon gas prices which are currently not provided for NI by the data source (BEIS, 2021b (BEIS, , 2021c .The baseline results without the instrumental variables are robust to the inclusion of Northern Ireland (NI). 6 We removed 71 individuals who participated in the main survey between March 2020 and May 2020 in order to avoid overlap with the COVID-19 pandemic. The sample statistics and estimates are quantitatively identical when including the 71 individuals (Table A5 , Column 1, Appendix A). More importantly, their removal provides a clean cut-off prior to the pandemic (January 2018-February 2020). 7 All sample statistics and estimation results presented in the paper are unweighted and consistent with cross-sectional survey weights adjusted for item and unit non-response. 8 Declaring being behind on bills is clearly less subjective than stating whether one's current (future) financial situation is difficult (expected to become worse). Nonetheless, we reserve the objective/subjective lexicon for fuel poverty indicators to avoid confusion and refer to the financial distress variables simply as 'self-reported measures' hereafter. and future expectations of their financial situation (Keese, 2012) . The sample statistics in Table 1 report that, on average, 5.4% of individuals were not up to date with all of their household bills, 7.5% find their current finances at least difficult, and 12.5% think they would be financially worse off a year from now. Whilst the measurement of fuel poverty remains somewhat contested (Deller et al., 2021; Thomson, 2020) , recent literature has drawn upon the strengths of objective and subjective approaches by employing both sets of indicators (see e.g. Awaworyi Churchill et al., 2020; Kahouli, 2020; Llorca et al., 2020; Awaworyi Churchill and Smyth, 2021) . The seminal work of Waddams Price et al. (2012) evaluates the positive yet complex overlap between official objective indicators and subjective indicators. The authors conclude the latter complements the former by way of informing energy policy on the extent to which it alleviates the feeling of being unable to afford energy. More recently, Llorca et al. (2020) argue for the use of subjective fuel poverty indicators, alongside objective indicators, in order to capture the personality underpinning self-reported outcomes and their covariates. We employ two objective indicators of fuel poverty, namely the 10% expenditure threshold FP10 and the low-income-high-cost indicator LIHC (Boardman, 1991; Hills, 2012) , as well as one subjective indicator, that is whether the household can afford to keep the home warm IHEAT (Waddams Price et al., 2012) . FP10 equals 1 if the individual's household spends more than 10% of their income on energy bills, and 0 otherwise. 9 LIHC takes a value of 1 if the individual's household meets two conditions: 1) they spend more than the national median on energy in the last year and 2) upon deducting energy expenditure and housing costs, their residual household net income falls below the poverty threshold (i.e. 60% of the national median household net income); and 0 otherwise. 10 The IHEAT indicator takes the value of 1 for those individuals (or a member of their household) who report inadequate heating during winter due to affordability issues, and 0 otherwise. On average, 11.2% of respondents are part of a fuel poor household according to LIHC, whereas FP10 and IHEAT identify 13.9% and 4.4% respondents as fuel poor respectively (Table 1) . Empirically the paper proceeds by estimating the probability of exhibiting financial distress using ordinary least squares regression. The general specification for the linear probability models (LPM) of financial distress on fuel poverty is defined as follows: where, FINDIS i * represents the latent variable for each of the financial distress measures (BEHINDBILLS, FINNOW or FINFUT) for individual i. FUELPOV i represents three separate models each containing a single fuel poverty indicator (LIHC, FP10 or IHEAT). Х i contains the economic and socio-demographic covariates identified as determinants of financial distress in the literature (e.g. Xu et al., 2017; Parise and Peijnenburg, 2019) . β and ρ are the estimated regression coefficients, with β being the set of parameters of interest. ω t is the vector of seasonal effects that capture the month and year in which the individual participated in the survey in the main wave. μ r represents the vector of 11 GB regional effects capturing England's nine government office levels, and one for Scotland and Wales respectively. ε i is the heteroskedastic robust error term. Table A1 (Appendix A) provides the definitions and summary statistics for the control variables. One potential concern regarding identification of the pathway between fuel poverty and financial distress through the above model is endogeneity. For example, reverse causality may exist if financial exclusion and debt arising from worsening economic conditions add to the precarious position of households, increasing the likelihood of falling into fuel poverty (Lacroix and Chaton, 2015) . As discussed above, whilst Gilbertson et al. (2012) argue that the most logical direction of causality runs from fuel poverty to financial stress (i.e. the "financial security pathway"), we cannot rule out that these variables are simultaneously determined or at least correlated via omitted variables (Liddell and Guiney, 2015) . A potential confounder is the lack of internal temperature readings for each homea variable often missing from national surveys. Internal temperatures may be linked indirectly to financial distress as suboptimal temperatures are linked directly to fuel poverty through expenditure shares. The bias attributed to internal temperatures is likely to be toward zero since, all else constant, it is reasonable to assume β INTERNALTEMP > 0 and Corr(INTERNALTEMP, FUELPOV) < 0. A third source of endogeneity could be attributed to measurement error. For instance, there may be a non-zero correlation between the errors made by households when self-reporting information underpinning fuel poverty indicators and financial distress measures. Unlike the omission of internal temperatures, one would expect the bias arising from self-reporting measurement error to be away from zero. 11 Therefore, in order to alleviate concerns surrounding endogeneity, we employ a suite of instrumental variable (IV) estimators. We add to the literature by implementing IVs based on the components of GB's nonlinear energy retail pricing system. It has been argued previously that exogenous movements in energy prices are a plausible instrument, similar to the use of other commodity prices (e.g. food) in the fuel poverty-health literature (Kahouli, 2020) . Indeed, energy prices have the potential to satisfy the exclusion restrictions condition. Not least because prices are assumed to work directly through fuel poverty, specifically the expenditure share of income, thereby indirectly affecting outcomes