Residential electricity consumption and adaptation to climate change by Colombian households

McRae, Shaun D. Working Paper Residential electricity consumption and adaptation to climate change by Colombian households IDB Working Paper Series, No. IDB-WP-1477 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: McRae, Shaun D. (2023) : Residential electricity consumption and adaptation to climate change by Colombian households, IDB Working Paper Series, No. IDB-WP-1477, InterAmerican Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0005017 This Version is available at: https://hdl.handle.net/10419/289924 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/3.0/igo/legalcode Residential Electricity Consumption and A daptation to Climate Change by Colombian Households Shaun McRae IDB WORKING PAPER SERIES Nº IDB-WP-1477 July 2023 Department of Research and Chief Economist Inter-American Development Bank July 2023 Residential Electricity Consumption and Adaptation to Climate Change by Colombian Households Shaun McRae Instituto Tecnológico Autónomo de México (ITAM) Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library McRae, Shaun D. Residential electricity consumption and adaptation to climate change by colombian households / Shaun McRae. p. cm. — (IDB Working Paper Series ; 1477) Includes bibliographical references. 1. Electric power consumption-Colombia. 2. Climatic changes-Colombia. 3. Electricity-Rates-Colombia. 4. Air conditioning-Climatic factors-Colombia. I. InterA merican Development Bank. Department of Research and Chief Economist. II. Title. III. Series. IDB-WP-1477 Copyright © Inter-American Development Bank. This work is licensed under a Creative Commons IGO 3.0 AttributionNonCommercial-NoDerivatives (CC-IGO BY-NC-ND 3.0 IGO) license (http://creativecommons.org/licenses/by-nc-nd/3.0/igo/ legalcode) and may be reproduced with attribution to the IDB and for any non-commercial purpose, as provided below. No derivative work is allowed. Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration pursuant to the UNCITRAL rules. The use of the IDB's name for any purpose other than for attribution, and the use of IDB's logo shall be subject to a separate written license agreement between the IDB and the user and is not authorized as part of this CC-IGO license. Following a peer review process, and with previous written consent by the Inter-American Development Bank (IDB), a revised version of this work may also be reproduced in any academic journal, including those indexed by the American Economic Association's EconLit, provided that the IDB is credited and that the author(s) receive no income from the publication. Therefore, the restriction to receive income from such publication shall only extend to the publication's author(s). With regard to such restriction, in case of any inconsistency between the Creative Commons IGO 3.0 Attribution-NonCommercial-NoDerivatives license and these statements, the latter shall prevail. Note that link provided above includes additional terms and conditions of the license. The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. http://www.iadb.org 2023 Abstract* This paper provides the first empirical estimates of the relationship between temperatures and household electricity consumption in Colombia, using electricity billing and weather data from 2010 to 2019. I find that higher temperatures (or higher values of the heat index) increase electricity consumption, with the largest effects observed for high-income households in regions with hot climates. However, I show that there has been partial convergence between lowand highincome households, with the effect of temperature on electricity consumption in lower-income neighborhoods more than doubling between 2011 and 2019. These results align with survey evidence of increased air conditioning adoption. Nevertheless, further growth in air conditioning adoption and use is required to alleviate the health effects of more frequent and severe heatwaves due to climate change. JEL classifications: L94, O13, Q41, Q54 Keywords: Electricity consumption, Climate change, Adaptation, Air conditioning * This work was financed with the support of the Latin America and the Caribbean Research Network of the InterAmerican Development Bank as part of the project “Implications of Climate Change and Natural Disasters for Latin America and the Caribbean.” The opinions expressed in this publication are those of the author and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. I thank the project organizers and other participants for many helpful comments and suggestions. Author contact information: Centro de Investigación Económica and Department of Economics, ITAM, [email protected]. 2 1. Introduction Global surface temperatures between 2011 and 2020 were 1.09 degrees Celsius higher than in the pre-industrial period. This increase has led to more frequent and severe heatwaves, including in Northwestern South America, where significant increases in extreme hot temperatures have already been observed (Castellanos et al., 2022). By the end of the century, there is high confidence that most regions in Central and South America will undergo extreme heat stress conditions much more often than in the recent past (Ranasinghe et al., 2021). Heatwaves have detrimental effects on human health and productivity, with informal housing and inadequate infrastructure exacerbating the vulnerability of the population in lowand middle-income countries. Using data for 40 countries, Carleton et al. (2022) show that the temperature-mortality gradient is steepest in low-income countries. For the specific case of Colombia, Helo Sarmiento (2023) finds that an additional day with temperatures above 27 degrees Celsius increases mortality rates by 0.72 percent, with children aged zero to nine at the highest risk. These heatwave-related risks in Colombia are expected to grow. Of the 20 countries studied by Guo (2018), Colombia has the highest expected increase in heatwave-related excess mortality for 2031–2080. In addition to their health effects, high temperatures have other economic consequences. For example, hot temperatures reduce worker productivity in factories (Adhvaryu et al., 2020; Somanathan et al., 2021), while excess heat in schools reduces student learning (Park et al., 2020) and reduces performance by students on high-stakes assessments (Park, 2022). Given the predicted rise in extreme heat conditions and heat-related mortality, even for ambitious climate policy scenarios, adaptation to the changing climate is essential. One form of adaptation to more frequent heatwaves is adopting air conditioning to cope with extreme hot temperatures. Barreca et al. (2016) demonstrate that adopting residential air conditioning played a vital role in flattening the temperature-mortality gradient in the United States. Interestingly, the recent estimates of this gradient