CAF - Asymmetric Adaptation to Heat and Energy Poverty

C A FW O R KI N GPA P E R# 2 0 2 6 / 0 5

F i r s t v e r s i o n : M a r c h 2 0 , 2 0 2 2

T h i s v e r s i o n : M a r c h 2 7 , 2 0 2 6

Asymmetric Adaptation to Heat and Energy Poverty Adriana Camacho1 | Leonardo Gasparini2 | Luis Laguinge3 | Jorge Puig4 | Hernán Winkler5 1Development Contributions and Impact Measurement, CAF. acamacho@caf.com 2CEDLAS-IIE-FCE, Universidad Nacional de La Plata & CONCIET. gasparinilc@gmail.com 3CEDLAS-IIE-FCE, Universidad Nacional de La Plata & CONCIET. luislaguinge4@gmail.com 4CEDLAS-IIE-FCE, Universidad Nacional de La Plata. jppuig@gmail.com 5Poverty and Equity Global Practice LAC, World Bank. hwinkler@worldbank.org We study the responses of household electricity consumption to temperature changes, focusing on asymmetries between welfare deciles. Our analysis exploits a unique panel dataset for Peru that links household survey microdata with repeated administrative records on energy use and local temperature. Using fixed-effects models, we estimate how electricity consumption varies with temperature, highlighting the unequal capacity of households across income deciles to adapt to climate change. Based on this evidence, we propose and implement a novel measure of adaptive energy poverty, which captures households’ ability to respond to rising ambient temperatures through increased electricity consumption.

K E Y W O R D S

electricity, temperature, energy poverty, Peru Small sections of text that are less than two paragraphs may be quoted without explicit permission as long as this document is acknowledged. Findings, interpretations and conclusions expressed in this publication are the sole responsibility of its author(s) and cannot be, in any way, attributed to CAF, its Executive Directors or the countries they represent. CAF does not guarantee the accuracy of the data included in this publication and is not, in any way, responsible for any consequences resulting from its use. ©2026 Corporación Andina de FomentoC A FD OC UM E NT O D E T R AB AJ O # 2 0 2 6 / 0 5

P r i m e r a v e r s i ó n : 2 0 d e m a r z o d e 2 0 2 2

E s t a v e r s i ó n : 2 7 d e m a r z o d e 2 0 2 6

Adaptación asimétrica al calor y pobreza energética Adriana Camacho1 | Leonardo Gasparini2 | Luis Laguinge3 | Jorge Puig4 | Hernán Winkler5 1Development Contributions and Impact Measurement, CAF. acamacho@caf.com 2CEDLAS-IIE-FCE, Universidad Nacional de La Plata & CONCIET. gasparinilc@gmail.com 3CEDLAS-IIE-FCE, Universidad Nacional de La Plata & CONCIET. luislaguinge4@gmail.com 4CEDLAS-IIE-FCE, Universidad Nacional de La Plata. jppuig@gmail.com 5Poverty and Equity Global Practice LAC, World Bank. hwinkler@worldbank.org Este artículo analiza cómo responde el consumo eléctrico de los

hogares a las variaciones de temperatura, con especial atención a las asimetrías entre deciles de bienestar. El estudio se basa en un panel de datos único para Perú que combina microdatos de encuestas de hogares con registros administrativos longitudinales sobre consumo energético y temperatura local. Mediante modelos de efectos fijos, estimamos de qué manera varía el consumo eléctrico ante cambios en la temperatura, poniendo en evidencia las diferencias en la capacidad de adaptación al cambio climático entre hogares pertenecientes a distintos deciles de ingreso. A partir de esta evidencia, proponemos y aplicamos una nueva medida de pobreza energética adaptativa, orientada a captar la capacidad de los hogares para responder al aumento de la temperatura ambiental mediante un mayor uso de electricidad. K E Y W O R D S electricidad, temperatura, pobreza energética, Perú Pequeñas secciones del texto, menores a dos párrafos, pueden ser citadas sin autorización explícita siempre que se cite el presente documento. Los resultados, interpretaciones y conclusiones expresados en esta publicación son de exclusiva responsabilidad de su(s) autor(es), y de ninguna manera pueden ser atribuidos a CAF, a los miembros de su Directorio Ejecutivo o a los países que ellos representan. CAF no garantiza la exactitud de los datos incluidos en esta publicación y no se hace responsable en ningún aspecto de las consecuencias que resulten de su utilización. ©2025 Corporación Andina de FomentoCAMACHO ET AL. 2 1|INTRODUCTION Electricity consumption is central to households’ capacity to adapt to weather conditions

