Policy Research Working Paper 10827 Conceptualizing and Measuring Energy Poverty in Bulgaria Monica Robayo-Abril Britta Rude Poverty and Equity Global Practice June 2024 Policy Research Working Paper 10827 Abstract Addressing energy poverty has emerged as one of the main instruments; and (3) providing policy recommendations challenges for Bulgaria’s poverty and social inclusion policy, for the measurement and monitoring of energy poverty. particularly in the context of the European Green Deal and The way energy poverty is measured and the overlap with the current crisis in Ukraine. To tackle the adverse impacts income poverty shape the types of policy solutions per- of energy poverty effectively, a crucial initial step involves ceived to be possible and appropriate to address it. The accurately defining and measuring this issue. Identifying evaluation supports the need to shift from single indica- households affected by energy poverty is essential for shap- tors to multidimensional approaches in measuring energy ing and implementing targeted policies. This study explores poverty. Additionally, enhancing the granularity, quality, various definitions of energy poverty within the Bulgarian and frequency of expenditure and income surveys can con- context by (1) systematically reviewing current methodolo- tribute to easier operationalization of these concepts and a gies and measures employed in the EU context; (2) assessing better understanding of the demographics of energy poverty. the feasibility of implementing these measures in Bulgaria The study proposes exploring alternative data generation based on data availability, comparing the incidence of methods, such as smart meters, further to enhance insights energy poverty using alternative measures, and presenting into the dynamics of energy poverty. characteristics of energy poverty to inform potential policy This paper is a product of the Poverty and Equity Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at mrobayo@worldbank.org and brude@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Conceptualizing and Measuring Energy Poverty in Bulgaria1 Monica Robayo-Abril Britta Rude JEL classification: O13, Q40, I32 Keywords: energy poverty, fuel poverty, poverty, vulnerability, EU-SILC, HBS, Bulgaria 1 This paper was prepared as part of the Poverty Program for Bulgaria in the World Bank's Poverty and Equity Global Practice. The study was carried out by a team composed of Monica Robayo-Abril (Senior Economist, World Bank) and Britta Rude (Young Professional, World Bank). We thank Carolina Diaz Bonilla (Senior Economist, World Bank), Kiyomi Cadena (Consultant, World Bank), Karen Bazex (Consultant, World Bank), Desislava Nikolova (Senior Economist, World Bank) and Mariano Gonzalez Serrano (Senior Energy Specialist, World Bank) for their helpful comments and feedback. 1. Introduction Energy poverty is a pressing social inclusion challenge in Bulgaria, gaining heightened significance after the European Green Deal and Russia’s invasion of Ukraine. The European Green Deal, introduced by the European Commission in December 2019, aims to ensure an equitable transition towards a climate-neutral Union by 2050. This transition, however, poses economic and social challenges, particularly with the potential disproportionate impact of rising fossil fuel prices on vulnerable groups who allocate a significant portion of their income to energy expenses. In response to the social and distributional consequences stemming from the cost impacts of decarbonization measures, the European Commission established a Social Climate Fund in 2021 2. Recent developments, such as the war in Ukraine, have led to steep increases in energy prices across most EU countries, including Bulgaria. Consequently, energy poverty, broadly understood as the inability of a household to attain a socially and materially required level of energy service, has emerged as a foremost and urgent concern in Bulgaria's social and economic inclusion efforts. Several directives in the EU legislative framework call for measures to address energy poverty 3; moreover, EU Member States are mandated to evaluate and tackle it within their national Energy and Climate Plans (NECPs). 4 If many households face energy poverty, Member States must include policies in their NECPs to mitigate it. Directives like the Recast Electricity Directive and Directive 2009/73/EC 5 require measures to combat energy poverty, protect vulnerable customers, and assess affected households. The Energy Efficiency Directive 6 prioritizes energy poverty considerations in energy efficiency obligations. The revised Energy Performance of Buildings Directive 7 mandates national measures to alleviate energy poverty in long-term renovation strategies. Strategies for achieving 2030 and 2050 energy efficiency targets should prioritize shielding energy poor households. The EU legislative framework ensures that addressing energy poverty aligns with smooth market operations, emphasizing the importance of fair retail markets for a smooth transition. The European Pillar of Social Rights underscores universal energy access 8, aligning with the December 2019 European Green Deal's goal of a fair transition to a climate-neutral Union by 2050. The Renovation Wave initiative, adopted concurrently, reinforces the call to combat energy poverty within the 'Clean Energy for All Europeans' legislative framework. Regulation (EU) 2018/1999 and the Recast Electricity 2 The Social Climate Fund is a financial tool established to address the social challenges and disparities arising from the transition to a green economy, including those exposed to energy poverty. 3 This includes the following directives: 1) The recast Electricity Directive and Directive 2009/73/EC, which are regulatory instruments within the EU framework addressing energy policy and market functioning; 2) The Energy Efficiency Directive, which aims to enhance energy efficiency and reduce energy consumption across the EU, and 3) the revised Energy Performance of Buildings Directive, which focuses on improving the energy performance of buildings within the EU. 4 Energy and Climate Plans (NECPs) are strategic documents outlining each EU Member State's energy and climate policies and targets. 5 The recast Electricity Directive and Directive 2009/73/EC are regulatory instruments within the EU framework addressing energy policy and market functioning. 6 The Energy Efficiency Directive aims to enhance energy efficiency and reduce energy consumption across the EU. 7 The revised Energy Performance of Buildings Directive focuses on improving the energy performance of buildings within the EU. 8 As shown in section II, access to energy can be measured at the extensive margin (having access to energy or not- a binary indicator) and the intensive margin (characteristics related to quality and reliability). 2 Directive mandate the Commission to guide the measurement of energy poverty and define a 'significant number of households in energy poverty. 9 To tackle energy poverty, EU Member states must formulate a national definition and criteria tailored to their context; the Bulgarian government has recently published an official definition. To meet its legal obligations, the Bulgarian government established an inter-ministerial working group 10 to work on changes to the Energy Act 11. The group has recently published an official definition of energy poverty and criteria for identification and assessment, endorsed by the Council of Ministers and the European Commission. These tools encompass short-term measures shielding energy- vulnerable households from rising prices and policies to enhance medium-term energy efficiency. Moreover, within the Bulgaria Recovery and Resilience Plan's (RRP) Low Carbon Economy Chapter, 12 approved by the European Commission in April 2022, Reform 3 focuses on defining energy poverty. This definition aims to facilitate electricity market liberalization and ensure priority treatment for energy poor households in energy efficiency measures and projects. The RRP allocates 59 percent of its funds to climate and clean energy support, covering decarbonization measures and interventions promoting energy efficiency. The challenge in measuring energy poverty begins with the absence of a common definition. The EU introduced the concept through the Directive on Common Rules for the Internal Electricity Market (2009/72/EC) and expanded it within a fair energy transition context. In 2020, the European Commission defined energy poverty as households unable to access essential energy services. 13 According to the EC, it arises when energy bills constitute a significant share of household income or when households curtail energy consumption, negatively impacting health and well-being (European Commission, 2023). The 2023 Social Climate Fund regulation and the revised Energy Efficiency Directive further define energy poverty as a lack of access to essential energy services, encompassing heating, hot water, cooling, lighting, and energy for appliances, tailored to the national context and existing social policies. Factors contributing to energy poverty include affordability issues, insufficient disposable income, high energy expenditure, and poor home energy efficiency. Alternative definitions, like the one used by the United Nations (United Nations, 2018), emphasize the lack of access to high- quality energy services. More recently, scholars have incorporated sustainability aspects, introducing energy efficiency and access to renewable energy into the concept of energy poverty (Siksnelyte- Butkiene, 2021). Measuring energy poverty presents challenges, yet there is a critical demand for accurate indicators at the country level. The complexity of energy poverty requires indicators that can capture the multifaceted nature of the issue. The urgency for precise measures has grown, propelled by the rising impact of climate change and the heightened push for a rapid, eco-friendly transition, particularly in light of the recent Ukraine crisis. The initial step involves defining SMART (Specific, Measurable, Achievable, Relevant, and Time-bound) indicators to monitor energy poverty and assess policy 9 Source: Commission Recommendation (EU) 2020/1563 of 14 October on energy poverty. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32020H1563&qid=1606124119302 10 The group includes the Minister of Labor and Social Policy, the Minister of Energy and the Minister of Regional Development and Public Works. 11 See updated Energy Act: https://dv.parliament.bg/DVWeb/showMaterialDV.jsp?idMat=201222 12 The Low Carbon Economy Chapter of the Bulgaria Recovery and Resilience Plan (RRP) focuses on climate goals and the use of clean energy, directing a significant portion of funds to related measures. 13 Source: Commission Recommendation (EU) 2020/1563 of 14 October on energy poverty. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32020H1563&qid=1606124119302 3 interventions promoting cleaner and more efficient energy use. A robust approach to measuring energy poverty enables policymakers to pinpoint those most in need and track progress over time. Furthermore, energy poverty has multiple drivers, including low income, poor energy efficiency, and high energy prices. As explored further in section 2, energy poverty manifests in various forms, including a lack of energy access due to power shortages or the inability to adequately heat one's home. Thomson et al. (2017) define energy poverty as inadequate access to home energy services, encompassing heating, cooling, lighting, and appliance use. The Energy Poverty Advisory Hub (EPAH) 14(2022) categorizes the drivers of energy poverty into contextual and individual factors, identifying low income, low energy efficiency, and high energy prices as overarching causes. The consequences of energy poverty on human welfare can be severe in both the short and long term (ibid). Specific population groups may experience varying impacts based on socio-demographic characteristics, household composition, health, energy literacy, and cultural factors. This paper systematically reviews existing literature on energy poverty measurement, assesses the suitability of potential measures for Bulgaria, and constructs feasible measures based on the availability and quality of existing micro datasets. This involves analyzing measures previously established in academic literature, white papers, and policy reports, focusing on the EU context. Then, utilizing available household-level microdata, we analyze the feasibility of implementing these indicators in Bulgaria. The examination involves several indicators using household expenditure data from the Household Budget Surveys and income and housing conditions from the European Living Condition Survey (EU-SILC). When selecting appropriate indicators, Palma and Gouveira (2022) recommend considering three key factors—context, scale, and availability. Additionally, we experiment with the potential use of big data, including satellite data. Subsequently, we compare these measures, highlighting their respective strengths and limitations. The main objective is to inform the Bulgarian government's efforts in measuring, operationalizing, and addressing energy poverty. Additionally, the paper examines the correlation between energy poverty and income poverty, 15 considering its policy implications. Our analysis involves characterizing energy poverty and identifying overlaps with monetary poverty to understand potential intersections between these vulnerable groups. Recognizing these overlaps is crucial for designing effective instruments to mitigate energy poverty. This includes measures to protect the energy poor and vulnerable populations in the face of rising energy prices or other forward-looking policy interventions addressing energy inefficiencies. Our results suggest substantial variations in the incidence rates of energy poverty depending on the chosen definition. Notably, the so-called 10 percent threshold, which defines the energy poor population as all those reporting an energy expenditure share in household income or expenditure of more than 10 percent, exhibits a significant deviation from other expenditure-based indicators, resulting in an incidence rate of approximately five times higher. This deviation could be attributed to measurement issues, and considerable criticism can be found in the existing literature on this indicator. Although this criticism is valid, the 10 percent measure is one of the few monetary measures of energy poverty that allow for comparisons across countries and over time. The remaining monetary-based measures show fluctuations ranging from 7.9 to 9.9 percent. Consensual measures, which rely on self- 14 EPAH is a leading EU initiative aiming to eradicate energy poverty and accelerate the just energy transition of European local governments, which replace the previous EU Energy Poverty Observatory. 15 Following the European Union standard, official poverty is measured by using relative-income poverty lines in Bulgaria. The at-risk- of -poverty rate (AROP) is the share of people living below the national poverty line, set at 60 percent of the equivalized median disposable income after social transfers. This is different from other upper- middle-income countries outside the EU, which use absolute measures to track monetary poverty. 