in Colombia resemble those in the United States prior to the widespread use of residential air conditioning (Helo Sarmiento, 2023). In this paper, I use electricity consumption and weather data to infer the adoption and use of air conditioning in Colombia. Specifically, I combine a large sample of monthly residential electricity billing data with hourly gridded weather data from 2010 to 2019. I estimate the effect of the monthly mean heat index on logged monthly electricity consumption—that is, the percentage increase in electricity consumption for a one-degree Celsius increase in the mean heat 3 index.1 A larger effect of the heat index on electricity consumption would indicate greater use of air conditioning during warmer weather. Furthermore, by following the same households for nine years, I can measure the short-term effect of the heat index on electricity consumption and the long-term changes in this effect. There are three main findings from my electricity consumption analysis. First, higher heat index values increase electricity consumption, especially in the hot climate regions of Colombia. Notably, there is no relationship between the heat index and electricity consumption in temperate regions. Second, the increase in electricity consumption for a one-degree increase in the heat index is largest for high-income households, where income is proxied by the socioeconomic stratum of the neighborhood used to assign electricity subsidies. The size of the estimated effect is economically meaningful: in hot regions, each one-degree increase in the mean heat index increases the electricity consumption of high-stratum households by about 6 percent. Finally, the effect of the heat index on electricity consumption is increasing for households in the lowest two strata, leading to partial convergence over time between lowand high-strata households. These results from the electricity consumption regressions are consistent with the change in fan and air conditioning ownership between 2011 and 2019, calculated from the annual Living Standards surveys (DANE, 2020). The surveys provide information about income and expenditure, asset ownership, and other household characteristics for a regionally representative sample of households. The national share of households with air conditioners increased from 3.8 percent to 4.5 percent between 2011–15 and 2016–19, with the largest increase observed in the hot Caribbean region. Moreover, the ownership of fans and air conditioners is increasing in household income, and there was an upward shift in the adoption curves of both appliances in hot regions between 2011–15 and 2016–19. The results in this paper provide the first evidence of a long-run change in the short-run relationship between residential electricity consumption and weather in a middle-income country. I show evidence, not just of the gap in air conditioning adoption and use between lowand highincome households, but of a narrowing in this difference over time. An advantage of my methodology compared to previous studies that rely on household survey data alone is that the 1 The heat index is a composite measure that combines air temperature and relative humidity to measure the perceived temperature after accounting for the cooling effect from the evaporation of perspiration, the rate of which decreases when humidity is higher. I show my results using both the mean temperature and heat index. The magnitude of the results is similar for both measures, but the results for the heat index are more precisely estimated. 4 estimates from the electricity consumption regressions reveal the use of air conditioning, not just the presence of an appliance in the household. Colombia is an ideal setting to study the adoption and use of air conditioning. It has vast disparities in adoption rates: in the Caribbean, more than 70 percent of households in the top five percent of income own air conditioners, compared to a share close to zero for the bottom five percent. This inequality in air conditioning ownership is a feature of all middle-income countries studied by Davis et al. (2021). Therefore, many of the findings may generalize to other middleincome countries such as Mexico, India, and China, especially given their similarities in climate. Moreover, given the expected growth in excess mortality from heatwaves in Colombia (Guo, 2018), understanding the role of air conditioning in climate change adaptation is essential. This paper contributes to the literature on the relationship between higher temperatures, air conditioning adoption, and electricity consumption. Most of this research studies the extensive margin of adoption, that is, how higher incomes and temperatures will lead to greater adoption of air conditioning in lowand middle-income countries. Auffhammer (2014) estimates diffusion curves for air conditioning in China as a function of income and temperature, finding that hot weather leads to increased adoption of air conditioning in subsequent summers. Davis et al. (2021) use data for 16 countries to show how air conditioning adoption increases with income, especially for hot regions in middle-income countries, then use their estimates to predict future adoption. As shown by Biardeau et al. (2020), there will be rapid growth in air conditioning adoption, given the unmet cooling potential at a global scale. In a rare study using quasi-experimental variation in income, Randazzo et al. (2023) use pooled data from household surveys in Mexico to show that higher remittance income leads to greater air conditioning adoption, but only in the warm coastal states. Fewer papers focus on the intensive margin relationship between higher temperatures and electricity consumption for cooling. Auffhammer (2022) most closely resembles the empirical strategy of this paper. He uses several years of billing data for residential electricity and natural gas to estimate the relationship between daily temperature observations and energy consumption in California. By comparing the temperature responsiveness of consumption across different climate regions in California, he recovers the extensive margin relationship between climate and air conditioning adoption. Another paper estimating temperature response functions from electricity consumption data is Li et al. (2019). They use daily electricity consumption for 5 households from one part of Shanghai to estimate the nonlinear relationship between temperature and electricity consumption, finding that for warm days above 25 degrees Celsius, a one-degree increase in temperature increases electricity consumption by 14.5 percent. Davis and Gertler (2015) is one of the few papers that studies both the intensive and extensive margins of air conditioning adoption and use. They use two years of household-level electricity billing data from Mexico to estimate a flexible relationship between local temperatures and