(Deschênes, 2014).1 Yet, this adaptive capacity is unequally distributed: income and other socioeconomic characteristics shape both households’ access to cooling and heating technologies and their patterns of energy use. In this context, understanding how electricity consumption responds to temperature changes —and how these responses vary across socioeconomic groups— is crucial for informing energy, social, and climate policies. This paper contributes to the growing literature on the relationship between temperature and electricity consumption, with a particular focus on distributional asymmetries. To that end, we study the case of Peru. Although electricity access in that Latin American developing country is widespread, substantial disparities persist in usage intensity, affordability, and vulnerability to climate-related risks (Tornarolli and Puig, 2023). We exploit a novel dataset that combines three sources of information: nationally representative household survey data, administrative records on electricity consumption, and high-frequency temperature information. This integration enables the construction of a monthly panel at the household level, covering the period 2015 to 2023. The richness of the dataset allows for identification of heterogeneous consumption responses to temperature changes across the well-being distribution. Methodologically, we rely on an econometric approach based on semi-parametric fixed effects models, allowing a flexible specification that captures nonlinear temperature effects while controlling for unobserved heterogeneity across time and space (Davis and Gertler, 2015; Harish et al., 2020; Zhang et al., 2022). The econometric results confirm that electricity consumption in Peru responds nonlinearly to temperature: it remains stable between mean monthly temperature levels of 16°C and 19°C, and rises sharply beyond that threshold, reaching increases of up to 25 kWh per month when temperatures approach 30°C. The analysis by welfare level reveals substantial

heterogeneity: while low-income households show almost no increase in consumption even under extreme heat, higher-income households react much more strongly once temperatures exceed 19°C, widening the energy gap across deciles. For instance, when monthly average temperatures reach 25ºC, electricity consumption in the bottom decile shows no response, whereas households in the top decile increase their usage by about 15 kWh. At 30ºC, the top decile adapts with an increase of nearly 30 kWh, while the bottom decile raises consumption by only around 4 kWh. The robustness of these findings is confirmed by alternative specifications and different measures of temperature exposure, underscoring both the asymmetric nature of the temperature–electricity relationship and the unequal adaptive capacity across socioeconomic groups. A natural extension of our analysis is to examine the implications of temperature shocks through the lens of energy poverty. Energy poverty broadly refers to households’ inability to secure adequate energy services for basic needs and well-being. We begin by applying conventional approaches to its measurement, focusing on both unidimensional and multidimensional indicators: the Ten Percent Rule Index (TPRI) and the Multidimensional Energy Poverty Index (MEPI). The TPRI remained relatively stable at around 3–4% of households between 2015 and 2019, but more than doubled in 2020 as household incomes fell sharply during the pandemic, while real energy expenditures remained broadly unchanged. Since then, it has stabilized at a higher level of 6–7%. The MEPI shows a smoother trajectory: energy poverty incidence gradually declined from 27% in 2015 to 23% in 2019, then rose again to almost 28% during the pandemic and has remained at that level since. 1Adaptation, according to the Intergovernmental Panel on Climate Change (IPCC), is defined as “adjustment in natural or human systems in response to actual or expected climatic stimuli or their effects, which moderates