of interest, in our case, financial distress (Awaworyi Churchill et al., 2020; Kahouli, 2020; Awaworyi Churchill and Smyth, 2021; Munyanyi et al., 2021) . Moreover, energy prices have further potential to satisfy the relevance condition, since one would expect energy prices to be positively and strongly associated with fuel poverty. However, the preceding literature acknowledges concerns about whether prices are exogenous to the error term from a statistical perspective (see e.g. Awaworyi Churchill et al., 2020; Kahouli, 2020) and about the potential weak correlation between the IVs (i.e. energy prices) and the endogenous variable (i.e. fuel poverty) (Munyanyi et al., 2021) . Considering such concerns, the present paper employs a novel yet complementary array of IVs, including: the marginal price M per unit of gas and electricity (£/kWh); the fixed charge F for supplying gas and/or electricity to the meter (£/year). Fixed charges are independent of consumption and typically cover the costs of the meter (e.g. maintaining connection to supply, meter reading and other customer account services); and, the fixed-to-marginal (FM) ratio. Davies et al. in 2014 introduce the FM ratio as a sufficient statistic that describes the time/regional evolution and asymmetry of two-part tariffs for representative consumers. 12 The regional variation in GB's retail energy pricing reflects the cost differences of incumbent companies (i.e. suppliers, distributed network 9 We further adjust FP10 by restricting the classification of fuel poverty to only those below the poverty threshold (60% of the national median household net income), negating the inclusion of relatively high-income high-energy expenditure households. 10 Income and energy are equivalisedsee Hills (2012). 11 If FINDIS + e = f(FUELPOV + v, X) and Corr(e, v) > 0, where e and v are measurement errors. 12 Like Davies et al. (2014) the fixed element of the ratio F is weighted by the variable price p for a median electricity (E) consumer (3600kWh, i.e., FM E = F E / 3600p E ) and median gas (G) consumer (13600kWh, i.e., FM G = F G /13600p G ). We use the most recent median typical domestic consumption values (BEIS, 2021b, 2021c). operators and transmission network operators). Since the 1990s wave of privatisation and liberalisation, the "Big 6" suppliers have dominated the GB retail energy market with 70% of consumers still supplied by the five electricity incumbents and the single gas incumbent (Ofgem, 2019). The retail suppliers also pass on transmission and distribution network costs charged by the regulated operators. The transmission and distribution network operators are monopolies regulated by the Office for Gas and Electricity Markets (Ofgem, 2015) . Three transmission operators (TOs) own and operate the national transmission (high pressure) gas and (high voltage) electricity networks. The low pressure and low voltage networks are split into fourteen electricity distribution networks (DNOs) and eight gas distribution networks (GDNs). Indeed, the number of DNOs and GDNs correspond to the locations managed by the regional gas and electricity boards that exist pre-privatisation (Ofgem, 2015) . The regulated part of prices reflects the regional differences in costs incurred by the network operators. The institutional and infrastructural legacy of GB's energy system allows us to exploit the regional differences in regional gas and electricity pricing (marginal and fixed) -oftentimes called the "postcode lottery" (Deller et al., 2020 ). The regional variation in GB energy pricing can be understood from two prevailing perspectives. On the one hand, according to Ofgem's study in 2015, differences in retail pricing are primarily attributed to national and local network charges i.e. the cost of building and maintaining the transmission and distribution network infrastructure (Ofgem, 2015) . Ofgem's report finds electricity network charges exert greater influence on retail prices than gas network charges. Nonetheless, Ofgem acknowledges that whilst some regions exhibit higher distributional charges they are, in some instances, partly offset by lower transmission charges. On the other hand, Davies et al. (2014) argue that the key source of price dispersion, in a given time period, is within-region (e. g. attributed to incumbent suppliers) rather than between-regions (e.g. associated with legacy networks). In fact, Davies et al. (2014) find over 63% of the variance in marginal prices and at least 82% of the variance in fixed charges can be explained by the variation within-region. Their study further suggests that asymmetric costs and other factors, including brand loyalty and market frictions, only partially influence price dispersion compared to tariff differentiation. Instead, dispersion arises through suppliers segmenting the market post-liberalisation into high (low) consumption consumers by charging high (low) fixed charges and low (high) marginal prices (Davies et al., 2014) . Our IVs therefore rely on the between-and within-region variation in GB nonlinear pricingas the first perspective most closely relates to fixed charges and second perspective relates to both the fixed and marginal components. Gas and electricity average retail marginal prices and fixed charges are collected annually for each GB region by the Department of Business, Energy and Industrial Strategy (BEIS, 2021b (BEIS, , 2021c . The data contains marginal prices and fixed charges by fuel type, region, and year. Moreover, the data further differentiate gas and electricity marginal prices and fixed charges by credit, direct debit, and prepayment methods of payment. For a given region and year, we calculate the fixed-marginal (FM) ratio by fuel and payment type. The data are matched to individuals in the UKHLS sample by region, year, fuel type and payment methodthe procedure is detailed in Appendix B (Table B1) . Table B2 (Appendix B) presents the definitions and summary statistics for the annual average gas and electricity prices between 2018 and 2020 (the years in which the respondents participated in the main survey) as well as between 2016 and 2018. It is important to note that our main IV results use prices from the period 2016-2018 for two key reasons: 1) prices 2-years prior to the year in which the respondents take part in the main survey have a stronger correlation with our indicators of fuel poverty. This is likely driven by the UKHLS asking participants to provide last year's household expenditure on gas, electricity or other fuels in their current residence. In addition, the individual's household representative is likely to be reporting the estimates of annual bills that appear on monthly/quarterly/annual statements and such billing estimates tend to be based on preceding years' consumption and prices determined at the start of a long-term contract 13 ; and 2) lagged prices will clearly be more exogenous than current prices (Charlier and Kahouli, 2018) . Hence, we can avert the issue