electricity consumption. They show that each additional day above 32 degrees Celsius increases monthly electricity consumption by 3.2 percent. A separate analysis uses Mexican household survey data to estimate the relationship between air conditioning adoption, income, and climate. In warm regions only, they show that an additional $10,000 in household income increases air conditioning adoption by 27 percentage points. Finally, they combine these two estimates to predict electricity consumption and emissions under future climate change scenarios with higher incomes and temperatures. This paper provides the first empirical estimates of the relationship between weather, air conditioning adoption, and electricity consumption in a South American country. Unlike the previous papers in the literature, this paper uses an extended sample period for each household to measure changes in the relationship between electricity consumption and weather. As a result, the methodology provides estimates of the intensive and extensive margin responses to warmer temperatures within a single empirical model. The rest of the paper is organized as follows. Section 2 describes the three main datasets used for the analysis and provides summary statistics about the relationship between income, climate, and electricity consumption. Section 3 provides the empirical methodology used for the analysis. There are two sets of results: Section 4 provides the main results for the electricity consumption estimation, which are used to motivate the appliance adoption results in Section 5. Section 6 concludes. 2. Data The three principal data sources for this paper are hourly gridded weather from climate reanalysis data, monthly household-level electricity consumption data, and household survey data on demographics, income, and appliance ownership. 12 higher in the eastern lowland regions, but not to the same extent as in the Caribbean. Despite their cold climate, high-altitude cities such as Bogotá, with many wealthy households, also have relatively high electricity consumption. Figure 5 illustrates the relationship between electricity consumption, climate, and the household stratum. Each panel shows consumption trends for one of the three climate regions: cold, temperate, and hot. Within each panel, the six lines correspond to the mean electricity consumption for the six strata. As in Table 1, within every climate region, electricity consumption monotonically increases with the stratum (that is, the highest line represents the highest stratum). The figures display the trend in mean consumption between 2010 and 2019.2 The difference in consumption between the strata is relatively low in the cold region, with a mean consumption for Stratum 1 in 2019 of about 100 kWh per month, compared to about 230 kWh per month for Stratum 6. The difference between the strata has fallen over time in the cold regions. Mean consumption for the lowest strata is relatively constant, while consumption in the higher strata has declined substantially since 2010. This decline may be due to improvements in energy efficiency, such as the adoption of LED lighting. The cross-strata variation is more prominent in the hot region, perhaps reflecting the greater use of air conditioning for households in higher strata. The mean consumption for Stratum 1 in 2010 was about 150 kWh per month, compared to about 450 kWh per month for Stratum 6. There has been a small degree of convergence between the strata over the decade from 2010 to 2019, as consumption for the highest strata has been relatively flat, while consumption in the lower two strata has increased. 3. Empirical Methodology The empirical analysis estimates the effect of the mean heat index or the mean temperature on electricity consumption using the panel dataset of electricity consumption and weather described in Section 2. The base case regression specification is provided in equation (1). log(𝑦𝑦𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖)=�𝛽𝛽𝑆𝑆𝐼𝐼[𝑠𝑠=𝑆𝑆]𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊ℎ𝑊𝑊𝑟𝑟𝑖𝑖𝑖𝑖𝑖𝑖 6 𝑆𝑆=1 +𝛾𝛾𝑍𝑍𝑖𝑖𝑖𝑖 +𝛼𝛼𝑖𝑖+𝜔𝜔𝑖𝑖𝑖𝑖 +𝜀𝜀𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 (1) 2 Unfortunately, the data for several electricity distributors are missing for seven months in 2018. For that reason, I exclude 2018 from the summary means in Figure 5, although I include all available data in the estimation. 13 In this equation, the dependent variable is the log of the electricity consumption for household i assigned to stratum s and living in municipality r in month-of-sample t.3 The main regressor of interest is Weatherirt, defined for household i in municipality r during the period t. In Section 4.1, I show results in which the weather variable is either the mean temperature or the mean heat index for the billing cycle. I use these results to justify using the mean heat index in the subsequent analysis. In Section 4.3, I show results using the proportion of days in a heatwave during the billing cycle. I also show the results for the mean, maximum, and minimum daily temperatures in Appendix B. In all cases, the weather variables differ across households in the same month and municipality due to differences in the billing cycle timing. The effect of the weather variables on electricity consumption is allowed to vary by the stratum s. Differences in the βS reflect variation across the strata in the effect of weather on electricity consumption. The base regression model includes household fixed effects αi to absorb the determinants of household electricity consumption that do not change over time, such as location and housing characteristics. Month-of-sample fixed effects ωst control for time-varying determinants of electricity consumption that are common to all households, including wholesale electricity prices and national economic shocks. These fixed effects vary by stratum s to account for potential differences in the effect of these factors by stratum. The month-of-sample for a particular observation is determined by the first date of the billing cycle. Household income is an important determinant of electricity consumption. The household fixed effect αi controls for the mean level of each household’s income during the sample period. However, there may be differences across households in the growth rate of income, and these may be correlated with the weather variables due to the correlation between weather and economic output in sectors such as agriculture and tourism. In the absence of time-varying income data at the household level, I include annual state-level Gross Domestic Product as a control variable Zit. This variable controls for differential regional trends in economic activity. The exact timing of the billing cycle may affect electricity consumption through the number of weekends and public holidays. For all regressions, I include a control variable in Zit for 3 As discussed in Section 2, I divide the metered electricity consumption by the number of billed days to adjust for differences in the number of days in each monthly billing cycle. 