harm or exploits beneficial opportunities” (Barreca et al., 2016).CAMACHO ET AL. 3 We propose a novel measure of energy poverty that complements existing access and expenditure-based indicators by capturing households’ limited capacity to increase electricity use in response to higher temperatures. The idea behind this behaviorally grounded concept of "adaptive energy poverty" is that energy deprivation arises not only from lack of access or affordability, but also from limited ability to adjust energy consumption when environmental conditions become more demanding. Households that fail to increase their consumption in response to heat exposure may face hidden vulnerabilities, even if they are formally connected to the grid. Applying this measure to Peru, we find that around 21% of households can be classified as energy poor under our benchmark parameters. The incidence is strongly graded by welfare: nearly 60% in the bottom welfare decile versus only 12% in the top decile. Substantial regional disparities also emerge, with the prevalence of adaptive energy poverty ranging from 8% in Arequipa to over 30% in La Libertad. Moreover, even among households with electricity access, 17% remain energy poor, reflecting constraints linked to low appliance ownership or financial limitations. Comparisons with standard metrics show moderate overlap with income poverty and the MEPI, but virtually no correlation with the Ten Percent Rule Index, underscoring the added value of this behavioral approach in uncovering forms of deprivation that remain invisible to conventional measures. As a byproduct, this paper underscores the importance of using administrative data on energy consumption. First, it corroborates the well-documented issue of substantial underreporting in national household surveys, which introduces substantial biases in estimating both the relationship between electricity use and temperature and the incidence of energy poverty. These distortions undermine the basis for sound policy design. Second, administrative records offer repeated observations of the same households over time, enabling panel data analysis. By contrast, survey-based data—even in a country like Peru with a short-panel structure—provide very few repeated observations per household, often restricted to the same month of the year, which severely limits the ability to study household responses to temperature variation. Our paper is related to at least three strands of literature. First, we speak to a growing body of work at the intersection of energy demand, climate change, and environmental economics that examines how households adjust electricity usage in response to temperature changes. Several studies have documented significant short-run responses of electricity consumption to temperatures, particularly in high-income countries with widespread access to cooling and heating technologies (Bessec and Fouquau, 2008; Auffhammer and Aroonruengsawat, 2011; Deschênes and Greenstone, 2011; Lee and Chiu, 2011; Kang and Reiner, 2022). More recent work has begun to explore these dynamics in developing countries, where behavioral and infrastructural constraints may limit households’ adaptive responses (Davis and Gertler, 2015; Li et al., 2018; Du et al., 2020; Harish et al., 2020; Zhang et al., 2022). Methodologically, this literature has employed a range of econometric strategies to test the asymmetric nature of the temperature–electricity relationship. Early studies often relied on parametric specifications—such as quadratic forms—that impose symmetric responses and strong functional assumptions. In contrast, more recent work has favored semi-parametric and non-parametric models that allow for flexible estimation without imposing such restrictions (Bessec and Fouquau, 2008; Deschênes and Greenstone, 2011; Gupta, 2012; Auffhammer, 2022). Within that body of work, evidence from high-income settings typically finds U-shaped or strongly asymmetric responses, with sharp increases

in electricity use on very hot (and, in some contexts, very cold) days. 2 Panel and smooth2For example, Deschênes and Greenstone (2011) use a semi-parametric panel approach for strong consumption responses to temperature extremes in the United States: annual residential energy consumption increases by approximately 0.4% for days with temperatures above 32ºC (90ºF), and by 0.3% for days below -12ºC (10ºF).CAMACHO ET AL. 4 transition approaches across the US and Europe reveal temperature thresholds and growing sensitivity to summer heat. 3 Specifically Bessec and Fouquau (2008) identified a clear heating effect in the European Union, whereas the cooling effect is less important. Bessec and Fouquau (2008) distinguish between Northern and Southern countries and show that the non-linear pattern is more pronounced in warmer countries (i.e., U-shaped relationship) than in the cold ones. In developing countries, where adaptation capacity is shaped by infrastructure gaps and income inequality, studies from Mexico, China, and India show muted reactions to cold, pronounced increases on hot days, and "hockey-stick" patterns concentrated among higherincome or better-equipped households—pointing to a central role for appliance ownership, housing quality, and related constraints.4 Our analysis contributes new evidence from Peru, a setting with marked climatic and socioeconomic heterogeneity.5 Second, our analysis contributes to the literature on economic inequality in developing countries—and in Peru, in particular. A large body of research has highlighted how inequalities in income, geography, and access to public goods shape household well-being and resilience to shocks (Escobal and Torero, 2005; Gasparini et al., 2013; Tornarolli and Puig, 2023). By documenting how well-being groups differ in their ability to adjust electricity