4 reported energy deprivations, demonstrate a wide variation, between 11 and 47 percent, depending on the metrics used. Our analysis also reveals that, except for one measure, the monetary poor and energy poor only overlap partially; this finding has important implications for policymaking. While the 10 percent measure has a considerable overlap of the income and energy poor, this relationship is not necessarily valid for the other expenditure-based measures. Depending on the measure employed, between 3.1 and 5.3 percent of the population experiences monetary and energy poverty. This finding has implications for policy interventions targeting poverty. The existing social protection scheme, which often relies on monetary definitions of poverty to develop targeting schemes, might not necessarily cover those who are energy poor. Separate targeting schemes should be developed to address energy poverty. Furthermore, we identify that different population subsets face distinct dimensions of energy poverty, further making the case for targeted interventions. We show that different population subsets differ in different aspects of energy poverty. For example, the population reporting leakages, one important indicator of energy efficiencies and energy poverty, does not generally coincide with the population reporting arrears on utility bills. Consequently, the government may need to consider different population segments for short- and long-term interventions targeting energy poverty through different types of policy interventions. Based on this assessment and recommendations established in the existing literature, we recommend an approach that moves away from relying on single poverty indicators and instead tracks a set of indicators, as done in other European countries with national Energy Observatories (Belgium, France). One possibility would be to adopt the approach used by previous initiatives, such as the Belgium Energy Observatory and the French National Energy Poverty Observatory (ONPE), which monitor multiple indicators. This also aligns with recommendations in the recent academic literature and the European Commission. Finally, multisector approaches are needed to measure and tackle energy poverty, given its multidimensional nature. Given its multidimensional nature, addressing energy poverty involves many stakeholders, including energy, transport, infrastructure, and social sectors, such as social protection, health, and education. It also requires interconnected approaches at different geographic levels, ranging from EU-wide policies and comparisons to monitoring at the local level. Given the multidimensionality of energy poverty, this note informs several public policies, including energy, social, and housing policy instruments. As such, a broad list of stakeholders is required to define and monitor energy poverty. The operationalization of energy poverty measures should involve a collaborative effort among various ministries, academic institutions, and local communities. Including marginalized and poor population groups in consultations that flow into developing a system that tracks, measures, and analyzes energy poverty is crucial. The paper is organized as follows. Section 2 presents the different concepts and related measurements of energy poverty in the EU. Section 3 constructs some of these measures for Bulgaria, based on data availability, and characterizes the energy poor. Section 4 concludes and provides policy recommendations. 5 2. Overview of Approaches for Measuring Energy Poverty in the EU 2.1. Conceptualizing Energy Poverty There is currently no consensus in the literature on how to best measure energy poverty due to a lack of a common understanding of energy poverty and a lack of a standardized set of metrics. There is no consensus on how to best measure energy poverty. The inexistence of a common measure has several reasons. First, there is no consensus on the basket of basic energy services (Culver, 2017). Second, the absence of such a measure might be partly related to the fact that energy usage varies significantly by country context and culture. Next, while the literature agrees that energy poverty should reflect a lack of energy services, it is unclear how to translate this concept into a common metric (ibid). Partly, this problem is related to the fact that energy services cannot be substituted for each other and that the poverty level for each energy service is arbitrary (ibid). Nevertheless, measures of energy poverty should be evidence-based and well-thought-through. What most papers agree on, however, is that energy poverty measures should not be chosen arbitrarily without reflecting on the shortcomings of each metric and the aspect of energy poverty, which is a specific indicator measure (Herrero, 2017). In the case of the European Union, Pye et al. (2017) conclude that the concept of energy poverty varies significantly across member states. They recommend that the European Commission take a leading role in achieving greater harmonization and less fragmentation in the EU concerning measuring and approaching energy poverty. Gouveia et al. (2022) divide energy poverty measures 16 in the EU into three types: expenditure- based, consensual indicators, and direct measures. Gouveia et al. (2022) take stock of energy poverty measures in the European Union and categorize them into three types: expenditure-based measures, consensual approaches, and direct measures (see Figure 1). Expenditure-based measures assess households' energy costs and compare them to absolute or relative thresholds. Consensual approaches are self-reported assessments of indoor housing conditions and/or the ability to meet necessities, such as keeping one's home warm in winter. These are then compared to relative values in the place of residence of a given household. Finally, direct measures measure the level of energy services (physical variables) and compare them to previously set standards, such as the reference comfort temperature by the World Health Organization (WHO) (Palma and Gouveira, 2022), or other indicators to determine the adequacy of energy services (e.g., room temperature). Figure 1: Types of energy poverty measures in the EU Expenditure-based Consensual approaches Direct measures •Energy costs of •Self-reported indoor •Level of energy services households housing conditions/Ability (e.g., heating; lighting) •Compared to absolute or to attain basic necessities •Relative to set standards relative thresholds •Relative to place of (e.g. reference comfort residence temperature) Source: Compilation of findings by Gouveia et al. (2022) When looking at factors that relate to energy poverty and vulnerability, Thomson et al. (2017) take a vulnerability approach to energy poverty and divide energy vulnerability into six factors. Thomson 16 Gouveia et al. (2022) limit their assessment to data from EUROSTAT, EU-SILC (European Union Statistics on Income and Living Conditions), HBS (Household Budget Survey), and BSO (Building Stock Observatory). 6 et al. (2017) approach energy poverty through a vulnerability lens and develop six factors that relate to energy poverty and vulnerability (Figure 2). First, households might face poor access to energy. However, in the context of EU countries, access to energy is usually not the main problem. Second, they might face constraints based on the limited affordability of energy. This constraint may be particularly binding in the current inflationary context because of the Ukraine conflict. Third, they might be subject to poor flexibility, which is the inability to switch to energy sources that are more appropriate to their needs. Fourth, energy efficiency is a potential vulnerability factor because households might experience high energy losses due to poor energy efficiency at home. Fifth, some households might not be able to meet their energy demands from a cultural or socioeconomic perspective. Lastly, households might suffer from a lack of or limited policies addressing energy poverty. With respect to measurement possibilities, Thomson et al. (2017) detail the same categories presented in Figure 1. Figure 2: Energy poverty through a vulnerability lens Access Affordability Flexibility Efficiency Needs Practices •Poor •Poor •Inability to •High loss of •Mismatch •Lack of availability affordability switch energy with needs policies Source: Compilation of findings by Thomson et al. (2017) In Table 1, we summarize different indicators measuring energy poverty gathered by Gouveia et al. (2022) and add several indicators identified from our review of the relevant literature. The table reveals that energy poverty indicators in the EU mainly rely on three different types of data sources: EU-SILC (European Union Statistics on Income and Living Conditions), HBS (Household Budget Survey), and BSO (Building Stock Observatory) (Gouveia et al., 2022). We add several indicators from our literature review, which we outline in the next subsections. Most indicators can be classified into three different measurement types (expenditure, consensual, and direct measurement), and some additional ones are more related to the building stock features and energy prices. Table 1: Different indicators measuring energy poverty Type of Availability – Indicator Definition Source Measure Bulgaria Indicators from Gouveia et al. (2022) Absolute (equivalized) household energy M/2 measure HBS Yes expenditure below half the median Expenditure- Share of household Based energy expenditure Yes, when (compared to disposable 2M measure HBS using HBS income/expenditure) income above twice the national median 7 Consumption expenditure on EUROSTAT Yes, but only Energy expenses electricity, gas, and other based on for 2005, by income fuels as a share of Harmonized 2010, 2015, quintile income-by-income HBS 17 and 2020 quintile Ability to keep Share of the population home adequately unable to keep home EU-SILC Yes warm adequately warm Arrears on utility Share of population with EU-SILC Yes bills arrears of utility bills Dwellings Share of the population comfortable cool Not available able to keep home EU-SILC Consensual in the for BGR 18 comfortably cool summertime Dwellings Share of the population comfortable Not available able to keep home EU-SILC warm in for BGR comfortably warm wintertime Presence of leak, Share of population EU-SILC Yes damp, rot. reporting leakage Average household price Biomass prices per kWh generated from BSO No biomass Average household price Coal prices per kWh generated from BSO No coal Average household prices Fuel oil prices per kWh generated from BSO No Energy fuel oil Prices Electricity price for Household household consumers electricity 19 EUROSTAT Yes (consumption band 2500- price 5000 kWh) Household gas Gas price (consumption EUROSTAT Yes price band 20-200 GJ) Average household price District heating per kWh generated from BSO No price district heating 17 https://ec.europa.eu/eurostat/web/household-budget-surveys/methodology 18 Information on ability to keep home cool is particularly important in the context of climate change and the increasing likelihood of heat waves. The World Bank rapid surveys collected in June 2023 collect information to construct this indicator in Bulgaria. This information could potentially be collected through specific ad hoc modules of the national EU-SILC, as recommended by the latest EC recommendation on energy poverty in October 2023. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=OJ:L_202302407 19 Household electricity and gas price are considered complementary indicators, as they affect affordability but do not indicate affordability per se. 8 Dwellings in Share of dwellings in densely BSO Yes densely populated areas population areas Dwellings in an Share of dwellings in intermediate intermediately populated BSO Yes population area areas Dwellings with Share of dwellings with BSO Yes energy label A energy label A Building Excess winter Share of excess winter Stock BSO Yes mortality/deaths mortality/deaths Features Equipped with Share of population living Not available EU-SILC heating in dwelling with heating for BGR Share of the population Equipped with Not available living in dwellings with air EU-SILC air conditioning for BGR conditioning Average number of Number of rooms per person in rooms per EU-SILC Yes rented/owned/all person dwellings Additional Indicators based on literature review Share of energy 10 percent expenditure (compared measure to overall expenditure or HBS Yes (Boardman, income) higher than 10 1991) percent The gap between theoretical standard Requires Energy energy consumption and Several 20 consumption deprivation actual consumption for data space heating and cooling Requires two conditions: Originated in Low-Income High Required energy England, Possible with Costs expenditure above the HBS when possible with Expenditure- median plus remaining relying on HBS HBS under Based income below the official income (Hills, 2011) conditions poverty threshold Household equalized income (possibly After-fuel-costs Originated in subtracting housing and Possible with poverty (AFCP) England, domestic energy costs) is HBS when indicator possible with below the standard relying on HBS HBS under threshold of 60% of income conditions equivalized national (Hills, 2011) income. Fuel costs (EHS) > Net Requires a Minimum household income (EHS) – combination Income Standard housing costs (EHS) – England of several (Moore, 2009) minimum living costs sources (MIS). 20 Energy needs of a dwelling, or data on indoor temperature (see Palma and Gouveira (2022)). 9 Disposable household Requires Theoretical income before energy standard on BSO energy demand expenditure for adequate adequate space heating space heating Type of dwelling Type of dwelling EU-SILC Yes Share of children (aged 0 to 17) living in a dwelling Children exposed with a leaking roof, damp EU-SILC Yes to leakages. walls, floors, or foundation, or rot in window frames or floor Ability to afford Share of the population to keep home able to afford to keep Eurobarometer Yes adequately warm home adequately warm Risk of falling Share of the population at behind with risk of falling behind with Eurobarometer Yes paying utility bills Consensual paying utility bills on time on time Dwelling is Share of the population leaking, damp, or who report leaking, Eurobarometer Yes rotting. damp, rotting Residential Energy usage and Energy-related data for Energy efficiency of housing units occupied as Consumption No housing and primary residences and Survey (United households their households States) Collects information English about people's housing Energy efficiency Housing circumstances and the No of housing Survey energy efficiency of (England) housing Several European indicators of Several Quality of Life Yes energy efficiency Survey and affordability Several Generation Several indicators of and Gender Several Yes energy efficiency Programme and affordability Survey Multidimensional Combination of several Several (see energy poverty Yes indicators main text) indicators Direct meters, Approximation of energy Satellite data, Energy Direct consumption, compared and other big Yes consumption to a norm data (see main text) Source: Compilation of findings by Gouveia et al. (2022) and additional literature (as shown in this section). 