14 the proportion of days that are Sundays or public holidays in each billing period. The historical dates of Colombian public holidays are from Festivos Colombia (2023). In Section 4.2, I estimate the base specification in equation (1) for different subgroups in the data. First, I estimate the model separately by climate region. I expect that the base effect of weather βS will be larger for hot regions due to higher levels of air conditioning adoption. Second, I estimate the model separately by stratum, allowing the coefficients on the weather regressor to vary by year. This specification is used to examine changes over time in the effect of weather on electricity consumption. The base specification assumes a linear relationship between the heat index (or temperature) and electricity consumption. As discussed in Section 2, most temperature variation in Colombia is across locations based on altitude, not within locations over time. Because no location experiences temperatures over the full range of the temperature distribution, estimating a flexible nonlinear regression may conflate the within-location and across-location variation in the temperature effects (Mendelsohn, 2016). One variable that is omitted from Equation (1) is the electricity price. I assume that the tariff nonlinearities do not have a measurable effect on the relationship between weather and household electricity consumption. This assumption is only relevant for the households in Strata 1 to 3, as the households in Strata 4 to 6 face a uniform tariff with no nonlinearities. The households in Strata 1 to 3 pay a higher marginal price above the subsidy threshold, but there is no discontinuous change in their total bill amount at the cutoff. 4 The subsidy thresholds were calibrated to provide households with a “lifeline” quantity of electricity consumption, so most households using air conditioning will be above the threshold and pay the full regulated price for their marginal consumption. The tariff structure, the subsidized quantities, and the subsidy amount have not changed during the sample period—in particular, the tariffs do not change in response to short-term weather conditions. The assumption that the nonlinear tariff can be ignored for this analysis is consistent with the existing literature on electricity consumption and weather (Davis and Gertler, 2015; Auffhammer, 2022). 4 The Colombian electricity tariffs are an example of an increasing block tariff. For this tariff type, the marginal price changes at the quantity thresholds, but households above each threshold keep the subsidy amount on their inframarginal units (Appendix A). An alternative tariff type is a quantity-differentiated tariff, for which households lose the subsidy on their inframarginal units if their consumption exceeds the quantity threshold (see, for example, Pellerano et al., 2017). Because of its “notch” in the total bill amount, a quantity-differentiated tariff would provide stronger incentives to keep electricity consumption below the threshold. 15 In all specifications, I cluster the standard errors by the municipality of the households. This allows for potential correlation in the error terms across households in the same municipality due to local economic shocks, electricity distribution outages, or disasters. Clustering by municipality also allows for correlation in the error term within a household over time. 4. Empirical Results I first present the baseline results for the sensitivity of electricity consumption to the mean temperature and the mean heat index in each billing cycle. These results support the use of the heat index in the subsequent analyses. I show the heterogeneity in the sensitivity to the heat index by stratum and climate zone, then examine the changes in the heat index sensitivity over time. Finally, I provide additional results on the relationship between heatwaves and electricity consumption. 4.1 Baseline Results Table 2 shows the results for 18 separate regressions of equation (1) for the log of monthly electricity consumption on alternative weather variables. There are three regressions for each stratum, differing by the definition of the weather variable: the mean temperature during the billing cycle (Model 1), the mean heat index during the billing cycle (Model 2), and both regressors combined in a single estimation (Model 3). The estimation uses the full sample of households from all climate regions. Higher temperatures during the month increase the electricity consumption of households in all six socioeconomic strata (Model 1). For Stratum 1 households, a one-degree Celsius increase in the temperature in the billing cycle increases electricity consumption by 2.4 percent. The magnitude of the temperature effect increases monotonically across the six strata. For the highest stratum, a one-degree temperature increase leads to a 6.0 percent increase in electricity consumption. All of the estimated effects are statistically significant at a 1 percent level. Results for the mean heat index during the billing cycle are quantitatively similar to the mean temperature results (Model 2). However, the statistical precision of the heat index estimates is greater, with smaller standard errors than in Model 1. The greater precision is especially notable for the Strata 4 to 6 estimates, which rely on fewer observations from a smaller number of municipalities. 16 The final model is an encompassing model that embeds both regressors from Models 1 and 2 in a single equation. The estimated coefficients on the mean heat index increase in magnitude but remain positive and strongly statistically significant. However, the coefficients on the mean temperature change their sign and are statistically significant and negative. From the definition of the heat index, if the heat index is held constant, an increase in the air temperature must imply a reduction in the relative humidity. In other words, the negative sign on temperature implies that lower humidity reduces electricity consumption (or, conversely, higher humidity increases electricity consumption). From a statistical perspective, the encompassing test strongly rejects both Model 1 and Model 2 in favor of the encompassing Model 3, implying that the temperature and heat index variables provide independent information about electricity consumption. However, the p-value for the encompassing test is lower for Model 1 than for Model 2. Similarly, goodness-of-fit measures like the R2, and model selection criteria like the AIC and BIC, all support the use of Model 2 over Model 1. Based on this evidence supporting the use of the heat index, I show results only for the heat index (Model 2) in the next subsection. 