consumption in response to climate events, we uncover a less explored yet critical dimension of inequality: differential adaptive capacity. Our approach complements traditional inequality metrics by using behavioral responses to exogenous variation as a window into the structural constraints faced by disadvantaged households. Finally, we offer new insights into the concept of energy poverty, broadly understood as the inability to obtain adequate energy services for basic needs and well-being (Boardman, 1991; Foster et al., 2000; Bouzarovski and Petrova, 2015). While most empirical work on the relationship between energy poverty and temperature variation has focused on access, affordability, or energy expenditures6, we argue that the limited responsiveness of electricity consumption to extreme temperatures may also signal energy poverty—reflecting constraints in the ability to increase usage when needs rise due to heat or cold. This behavioral perspective, grounded in observed reactions to climate variation, complements existing static indicators and may be particularly valuable in identifying hidden vulnerabilities 3Bessec and Fouquau (2008) and Lee and Chiu (2011) adopt panel smooth transition models to estimate temperature thresholds across European and OECD countries, revealing substantial regional heterogeneity and a growing sensitivity to summer temperatures over time. 4In Mexico, Davis and Gertler (2015) finds virtually no response to cold days —consistent with the absence of electric heating— but estimates that for days with temperatures above 32ºC (90ºF), monthly electricity consumption increases by 3.2%, with stronger effects in regions with greater air conditioning prevalence. In China, several studies report asymmetric patterns: electricity use rises significantly on hot days but shows muted or no response to cold days due to the widespread use of centralized heating in the north of the country (Li et al., 2018; Du et al., 2020; Zhang et al., 2022). For instance, Zhang et al. (2022) find that each additional

day above 32ºC raises annual electricity consumption by 8.9%, with a disproportionately smaller effect among rural and low-income households. These differences reflect variations in air conditioning ownership and usage, which are strongly correlated with income and location. Similar heterogeneity is documented in India by Harish et al. (2020), who show that electricity demand in Delhi increases by 30–43% on days between 30ºC and 39ºC among high-income households. In contrast, low-income groups show little to no increase, constrained by limited appliance ownership and poorer housing conditions. 5Miranda Montero and Contreras (2025) evaluate the impact of higher temperatures in Peru, not on electricity consumption but on a different outcome: learning. Their results suggest that one degree above 20°C is equivalent to 7 and 6% of a standard deviation of what a student learns in a year for math and reading tests, respectively. 6Most studies in this line of research estimate the impact of temperature anomalies on conventional measures of energy poverty—such as the 10% income threshold rule proposed by Boardman (1991), the Multidimensional Energy Poverty Index (MEPI), or the Low Income–High Cost (LIHC) indicator (e.g. Feeny et al. (2021), Awaworyi Churchill et al. (2022), and Li et al. (2023)).CAMACHO ET AL. 5 among formally connected but underserved households. The remainder of the paper is organized as follows. Section 2 describes the main data sources used in the analysis. Section 3 presents a descriptive analysis of electricity consumption at the household level, exploiting matched data from national household surveys and administrative records. Section 4 expands the analysis using an econometric approach, presenting the main results together with heterogeneity analyses and extensions. The next two sections address the issue of energy poverty. Section 5 presents estimates of two traditional measures: the Ten Percent Rule Index and the Multidimensional Energy Poverty Index. Section 6 introduces and implements a novel measure of adaptive energy poverty, which reflects households’ ability to cope with rising temperatures through increased electricity consumption. Finally, Section 7 provides a summary of the results and a concluding discussion. Additional materials are provided in the Appendix. 2|DATA We draw information from three main sources. First, we use Peru’s national household survey (ENAHO), which, in addition to providing standard socioeconomic information on households, includes detailed questions on expenditures, including electricity consumption. Second, we have access to unique administrative data that matches households in ENAHO with records of residential electricity meters. Finally, we exploit information on climate changes (mostly temperature) mapped at the household’s Primary Sampling Unit (PSU) level. The remainder of this section provides more details on each of these data sources. 2.1|National household survey The ENAHO is Peru’s primary nationally representative household survey, conducted continuously by the Instituto Nacional de Estadística e Informática (INEI). It collects detailed information on demographic characteristics, income, labor market outcomes, housing conditions, and household expenditures. The survey covers both urban and rural areas across all 24 departments and the Constitutional Province of Callao. Each year, ENAHO interviews over 35,000 households, which, when weighted, represent approximately 10 million households nationwide. For this study, we use the waves from 2015 to 2023. ENAHO’s stratified probabilistic sampling design ensures representativeness at the national, regional, and urban/rural levels, making it well-suited for distributional analysis and for linkage with external sources at the PSU level. Although ENAHO has a rotating short-panel structure, most of the analysis in this paper exploits it as a series of repeated cross-sections. The