10 Our review reveals that direct measures of energy poverty are less used in the European Union, although the recent introduction of smart meters could change that. Given the limited availability of datasets on temperature or energy services, direct measures are more challenging to implement. Gathering this data involves visiting households and taking stock of indoor temperatures, for example, a procedure that might be burdensome. In addition, it is challenging to select appropriate standards, given that they might be subjective by nature. Also, standards can vary by context and individual factors (Palma and Gouveira, 2022). One potential solution to the lack of reliable data on energy consumption is the introduction of smart meters (European Commission, 2023b). Although smart meters potentially increase access to timely and more accurate information on energy usage, they are still subject to shortfalls, such as energy sources that are not metered (e.g., wood, heating oil, coal) (Frei et al., 2021). Research to date has mainly relied on temperature sensor deployment (ibid). Based on our review, energy poverty indicators are mainly only available at aggregated levels; generating them at disaggregated levels would be useful as tackling energy poverty requires local policy engagement. The European Commission recommends a local approach to tackle energy poverty, meaning that solutions must be adapted to national and local conditions. Therefore, it is worth generating more data and information on local (disaggregated) energy poverty. The geographical distribution and identification of energy poverty are critical for tackling energy poverty. The Energy Poverty Advisory Hub (2022) advises tackling energy poverty via a three-step process at the local level. 21 Given the local approach towards energy poverty, there is a need for indicators at the disaggregated geographic level. Palma and Gouveira (2022) review the literature on subnational analyses around energy poverty. They found that studies mainly focused on Southern Europe. While they do not mention any studies on Bulgaria, some generated insights could serve as inputs to generate geographically disaggregated analyses in Bulgaria. Although there is consensus that the geographical (subnational) distribution and identification of energy poverty is vital for mitigation, most subnational analyses rely on census data, and energy poverty indicators are often not collected as part of the population census. According to the review by Palma and Gouveira (2022), most subnational studies rely on data collected as part of the national census, although energy poverty indicators are often not included. While other studies take advantage of HBS or EU-SILC for regional analyses, the data is usually only representative at the NUTS 1 or NUTS- 2 level (depending on the country). Therefore, direct estimates for higher disaggregation levels are unreliable (ibid). Small-area estimates of energy poverty are needed but not necessarily available. Alternatively, many studies incorporating subnational indicators included in the review by Palma and Gouveira (2022) rely on new surveys collected by researchers. Nevertheless, the increasing usage of big data and satellite data generates new opportunities for subnational analyses of energy poverty. 22 National research institutes, for example, might be able to share meteorological data and facilitate the creation of subnational energy poverty indicators based on climate data. 21 The first step is an energy poverty diagnosis. During this first phase, policy makers should try to understand local energy poverty more closely. The second step is the design of a Local Climate Plan, which is a plan of how to best tackle energy poverty by taking an integrated approach at the local level. It involves setting up a detailed strategy and defining short- and long-term goals. The last step is the implementation of the Local Climate Plan. 22 In this context, the World Bank is exploring the development of a methodology for small-area estimates of energy poverty, under a DG Regio Trust Fund. The project aims to establish a methodology to measure energy poverty at the sub-national level using small-area estimation techniques and assess data availability across selected member states. The methodology will use a specific indicator of energy poverty (e.g., utility arrears, the share of energy expenditures within the household budget) and will then develop and apply an appropriate empirical methodology to project estimates of energy poverty from survey data into census data. 11 Based on the conceptualization of energy poverty, we describe these approaches in more detail but rely on approaches we can feasibly measure and monitor with the available data in Bulgaria. As we are ultimately interested in recommending a measure of energy poverty in Bulgaria that can be operationalized for monitoring and evaluation purposes, we rely on those approaches that we can feasibly measure and monitor through household expenditure surveys and the EU-SILC (see Column 4 in Table 1). We also explore the potential of big data sources. This leaves us with the following approaches: the access-based approach, the consensual-based approach, the outcome-based approach, the big data approach, and the expenditure-based approach. Finally, we analyze the strengths and weaknesses of each of these approaches. 2.2. Access-Based Approach The most straightforward approach to measuring energy poverty is to analyze the share of the population without access to energy. The worst form of energy poverty is not having access to energy, which has important implications for overall well-being and monetary poverty. Previous research shows a significant relationship between access to electricity, for example, and poverty rates (Asghar et al., 2022). A common approach to measuring this aspect of energy poverty is to look at the share of the population without access to certain energy services, such as electricity, natural gas, or other forms of energy. There are binary access-based approaches (e.g., having access to electricity or not) and those using a threshold (e.g., having more than 50 kWh of electricity per capita) (Nussbaumer et al., 2012). Others have constructed composite measures such as the Energy Access Index (EAI), which measures access to six essential energy services: lighting, cooking, water heating, space heating/cooling, information/communication, and mobility (Practical Action, 2010). Access to energy services can be measured at the extensive but also the intensive margin 23 and can, additionally, refer to access to clean and sustainable energy. Most measures focusing on energy access measure households' access to energy at the extensive margin, meaning that they analyze whether households have access to energy. Moreover, quality aspects in households' energy access and reliability also play a role in energy poverty (the intensive margin). Bhatia and Angelou (2015) present a multi-tier framework to analyze the quality of energy access in the case of electricity. They measure reliability (the number of disruptions per week), legality, the health and safety aspects of the available energy source, peak capacity, and duration, among other factors. Another aspect one can measure with access-based approaches to energy poverty is access to certain types of energy, such as clean energy (Ullah et al., 2021). Access-based measures are helpful in identifying the most deprived populations and improving energy poverty around lack of access, but they also suffer from limitations. While access-based approaches are helpful in detecting the most deprived populations as they lack access to energy altogether, they abstract from other important aspects of energy poverty, such as variations in energy needs and preferences or the affordability and sustainability of energy services. Measures at the extensive margin also abstract from the quality and reliability of energy services. 2.3. Expenditure-Based Approaches The discussion on how to best measure energy poverty does not stop with the type of approach one should best use but extends to questions around how to best construct each metric within a given 23 By extensive margin, we refer to having access to energy or not (a binary description). By intensive margin, we refer to various characteristics of energy poverty related to its quality and reliability. For example, how many hours per day the household receive electricity, the number of blackouts per month, etc. This is more difficult to quantify. 12 approach. This problem is especially apparent within the literature using expenditure-based approaches. Often, even if researchers use the same metric, approaches on how to construct this metric deviate; this problem is especially apparent within the literature relying on expenditure data focusing on the affordability aspect of energy poverty. We identify the following inconsistencies in the literature: • On scaling energy expenditure and measuring energy costs: While some scholars divide energy expenditure by adult equivalents, others scale it by the number of household members (Schulte and Heidl, 2016) or the OECD equivalence scale (Hills, 2012). A third possibility is to use absolute values altogether (Thema and Vondung, 2020). The rationale behind this debate is that residential energy use is subject to important economies of scale (Schulte and Heidl, 2016). There is no consensus in the literature on how to address this point. Moreover, different researchers consider different components of energy costs. For example, one could include transport-related energy costs or abstract from them. Finally, there is debate on whether theoretical energy needs for a household with certain characteristics at current prices or actual energy expenditures reported by households should be used. • On using household income or expenditure: In addition, there is variation in the literature on defining the denominator of the energy spending share. Several papers relate energy expenditure to household income (e.g., Thema and Vondung (2020); Hills (2012); Saghir (2005)) while others relate it to household expenditure (e.g., Freund and Wallich (1995); Schulte and Heindl (2012); Laderchi et al. (2013)). At the same time, several measures abstract from scaling household energy spending altogether and use absolute energy costs (Thema and Vondung, 2020). When using income as the denominator, there is again an inconsistency with respect to whether to consider disposable income before or after housing costs (Herrero, 2017).24 • The threshold used to define energy poverty. Lastly, there is no consensus in the literature to date on the threshold one should use to define those who are energy poor. While energy poverty is multidimensional, in practice, most definitions of energy poverty specify a resource threshold in terms of a maximum acceptable proportion of household income devoted to energy consumption. The threshold can be based on an absolute or relative measure. In an absolute measure, the threshold is a fixed percentage of the income/expenditure spent on energy, whereas a relative threshold is based on a median or average energy burden. The most used thresholds are the 10-percent measure, the twice median (M2) threshold, the half median (M/2) threshold, and the Low-Income High Costs (LIHC) measure (Thomson et al., 2022). In some cases, researchers and policymakers also use thresholds for subgroups of the population, such as households with a similar composition or the same number of rooms, or might apply several thresholds simultaneously, such as a threshold for energy expenditure and income (King Baudouin Foundation, 2018). • National vs. subnational threshold. Using a consistent nationwide threshold may also be problematic, as it overlooks the variations in energy requirements between households living in different regions. Using the same threshold can lead to lower energy poverty rates in warmer zones and higher rates in colder zones. The rationale behind this pattern lies in the higher energy needs for domestic heating services prevalent in colder regions, thus leading to increased domestic energy usage. Conversely, individuals residing in warmer regions exhibit lower heating energy needs, resulting in decreased domestic energy usage. Consequently, applying a uniform national threshold indicates a higher likelihood of surpassing the threshold 24 For example, the Energy Poverty Barometer developed in the case of Belgium relates energy expenditure to disposable income after deduction of housing costs (King Baudouin Foundation, 2018). 13 for those in colder regions and vice versa. This underscores the necessity to appropriately consider variations in energy needs due to climatic and circumstantial factors when establishing levels of 'adequate energy services' (Boardman, 1991, p.227) or a socially and materially necessitated level of domestic energy services' (Bouzarovski and Petrova, 2015, p.31). In summary, while expenditure-based measures might seem straightforward at first, there is a lot of discussion about how exactly to measure these indicators. In addition, these types of indicators might simplify the concept of energy poverty and reduce it to the affordability aspect of energy poverty. In the following, we explore four expenditure-based measures in more detail. Ten Percent Measure The 10-percent measure relies on an absolute threshold, defining energy poverty as the share of households that spend a significantly high portion of their household budgets (10 percent or more) on energy. For monitoring progress over time, it is recommended to use absolute thresholds instead of relative thresholds, or in other words, a fixed standard of what households should be able to count on to meet their basic energy needs, rather than a cutoff point about the overall distribution of energy spending in a country. 