4.2 Heterogeneity by Climate and Year In this subsection, I show the heterogeneity, by climate region and year, in the heat index results in Model 2 of Table 2. Each set of results is shown in both tabular and graphical form. First, for heterogeneity by climate region, Table 3 and Figure 6 show the results for separate estimations of Equation (1) by climate region. In the table, the uninteracted coefficient on the heat index variable shows the effect of a one-degree Celsius increase in the mean heat index on the logged electricity consumption of households in Stratum 1. The other coefficients show the interaction of the heat index with the stratum indicator variables. The overall heat index effect by stratum is the sum of the uninteracted heat index coefficient and the coefficient on its interaction with the stratum indicator. Figure 6 shows this combined effect, with its 95 percent confidence interval, for each climate region. The mean heat index has a positive but relatively small effect on electricity consumption in cold regions. The coefficient on the mean heat index is 0.013, meaning that a one-degree increase in the mean heat index during the billing cycle is associated with a 1.3 percent increase in the electricity consumption of Stratum 1 households. With the exception of Stratum 5, there is 17 no statistically significant difference in this effect across the other strata. For Stratum 5 households in cold regions, the effect of a temperature increase is larger: a one-degree increase in the mean heat index is associated with a 3.3 percent increase in electricity consumption. A counterintuitive aspect of the cold region results is the absence of a negative coefficient on the heat index. We would expect that if households use electric heating in cold regions, an increase in the temperature or the heat index would reduce the heating requirement and thus decrease electricity usage. This effect was observed by Berkouwer (2020), who found a negative relationship between temperature and electricity consumption for households in South Africa, up to 23 degrees Celsius.5 The small positive coefficient on the heat index suggests that electric heating is less common in Colombia than in South Africa and, instead, there is a more nuanced relationship between electricity consumption and weather in cold regions. One explanation consistent with the results is that households have multiple heating sources, other fuels are used for heating at low temperatures, and electricity is used when temperatures are relatively warm. Another possibility is that the usage of other appliances is correlated with the weather. For example, households in cold regions might use their clothes washers more when the weather is warmer. For temperate regions, the effect of the heat index on electricity consumption is close to zero and, in all but Stratum 2, statistically indistinguishable from zero. Even for Stratum 2, the effect of 0.01 is smaller than the coefficients for any of the strata in cold regions. The point estimate for Stratum 6 is negative (-0.017), but its 95 percent confidence interval is wide and includes zero. These results suggest that there is little need to own and use heating and cooling appliances in temperate regions of Colombia. Finally, there is a large, positive, and statistically significant relationship between the heat index and electricity consumption in hot regions. For Stratum 1 households in hot regions, a onedegree increase in the mean heat index increases electricity consumption by 2.5 percent (with a 95 percent confidence interval of 1.9 to 3.1 percent). Electricity consumption is more sensitive to the heat index for the higher strata. For Stratum 5 households, a one-degree increase in the heat index increases electricity consumption by 6.4 percent. The greater effect of the heat index on electricity 5 Conversely, Davis and Gertler (2015) did not find any relationship between temperature and electricity consumption at low temperatures, which they attributed to the limited use of electric heating in Mexico. 18 consumption for the higher strata is consistent with the result in Table 1 that households in higher strata are more likely to own an air conditioner. The second set of results reveals how the relationship between electricity consumption and the heat index changes over time for the different strata (Table 4 and Figure 7). For these results, I focus only on those households living in the hot climate region. Each column in Table 4 and each panel in Figure 7 shows the results for a separate regression for the households in a different stratum. The uninteracted coefficient on the heat index variable shows the effect of a one-degree increase in the mean heat index on the logged electricity consumption of households in the base period: the last five months of 2010 and all of 2011. The other coefficients in the table show the interaction of heat index with the year indicator variables. The overall heat index effect by year, shown in Figure 7, is the sum of the uninteracted heat index coefficient and the coefficient on its interaction with the year indicator. The results show that the effect of the heat index on electricity consumption has increased over time for Strata 1 to 3 households, with the largest change for Stratum 1 households. In the 2010-11 base period, a one-degree increase in the heat index increased the electricity consumption of a Stratum 1 household in the hot region by 1.4 percent. This effect more than doubled to 3.3 percent by 2019. For Stratum 2 households, the effect of a one-degree change in the heat index increased from 2.4 percent in 2011 to 3.2 percent in 2019. The changes for both strata were statistically significant at the 1 percent level. Stratum 3 households had a slightly higher gradient between 2013 and 2017 relative to 2011, but the increase was no longer statistically significant by 2019. As shown in Figure 6, the electricity consumption of Stratum 4 to 6 households in hot regions is especially sensitive to the heat index, with overall coefficient estimates between 0.057 and 0.064. For these strata, there is no statistically significant change in the effect between 2011 and 2019 (Table 4). Although there are small changes in the point estimates from year to year, the confidence intervals for the heat index coefficients are relatively wide because of the smaller number of municipalities containing households from the higher strata. Because of the imprecise estimates, I cannot reject that there was no change in the effect for these strata. Overall, the results in Figure 6 show substantial convergence across the strata in the effect of the heat index on electricity consumption. 