short-panel dimension is used only for selected robustness exercises. Our main source of variation over time comes from administrative records on electricity consumption, which we describe in the following subsection. 2.2|Electricity consumption We use administrative records from the Organismo Supervisor de la Inversión en Energía y Minería (OSINERGMIN), which provide monthly data on residential electricity consumption, billing amounts, tariff codes, and supply characteristics for the period 2015–2023. This dataset contains over 11 million observations at the household-meter level, and includes key variables such as energy consumption in kilowatt-hours (kWh), total billing in Peruvian soles, and collective supply factors. To incorporate this information into our analysis, the National Statistical Institute ofCAMACHO ET AL. 6 Peru (INEI) matched the OSINERGMIN records with the ENAHO household sample (over 200 thousand observations). The primary linkage was performed using household addresses, leveraging fields such as street name, lot number, and door number. When available, national ID numbers (DNI) of the account holder were utilized. The matching process involved multiple stages: an exact match on addresses where possible, followed by approximate string matching techniques (Bigram similarity >0.85 and Jaro-Winkler>0.9) to identify additional matches based on the textual similarity of street names. At later stages, unmatched households were assigned average electricity consumption at the block (manzana) or district level using verified geospatial identifiers (UBIGEO codes). Overall, this procedure allowed us to successfully link approximately 65% (average for the period 2015-2023) of ENAHO surveyed dwellings to administrative electricity consumption records. The quality of the linkage varies across years and matching levels. On average, about 31% of dwellings were matched at the exact household level (same dwelling), around 4% at the block level, and close to 63% at the district level. Table A.1 in

the Appendix provides the annual breakdown of matches by level of precision. To ensure internal consistency, we performed additional data cleaning steps: we excluded households that reported having access to electricity in ENAHO but registered zero consumption in the administrative data, and we truncated extreme or implausible values. For reference, the highest monthly residential consumption reported in the Encuesta Residencial de Consumo y Usos de Energía (ERCUE) conducted by OSINERGMIN is around 1,100 kWh; we dropped a small number of outliers that exceeded this threshold. 2.3|Temperature changes Temperature data is sourced from the Climate Hazards Center InfraRed Temperature with Stations – daily (CHIRTS-daily) dataset, developed by the Climate Hazards Center at the University of California, Santa Barbara.7 CHIRTS-daily is a quasi-global (60°S–70°N), highresolution dataset (0.05° × 0.05°, approximately 5 km) that combines satellite infrared data with ground station observations and ERA5 reanalysis fields. The dataset was developed to address the scarcity of reliable temperature data in regions with limited station coverage, and is particularly suited for evaluating climate-related risks in areas vulnerable to food insecurity and extreme weather events. To assign temperature values to each household, geospatial buffers are constructed around the centroid of each PSU.8 Each buffer represents a 10-kilometer radius around the PSU centroid, and is used to extract localized climate conditions from the CHIRTS-daily gridded surface. A separate shapefile with approximately 5,000 PSU-level polygons is used for each year in the dataset. These buffers allow us to match temperature data with sufficient spatial precision, reflecting meaningful variation in climatic exposure across the Peruvian territory. For simplicity, most of the analysis relies on monthly averages of daily minimum and maximum temperatures for each location. For robustness, we also consider alternative