of tariffs and thus energy expenditure that is contemporaneously influenced by either local or national demand and supply forces dictated in the wholesale and retail energy markets. Table B2 shows that average marginal gas prices have decreased slightly over the two time periods, whilst electricity prices have increased, in line with movements in the wholesale markets. Gas and electricity fixed charges have increased, driving up the fixedmarginal ratio between 2016 and 2020. It is important to note that the between-region variation (represented by the R 2 in Table B2 ) shows that regional variation is not constant over time and varies across the three price measures. Indeed, in-line with previous studies (Davies et al., 2014; Deller et al., 2020) , 14 within-region and time variation explains most of the price dispersion and provides further support as to why differentiating prices by payment method in the IV procedure is of importance. The first stage regression of the IV estimator, estimated using LPM, involves a reduced form equation specified as follows: Where PRICES i represents the vector of gas (G) and electricity (E) prices. The prices (M, F and FM) enter as separate pairs in order to reduce multicollinearity between the gas and electricity marginal prices and fixed charges. Hence, we employ three specifications which separately include the pairs M G and M E , F G and F E , or FM G and FM E . γ denotes the vector of coefficients for the respective pairs of prices, whilst u i represents the first stage regression error term. All other variables and coefficients are as defined in the second stage regression (Eq. (1)). This section first investigates the relationship between financial distress and fuel poverty prior to the pandemic using the main survey. These findings are scrutinised using a suite of specification and robustness checks in order to alleviate concerns about endogeneity. Next, this section explores the role of fuel poverty in determining financial distress during the pandemic. Table 2 presents the coefficients associated with fuel poverty using our baseline (LPM) specifications outlined in Eq. (1). The models either include objective indicators of fuel poverty, LIHC (Columns 1 and 2) and FP10 (Columns 3 and 4) or a subjective indicator of fuel poverty IHEAT (Columns 5 and 6). All even Columns (2, 4 and 6) include economic and socio-demographic controls and regional/time fixed effects. There is a clear positive association between the indicators of fuel poverty and measures of financial distress, either in the form of being behind on bills (Panel A), finding current finances difficult (Panel B) or expecting future finances to be worse in a year's time (Panel C). Focusing on the specifications that include controls, the objective (subjective) indicators suggest that fuel poverty, compared with not being in fuel poverty, is associated with an increased probability of falling behind on bills by 4.1 and 4.3 (21.6) percentage points (Panel A), finding the current financial situation difficult by 6.4 and 6.9 (22.6) percentage points (Panel B) and expecting future finances to become worse by around 1.5 and 1.9 (13.0) percentage points (Panel C), on average, ceteris paribus. 13 Contracts are set typically set between 12 and 24 months. There is no set price or contract for standard variable tariffs. 14 For example, Deller et al. (2020) show that regional price differences represented around a third of the average electricity bill in the 1970s and 8-18% of the average bill since 2009. Hence, for all measures of financial distress, we find the estimated probabilities are consistent in magnitude and in significance levels for both objective indicators (LIHC, FP10) despite their different definition. By contrast, the magnitude of the coefficient associated with these indicators appears smaller than that associated with the subjective indicator (IHEAT). This is consistent with relevant literature which finds a more pronounced relationship between self-assessed (health) outcomes and subjective, rather than objective, indicators of fuel poverty (see e.g. Awaworyi Churchill et al., 2020; Kahouli, 2020; Llorca et al., 2020) . To help address endogeneity concerns, we instrument the fuel poverty indicators (LIHC, FP10, IHEAT) sequentially by employing three separate pairs of gas and electricity prices i.e. marginal prices (M G and M E ), fixed charges (F G and F E ), and the fixed-marginal ratio (FM G and FM E ). 15 All specifications include economic and socio-demographic controls and regional/time fixed effects. For each measure of financial distress, at least one pair of instruments (M G and M E , F G and F E , or FM G and FM E ) is valid according to the Sagan-Hansen test (i.e. the null of exogeneity cannot be rejected). Not only is F G and F E the most relevant pair according to the first stage Fstatistic, but also in all but one specification this pair of instruments appear valid. In Table 3 , we focus on the specifications with the most relevant pair of instruments (i.e. the largest F-statistic reported in the first stage regressions) that are also valid (i.e. the J-statistic p-value >0.1 in the second stage regressions). The complete set of IV results are presented in Table A2 (Appendix A). The first stage regression results are contained in upper panel in Table 3 . The second stage regressions, which estimate the instrumented relationship between the fuel poverty indicators and our three selfreported measures of financial distress, are placed below. Column 1 presents the instrumented results for the LIHC indicator, followed by FP10 in Column 2 and finally IHEAT in Columns 3-4. The results for BEHINDBILLS, FINNOW and FINFUT are displayed in Panels A, B and C, respectively. As expected, in the first stage, increases in energy prices increase the likelihood of fuel poverty. For example, according to the LIHC indicator, the probability of being identified as fuel poor (c.f. nonfuel poor) increases between 0.87 and 2.92 percentage points given a respective £10/year rise in F G and F E (Column, 1). Similarly, turning to IHEAT (Column 3), increasing M G and M E by 0.01p/kWh increases the probability of being fuel poor by around 7.4 and 0.37 percentage points respectively, on average, ceteris paribus. Across all models, the strength of the instruments is markedly improved when fixed charges either enter exclusively or working as part of the FM-ratio (Table 3; Table A2 , Appendix A). The first stage F-statistic is consistently greater than 10, in-line with the Staiger and Stock (1997) rule-of-thumb. However, they fall below the level of 104.7, which recent literature suggests the first stage Fstatistic should exceed (Lee et al., 2020) . For each given F-statistic therefore, we correct the critical values and calculate "tF 0.05 standard errors" proposed by Lee et al. (2020: 21) . Compared to the true standard errors, Lee et al. (2020) consider these values to be somewhat conservative. Despite the conservative nature of this correction, the Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors in parentheses. Controls include economic and socio-demographic characteristics and regional/time fixed effects (Table A1 , Appendix A). a Future financial situation time horizon is "a year from now". 