25 A common approach used is to rely on a threshold of 10 percent to define the energy poor. Under this definition, energy poverty is the share of households that spend a significantly high portion of their household budgets (10 percent or more) on energy. Ten percent is a threshold related to a minimum level of energy consumption to maintain an adequate level of warmth, originally measured as 10 percent for the UK. A broad set of studies uses this definition because it allows tracking progress over time and comparing energy poverty rates across countries. This measure has also been used in several World Bank studies on energy prices and was used in a World Bank report across a large set of countries in Europe and Central Asia (Olivier & Ruggeri Landerchi, 2017). The 10-percent measure has been widely used in the literature despite being criticized for several reasons. The measure was originally developed for the UK and, therefore, lacks strong empirical justification for other countries, and it also fails to account for households in precarious situations that restrict their energy usage and consume less energy than needed. While many reports rely on a threshold of 10 percent (hereinafter: 10-percent measure), this measure was originally developed for the UK context (Hills, 2012). It originally was the 2M measure (twice the national median) of the energy expenditure ratio (energy expenditure (modeling consumption and prices) and total UK people's income) in 1990 (King Baudouin Foundation, 2018). The application of this measure to other countries' contexts has been highly criticized to date (see, for example, Thomson et al. (2017) and Schuessler (2014)). More recently, experts have even recommended moving away from this definition in the UK, where the measure was originally developed (Hills, 2012). This method is also problematic as it leads to a binary distinction of energy poor/non-poor, while energy poverty may also be a continuum. Additionally, it does not capture vulnerable households that limit their energy use and under-consume energy (or the "hidden" energy poor) 26. This shortfall can be overcome by using theoretical energy needs instead of actual energy bills, but calculations of theoretical energy bills are data-intensive and may require additional modeling tools. The 10-percent measure can be beneficial in comparing energy 25 A similar approach is used to track changes in income poverty over time based on absolute measures. 26 Energy poor households may go unnoticed as they limit their residential heating to save energy. These households may tolerate colder room temperatures to prevent high energy bills. However, under this definition, they are no considered energy poor and therefore, may not meet eligibility criteria for social policies designed to protect them. 14 poverty incidence rates across countries, as they are compared against a fixed standard, although this comparison still suffers from limitations. To estimate this measure, one can divide the household's overall energy expenditure by the household income or household expenditure. 27 The measure can be estimated as a share of household expenditure or income. Still, income measures are usually not captured accurately in household budget surveys, as the instruments are generally designed to capture consumption and expenditure patterns rather than income sources. Therefore, it is recommended that these measures work appropriately and are reliable when deciding how to scale energy expenditure in the case of the relative measures. Measure M/2 and 2M Another relative expenditure-based measure, next to the 10-percent measure, is the so-called 2M measure. The "2M" indicator captures abnormally high energy expenditure. It represents the percentage of households whose energy expenditure share (in income or consumption) is more than twice the national median share (Energy Advisory Hub, 2022). This metric gauges the burden of energy bills on households in relation to their incomes or overall expenditure, using the national median as a reference. Therefore, poor households that limit their energy consumption due to budgetary constraints may not be captured in this measure. In contrast, the M/2 measure follows the concept of hidden energy poverty and measures abnormally low absolute energy expenditure. M/2 is an indicator that captures abnormally low energy expenditure, consisting of the share of households whose absolute energy expenditure is below half the national median. The M/2 measure draws from the "hidden" energy poverty concept, which argues that households might decrease their energy consumption so much that it becomes abnormally low and could result in negative health outcomes (Nguyen, 2022). Notice that households living in highly efficient buildings that technically require less energy may be considered energy poor under this definition. Both of these are relative measures, as the thresholds moved with the distribution, which makes it challenging to compare them over time. Relative thresholds are a cutoff point in relation to the overall distribution of income or consumption in a country (in this case, they are linked to the median of the distribution). Consequently, if the entire distribution moves up or down, the M2 and M/2 measures might still not improve. It might, therefore, be difficult to use these measures to track progress over time. Moreover, they might suffer from measurement errors, and it might be difficult to compare these measures across countries as it requires harmonizing income, expenditure, and the approach toward measuring energy expenditure. Low-Income-High-Cost Measure The expenditure-based indicator "Low Income-High Costs (LIHC)" identifies energy poor households as those pushed into energy poverty due to high fuel costs and low incomes. This measure was officially adopted in England to measure energy poverty and replace the original 10 percent measure. Based on this indicator, a household is considered energy poor if two conditions are met: 1) if they have an energy expenditure above the national median, and 2) if the remaining income after they spend that amount (in equivalized terms) is below the official at-risk-of-poverty (AROP) threshold (Hills, 2012) (Figure 3). 28 This approach directly considers households affected by the problem, 27 In the case of the application at hand, the overall household expenditure is harmonized following World Bank ECAPOV Harmonization Guidelines. 28 This is equivalent to having an income threshold equal to the AROP plus the energy expenditure. This includes the income poor (those with incomes below AROP) and the vulnerable that fall into poverty because of their 15 specifically those with low income and high costs. To be more precise, it includes households with energy costs exceeding a 'reasonable level' (the median) and whose residual equivalized income would fall below the official poverty line if they were to spend that amount. Notice that two types of households meet these conditions: 1) Households that are income poor and have relatively high energy expenditures; and 2) Households that are just above the poverty line and would fall into income poverty after considering energy expenses (households vulnerable to poverty), with relatively high energy expenditures. This definition excludes income-poor households with relatively low energy expenditures and households above the poverty line that do not fall into poverty when energy expenses are taken into account. Similarly to the M/2 and the 2M metrics, the “LIHC” also uses relative thresholds, making comparisons over time difficult unless the thresholds are anchored. A similar definition is the after-fuel-costs poverty (AFCP) indicator, developed by Hills (2011). Still, it considers only one condition: household equalized income (possibly subtracting housing costs and domestic energy costs) is below the standard threshold of 60% of equivalized national income (Figure 4). 29 In addition to the incidence of energy poverty, the energy poverty gap can also be estimated. The depth of energy poverty is defined as the amount by which the assessed energy requirements of fuel- poor households exceed the threshold for reasonable costs. The energy expenditure can be the observed energy expenditure, for example, from the HBS, or it can be modeled. In the UK, the energy cost is modeled rather than based on observed spending. This ensures that one does not overlook households with low energy bills simply because they actively limit their energy use at home, for example, by not heating their home. It is calculated by combining the fuel requirements of the household with corresponding fuel prices. This approach is data intensive, as these required costs capture four areas of energy requirements: a. Space heating; b. Water heating; c. Lights and appliances; and d. Cooking. The modeling ensures that the household achieves adequate warmth, subject to various characteristics concerning the dwelling and its occupants. Figure 3: Low Income-High Costs (LIHC) Measure energy expenditures. Several EU countries have adopted different versions of this definition, including France. Modifications include different treatment of housing costs (included or excluded), expressing the energy spending in absolute terms or as a share, etc. 29 This measure is close to the LIHC definition, but it does not include an energy expenditure threshold. A limitation of the AFCP is that it categorizes a significant number of households with extremely limited incomes as being in energy poverty, regardless of their actual fuel needs. Under this definition all income poor households are considered energy poor, regardless of their energy use. This is a group that it is normally excluded under the LIHC definition, but is considered energy poor under this definition. 16 Source: Author's own elaboration Figure 4: After-fuel-costs poverty (AFCP) Measure Source: Author's own elaboration The income and expenditure measures used in this indicator must use the same scale and reference period. The at-risk-of-poverty threshold (AROP) is set at 60 % of the national median equivalized disposable income after social transfers. Household members are made equivalent by weighting each according to age, using the so-called modified OECD equivalence scale. Therefore, when comparing energy expenditures and income to this threshold, they must be in the same unit (equivalized with the same equivalent scale). The reference period should also be the same. For example, the reference period for household incomes in the SILC surveys is the last 12 months, so expenditures must also be annualized compared to annual incomes. Moreover, the AROP line is a relative measure that adheres to EU standards for officially measuring poverty, using a relative poverty line. This standardized approach is employed across all EU countries for national and EU poverty monitoring purposes. However, to monitor changes over time, fixing or anchoring the relative poverty (AROP) lines at a 17 specific point in time is recommended, effectively converting them into an absolute measure. These are referred to as anchored AROP poverty lines. In the case of Bulgaria, we construct the low-income-high-cost measure drawing from both observed expenditure and income information gathered as part of the HBS. We estimate overall household energy expenditure and subtract this amount from the net total household income reported in the household budget survey (HBS). 30 We then compare the remaining household income (the net disposable household income after subtracting energy expenditures) to the official at-risk-of-poverty line (from the EUSILC survey). Income poverty lines and rates are usually estimated using the EU-SILC survey, but this survey does not collect expenditure information. Therefore, we rely on use income and expenditures from the HBS and the official poverty line from the SILC. Note that the at-risk-of- poverty line differs from the official one reported by Eurostat, which relies on data from EU-SILC. 31 2.4. Consensual Approaches The set of indicators often denoted as consensual approaches rely on self-reported information by households. We next explore consensual approaches toward measuring energy poverty. Thema and Vondung (2020) define consensual measures as those that rely on self-reported assessments of housing conditions or the inability to cover basic energy needs because of a lack of access or affordability. Examples of such metrics are the share of households that cannot keep their home adequately warm/cool or those reporting leakages. Consensual approaches are direct and personal measures but are subjective. Consensual approaches are easy to measure and might reflect different energy needs and demands. They also intrinsically capture some of energy poverty's negative social and psychological effects. However, they also suffer from important flaws. First, they are subjective measures, and it might be difficult to compare them across different households, regions, or countries. Next, they might also be influenced by household expectations, preferences, and social norms. These factors might significantly influence the reliability and comparability of consensual approaches. 2.5. Outcome-Based Approaches Another possibility for measuring energy poverty is to focus on the outcomes that occur because of energy poverty (Thema and Vondung, 2020). Energy poverty leads to unfavorable health, socioeconomic, and environmental impacts (Halkos and Gkampoura, 2021), and measuring these impacts can then again shed light on the severity of energy poverty experienced by the population of interest. Examples of such measures are the number of cold- or heat-related mortalities or arrears on energy/utility bills. Outcome-based approaches might be easy to measure but, at the same time, could be confounded by other factors, such as lack of appropriate public infrastructure or income. While it might be easy, conceptually speaking, to measure the number of cold-related deaths, these measures might be influenced by other factors that we might not be able to observe, such as the development of the public infrastructure or the available income at the household level. 2.6. Multidimensional Measures Given the multidimensional nature of energy poverty, several scholars propose multidimensional indicators to measure energy poverty. Given the multidimensional nature of energy poverty and the 30 As shown later, the AROP estimate based on the HBS is significantly below the estimate from the EU-SILC. 31 The at-risk of poverty rate constructed for this measure is equal to 60 percent of the median value of the equivalized household income after social transfers in the HBS. 18 difficulty of capturing it with single indicators, several scholars propose to create multidimensional indicators. A recent review by Siksnelyte-Butkiene (2021) for 1994 to 2020 identified 34 indicators. For example, Day et al. (2016) embark on energy poverty from a capability framework. Similarly, Sokolowski et al. (2020) constructed a multidimensional indicator based on five dimensions of energy deprivation in Poland. Bollino and Botti (2017) generated a multidimensional energy poverty indicator based on five indicators from EU-SILC data, which is in line with this methodology. Similarly, Bouzarovski and Herrero (2016) constructed an energy poverty index for EU countries using three indicators from the EU-SILC (European Union, 2016). Pelz et al. (2018) evaluate several additional indicators that try to capture the multidimensional nature of energy poverty and evaluate their strengths and weaknesses. Further work exploring multidimensional energy poverty indicators in the EU is by Maxim et al. (2016) and Herrero and Bouzarovski (2014), among others. Other practical examples are from the Philippines (Mendoza et al., 2019), several countries in Asia (Lan et al., 2022), Africa (Nussbaum et al., 2012), and Uganda (Ssennono et al., 2021), among others. While multidimensional measures of energy poverty have gained increasing attention in the literature, these measures are also subject to downfalls. Nussbaum et al. (2012) evaluate composite energy poverty metrics and conclude that these measures are useful but might be subject to important methodological shortfalls. While they help incorporate multidimensionality into one metric, there needs to be a consensus around which indicators to include, how to weigh them, and which thresholds to use for each subdimension to define energy poverty. 