measures based on daily temperature data, such as the number of days per month exceeding a given threshold. Figure A.1a in the Appendix shows the histogram of the monthly mean of the daily average temperature —calculated as the mean between the daily minimum and maximum—for each location, expressed in degrees Celsius (°C). In most locations, mean temperatures fall between 10°C and 30°C. For this reason, we restrict the sample to that range in parts of our analysis. Most locations exhibit mean daily minimum temperatures between 0°C and 26°C (Figure A.1b), and mean daily maximum temperatures between 7https://www.chc.ucsb.edu/data/chirtsdaily 8We also have data on precipitation, although it is not the focus of our analysis and is used only as a control variable.CAMACHO ET AL. 7 19°C and 29°C (Figure A.1c). Figure A.2 displays a map of Peru showing mean monthly temperatures, in degrees Celsius by location. 2.4|Data integration and structure To construct our dataset, we first append all available waves of the ENAHO survey from 2015 to 2023. Each household is then matched with OSINERGMIN administrative records of electricity consumption, billing amounts, and tariff structure, as described in the previous subsections. The resulting dataset is reshaped into a panel format, where the unit of observation is a household–month. This structure allows us to track the monthly electricity use for each surveyed household over time. Finally, we merge high-frequency weather data using the geographic location of the household’s PSU, which captures localized temperature variation. This integrated panel allows us to exploit within-household temporal variation in both electricity consumption and temperature exposure. In addition, the richness of the ENAHO survey enables us to capture substantial heterogeneity in regional conditions and socioeconomic characteristics. In particular, we classify households into well-being strata based on

per capita household total expenditure. 9 To ensure comparability over time, these monetary variables are deflated using the Consumer Price Index (CPI, base year 2023), yielding real, time-consistent measures of household welfare. We then construct expenditure deciles for the full sample, which serve as key stratification variables in our empirical analysis. 3|DESCRIPTIVE EVIDENCE In this section, we present descriptive evidence on electricity consumption in Peru over the period of study, highlighting changes associated with temperature changes, as well as asymmetric responses across the household per capita expenditure distribution. The evidence provided here offers suggestive insights into patterns that will be examined in greater detail in the following section. For simplicity, the evidence in this section is constructed using pooled data for the period 2015–2023. All monetary values are expressed in soles at constant 2023 prices. Figure 1 shows the access to electricity and to two appliances that require electricity to operate—a TV and a refrigerator—across the distribution of per capita household expenditures.10 As expected, the gradient is positive in all cases, especially for the appliances. This is relevant to our discussion because it suggests that vulnerable population may have access to electricity (one of the traditional indicators of energy poverty) but may lack very basic appliances, such as a TV and a fridge, to fully benefit from that service.This suggests a form of hidden energy deprivation: households technically connected to the grid but lacking appliances to utilize electricity for cooling or food preservation. This evidence is consistent with the findings of Tornarolli and Puig (2023), who document large inequalities in the quality and intensity of electricity use across income groups in Peru, even among households formally connected to the grid. 9Throughout most of the analysis, we use per capita household total expenditures as a proxy for well-being. Results are robust to the use of per capita income as an alternative well-being measure.

10Regarding appliances, we use TV and refrigerator, given that the survey does not include specific devices to control temperatures, such as fan or air conditioning.CAMACHO ET AL. 8 F I G U R E 1Access to electricity, refrigerator, and TV (%).Notes:all values are weighted using the population expansion factor. Deciles constructed from household per capita expenditure.

Source: own calculations based on a pool dataset from ENAHO 2015-2023.

F I G U R E 2Monthly electricity consumption.Notes:all values are weighted using the population expansion factor. Deciles constructed from household per capita expenditure.Source:own calculations based on a pool dataset from matched ENAHO–OSINERGMIN data. The matched ENAHO-OSINERGMIN data enables a more detailed and precise analysis of electricity consumption. Figure 2 shows the average monthly consumption in kWh by deciles. On average for the matched sample, mean monthly electricity consumption is 126.5 kWh. To validate this result, we compare it with those obtained from the ERCUE. The last edition (2023) shows that the monthly average consumption of electricity was 133 kWh at the national level. These similarities persist when the sample is restricted to specific regions or areas. For example, the average consumption reported in urban (rural) areas is 151 kWh (40 kWh) in the matched data and 145 kWh (40 kWh) in ERCUE. The case of the metropolitan area of Lima is also illustrative: although the average consumption reported in the matched data is higher than that reported in ERCUE (181 vs. 235 kWh), both surveys are able to capture the higher consumption in this region. The consistency betweenCAMACHO ET AL. 9 the c