15 The results are consistent with the use of current prices (2018-2020) (Table A3 , Appendix A) and a one-year lag in prices (2017-2019) -for brevity these results are available upon request. statistically significant findings remain so at the 5% level. 16 Homing in on the preferred specifications in Table 3 , fuel poverty exerts a positive and significant impact on falling behind on bills (Panel A) and whether individuals consider their current financial situation to be at least difficult (Panel B). These findings exhibit the same sign as our baseline results and remain statistically significant at the 5% level when employing the more conservative (tF 0.05) standard errors. For example, according to the FP10 indicator, fuel poverty increases the probability of being behind on bills by 84.4 percentage points, on average, all else constant. In addition, the probability of finding current finances at least difficult increases by 24.8 percentage points if fuel poor (c.f. non-fuel poor). In contrast with our baseline results, Panel C suggests that fuel poverty does not exert a significant influence on future expectations of financial distress. 17 Concerns may remain about endogeneity or about the validity of instruments. Overall, the baseline coefficients presentedwith and without controlsin Table 2 are relatively stable, particularly in the case of FINNOW (Panel B) and FINFUT (Panel C). Whilst coefficient stability has been used as an indication of the limited influence of omitted variable bias (Altonji et al., 2005) , Oster in 2019 acknowledges that this argument overlooks the concomitant movements (or lack thereof) in the R 2 i.e. whether (or not) the controls are informative. Utilising movements in coefficients and in the R 2 , Oster (2019) formalises an approach that exploits the relative degree of selection on observed and unobserved variables to evaluate the pervasiveness of omitted variables bias in linear models. We therefore implement Oster's approach to further assess the robustness of the baseline results to selection on unobserved variables. Oster (2019) defines the relative degree of selection on observed and unobserved variables as δ and equates this to unity if the observed variables are of equal importance to those unobserved. This is an innocuous assumption if the observed variables have been carefully collected based on the relevant literature and given that their inclusion partitions out their effect captured by the unobserved variables. Therefore, we set δ = 1. 18 In addition, Oster (2019) proposes that whilst the R 2 has a limit of one, practically, due to measurement error, its theoretical maximum (R MAX 2 ) is likely to fall below unity. Appealing to the survival rate of experimental studies in top journals, upon applying her bounding approach, Oster (2019) . 19 β* can be estimated as: β denotes the sample estimate of β using Eq. (1) (setting δ = 0). Respectively, β and Ṙ 2 represents the sample estimate of β and the coefficient of determination obtained from specification (1) without controls. The bounding set contains the true β, therefore if zero falls within this bound the causal effect can be interpreted as non-statistically significant. Table 4 presents the bounding sets. For comparison purposes, the baseline estimates β (setting δ = 0) are taken from the regressions with controls as presented in Table 2 . Oster's approach consistently provides a lower bound to our baseline results for current measures of financial distress (BEHINDBILLS, FINNOW). In contrast, an upper bound is established relative to the baseline estimates for expectations of future financial distress (FINFUT). All point estimates are statistically significant at least at the 5% level and the bounding sets do not contain zero. In addition, Table 4 presents the estimated δ that would be required to force the causal effect to be zero. This is positive for current measures of financial distress (BEHINDBILLS, FINNOW), consistent with downward bias, and ranges between 2 and 4.1. In contrast, δ is negative for FINFUT, in line with the upper bound estimated. In two out of three cases |δ| exceeds 20 and 80. Therefore, altogether, since it is unlikely that the selection on unobserved variables is between 2 and 80 times greater than the observed variables, and the bounded sets do not contain zero, the baseline results can be interpreted as robust to selection on Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors in parentheses. All models (first and second stage) include economic and socio-demographic controls and regional/time fixed effects (Table A1 , Appendix A). FP denotes fuel poverty. M, F and FM refer to marginal prices, fixed charges and the fixedmarginal ratio respectively. The most relevant pairs (the largest F-Statistic reported in the first stage regressions) out of the valid IVs (J (p-value) > 0.1 in the second stage regressions) are presented here (see Table A2 , Appendix A, for complete table of IV results). a Future financial situation time horizon is "a year from now". 16 The findings hold using models that correct for potentially weak instruments including the limited information maximum likelihood (LIML) and jackknife IV estimators (see e.g. Angrist et al., 1999) . For brevity these results are available upon request. 17 We also implement the IV estimators whilst balancing the covariates using inverse-propensity score weighting to further assess potential selection bias (Aizer and Doyle Jr., 2013). Table A4 shows estimates consistent in significance, albeit smaller in magnitude, with the main IV results (Table 3) . 18 Otherwise, if the unobserved variables are of greater (lesser) importance than the observed in explaining the outcomes then δ > 1 (0 < δ < 1). On average, Oster (2019)'s examination of studies published in top journals found δ < 1 hence setting δ = 1 provides a more conservative approach. 19 The converse is true for β<0. unobserved variables. Moreover, it is important to note that our IV estimates are consistent if we relax the underlying assumption that the bias arising from unobserved variables is in the same direction as the observed variables (or the size of the bias is so small the overall direction of bias is unphased). Table 4 presents the estimate value of β* upon relaxing this assumption. The significant IV estimates (Panels A and B) fall within the upper bound. As with the IV estimates, there is evidence to suggest that fuel poverty has a deleterious impact on current measures of financial distress (BEHINDBILLS, FINNOW) yet may not alter expectations of future financial distress since the FINFUT bounding sets include zero. In addition, the results from the main survey remain robust upon further sensitivity checks (Table 5) . We assess whether fuel poverty has a persistent effect on financial distress by including the lag (t− 1) of fuel poverty indicatorsthis represents fuel poverty in main survey Wave 9 (January 2017-May 2019). Table 5 (Column 1) shows, as one may expect, that the coefficients are generally smaller than in the 'static' models, not least because the impact of fuel poverty is somewhat attenuated over time. The findings related to falling behind on bills and current finances remain statistically significant. Lags of fuel poverty provide some additional assurance that the direction of the effect flows from fuel poverty to financial distress rather than vice versa. 