2.7. Big Data More recently, big and open-source data started to play a significant role in direct measures of energy consumption, which can also shed light on energy poverty and vulnerability. Access to satellite data has facilitated the measurement of direct indicators without the necessity of implementing time-consuming home visits to employ sensors or meters. Xinyi et al. (2018) use freely available satellite data to construct a residential building stock model for energy consumption. Similarly, Fehrer and Krarti (2018) use nighttime light data to generate US electricity and fuel consumption maps. Sun et al. (2022) developed a building energy efficiency prediction algorithm to predict the energy efficiency of buildings by combining data from the Energy Performance Certificate (EPC) database and Google Street View (GSV) building façade images in the case of Glasgow. Another example is by Berger and Worlitschek (2018), who generate energy end-use data for heating in the residential sector by combining data on population, norm temperatures, energy supply, consumption data, and buildings. Table 2 summarizes this evidence. Table 2: Measurement strategies for energy consumption of buildings using open-source and satellite data – 4 examples Authors/Source Summary Data Place Definition of representative - Satellite data (Google buildings of a built-up residential Earth) People’s building stock, using satellite - Ground survey Republic of Xinyi et al. (2018) images and cluster analysis China methods, to generate energy Use (Districts) Intensity (EUI) indicators Application of deep learning- - Energy Performance Sun et al. (2022) Glasgow, UK based framework using street Certificate (EPC) 32 This type of information is also gathered in Bulgaria, at least for new buildings and old buildings that undergo 32 major renovations (iBroad, 2018). 19 view images for overall energy - Morphology efficiency predictions and description building attributes - Google Street View (GSV) Fehrer and Krarti Use of nighttime light imagery - Satellite data from (2018) from VIIRS as a predictor of VIIRS US (several building energy use intensity levels) within states, counties, and cities in the United States Berger and Estimating residential space - Population Worlitschek heating demand via maps of - Temperature (2018) population density distribution - Energy Switzerland combined with norm supply/consumption temperature profiles and the - Buildings definition of heating degree days Source: Own elaboration based on literature review. Big data sources might also help generate new insights into energy spending, consumption patterns, and related social norms and behaviors. For example, Fergus and Chalmers (2021) use data from smart meters, consumer access device (CAD) data, and machine learning algorithms to detect and monitor fuel poverty. They also generate new insights into energy consumption, behavior, social norms, and expectations of energy sobriety. Big data approaches are promising but require advanced data science skills and tools. Big data approaches allow for real-time tracking and extremely granular measurement of energy poverty. At the same time, they require advanced data science skills and tools. In addition, they might raise concerns about data privacy, ethics, and security. Lastly, they might still abstract from the multidimensionality of energy poverty, especially considering quality, affordability, or differences in energy demands and needs. 2.8. Summary In summary, there are a variety of approaches to measuring energy poverty. Each approach highlights different aspects of energy poverty, including strengths and weaknesses. Having these strengths and weaknesses in mind when designing operational measures of energy poverty and having clarity of what aspect of energy poverty one would want to measure can result in more effective and impactful measuring strategies. We summarize each of the measures and their respective strengths and weaknesses in Table 3. Table 3: Summary of energy poverty measures and strengths and weaknesses of each approach Name Summary Strengths Weaknesses Access-based Lack of access to a Abstracts from quality and Useful to improve measures minimum level of energy reliability; does not factor energy access; useful services, such as in variations in energy to identify the most electricity and clean deprived needs and preferences; cooking fuels population/region does not account for sustainability and affordability. Expenditure- Energy poverty is Easy to calculate and Abstracts from quality and based measures spending more than a simple measure quantity of energy services certain percentage of and different household income or expenditure realities; no consensus in 20 on energy consumption the literature on how or having absolute exactly to measure this; no energy expenditures incorporation of minimum below a threshold adequate levels of energy (hidden energy poverty) (energy basket) Consensual Self-reported perception Direct and personal; Subjective and personal; approaches of energy deprivation incorporates different lower comparability and energy demands and higher measurement necessities. errors Outcome-based Measures outcomes Direct and observable This might be confounded approaches based on energy by other factors, such as poverty, such as cold- low savings, low public related deaths or arrears infrastructure, and low on utility bills awareness. Multidimensional A combination of Nuanced and holistic, Subjective choices on the measures multiple indicators that it captures the inclusion of indicators and reflect different aspects complexity and weighting require more of energy deprivation multidimensionality of data. energy poverty. Approaches that Measures that use big Innovative, real-time, Abstracts from quality, use big data data, such as satellite, and granular analysis affordability, and mobile phones, social reliability; require media, etc. advanced data science skills and tools; may raise ethical and data privacy concerns Source: Own compilation based on literature (2023). 3. Energy Poverty Measures in Bulgaria We next explore several energy poverty measures in Bulgaria. We start by looking at purely expenditure-based measures. We then explore consensual and outcome-based approaches and the potential of big data sources. We mainly rely on data from the household budget survey (2021) and the EU-SILC (2020), the most recent data for this study. 3.1. Expenditure-based measures We start by exploring expenditure-based measures of energy poverty using survey data from 2021. We explore the three expenditure-based approaches detailed previously and also analyze the sensitivity of these measures to varying some of the underlying parameters. For the relative measure, we start by defining energy spending shares as follows: ES = , where E is the total energy spending category reported in the household budget survey (electricity, natural gas, town gas, liquified energy sources, liquid fuels, coal, other solid fuels, and heat energy). We divide it by total household expenditure and alternatively disposable household income and then compare both measures. Importantly, we rely on net income, the total income from all sources, 21 including non-monetary components minus income taxes. We do not consider imputed rent 33. We scale the nominator and denominator by the same measure (e.g., the number of total household members or the adult equivalent), which means they cancel each other out. We also explore the possibility of only scaling the denominator, but estimates are significantly higher than benchmarks from the literature, which is why we discard this approach. The average energy expenditure shares are close to each other, but the income-based measure is significantly less dispersed; the energy poverty rates show similar proximity across the three measures: the 2M, M/2, and equivalized M/2. While the average energy expenditure shares are close to each other (13.5 percent versus 12.7 percent using expenditure and income, respectively) 34, the distribution depicted in Figures 5 and 6 looks different, which could impact energy poverty measures based on energy shares. Figures 7 and 8 reveal that measures 2M and 10 percent have significant differences, likely due to the differences in the distributions. M/2 measures do not vary, as they are absolute measures that do not rely on the shares. The graphs also show that the 10 percent measure is significantly above the other expenditure-based measures, probably related to the methodological shortfalls detailed previously. 35 While the 10-percent measure results in an energy poverty incidence rate above 50 percent, the incidence rates of the other measures fluctuate between 8 and 10 percent. Figure 5: Energy expenditure share (energy over Figure 6: Energy expenditure share (energy over expenditure) (2021) income) (2021) Source: Own estimates based on HBS (2021) 33 Rent is treated separately in the income aggregate due to the complexity of its inclusion. When rent is included, it can create a situation where a tenant appears less economically disadvantaged than a homeowner with similar characteristics living in the same area, even if their incomes in other categories are identical. To address this disparity and treat these individuals equally, one option is to exclude rents altogether, or alternatively, to devise a method for measuring the utility flow generated by owner-occupied housing. In alignment with World Bank ECAPOV guidelines (World Bank, 2022), imputed rent is not included in the income aggregate. 34 These values are similar to estimates of the energy poverty share reported previously in the literature (Laderchi et al., 2013). 35 Two notable effects can explain this result. First, the 10 percent measure, relying on actual energy expenditures, fails to account for the concept of "hidden poverty," which pertains to individuals who curtail their usage and consume less energy, thus resulting in an undercounting of energy poverty. Second, the choice of a 10 percent threshold is arbitrary and based on UK standards. According to Thomson (2017), researchers risk producing invalid results by using a 10 percent actual expenditure threshold that is not grounded in the specific context of the country under consideration. In the case of Bulgaria, this threshold seems exceptionally low, leading to a higher proportion of individuals being classified as energy poor. These two effects operate in opposing directions, with the dominance of the second effect resulting in a relatively high incidence of energy poverty. Moreover, our estimate of energy poverty using the p10 measure is close to previous estimates (Laderchi et al., 2013). 22 Note: Energy includes electricity, natural gas, town gas, liquified energy sources, liquid fuels, coal, other solid fuels, and heat energy Figure 7: Energy spending share (over expenditure) Figure 8: Energy spending share (over income) and and energy poverty incidence rates, by definition energy poverty incidence rates and, by definition (2021) (2021) 70 62.3 60.0 51.9 60 50.0 50 40.0 40 30.0 30 20 13.5 20.0 12.7 12.7 9.9 9.5 7.9 9.5 7.9 10 10.0 0 10 percent measure 0.0 2M M/2 M/2 (equ.) Energy spending share 10 percent measure 2M M/2 M/2 (equ.) Energy spending share (expenditure-based) (income-based) Notes: The energy spending share is the average expenditure on energy divided by the average household income (on the right) and by the average household expenditure (on the left). We consider the following measures of energy poverty: (1) A household is considered energy poor as soon as the share of energy expenditure (compared to equivalised disposable income) is above twice the national median of energy spending shares (2M); (2) a household is energy poor as soon as absolute energy expenditure is below half the median of absolute energy expenditure (M/2); (3) a household is energy poor as soon as equivalized energy expenditure is below half the median of absolute energy expenditure (M/2 (equ.)); (4) A household is considered energy poor as soon as the share of energy expenditure is above 10 percent. Source: Own estimates based on HBS (2021). An additional measure of energy poverty that is increasingly used, the low-income-high-cost measure (LIHC), is below the 2M and (equivalized) M/2 measures in Bulgaria. This measure is close to the newly announced official measure but uses observed energy expenditures from the household budget survey rather than modeled energy expenditures. The energy poverty rate resulting from the low-income-high-cost measure is 7.1 percent (Figure 9), below the 2M and M/2 measures. However, it is important to note that the at-risk of poverty rate using income reported in the HBS 2021 results in slightly higher rates than the official measures from the EU-SILC survey (25 percent versus 22.1 percent). 36 The difference in the estimates reflects differences in the income distribution generated by the household budget survey compared to the EU-SILC. 36 We create the LIHC measure of energy poverty by utilizing the official poverty line from the European Union Statistics on Income and Living Conditions (EU-SILC) rather than the poverty line derived from the Household Budget Survey (HBS). The purpose is to align our measure more closely with the official definition, which relies on the poverty line from the SILC as a benchmark. Using the poverty line from the HBS could potentially result in a higher incidence of energy poverty. This is because the HBS poverty line is higher than the one estimated using SILC, leading to higher income poverty rates. 