19 The magnitude of the heat index effect for Colombian households in Strata 5 and 6 is comparable to previous results in the literature for Mexican households. Davis and Gertler (2015) find that moving one day in a month from 18-to-21 degrees Celsius to a temperature above 32 degrees Celsius will increase monthly electricity consumption by 3.2 percent. This change in the temperature distribution corresponds to an increase in the monthly mean temperature of approximately 0.45 degrees Celsius. Equivalently, the Davis and Gertler (2015) results can be interpreted as a 7.1 percent increase in electricity consumption for a one-degree Celsius increase in mean temperature.6 This result is slightly higher than the Strata 5 and 6 point estimates for heat index in Figure 6, but is within the 95 percent confidence interval.7 Conversely, the estimated effect for Strata 1 to 3 households is lower than the average for Mexican households in Davis and Gertler (2015). Although the magnitude of the effect of the heat index on electricity consumption is smaller for Strata 1 to 3 households, the nonlinearity in the electricity tariff structure means that its effect on the electricity bill amount may be even larger than for the higher strata. For example, for Stratum 1 households in 2019, a one-degree Celsius increase in the mean heat index increases monthly electricity consumption by 3.3 percent. However, if the increase in consumption occurs at the unsubsidized price (for example, if monthly consumption exceeds 173 kWh in lowland regions), then the household electricity bill may increase by more than 7 percent. Moreover, as shown in Table 1, electricity bills are a larger share of income for households in Stratum 1 than for households in Stratum 6. As a result, this increase in the bill may be especially salient. 4.3 Alternative Weather Variables In this subsection, I provide additional robustness checks for the results in Section 4.2. I show that the qualitative results are not affected by the exact definition of the weather variables for the analysis. Figure 8 shows the relationship between electricity consumption and the proportion of days with heatwave events (as defined in Section 2) for hot regions in Colombia. Heatwaves have a positive and statistically significant on electricity consumption for all six strata, with the estimates 6 This calculation assumes a change from the midpoint of the lower bin (19.5 degrees) to a temperature of 33 degrees. Assume 30 days in the month: (33 − 19.5)/30 = 0.45. The result for a one-degree increase is 0.032/0.45 = 0.071. 7 The results in Table 6 are not strictly comparable to Davis and Gertler (2015), because they measure the sensitivity to the heat index and not to temperature. However, the temperature results for Strata 5 and 6 in Figure A3 are very similar, just with larger confidence intervals. 20 increasing across the first five strata and the largest effect observed for households in Stratum 5. For Stratum 5 households, one additional day with a heatwave in a month increases electricity consumption for the month by 1.1 percent.8 I repeat the analysis in Section 4.2 using three alternative regressors instead of the heat index: the mean temperature, the mean of the daily maximum temperature, and the mean of the daily minimum temperature. Appendix B provides two figures for each of these regressors, equivalent to the results in Section 4.2. The mean temperature results by climate region are very similar to the heat index results (Figure A3). There is a small positive relationship between mean temperature and electricity consumption in cold regions, no relationship in temperate regions, and a large positive relationship in hot regions that is increasing in the household stratum. Similarly, the annual results from 2011 to 2019 show a statistically significant increase in the effect of the mean temperature on electricity consumption for households in Strata 1 and 2 (Figure A4). The main difference between the heat index and the temperature results is in the precision of the estimates, as discussed in Section 4.1. Because the mean temperature coefficients are less precisely estimated, the confidence intervals for Strata 4 to 6 households are particularly wide in Figures A3 and A4. In theory, the minimum and maximum daily temperatures might matter more for electricity consumption than the mean temperature. This would be the case if, for example, electric heaters or air conditioners are only turned on during the coldest or hottest hours of the day. In practice, there is much autocorrelation in within-day temperatures, so days in which the maximum temperature is higher will also tend to have higher mean and minimum temperatures too. This makes it difficult to separate the effect of the daily extreme temperatures from the mean temperature. The overall results using the mean of the daily minimum and maximum temperatures are consistent with the results in Section 4.2. One notable exception is the result for cold regions in Figure A5. Although the coefficients on the mean maximum temperatures are all positive, they are much smaller in magnitude than the corresponding coefficients on the heat index or mean temperature. For Stratum 1, the coefficient is not statistically different from zero. Conversely, the effects of minimum temperature on electricity consumption in cold regions are large in magnitude 8 The Stratum 5 estimate is 0.285, so the effect of one extra day of a heatwave on electricity consumption is (1/30)× (exp(0.285)− 1). 21 and statistically significant, except for an anomalous negative and insignificant coefficient for Stratum 6 (Figure A7). These results provide suggestive support for the positive effect of temperature on electricity consumption in cold regions being driven by household heating choices, for which the minimum daily temperature is more relevant than the maximum. The final set of supplementary results uses the detailed hourly temperature data to estimate separate effects of temperature during hours when household members are more or less likely to be at home. For each billing cycle, I calculate the mean temperature and mean heat index for three different time periods: daytime hours (assumed to be from 8 a.m. to 6 p.m. on non-holiday weekdays and Saturdays), nighttime hours (6 p.m. to 8 a.m. on non-holiday weekdays and Saturdays), and any time on public holidays and Sundays. I then estimate a version of equation (1) including the three mean temperature or mean heat index variables in a single regression, restricting the sample to households living in the hot climate region. Table A1 shows the results from splitting the overall mean temperature into the mean temperature for the three time periods. For every stratum, the coefficients on the mean daytime temperature are small in magnitude and statistically indistinguishable from zero. In contrast, the coefficients on the mean nighttime temperature are large, statistically significant at the 1 percent level, and increasing with the household stratum. The holiday temperature coefficients are smaller than the nighttime coefficients, though (except for Strata 4 and 5) still statistically significant. These results suggest that higher temperatures during hours when household members are more likely to be at home have the greatest effect on electricity consumption. The results from splitting the mean heat index into the three time periods are more ambiguous (Table A2). For Strata 1 to 4 households, the coefficients on both the daytime and nighttime heat index variables are positive and statistically significant, with the nighttime coefficients typically smaller in magnitude. The Strata 5 and 6 coefficients are similar in magnitude but less precisely estimated. These results indicate that a higher heat index increases electricity consumption in hot regions, regardless of the time of day. The discrepancies in the results between Tables A1 and A2 may be caused by the high correlation between the three heat index or temperature variables. This collinearity makes it empirically challenging to separate their effect on a single outcome variable observed at a monthly frequency. 