20 Like in the IV results, the relationship between fuel poverty and expectations of future financial distress is attenuated. This is a further indication that baseline findings relating fuel poverty to FINFUT may be picking up confounding factors. To further assess potential confounding variables, we draw upon two additional sets of controls: 1) subjective well-being (SWB) and psychological distress (PD) (Table 5, Column 2); and 2) the Big 5 personality traits (Table 5 , Column 3). The variable descriptions are detailed in Table A6 . We examine whether the relationship between fuel poverty and financial distress is mediated by levels of psychological distress and life satisfaction. On the one hand, self-reported financial distress has been associated with psychological distress during the COVID-19 pandemic (Davillas and Jones, 2020) and life satisfaction prior to and during the financial crisis (Keese, 2012; Arampatzi et al., 2014) . On the other hand, as noted in Section 1, fuel poverty has been reported to affect subjective measures of health and well-being. The findings presented in Table 5 (Column 2) show that the impact of fuel poverty on current measures of financial distress (BEHINDBILLS, FINNOW) remains statistically significant (Panels A and B). Whilst the link between objective indicators of fuel poverty and expectations of future financial distress (FINFUT) are mediated and consistent with the conclusions drawn from the IV estimates, the relationship remains statistically significant for the subjective indicator of fuel poverty. Table 5 (Column 3) utilises data contained in the UKHLS Wave 3 (January 2011-May 2013), the only UKHLS survey containing the Big 5 personality traitsagreeableness, conscientiousness, extraversion, neuroticism and openness. The Big 5 personality traits are considered important factors for economic outcomes, including financial distress (Xu et al., 2015; Parise and Peijnenburg, 2019; Liao, 2020) . Unlike SWB and PD, these controls can be considered exogenous as they are Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors in parentheses. Standard errors in square brackets are bootstrapped for 1000 replications. δ and β* are estimated using Oster (2019)'s psacalc Stata Code. All models include economic and socio-demographic controls and regional/time fixed effects (Table A1 , Appendix A). a Future financial situation time horizon is "a year from now". Table 5 Baseline specification checks of (LPM) regressions of financial distress on indicators of fuel poverty: UKHLS Main survey (Jan/2018-Feb/2020). Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors in parentheses. All models include economic and socio-demographic controls and regional/time fixed effects (Table A1 , Appendix A). Specifications: (1) lags fuel poverty in the baseline model (Eq. (1)); (2) adds subjective well-being (SWB) and psychological distress variables (PD) to the baseline model (Eq. (1)); and (3) adjusts the baseline model by including the Big 5 personality traits using answers provided in UKHLS Wave 3 -see Table A6 for variable definitions. a Financial future situation time horizon is "a year from now". generalisable across the life course (Xu et al., 2015) . 21 The baseline findings hold upon inclusion of the Big 5 personality traits. Overall, there is some evidence to suggest that the link between objective indicators of fuel poverty and FINFUT is attenuated by measures of subjective well-being and psychological distress. In contrast, there is no evidence to suggest this is the same for subjective indicators of fuel poverty. Hence, in light of the IV results, unobserved factors (e.g. internal temperatures) and/or self-assessed measurement error(s) may be driving the baseline association between subjective fuel poverty and expectations about future financial distress. Indeed, there is an argument for the inclusion of non-financial factors in order to subvert potential biases related to self-reported measures of financial distress (Keese, 2012; Kellstedt et al., 2015) . However, since the literature discussed earlier has established a causal link between fuel poverty and health outcomes (see e.g. Awaworyi Churchill et al., 2020; Kahouli, 2020) , these controls (SWB and PD) are clearly endogenous and such specification checks should be viewed with caution. To investigate the relevance of fuel poverty during the current COVID-19 pandemic, we employ UKHLS' COVID-19 web surveys (University of Essex, 2021). We rely on the surveys which take place in April, May and July 2020 as those carried out in June and September 2020 do not contain measures of financial distress. It is important to note that these surveys map onto the peak, decline and trough of the first wave of the pandemic. The number of admissions to hospital peaks at 3,115 patients (7-day average) on 4th April 2020, followed by the 7-day average falling to 1,199 patients on the 4th May 2020, which then starts to approach the trough of admissions by 4th July 2020 with numbers falling further to 216 patients (HM Government, 2021) . The 4th July 2020 coincides with the easing of national lockdown restrictions in the UKfor example, salons and beauty services reopen on 13th July 2020 and the use of public transport for non-essential journeys is permitted by 17th July 2020. The questions underpinning BEHINDBILLS and FINNOW are identical to the main survey. The time horizon for FINFUT changes from 1 year to 1 month. Table 6 shows that the proportion of individuals experiencing financial distress declines from April to July 2020 in line with the pandemic's first wave coming to an end. Individuals are identified as fuel poor based on their responses and information contained in the main survey data (Table 6 ). Although the COVID-19 surveys do not contain income or expenditure information, this approach allows us to explore whether those individuals identified as fuel poor prior to the pandemic are more likely to experience financial distress during the pandemic. In the COVID-19 regressions, we include time effects that represent the year in which the individual participates in the main survey (Wave 10) in order to control for annual variation in energy bills, income and therefore fuel poverty. The proportion of individuals we identify as fuel poor in the main