23 Figure 9: At-risk of poverty rate (measured via the Figure 10: Overlap of at-risk-of-poverty rate (measured equivalized income reported in the HBS) and energy via the equivalized income reported in the HBS) and poverty rate (low-income-high-cost measure) energy poverty rate (low-income-high-cost measure) 30.0 100% 25.0 80% 25.0 60% 71.5 20.0 40% 3.5 15.0 20% 21.4 0% 3.6 10.0 7.1 Neither nor Only energy poor (LIHC) 5.0 Only monetary poor 0.0 Energy (LIHC) and monetary poor At-risk-of poverty Energy poor (LIHC) Notes: The at-risk-of-poverty rate is calculated using household income from the 2021 household budget survey and, therefore, does not correspond to the official poverty estimate. The resulting estimate is slightly above the official poverty estimate from the EU-SILC (22.9% in the 2022 survey year and 2021 income reference year). A household is considered energy-poor if two conditions are met: 1) if they have an energy expenditure above the national median, and 2) if the remaining income after they spend that amount (in equivalized terms) is below the official at-risk-of-poverty threshold. By the definition of the LIHC, all monetary income poor are also energy-poor if they have energy spending shares above the national median. Source: Own estimates based on HBS (2021). The overlap between energy and monetary poverty varies considerably between the different measures. By construction, all monetary poor are energy poor in the low-income-high-costs (LIHC) measure only if they have energy spending above the national median 37 (see Figure 10). An additional share of vulnerable households falls below the threshold (AROP poverty line) when subtracting their household energy expenditure from their available household income. Still, some have energy spending shares below the national median, so these households remain unaffected by energy poverty. The P10 measure highlights the most significant intersection between households facing energy and monetary challenges. Specifically, it reveals that 20% of all Bulgarian households experience both energy and monetary poverty. Additionally, a striking 80% of those classified as income poor also fall into the category of energy poverty. The overlap is significantly lower among other measures. For example, only 14.3% of income poor households are also energy poor under the LIHC measure. The overlap between the different measures and at-risk of poverty rates only varies slightly when using income or expenditure as the denominator (compare Figures 11 and 12) 38. The limited overlap in the LIHC, the 2M, and M/2 measures means that policies targeting the energy poor 37 Households are considered poor when their income falls below the AROP poverty line. The at-risk-of-poverty threshold is set at 60 % of the national median equivalized disposable income after social transfers. Household members are made equivalent by weighting each according to their age, using the so-called modified OECD equivalence scale. Therefore, when subtracting energy expenditures from their income, the remaining income will always fall below the AROP poverty line for these households that have energy spending above the median (condition 2). 38 Note that the M/2 measures are the same in both figures, as they, by definition, do not rely on the denominator. 24 using these definitions must go beyond income-support measures designed to target the income poor, as they will not reach a large share of this group. Figure 11: Overlap of energy and monetary poor using Figure 12: Overlap of energy and monetary poor expenditure-based measures (over expenditure, when using expenditure-based measures (over income, applicable) when applicable) 100% 100% 80% 80% 60% 60% 40% 40% 20% 20% 0% LIHC 2M M/2 M/2 P10 0% (equ) LIHC 2M M/2 M/2 P10 Neither nor Neither nor (equ) Only energy poor Only energy poor Only monetary poor Only monetary poor Energy and monetary poor Energy and monetary poor Notes: Monetary poverty is the at-risk poverty rate using the income measure reported in the HBS 2021 and, therefore, does not correspond to the official poverty measure. The resulting poverty estimate is slightly above the estimate from the EU-SILC. Source: Own estimates based on HBS (2021). In summary, the large differences in the incidence of energy poverty between the alternative expenditure-based measures demonstrate how sensitive these measures are to the underlying definition; the most effective public policies that address energy poverty will, therefore, be highly conditioned by the definition adopted by the government. The differences in energy poverty incidence rates between the measures confirm the challenge of identifying sound and precise indicators. If policymakers plan to target those who experience low income and high costs, the low- income-high-costs measure might be most appropriate. If they want to generate indicators that can be compared across countries, the 10 percent measure might be most appropriate. Suppose the goal is to capture those that under-consume energy in absolute terms or abnormally high energy expenditure relative to the distribution. The M2 and ( equivalized) M/2 measures might be best suited in that case. Finally, since these expenditure-based measures rely on data from the Household Budget Survey (HBS), and this data is only representative at the national, urban, and rural levels, additional geographical disaggregation (NUTS2 and below) is not possible. Characterizing energy affordability within the group experiencing income poverty or estimating the intersection between official income poverty and energy poverty poses challenges. In particular, the above indicators rely exclusively on the Household budget survey, the main source of energy expenditures among households, but use the income reported in this survey. However, official income poverty in Bulgaria, as in other EU countries, relies on the EUSILC survey. A different approach to this problem involves the statistical matching of these datasets (Rude and Robayo-Abril, 2024). 39 Using 39 Through data fusion, they create a unique dataset incorporating details on energy spending shares, income- based poverty indicators, inequality measures, and additional variables related to households' living conditions 25 imputation methods, energy shares can be imputed into the EU-SILC survey. Utilizing the resulting synthetic dataset, one can overlay energy poverty with official monetary poverty. The findings reveal a substantial portion of the energy-poor population (using the LIHC measure) that does not fall under the category of official income poverty. Consistent with the findings in this paper, the results vary significantly depending on the energy measure used. This research contributes to the expanding body of literature exploring the potential of statistical matching to enhance the current data landscape in the European Union. 3.2. Consensual Approaches and Outcome-based Metrics We next explore consensual and outcome-based measures in Bulgaria and rely on those indicators mentioned in Table 1, which are included in the EU-SILC and available in Bulgaria. Two aspects of the main EU indicators for material and social deprivation (MSD) and severe material and social deprivation (SMSD) are directly applicable to the monitoring of energy poverty using this approach. They are commonly employed as proxies for assessing energy poverty in the EU: the inability to adequately heat one's home and the incidence of arrears on utility bills. 40 Both factors are integral components of the AROPE headline indicator, which has been fundamental to the EU2020 poverty and social exclusion target. It continues to play a key role in supporting the 2030 EU poverty and social exclusion reduction target. Figure 13 shows that the energy poverty incidence rate fluctuates between 11.0 percent and 46.5 percent, depending on the indicator used. Consequently, the overlap between monetary and energy poor differs significantly by indicator (Figure 14). Not surprisingly, the leakage and type of dwelling indicators 41 report relatively low overlaps. At the same time, nearly half of the monetary poor are also energy poor in the case of the affordability indicators. A high share of households are only energy poor on the building indicator (40.2 percent). These measures can be disaggregated at the NUTS3 level, as the SILC survey is representative at this level. and welfare is created. To achieve this, the authors construct a harmonized dataset by combining EU-SILC and household budget survey data. Subsequently, they employ various imputation models and identify the most effective one to impute energy spending shares into the EU-SILC data. 40 Arrears on utility bills are captured in the SILC survey question: “During the last 12 months, did your household face some problems with payment on time of the expenditure on the dwelling you are living in? Ability to keep the home warm is captured in the SILC survey question: “Can you afford adequate heating of your dwelling?“ 41 Privately-owned inefficient multi-family apartment blocks with insufficient maintenance may have high energy poverty rates. 26 Figure 13: Energy poverty indicators based on Figure 14: Overlap of monetary and energy poverty based on consensual approaches and outcome-based consensual approaches metrics Apartment in multi- Apartment in multi- household building 46.5 household building Leakage (roof, wall, floor, Leakages (children) foundation, window) - 12.3 children Leakages Leakage (roof, wall, floor, foundation, window) 11.0 Arrears on utility bills Arrears on utility bill 22.2 Inability to keep home adequately warm Inability to keep home 0% 20% 40% 60% 80% 100% adequately warm 27.5 Energy and monetary poor Only monetary poor 0.0 10.0 20.0 30.0 40.0 50.0 Only energy poor Neither nor Notes: Monetary poverty is defined via the at-risk-of-poverty measure using income measures reported in the EU- SILC. Income refers to the year 2019. Source: Own estimates based on EU-SILC (2020). Overlapping the different consensual measures confirms findings from the previous literature on households experiencing different dimensions of energy poverty. For example, only half of households that report arrears on their utility bills also report an inability to keep their home warm (EU-SILC, 2020). Similar observations can be made when overlapping households that report leakages with those unable to keep their home warm (EU-SILC, 2020). These results show that various population segments encounter distinct aspects of energy poverty. Efficient targeting strategies must consider which part of the population is affected by what type of energy poverty. 3.3. Multidimensional Measures Given the multidimensional nature of energy poverty, one approach would involve using a comprehensive array of indicators to construct a thorough understanding of energy. Among the wide range of multidimensional measures identified in the literature, we construct a multidimensional energy poverty index following the methodology developed by Bouzarovski and Tirado Herrero (2017) and Rodriguez-Alvarez et al. (2021). The advantage of this index is its simplicity and the fact that all the dimensions can be measured with one data source, the EUSILC, and, therefore, can easily be constructed and tracked over time. Tirado Herrero (2017) emphasizes that the suggested index operates under the assumption that relying solely on consensual measures (like self-reported inability to maintain warmth) may not adequately grasp the intricate nature of energy poverty. They suggest integrating indicators describing the population's housing and financial circumstances to comprehensively understand them. The multidimensional energy poverty index is constructed as a composite index using three indicators from the EU-SILC and weighting them accordingly as follows: = (0.5 ∗ + 0.25 ∗ + 0.25 ∗ ) ∗ 100 Where: 27 • Inability indicates the percentage of the total population who cannot keep their homes adequately warm. • Arrears is the percentage of the total population with arrears on utility bills (heating, electricity, gas, water) on time due to financial difficulties. • Leakages capture the percentage of the population living in a dwelling with a leaking roof, damp walls, floors, or foundation, or rot in window frames or floors. In this case, the higher the value of the EP index, the higher the degree of energy poverty. The researchers give the inability to keep the home adequately warm a higher value as they argue that this indicator has greater importance, as proposed in the literature (Thompson and Neil, 2013; Bouzarovski and Tirado Herrero (2017)). According to this indicator, nearly half of the population is affected by at least some dimension of energy poverty (Figure 15). In this case, a significant share (seven out of 10) of the poor based on the at-risk of poverty are also energy poor (Figure 16). Figure 15: Multidimensional energy poverty rate Figure 16: Overlap of at-risk of poverty and (2019/20) multidimensional energy poverty rate (2019/20) 45.0 42.8 100% 40.0 0.504 35.0 50% 30.0 0.258 25.0 0.068 0.170 0% 20.0 15.0 Neither nor 10.0 Only energy poor 5.0 Only monetary poor 0.0 Energy and monetary poor Energy poverty index Source: Own estimates based on EU-SILC (2020) 3.4. Big Data We lastly experiment with big data to explore the potential of this data type to measure energy poverty. We note that this approach is experimental and should be cautiously taken. Nighttime lights have been used to measure urbanization dynamics (Zhou, 2012), population, local economic activity, and energy use (Goldblatt, 2016; Shi et al., 2014; N. Zhao,2017; Amaral et al., 2005; Townsend et al. 2010, Coscieme et al. 2014; Shi et al., 2014; Fehrer and Krarti (2018)). We follow the approach of Fehrer and Krarti (2018) and argue that nighttime light satellite data could be used as a proxy for building energy use intensity. The methodology presented in this study offers a replicable approach globally, providing opportunities for geospatial energy analysis without the challenges associated with disaggregated building energy use data collection. 42 We use the data made available by Li and Zhou (2017). Applying zonal statistics, we calculate the mean nighttime light data per municipality and plot it in Figure 17. 42 The study addresses the challenge of limited data for geospatial analysis of energy use in buildings by proposing the use of satellite-derived information, specifically nighttime light imagery from the VIIRS instrument aboard the SUOMI NPP satellite. Researchers evaluated this imagery as a predictor of building energy use intensity across states, counties, and cities in the United States. The findings indicate that nighttime lights can explain a significant portion of the variability in energy consumption, reaching up to 90%, depending on conditions and geospatial scale. The results are applied to generate detailed electricity and fuel consumption maps for the United States at a resolution of less than 200 square meters. 