28 Mendelsohn, R. 2016. “Measuring Weather Impacts Using Panel Data.” New Haven, United States: Yale University. Manuscript. Mistry, M.N. et al. 2022. “Comparison of Weather Station and Climate Reanalysis Data for Modelling Temperature-related Mortality.” Scientific Reports 12(1): 1–14. Muñoz Sabater, J. et al. 2019. “ERA5-Land Hourly Data from 1981 to Present.” Copernicus Climate Change Service (C3S) Climate Data Store (CDS) 10. National Weather Service. 2023a. “Cold & Warm Episodes by Season.” https://www. weather.gov/ama/heatindex. ---- 2023b. “What Is the Heat Index?” https://origin.cpc.ncep.noaa.gov/products/ analysis_monitoring/ensostuff/ONI_v5.php. Noy, Ilan, and Eric Strobl. 2022. “Heat Waves and Innovation in Air Conditioning in the United States.” Weather, Climate, and Society 14(1): 143–154. Park, R. Jisung. 2022. “Hot Temperature and High-Stakes Performance.” Journal of Human Resources 57(2): 400–434. Park, R.J. et al. 2020. “Heat and Learning.” American Economic Journal: Economic Policy 12(2): 306–39. Pellerano, J.A. et al. 2017. “Do Extrinsic Incentives Undermine Social Norms? Evidence from a Field Experiment in Energy Conservation.” Environmental and Resource Economics 67:413–428. Ranasinghe, R. et al. 2021. “Climate Change Information for Regional Impact and for Risk Assessment.” In: V. Masson-Delmotte et al., editors. Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge, United Kingdom and New York, United States: Cambridge University Press. Randazzo, T., F. Pavanello, and E. De Cian. 2023. “Adaptation to Climate Change: Airconditioning and the Role of Remittances.” Journal of Environmental Economics and Management; 102818. Somanathan, E. et al. 2021. “The Impact of Temperature on Productivity and Labor Supply: Evidence from Indian Manufacturing.” Journal of Political Economy 129(6): 1797–1827. Tiecke, T.G. et al. 2017. “Mapping the World Population One Building at a Time.” Available at: https://arxiv.org/pdf/1712.05839.pdf 29 Tables and Figures Figure 1. Calculation of Hourly Municipality Weather Variables from Gridded Population and Weather Data Notes: Each square represents an approximately 9x9 km grid cell from the ERA5-Land dataset, with colors indicating air temperature at midday on September 15, 2015. The white outline shows the Santa Marta municipality in northern Colombia, and black pixels represent inhabited locations (Data for Good, 2020). The population-weighted average temperature for Santa Marta during this hour is 23.4 degrees Celsius. 30 Figure 2. Average Heat Index and Electricity Consumption by Municipality, 2010–19 Notes: The left panel shows the mean hourly heat index for each municipality from 2010 to 2019, calculated using ERA5-Land air temperature and dewpoint temperature data, aggregated as shown in Figure 1. The right panel shows the mean monthly residential electricity consumption for each municipality from August 2010 to November 2019, based on the estimation dataset. Black outlines represent Colombia’s 32 departments, while light gray outlines represent municipalities. Gray municipalities in the right panel are not included in the estimation dataset, mostly because they are not connected to the national transmission network. 31 Figure 3. Frequency and Location of Heatwaves, 2010–19 Notes: Each panel displays the annual proportion of heatwave days for each municipality. A heatwave is defined as two or more consecutive days with a maximum heat index above 32 degrees Celsius. 32 Figure 4. Distribution of Household Income per Capita by Stratum Notes: Income per capita data is from the Living Standards Surveys for 2011 to 2019 (excluding 2017). Nominal Colombian pesos were converted to real 2018 United States dollars using the Colombian Consumer Price Index deflator and the 2018 exchange rate. Values of income per capita exceeding $2,000 were set to $2,000 to maintain a reasonable vertical scale. The horizontal line shows the median income per capita for each stratum. 33 Figure 5. Electricity Consumption by Climate Region and Stratum, 2010–19 Notes: Each panel represents one of three climate regions, defined for each municipality based on the classification in IDEAM (2023). Within each panel, lines depict the mean monthly electricity consumption for households in different strata, with the lowest line for Stratum 1 and the highest line for Stratum 6. Annual means are based on observations in the estimation dataset. No results are shown for 2018 because of many missing observations. 34 Figure 6. Effect of the Heat Index on Household Electricity Consumption by Climate Zone and Stratum Notes: Each panel corresponds to one column in Table 3 and shows the coefficients for the interaction between stratum indicators and the heat index. The error bars show 95 percent confidence intervals, calculated using standard errors clustered by municipality. See also the notes to Table 3. 35 Figure 7. Effect of the Heat Index on Household Electricity Consumption in Hot Regions, by Year and Stratum Notes: Each panel corresponds to one column in Table 4 and shows the coefficients for the interaction between year indicators and the heat index. The error bars show 95 percent confidence intervals, calculated using standard errors clustered by municipality. See also the notes to Table 4. 36 Figure 8. Effect of Heatwaves on Household Electricity Consumption Notes: The figure shows results from a regression of log electricity consumption on an interaction between stratum indicators and the proportion of heatwave days in the month. Heatwaves are defined as two or more consecutive days with a maximum heat index exceeding 32 degrees Celsius. The sample consists only of households living in a hot climate region. The error bars show 95 percent confidence intervals, calculated using standard errors clustered by municipality. 