survey are similar across the April to July 2020 samples (Table 6 ). This is supported by the notable stability in the economic and sociodemographic statistics collected from the COVID-19 surveys (Table A7 , Appendix A). The controls collected for the baseline results in the COVID-19 surveys matches those specified in Eq. (1) with the addition of a variable controlling for individuals mandated to stay at home in accordance with the UK's Coronavirus Job Retention Scheme (CJRS) ( Table A7 , Appendix). This is crucial since CJRS helps facilitate the transition into lockdown during the first wave of the pandemic, supporting the households' adjustment to the changes in living and working arrangements at home. Figs. 1A-1C (Table A8 , Appendix A) below present the COVID-19 surveys' lower bound (BEHINDBILLS, FINNOW) and upper bound (FIN-FUT) according to Oster 's (2019) approach as outlined in Eq. (3). Similar to the results for the main survey, we generally find that fuel poverty continues to exert a positive influence over financial distress during the pandemic. We also observe a similar pattern in terms of the objective fuel poverty indicators exhibiting smaller effects than the subjective indicator. 22 Nonetheless, the confidence intervals presented in Figs. 1A-1C suggest that the differences across the first wave of the pandemic and prior to the pandemic (main survey) are statistically insignificant. During these months, the cost of changes in electricity consumption attributable to working at home could be partly recovered by claims for tax relief for additional work-related expenses (around £6/ week). Moreover, expenditure on energy and other necessities is indirectly supported through the UK's Coronavirus Job Retention Scheme for workers on furlough, which paid 80% of the regular wage of employed individuals (up to £2500/month). Whilst these schemes provide further assurance that the energy bills and income information used herein are relevant to the first wave of the COVID-19 pandemic, they potentially worked effectively to dampen the financial impact on those identified as fuel poor, relative to those not in fuel poverty, prior to the pandemic. As a final robustness check, we restrict the main survey to individuals participating in the COVID-19 May 2020 survey (Table A5 , Column 2, Appendix A). 23 There is a stark similarity in the economic and statistical significance of the coefficients in Table A5 (Column 2) and those from the main survey (Table 2 , Even Columns). This helps to avert concerns that the overlap in the findings prior to and during the pandemic could arise from attrition or potential changes in the sample composition in the COVID-19 surveys. Fuel poverty is an increasingly relevant dimension of social a Future financial situation time horizon is "a month from now". 21 The key pitfall arises from attrition as the number of observations decreases by 6000 individuals, therefore this specification is used as a robustness check rather than a baseline finding. We also used numerical cognitive and verbal ability data taken from Wave 3 (see e.g. Xu et al., 2015; Liao, 2020) , however these variables are non-generalisable across one's life course. Nonetheless, the baseline results remain intact upon their inclusion and are available upon request. 22 Moreover, concerns surrounding the impact of a change in time horizon is alleviated by the fact that the relationship between fuel poverty and FINFUT is similar to the main survey by the end of the first wave of the pandemic. 23 Whilst the results are robust when restricting the sample to individuals participating in either April, May or July 2020, only the results for May 2020 are presented in the Appendix for brevity. Results for April and July 2020 are available upon request. deprivation which is observed and monitored in many high-income countries where economic inequality is persistent or even growing. In most of these countries policy measures are in place to reduce the extent and the effects of this social inequity. These policies have achieved mixed results in the past due to the complex and multidimensional nature of the issues being addressed by policy makers. The adoption of well targeted and effective policy measures aims at tackling fuel poverty and its effects on the mental and physical wellbeing of the individuals who are affected by it will be even more important during the economic recovery from the current pandemic, as many households will have suffered losses or reductions in income and potentially also increases in expenditure due to the effect of lockdowns on mobility and travel. This paper investigates the relationship between fuel poverty indictors (both objective and subjective) and self-reported measures of financial distress. While fuel poverty in itself is a source of concern in society, its broader effects are also concerning due to their potential long-term effect on health and wellbeing. The literature on fuel poverty, which has been briefly discussed in the paper, has identified a link between fuel poverty and health outcomes and has suggested two potential pathways through which the link can be established. On the one hand the "living conditions" pathway could impact health, via anxiety and depression or as a result of insufficient thermal comfort. On the other hand, the "financial security" pathway can affect individuals' wellbeing as a result of financial stress. This latter relationship is investigated empirically in our paper based on the responses to nationally representative surveys of GB held between January 2018 and February 2020. The responses to surveys run between March and May 2020 are instead used to extend the analysis to the early phases of the Covid-19 pandemic. The paper therefore offers an original contribution to knowledge by investigating intermediate links within the recognised relationship between fuel poverty and health and wellbeing outcomes, via the role of financial distress. Our results are obtained using econometric methodologies aimed at dealing with the effects of potential sources of endogeneity. Our results have identified a statistically significant and positive relationship between objective and subjective measures of fuel poverty and current situations of financial distress among fuel poor households. The link between fuel poverty and expectations about future financial circumstances however is less statistically robust. Our results are confirmed, but not necessarily, strengthened for the Covid-19 period. Hence, according to our instrumental variable estimates, those identified as fuel poor find managing their current finances more difficult yet are no more likely to think that their financial situation will be worse off, in the future, than those who are not considered as fuel poor. This finding accords with scarcity theory, which predicts that poverty leads to reinforcing behaviour (e.g. overborrowing), since "attention is allocated to the most pressing financial problems and needs. Future