28 Figure 17: Average Nighttime Light Data (2021) Source: Own elaboration based on data of average nighttime lights in 2021 published by Fehrer and Krarti (2018) by applying the zonal statistic package in Python. Overall, using nighttime light data to measure energy poverty can provide valuable insights into the spatial distribution and dynamics of energy usage and can help policymakers and practitioners better understand and address the challenges faced by vulnerable communities. These types of maps could be helpful in identifying areas with the most significant demands for interventions while also indicating areas with low population density or limited energy use. The map in Figure 17 indicates a significant variation of nighttime In addition, we show that different subsets light data across municipalities. Given Bulgaria's very high electrification rate (effectively 100%), darker areas (with low levels of nighttime light intensity) can be indicative of low population density, energy usage, or economic activity. Similarly, areas that report the highest values could experience high energy usage. Therefore, more detailed studies are required when using these types of datasets, and overlaying nighttime light data with data on income levels, population density, housing conditions, or access to basic services can be particularly useful in understanding energy poverty patterns at the local level. Combining big data with traditional data sources can increase the informativity of these sources. Another possibility to use these maps is to scale them by population or square meters of the ground floor, for example, to detect abnormalities. Combining big data sources with these types of data might increase their informativity. Nevertheless, when using these types of alternative data sources, it is recommended to validate the information – at least for a subset of the data - by alternative data sources, such as surveys or administrative data. 3.5. Characteristics of Energy Poor Energy poverty might have several correlates, both at the individual and contextual levels. Identifying the drivers behind energy poverty and characterizing the energy poor can be beneficial for designing policy interventions that try to target the energy poor on the one hand and tackle the mechanisms behind energy poverty on the other. Energy poverty is influenced by many factors, which exhibit variations between countries and even within geographically disaggregated regions. According 29 to the Energy Poverty Advisory Hub's report (2022), the driving forces behind energy poverty can be categorized into contextual and individual factors. Additionally, the report identifies three principal causes responsible for the occurrence of energy poverty: low income, energy inefficiency, and energy prices. To better understand what could be associated with energy poverty, we next describe the energy poor, using the multidimensional energy poverty index; characterizing the energy poor can also be helpful for the design of targeting strategies. Table 4 shows that the energy poor are more likely to reside in households with at least one unemployed, with more than five members, and with at least one pensioner when compared to households that are not energy poor. They are more likely to live in rural areas, be elderly, and disabled. We confirm that these characteristics are correlated with energy poverty incidence rates by estimating a Probit regression, using the probability of energy poverty based on the energy poverty index as a binary outcome variable. Table 5 shows that these groups are still more likely to be energy poor, even when compared with households with similar household characteristics 43. The fact that these characteristics relate to a higher probability of being energy poor can help with targeting mechanisms, such as proxy mean testing, and generate insights on what might be driving energy poverty in the first place (e.g., increased vulnerabilities, more generally speaking). Table 4: Descriptive Statistics of Energy Poor and Non-Energy Poor (1) (2) Energy poor Non-energy poor Mean sd mean sd Members 2.56 1.31 2.55 1.10 Adult equivalent 2.73 1.61 2.64 1.15 Female-headed 0.45 0.50 0.43 0.50 HH with +1 unemployed 0.25 0.43 0.12 0.33 Single-elderly HH 0.13 0.33 0.06 0.24 HH with +1 child 0.39 0.49 0.41 0.49 HH with +1 disabled 0.59 0.49 0.57 0.50 HH with +5 members 0.08 0.27 0.05 0.22 HH with +1 pensioner 0.41 0.49 0.38 0.48 Social-aid recipient HH 0.90 0.30 0.89 0.31 Urban 0.64 0.48 0.72 0.45 Female 0.45 0.50 0.43 0.49 Unemployed 0.09 0.29 0.04 0.20 Age 44.23 23.43 42.61 22.04 Elderly 0.25 0.43 0.19 0.39 Child 0.14 0.35 0.15 0.35 Disabled 0.28 0.45 0.25 0.43 Pensioner 0.28 0.45 0.24 0.43 Social aid 0.62 0.48 0.60 0.49 Observations 7,008 9,617 Notes: We define energy poverty using the multidimensional energy poverty index. "SD" is the standard deviation. Source: Own estimates based on EU-SILC (2020). 43 Unfortunately, we do not have sources of heating and cooking in the data to control for these characteristics. 30 Table 5: Determinants of energy poverty (Marginal Effects) (measured via the energy poverty index) (1) VARIABLES Energy Poverty Members -0.135*** (0.000569) Adult equ. 0.139*** (0.000554) Female-headed 0.0402*** (0.000409) HH with +1 unemployed 0.203*** (0.000641) Single-elderly HH 0.197*** (0.000907) HH with +1 child -0.161*** (0.000795) HH with +1 disabled 0.0705*** (0.000723) HH with +5 members 0.0480*** (0.00105) HH with +1 pensioner 0.00778*** (0.000631) Social-aid recipient HH -0.0316*** (0.000831) Urban -0.0425*** (0.000422) Female 0.0197*** (0.000434) Unemployed 0.0653*** (0.000981) Age -0.000325*** (1.90e-05) Elderly 0.0230*** (0.000995) Child -0.0220*** (0.00116) Disabled 0.0411*** (0.000823) Pensioner 0.0190*** (0.00107) Social aid -0.0131*** (0.000612) Observations 6,962,414 Notes: The table presents marginal effects from a Probit model that estimates the correlation between individual and household characteristics on energy poverty. Energy poverty relies on the definition of the energy poverty index (see Section 3.3). Source: Own estimates based on EU-SILC (2020). 31 4. Conclusion and Recommendations In this paper, we review existing energy poverty indicators in the academic literature, white papers, and policy reviews and then explore the employability of these indicators in Bulgaria's case. Energy poverty is a complex concept that lacks a common definition and is multidimensional by nature. It involves multisectoral and multilevel approaches. Based on these factors, measuring energy poverty is complex, and there are many different approaches to its measurement. In the work at hand, we review existing approaches and then explore their applicability to the concept of Bulgaria based on data availability and quality. We show that Bulgaria's energy poverty incidence rate fluctuates significantly by definition. The 10 percent measure deviates significantly from the other expenditure-based indicators, resulting in an approximately five times larger incidence rate. This could be related to the measurement issues and the criticism raised in the literature about this indicator. The rest of the monetary measures fluctuate between 9 and 13 percent. The incidence rates of consensual measures are slightly larger, varying between 11 and 47 percent. Our evidence demonstrates that the monetary poor and energy poor only partly overlap. 44 By construction, the monetary poor are also energy poor in the low-income-high-cost measure. If they report energy expenditure shares above the national median, this is not necessarily true for the rest of the measures. We find that, depending on the measure, only 1 to 12 percent of the population is affected by monetary and energy poverty. This has important implications for the targeting of policy interventions. As the monetary and energy poor only overlap partly, traditional social protection strategies might not target those who are energy poor, and additional policy instruments should be implemented when governments try to tackle energy poverty. In addition, we show that different subsets of the population face different dimensions of energy poverty. This means that the government might want to include a different population set for interventions that target energy poverty in the short and long-run, for example. Those needing renovation programs and interventions that increase their energy efficiency might differ from those who cannot afford their energy bills. Based on our literature review, we recommend moving away from single-indicator energy poverty metrics toward multiple-indicator approaches, which are consistent with previous studies. We show that the incidence rate of energy poverty fluctuates significantly depending on the underlying methodology. Based on this pattern of results, we recommend moving away from single-indicator energy poverty metrics towards multiple-indicator approaches. This recommendation is in line with recommendations made previously in the literature (see, for example, Herrero (2017) and Siksnelyte- Butkiene (2021)). It is also consistent with the latest EC recommendation developed by the Energy Poverty Advisory Hub, which emphasizes the importance of utilizing a mixed indicators approach that covers all contributing factors and accurately identifies vulnerable populations, given the complex nature of energy poverty 45. One example would be to follow previous pilots like the Belgium Energy Observatory or the French National Energy Poverty Observatory (ONPE), which both track more than one indicator. The operationalization of energy poverty measures should involve a cooperative approach between several ministries, academics, and local communities. In addition, big data sources 44 Importantly, results differ from some of the results detailed in Rude and Robayo-Abril (2024) because we use income from the EU-SILC after a statistical matching procedure in this case, while we rely on income measures reported in the HBS in the work at hand. 45 Source: Energy Poverty Advisory Hub (2023). 32 can be useful, especially when considering disaggregated statistics, but should – at least for a subset of the data – be validated by traditional data sources. To operationalize the measurement of energy poverty, Bulgaria could follow the model previously explored in Belgium. In Belgium, the King Baudouin Foundation has developed an Energy Poverty Barometer, considering the multidimensional nature of energy poverty (Meyer et al., 2018).46 The barometer considers several complementary measures of energy poverty and considers that different subsets of the population might suffer from different aspects of the overall concept of energy poverty (ibid). The Energy Poverty Barometer considers the following dimensions of energy poverty: measured energy poverty, hidden energy poverty, and perceived energy poverty measured, respectively, by the share of households reporting excessive energy bills compared to available income, restriction in energy consumption below basic needs, and self-reported difficulties to heat the housing adequately. In addition, Bulgaria could learn from success stories and challenges identified in other countries, such as France, when operationalizing energy poverty. The EU Energy Poverty Observatory (EPOV) 47 (2019) published a case study outlining France's successfully implemented energy poverty measures. This initiative relied on two complementarity measures: the energy stress indicator, based on household energy expenditure and income ratio, and the indicator on the ability to keep the home adequately warm. The main challenges identified in the case study were little awareness of energy poverty, lack of data, insufficient evaluations of existing energy poverty policies to identify what works and does not work, and lack of coordination between the involved stakeholders. Moreover, the chosen indicator tracked to measure progress in the fight against energy poverty should reflect the aspect of energy poverty that the government aims to address. Suppose the government aims to implement long-term approaches to address energy poverty via renovation programs, for example. In that case, observational indicators about the square meters renovated or the energy efficiency of buildings and consensual metrics on the share of households experiencing leakages might be most appropriate. If governments are most interested in tackling the income component of energy poverty via social protection programs, for example, a measure incorporating total household income might be most appropriate. Similarly, if policymakers wish to address the dimension around low access to high-quality energy, measures that track connectivity to energy might be most suited. Regarding expenditure-based approaches, we recommend relating energy expenditure to the measure used to define monetary poverty. This means that countries should use total household expenditure to scale energy expenditure in the case of relative measures when they also use consumption to define poverty. If their welfare measures rely on income-based approaches, they should relate energy expenditure to household income. Nevertheless, this approach should only be applied if income information from expenditure surveys is reliable. If income is unreliable, researchers and statisticians can explore alternative approaches to combine information on expenditure and 46 For more information, see the website: https://kbs-frb.be/fr/barometre-de-la-precarite-energetique-2009-2017 47 The European Energy Poverty Observatory (EPOV) initiative receives funding from the European Commission and is executed by a consortium of 13 diverse organizations, including universities, advocacy groups, think tanks, and representatives from the business sector. The University of Manchester leads this consortium. EPOV has several objectives, including enhancing information transparency and policy effectiveness by consolidating diverse data and knowledge sources existing across the EU. Additionally, it aims to offer an easily accessible resource that fosters public engagement and supports well-informed decision-making for local, national, and EU- level authorities. The Observatory strives for a comprehensive understanding of energy poverty in the European Union and intends to aid various stakeholders in their efforts. 33 income collected in different surveys, such as statistical matching. For the application in Bulgaria and the related performance evaluation, see Rude and Robayo-Abril (2024). Moreover, Bulgaria would benefit from improving the data environment around measuring energy poverty based on three factors. First, greater granularity, both geographically speaking and concerning the frequency of data collection, could help with local implementation approaches and tracking of energy poverty over time. Next, harmonizing datasets across countries could facilitate the comparison of energy poverty rates across countries and identify those most in need at the European level. Third, close collaborations between academia, research institutions, analysts, and the government could create valuable knowledge and innovative approaches in this area, especially using big data sources. The measurement of the definition of energy poverty and the development of all energy poverty policies should be based on meaningful and transparent processes of public involvement and inclusive engagement with various stakeholders. Establishing a common understanding of what energy poverty entails and how it is measured helps ensure that different stakeholders, including policymakers, researchers, and communities, are on the same page. This shared understanding forms the foundation for policies aiming to reduce energy poverty. Given the multidimensional nature of the issue, appropriate coordination and involvement of relevant stakeholders in energy poverty reduction efforts is critical, including national statistical offices, finance ministries, sectoral ministries, and other significant entities (at local and national levels). A broad consultation process can improve the acceptance of a methodological approach, including consensus on the minimum comfort levels. Finally, energy poverty observatories are also invaluable tools for addressing and combatting energy poverty; in the EU, several countries have established country-specific energy poverty observatories. Well-designed energy poverty observatories work as a platform that unites practitioners, decision- makers, and researchers addressing energy poverty in the country and beyond. The growing importance of evidence-based energy poverty policy-making at the European Union (EU) level underscores the necessity for establishing a dedicated decision-support hub in Bulgaria. As an EU Member State with some of Europe's highest energy poverty rates, Bulgaria faces pronounced structural challenges at the intersection of energy, low incomes, and poor housing. The Energy Poverty Observatory can help comprehensively tackle these issues. First, it provides a means to accurately quantify the current levels of energy poverty within a region or country. This quantification is vital for understanding the extent of the issue, identifying the affected population, and enabling targeted interventions. In the Bulgarian context, the observatory can help develop a rigorous and comprehensive set of indicators to operationalize the new legal energy poverty definition and create a central repository of these indicators. Second, the observatory can facilitate ongoing monitoring of energy poverty trends and developments. This longitudinal or temporal analysis helps assess the effectiveness of implemented policies and interventions, guiding adjustments or improvements where necessary. Third, it can be crucial to identify the multifaceted factors contributing to energy poverty and develop evidence-based policy recommendations. Lastly, it plays a crucial role in convening key stakeholders to understand energy poverty comprehensively across different sectors and levels48. By pinpointing these factors, policymakers and stakeholders can tailor strategies to mitigate the root causes and alleviate the burden of energy poverty on vulnerable 48 This involves a broad spectrum of national and local decision-makers, encompassing the energy regulator, the Ministry of Labour and Social Policy, the Social Protection Agency, the Ministry of Health, Ministry of Regional Development and Public Works, the Ministry of Energy, the Sustainable Energy Development Agency, local authorities, the statistical office, and the Energy and Water Regulatory Commission (EWRC). Additionally, it includes practitioners from the energy, housing, and social sectors, advocacy groups, the media, other EU observatories, development support agencies, the private sector, and the research community. 