37 Figure 9. Share of Households with Cooling Appliances by Climate Region, Time Period, and Income per Capita Notes: Each panel shows the share of households owning a fan (top) and an air conditioner (bottom), split by the climate region. See the notes for Table 5 for the definition of the regions. For each period and region, households are divided into 20 equally sized bins of their income per capita. The dots on the graph show the mean share of households owning the appliance for the period, region, and income per capita bin, calculated using the household sampling weights. The lines show a local linear regression fitted through the binned points. 44 The figure includes the total bill amounts for three households with monthly consumption of 200 kWh. Consider a Stratum 1 household living at an altitude above 1,000 meters. It receives a 55 percent subsidy for its first 130 kWh of consumption, then pays 18.6 cents for each additional kWh. The total bill for monthly consumption of 200 kWh is US$23.90: Bill1 = 130 × (1 − 0.55) × 0.186 + (200 − 130) × 0.186 = 23.90 A Stratum 1 household living at an altitude below 1,000 meters receives the 55 percent subsidy for its first 173 kWh of consumption. The total bill for monthly consumption of 200 kWh is US$19.50: Bill2 = 173 × (1 − 0.55) × 0.186 + (200 − 173) × 0.186 = 19.50 For a Stratum 6 household, the price per kWh is the same for all units and includes a 20 percent contribution above the regulated base tariff. The total monthly bill for a consumption of 200 kWh is US$44.64: Bill3 = 200 × (1 + 0.2) × 0.186 = 44.64 45 Appendix B. Additional Tables and Figures Figure A2. Classification of Climate Regions Notes: The climate classification is based on the 23 climate zones in IDEAM (2023), aggregated to the three temperature categories of cold, temperate, and hot. Each municipality is assigned to the climate of the largest urban center in the municipality. 46 Figure A3. Effect of Mean Temperature on Household Electricity Consumption by Climate Zone and Stratum Notes: Each panel shows the coefficients for the interaction between stratum indicators and the mean temperature. The error bars show 95 percent confidence intervals, calculated using standard errors clustered by municipality. See also the notes to Table 3. 47 Figure A4. Effect of the Mean Temperature on Household Electricity Consumption in Hot Regions, by Year and Stratum Notes: Each panel shows the coefficients for the interaction between year indicators and the mean temperature. The error bars show 95 percent confidence intervals, calculated using standard errors clustered by municipality. See also the notes to Table 4. 48 Figure A5. Effect of the Daily Maximum Temperature on Household Electricity Consumption by Climate Zone and Stratum Notes: Each panel shows the coefficients for the interaction between stratum indicators and the mean of the daily maximum temperatures during the billing cycle. The error bars show 95 percent confidence intervals, calculated using standard errors clustered by municipality. See also the notes to Table 3. 49 Figure A6. Effect of the Daily Maximum Temperature on Household Electricity Consumption in Hot Regions, by Year and Stratum Notes: Each panel shows the coefficients for the interaction between year indicators and the mean of the daily maximum temperature. The error bars show 95 percent confidence intervals, calculated using standard errors clustered by municipality. See also the notes to Table 4. 50 Figure A7. Effect of the Daily Minimum Temperature on Household Electricity Consumption by Climate Zone and Stratum Notes: Each panel shows the coefficients for the interaction between stratum indicators and the mean of the daily minimum temperatures during the billing cycle. The error bars show 95 percent confidence intervals, calculated using standard errors clustered by municipality. See also the notes to Table 3. 51 Figure A8. Effect of the Daily Minimum Temperature on Household Electricity Consumption in Hot Regions, by Year and Stratum Notes: Each panel shows the coefficients for the interaction between year indicators and the mean of the daily minimum temperature. The error bars show 95 percent confidence intervals, calculated using standard errors clustered by municipality. See also the notes to Table 4. 52 Table A1. Effect of the Mean Temperature on Household Electricity Consumption in Hot Regions, Split by Time-of-Day and Holiday Periods Stratum 1 2 3 4 5 6 Daytime temperature 0.000 (0.003) 0.002 (0.003) 0.002 (0.006) -0.002 (0.007) -0.005 (0.018) -0.024 (0.021) Nighttime temperature 0.016*** (0.006) 0.025*** (0.006) 0.045*** (0.010) 0.075*** (0.013) 0.084*** (0.025) 0.080*** (0.028) Holiday temperature 0.012*** (0.003) 0.008*** (0.002) 0.007** (0.003) 0.002 (0.004) 0.004 (0.007) 0.023* (0.012) Prop. holidays -0.093*** (0.012) -0.043*** (0.011) -0.055** (0.021) 0.004 (0.055) 0.001 (0.106) 0.181 (0.221) Log(state GDP) 0.151*** (0.045) 0.062* (0.037) 0.034 (0.038) 0.034 (0.038) 0.090* (0.046) -0.054 (0.122) Fixed effects Month of sample (112) Y Y Y Y Y Y Household Y Y Y Y Y Y # Household 134,660 135,658 63,555 18,385 7,997 3,993 Observations 13,069,647 13,438,138 6,315,522 1,820,100 789,540 389,028 Notes: The dependent variable in all regressions is the log of monthly electricity consumption for one household. Each column presents the results for one of the six strata. The three main regressors of interest are: the mean temperature during daytime hours (8 a.m. to 6 p.m.) for non-holiday weekdays and Saturdays, the mean temperature during nighttime hours (6 p.m. to 8 a.m.) for non-holiday weekdays and Saturdays, and the mean temperature on Sundays and public holidays. All models include the proportion of Sundays and public holidays, annual state-level GDP, month-of-sample fixed effects, and household fixed effects. Standard errors in parentheses are clustered by municipality. 53 Table A2. Effect of the Mean Heat Index on Household Electricity Consumption in Hot Regions, Split by Time-of-Day and Holiday Periods Stratum 1 2 3 4 5 6 Daytime heat index 0.007*** (0.003) 0.016*** (0.002) 0.021*** (0.003) 0.031*** (0.009) 0.032 (0.019) 0.025 (0.029) Nighttime heat index 0.008** (0.004) 0.007* (0.004) 0.019*** (0.005) 0.023** (0.009) 0.027 (0.021) 0.021 (0.031) Holiday heat index 0.010*** (0.002) 0.007*** (0.002) 0.004 (0.003) 0.002 (0.003) 0.005 (0.006) 0.014* (0.007) Prop. holidays -0.094*** (0.012) -0.042*** (0.011) -0.059*** (0.021) 0.008 (0.056) -0.002 (0.106) 0.201 (0.229) Log(state GDP) 0.140*** (0.045) 0.049 (0.037) 0.019 (0.036) 0.006 (0.036) 0.067 (0.056) -0.090 (0.130) Fixed effects Month of sample (112) Y Y Y Y Y Y Household Y Y Y Y Y Y # Household 134,660 135,658 63,555 18,385 7,997 3,993 Observations 13,069,647 13,438,138 6,315,522 1,820,100 789,540 389,028 Notes: See the notes to Table A1. The only difference is that the regressors in this table are based on the mean heat index during daytime, nighttime, and holidays, not the mean temperature.