needs loom far away." (de Bruijn and Antonides, 2021: 10) . Whilst scarcity increases focus on limited resources, attentional focus on pressing present outgoings (e.g. utility expenses, groceries, rent) may come at the expense of neglecting future outgoings (Shah et al., 2012; Shah et al., 2018) . This line of thought is consistent with (but does not necessarily imply) low-income consumers behaving as if they employ larger intemporal discount rates than high-income consumers (Train, 1985; Lawrance, 1991; Shah et al., 2012; de Bruijn and Antonides, 2021) . The key policy implications of our empirical analysis are that the evaluation of the effectiveness and potential benefits of policy measures aimed at addressing situations of fuel poverty should be assessed by taking into consideration the avoidance of, or reduction in, financial distress among fuel poor households, with indirect individual and societal benefits in terms improved health and wellbeing outcomes. While fortunately the impact of the first wave of the COVID-19 pandemic does not seem to have significantly worsened the situation of financial distress among fuel poor households, this may be due to the extraordinary support measures put in place by the Government and the energy regulator in order to mitigate the worst financial effects of the pandemic, including a furlough scheme and a ban on evictions and disconnections. It is therefore important that any future policy of recovery from the pandemic continues to shelter these vulnerable individuals in order to make sure that any adverse impact of financial distress and eventually health has not simply been delayed through the existing measures. Indeed, National Energy Action (2020b) has argued for utility debt reform in order to protect households, energy suppliers and the economy from the "gathering storm" of utility debt that has been either been exacerbated or newly accrued during the pandemic. Looking more broadly to the energy and environmental policy landscape, it is important to point out that the recently adopted net zero objectives and the associated strategies aimed at meeting them need to take into account the potential implications for individuals who find themselves in fuel poverty or are at risk of it. Indeed, the ambitious environmental objectives currently being adopted by many countries might actually increase the risk of excluding parts of society from access to affordable fuels and appliances, or even of eliciting the exploitation of the most vulnerable in society if they are unable to take advantage of the sustainable and energy efficient technologies that will make the achievement of those objectives possible. Control variable definitions and summary statistics UKHLS: main survey (Jan/2018-Feb/2020). Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors in parentheses. All models (first and second stage) include economic and socio-demographic controls and regional/time fixed effects. a Future financial situation time horizon is "a year from now". Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors in parentheses. All models (first and second stage) include economic and socio-demographic controls and regional/time fixed effects. a Future financial situation time horizon is "a year from now". IV (LPM) regressions of financial distress on indicators of fuel poverty using prices (M, F, FM) between 2016 and 2018 using inverse-propensity score weighting: UKHLS main survey (Jan/2018-Feb/2020). Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors in parentheses. All models (first and second stage) include economic and socio-demographic controls and regional/time fixed effects. a Future financial situation time horizon is "a year from now". Baseline specification checks of (LPM) regressions of financial distress on indicators of fuel poverty: UKHLS Main survey (Jan/2018-Feb/2020). (1) Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors in parentheses. All models include economic and socio-demographic controls and regional/time fixed effects. Specifications: (1) reintroduces 71 individuals participating in the main (Wave 10) survey during the COVID-19 pandemic; and (2) restricts the sample to only include participants of the COVID-19 May survey. a Financial future situation time horizon is "a year from now". Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors in parentheses. Standard errors in square brackets are bootstrapped for 1000 replications. δ and β* are estimated using Oster (2019)'s psacalc Stata Code. All models include economic and socio-demographic controls and regional/time fixed effects. a Future financial situation time horizon is "a month from now". We match gas and electricity average retail marginal prices and fixed charges, collected annually for each GB region by the Department of Business and Industrial Strategy (BEIS, 2021b (BEIS, , 2021c , to individuals in our UKHLS sample. Table B1 presents the time, regional and payment method matching process. As discussed in Section 2, the year individuals participated in the main survey (2018-2020) is either matched to prices from the current year(s) (2018-2020) or matched using prices from 2 years prior (2016-2018). Prices are matched by geographical region. For the most part, this is a straightforward match between the 14 regional distribution networks and 12 government office regions (Table B2 ). In the case of Scotland and Wales, the arithmetic mean of North/South sub-regions is used. Whilst the Northern Wales distribution network also extends across Merseyside, we do not believe this negatively affects the overall results based on the matching process. Individuals can pay by credit (i.e. the default standard variable supplier and/or tariff), direct debit (i.e. a fixed or variable tariff allocated after switching supplier and/or tariff) or prepayment (i.e. pay-as-you-go typically using a key card or token). UKHLS does not declare as to whether electricity consumers use time-of-use (Economy 7) tariffs. Nonetheless, the payment methods remain the same for Economy 7 consumers of whom represent only 6% of meters in Wales and 14% of meters in England and Scotland (BEIS, 2020b). Credit prices are matched to those paying each quarter/year (the default method) and other non-standard methods of payment (including frequent cash payments, government schemes). Direct debit prices are allocated to those paying a fixed amount each month by standing order or monthly by direct debit. Prepayment prices are allocated to consumers who pay-as-they-go using a prepaid key, card or token (Table B1 ). Other configurations of credit and debit prices reveal consistent findings but perform weaker as instruments (i.e. less correlated with the fuel poverty indicators). Table B2 presents the gas and electricity average retail marginal prices, fixed charges, and fixed-marginal ratio. There are 99 prices in total as we have 11 regions, 3 years and 3 payment methods. The proportion of total variation in prices explained by within-region