34 communities. Overall, energy poverty observatories serve as indispensable tools in the fight against energy poverty, aiding in informed decision-making and the design of effective interventions. References Amaral, S., Câmara, G., Monteiro, A. M. V., Quintanilha, J. A., & Elvidge, C. D. (2005). Estimating population and energy consumption in Brazilian Amazonia using DMSP night-time satellite data. Computers, Environment and Urban Systems, 29, 179–195. http://dx.doi.org/10.1016/j.compenvurbsys.2003.09.004. Asghar, N., Amjad, M. A., ur Rehman, H., Munir, M., & Alhajj, R. (2022). Achieving sustainable development resilience: Poverty reduction through affordable access to electricity in developing economies. Journal of Cleaner Production, 376, 134040. Bhatia, M. and N. Angelou (2015). Beyond Connections Energy Access Redefined. Tech. Rep. Washington, DC: Energy Sector Management Assistance Program, World Bank Group, pp. 1–244. Berger, M., & Worlitschek, J. (2018). A novel approach for estimating residential space heating demand. Energy, 159, 294-301. Boardman, B. (1991). Fuel poverty: from cold homes to affordable warmth. Pinter Pub Limited. Bollino, C. A., & Botti, F. (2017). Energy poverty in Europe: A multidimensional approach. PSL Quarterly Review, 70(283). Bouzarovski, S, Petrova, S, (2015), A global perspective on domestic energy deprivation: Overcoming the energy poverty–fuel poverty binary, Energy Research & Social Science, Volume 10, Pages 31-40, ISSN 2214-6296, https://doi.org/10.1016/j.erss.2015.06.007 Bouzarovski, S., & Tirado Herrero, S. (2017). The energy divide: Integrating energy transitions, regional inequalities, and poverty trends in the European Union. European Urban and Regional Studies, 24(1), 69-86. https://doi.org/10.1177/0969776415596449. Coscieme, L., Pulselli, F. M., Bastianoni, S., Elvidge, C. D., Anderson, S., & Sutton, P. C. (2014). A thermodynamic geography: Night-time satellite imagery as a proxy measure of emergy. Ambio, 43, 969–979. http://dx.doi.org/10.1007/s13280-013-0468-5. Culver, L. C. (2017, May). Energy poverty: What you measure matters. In Proceedings of the reducing energy poverty with natural gas: Changing political, business and technology paradigms symposium, Stanford, CA, USA (pp. 9-10). Day, R., Walker, G., & Simcock, N. (2016). Conceptualizing energy use and energy poverty using a capabilities framework. Energy Policy, 93, 255-264. Energy Poverty Advisory Hub (2022). Introduction to the Energy Poverty Advisory Hub (EPAH) Handbooks: A Guide to Understanding and Addressing Energy Poverty Published by the Energy Poverty Advisory Hub 35 Energy Poverty Advisory Hub (2023). National Indicators: "Uncovering New Possibilities Expanded Knowledge" https://energy-poverty.ec.europa.eu/system/files/2023- 10/EPAH2023_2nd%20Indicators%20Report_Final_0.pdf EU Energy Poverty Observatory (EPOV) (2019). Case Study: French National Energy Poverty Observatory (ONPE). Link: https://energy-poverty.ec.europa.eu/system/files/2021- 10/epov_case_study_france.pdf European Commission (2023). Energy poverty in the EU. Link: https://energy.ec.europa.eu/topics/markets-and-consumers/energy-consumer-rights/energy- poverty-eu_en European Commission (2023b). Smart grids and meters. Link: Smart grids and meters (europa.eu) European Union (2016). Energy Poverty Handbook. Link: https://bpie.eu/wp- content/uploads/2016/11/energypovertyhandbook-online.pdf Fehrer, D., and Krarti, M. (2018). Spatial distribution of building energy use in the United States through satellite imagery of the earth at night, Building and Environment, Volume 142, Pages 252-264, ISSN 0360-1323, https://doi.org/10.1016/j.buildenv.2018.06.033. Fergus, P., & Chalmers, C. (2021). Energy sobriety: a behaviour measurement indicator for fuel poverty using aggregated load readings from smart meters. In Towards Energy Smart Homes: Algorithms, Technologies, and Applications (pp. 21-47). Cham: Springer International Publishing. Frei et al. (2021). Front. Built Environ., 10 February 2021 Sec. Sustainable Design and Construction Volume 6 - 2020 | https://doi.org/10.3389/fbuil.2020.609877 Freund, C. L., & Wallich, C. I. (1996). The welfare effects of raising household energy prices in Poland. The Energy Journal, 17(1). Goldblatt, R., Hanson, G., Heilmann, K., & Khandelwal, A. (2016). What's the matter with nightlights? Using Landsat imagery to improve nightlight-based measures of local economic activity with an application to India (pp. 1–18).González-Eguino, M. (2015). Energy poverty: An overview. Renewable and Sustainable Energy Reviews, 47, 377-385. https://doi.org/10.1016/j.rser.2015.03.013 Gouveia et al. (2022). Energy Poverty. National Indicators. Insights for a more effective measuring. Link: https://energy-poverty.ec.europa.eu/system/files/2023- 01/EPAH_Energy%20Poverty%20National%20Indicators%20Report_0.pdf Government of the UK. English Housing Survey. Link: https://www.gov.uk/government/collections/english-housing-survey Halkos, G. E., & Gkampoura, E. C. (2021). Coping with energy poverty: Measurements, drivers, impacts, and solutions. Energies, 14(10), 2807. Herrero, S. T., & Bouzarovski, S. (2014). Energy transitions and regional inequalities in energy poverty trends: Exploring the EU energy divide. Herrero, S. T. (2017). Energy poverty indicators: A critical review of methods. Indoor and Built Environment, 26(7), 1018-1031. Hills, J. (2011). Fuel poverty: the problem and its measurement. Interim report of fuel poverty. URL http://sticerd.lse.ac.uk/dps/case/cr/CASEreport69.pdf 36 885 January_27,_2016]. Center of analysis of social exclusion. Report 69. ISSN 1465-3001. Hills, J. (2012). Getting the measure of fuel poverty: Final Report of the Fuel Poverty Review. Link: https://sticerd.lse.ac.uk/dps/case/cr/casereport72.pdf IBroad (2018). Factsheet: Bulgaria. Current use of EPCs and potential links to iBRoad. Link: http://ibroad-project.eu/wp-content/uploads/2018/01/iBROAD_CountryFactsheet_BULGARIA.pdf King Baudouin Foundation, 2018. The energy poverty barometer. Link: https://media.kbs- frb.be/en/media/7604/20180315NT.pdf Laderchi et al. (2013). Balancing Act: Cutting Energy Subsidies While Protecting Affordability. https://books.google.de/books?hl=de&lr=&id=D9hOnwAFBVQC&oi=fnd&pg=PP2&dq=balancing+act +2013+energy+poverty&ots=NYVSy6Y91U&sig=P7hQH_rJSBxeC1KEamqaG0W0ENc#v=onepage&q&f =false Lan, J., Khan, S. U., Sadiq, M., Chien, F., & Baloch, Z. A. (2022). Evaluating energy poverty and its effects using multidimensional based DEA-like mathematical composite indicator approach: Findings from Asia. Energy Policy, 165, 112933. Li, Xinyi, et al. (2018) "Developing urban residential reference buildings using clustering analysis of satellite images." Energy and Buildings 169 (2018): 417-429. Li, X., & Zhou, Y. (2017). A stepwise calibration of global DMSP/OLS stable nighttime light data (1992– 2013). Remote Sensing, 9(6), 637. Ma, T., Zhou, C., Pei, T., Haynie, S., & Fan, J. (2012). Quantitative estimation of urbanization dynamics using time series of DMSP/OLS nighttime light data: A comparative case study from China's cities. Remote Sensing of Environment, 124, 99–107. http://dx.doi.org/10.1016/j.rse.2012.04.018 Mendoza Jr, C. B., Cayonte, D. D. D., Leabres, M. S., & Manaligod, L. R. A. (2019). Understanding multidimensional energy poverty in the Philippines. Energy Policy, 133, 110886. Meyer, S., Laurence, H., Bart, D., Middlemiss, L., & Maréchal, K. (2018). Capturing the multifaceted nature of energy poverty: Lessons from Belgium. Energy research & social science, 40, 273-283. Moore, R. (2009). A New Approach to Assessing Fuel Poverty. Energy Action, 108. Nguyen (2022). Energy poverty in Belgium: are you affected? Link: https://www.energyprice.be/blog/energy-poverty-in-belgium/ Nussbaumer, P., Bazilian, M., & Modi, V. (2012). Measuring energy poverty: Focusing on what matters. Renewable and Sustainable Energy Reviews, 16(1), 231-243. Palma and Gouveira (2022). Bringing Energy Poverty Research into Local Practice: Exploring Subnational Scale Analyses Published by the Energy Poverty Advisory Hub. Link: https://energy- poverty.ec.europa.eu/system/files/2022- 03/EPAH_Bringing%20Energy%20Poverty%20Research%20into%20local%20practice_final.pdf Pelz, S., Pachauri, S., & Groh, S. (2018). A critical review of modern approaches for multidimensional energy poverty measurement. Wiley Interdisciplinary Reviews: Energy and Environment, 7(6), e304. Practical Action, 2010. Poor People's Energy Outlook 2010: Energy for Cooking, Lighting and Communication. Practical Action Publishing. Pye, S., Dobbins, A., Baffert, C., Brajković, J., Deane, P., & De Miglio, R. (2017). Energy poverty across the EU: Analysis of policies and measures. In Europe's Energy Transition (pp. 261-280). Academic Press. 37 Rodriguez-Alvarez, A. Manuel Llorca,M., and Jamasb, Tooraj, 2021. Alleviating energy poverty in Europe: Front-runners and laggards, Energy Economics, Volume 103, 105575, ISSN 0140-9883, https://doi.org/10.1016/j.eneco.2021.105575. Rude, B. L., & Robayo, M. (2024). Statistical Matching for Combining the European Survey on Income and Living Conditions and the Household Budget Surveys: An Evaluation of Energy Expenditures in Bulgaria. Policy Research Working Paper No. WPS 10818, PROSPERITY; RRR. Washington, D.C.: World Bank Group. http://documents.worldbank.org/curated/en/099918006242435645/IDU18642b8e2160eb142ff18e2 41ebb95b21d260 Ruggeri Laderchi, Trimble, C, & Olivieri, A. "Balancing Act, Cutting Energy Subsidies While Protecting Affordability," 2013. Saghir, J. (2005). Energy and poverty: myths, links, and policy issues. Shi, K., Yu, B., Huang, Y., Hu, Y., Yin, B., Chen, Z., ... Wu, J. (2014). Evaluating the ability of NPP-VIIRS nighttime light data to estimate the gross domestic product and the electric power consumption of China at multiple scales: A comparison with DMSP-OLS data. Remote Sensing, 6, 1705–1724. http://dx.doi.org/10.3390/rs6021705Schuessler, R. (2014). Energy Poverty Indicators: Conceptual Issues-Part I: The Ten-Percent-Rule and Double Median/Mean Indicators. ZEW-Centre for European Economic Research Discussion Paper, (14-037). Schulte, I., & Heindl, P. (2017). Price and income elasticities of residential energy demand in Germany. Energy Policy, 102, 512-528. Siksnelyte-Butkiene, I. (2021). A systematic literature review of indices for energy poverty assessment: a household perspective. Sustainability, 13(19), 10900. Ssennono, V. F., Ntayi, J. M., Buyinza, F., Wasswa, F., Aarakit, S. M., & Mukiza, C. N. (2021). Energy poverty in Uganda: evidence from a multidimensional approach. Energy Economics, 101, 105445. Sokołowski, J., Lewandowski, P., Kiełczewska, A., & Bouzarovski, S. (2020). A multidimensional index to measure energy poverty: the Polish case. Energy Sources, Part B: Economics, Planning, and Policy, 15(2), 92-112. Sun, M., Han, C., Nie, Q., Xu, J., Zhang, F., & Zhao, Q. (2022). Understanding building energy efficiency with administrative and emerging urban big data by deep learning in Glasgow. Energy and Buildings, 273, 112331. https://doi.org/10.1016/j.enbuild.2022.112331 Thema, J., & Vondung, F. (2020). Expenditure-based indicators of energy poverty—An analysis of income and expenditure elasticities. Energies, 14(1), 8. Thomson, H. Snell, C, 2013, "Quantifying the prevalence of fuel poverty across the European Union, Energy Policy, Volume 52, Pages 563-572, ISSN 0301-4215, https://doi.org/10.1016/j.enpol.2012.10.009 Thomson H, Bouzarovski S, Snell C. 2017. Rethinking the measurement of energy poverty in Europe: A critical analysis of indicators and data. Indoor and Built Environment.;26(7):879-901. doi:10.1177/1420326X17699260 Thomson, H., Day, R., Ricalde, K., Brand-Correa, L. I., Cedano, K., Martinez, M., ... & Pereira, M. G. (2022). Understanding, recognizing, and sharing energy poverty knowledge and gaps in Latin America and the Caribbean–because conocer es resolver. Energy Research & Social Science, 87, 102475. 38 Townsend, A. C., & Bruce, D. A. (2010). The use of night-time lights satellite imagery as a measure of Australia's regional electricity consumption and population distribution. International Journal of Remote Sensing, 31, 4459–4480. http://dx.doi.org/10.1080/01431160903261005 Ullah, S., Khan, M., & Yoon, S. M. (2021). Measuring Energy Poverty and Its Impact on Economic Growth in Pakistan. Sustainability, 13(19), 10969. United Nations (2018). Policy brief 08. Interlinkages among energy, poverty, and inequality. Link: https://sustainabledevelopment.un.org/content/documents/17480PB8.pdf US Energy Information Administration. Residential Energy Consumption Survey. Link: https://www.eia.gov/consumption/residential/data/2020/index.php?view=characteristics World Bank (2022) “ECAPOV 3.0 Harmonization Guidelines” Zhao, N., Hsu, F., Cao, G., & Samson, E. L. (2017). Improving accuracy of economic estimations with VIIRS DNB image products. International Journal of Remote Sensing, 38, 5899–5918. http://dx.doi.org/10.1080/01431161.2017.1331060 39