variation (i.e. the R 2 ) is estimated using a simple linear regression of prices on a vector of regional indicators. b BEIS does not collect gas price data for NI, therefore GB only. We use the most recent median typical domestic consumption values (BEIS, 2021b (BEIS, , 2021c . Notes: N = 11 (regions) × 3 (years) ×3 (methods of payment). Gas (G) and electricity (E) pricesmarginal (P), fixed (F) and fixed-marginal ratio (FM). All statistics are adjusted to 2016 prices using the retail price all items index (ONS, 2021). BEIS UKHLS Year Current prices ➔ Interview year Lagged prices ➔ Interview year Juvenile incarceration, human capital and future crime: evidence from randomly-assigned judges Selection on observed and unobserved variables: assessing the effectiveness of catholic schools Coping with low incomes and cold homes Jackknife instrumental variables estimation Financial distress and happiness of employees in times of economic crisis Subjective financial well-being, income and health inequalities in mid and later life in Britain Energy poverty and health: panel data evidence from Australia Fuel poverty and subjective wellbeing Fuel Poverty: From Cold Homes to Affordable Warmth Does education loan debt influence household financial distress? An assessment using the 2007-2009 Survey of Consumer Finances panel From residential energy demand to fuel poverty: incomeinduced non-linearities in the reactions of households to energy price fluctuations Hidden Debts the Growing Problem of Being behind on Bills and in Debt to the Government COVID-19: The Impact on Energy Bills and Fuel Poverty The health benefits of a targeted cash transfer: the UK winter fuel payment Nonlinear pricing and tariff differentiation: evidence from the British electricity market The COVID-19 pandemic and its impact on inequality of opportunity in psychological distress in the UK. Health Econ Poverty and economic decision making: a review of scarcity theory A postcode lottery? Regional variations in electricity prices for inactive consumers. CCP Working Paper 18-10 Fuel poverty: potentially inconsistent indicators and where next ), 2020b. Subnational Electricity and Gas Consumption Statistics. HM Government Average Unit Costs and Fixed Costs for Gas for GB Regions (QEP 2.3.4). HM Government Average Unit Costs and Fixed Costs for Electricity for UK Regions (QEP 2.2.4). HM Government The economic consequences of hospital admissions Outsized impacts of residential energy and utility costs on household financial distress Student debt effects on financial well-being: research and policy implications The long-term distributional and welfare effects of Covid-19 school closures Debt and depression: causal links and social norm effects Psychosocial routes from housing investment to health: evidence from England's home energy efficiency scheme Warm Front Better Health: Health Impact Evaluation of the Warm Front Scheme Cold homes, fuel poverty and energy efficiency improvements: a longitudinal focus group approach Keeping warm and staying well: findings from the qualitative arm of the Warm Homes Project Fuel poverty: the problem and its measurement. CASE Rep Getting the measure of fuel poverty: final report of the Fuel Poverty Review Fuel Poverty. Briefing Paper, Number 8730 Coronavirus (COVID-19) in the UK The effect of the affordable care act Medicaid expansions on financial wellbeing Institute of Fiscal Studies (IFS), 2020. Household spending and coronavirus An economic approach to the study of the relationship between housing hazards and health: the case of residential fuel poverty in France Who feels constrained by high debt burdens? Subjective vs. objective measures of household debt The usefulness of consumer sentiment: assessing construct and measurement Fuel poverty as a major determinant of perceived health: the case of France Poverty and the rate of time preference: evidence from panel data Valid T-Ratio Inference for IV (Working Paper) Measuring fuel poverty in France: which households are the most fuel vulnerable? ADHD symptoms and financial distress Living in a Cold and Damp Home: Frameworks for Understanding Impacts on Mental Well-Being Fuel poverty and human health: a review of recent evidence Objective vs. subjective fuel poverty and self-assessed health The Health Impacts of Cold Homes and Fuel Poverty. Friends of the Earth and the Marmot Review Team The effects of the Massachusetts health reform on household financial distress Household energy efficiency interventions: a systematic literature review Necessity or Luxury Good? Household Energy Spending and Income in Britain Definitions of fuel poverty: implications for policy National Energy Action (NEA), 2020b. The Gathering Storm: Utility Debt and COVID-19 Office for Gas and Electricity Markets (Ofgem), 2015. Regional Differences in Network Charges. Ofgem Consumer Vulnerability Strategy 2025 Office of National Statistics (ONS), 2021. RPI all Items Index Gauging energy poverty: a multidimensional approach Household financial distress and initial endowments: evidence from the 2008 financial crisis Health and thermal comfort: from WHO guidance to housing strategies Unobservable selection and coefficient stability: theory and evidence Noncognitive abilities and financial distress: evidence from a representative household panel Minimum Home Temperature Thresholds for Health in Winter: A Systematic Literature Review Addressing self-disconnection among prepayment energy consumers: a behavioural approach Experimental Analysis of the Impact of COVID-19 on Fuel Poverty Rates Some consequences of having too little Money in the mental lives of the poor Instrumental variables regression with weak instruments The Coronavirus Could Devastate Poor Countries: It is in the Rich World's Self-Interest to Help The End Fuel Poverty Coalition writes to the Prime Minister Quantification beyond expenditure Health effects of housing improvement: systematic review of intervention studies Health, well-being and energy poverty in Europe: a comparative study of 32 European countries Discount rates in consumers' energy-related decisions: a review of the literature php?id=37072&src=%E2%80%B9%20Consumption% 20%20%20%20%20%20Residential%20Energy%20Consumption%20Survey%20 (RECS)-b4 Understanding Society: Waves 1-10, 2009-2019 and Harmonised BHPS: Waves 1-18 Objective and subjective measures of fuel poverty Racial disparities in energy poverty in the United States COVID-19 and income inequality in OECD countries Personality and young adult financial distress Genetic and environmental influences on household financial distress A multidimensional measure of energy poverty in China and its impacts on health: an empirical study based on the China family panel studies The authors thank Professor Catherine Waddams Price for the helpful suggestions, and the guest editor and two referees for very helpful comments. Supplementary data to this article can be found online at https://doi.org/10.1016/j.eneco.2021.105464.