Income, extreme temperature, and residential electricity consumption in the Spanish provinces

1 Income, extreme temperature, and residential electricity consumption in the Spanish Provinces María P. Pablo-Romeroa*, Antonio Sánchez-Brazaa and Claudia P. Izquierdoa aDepartment of Economic Analysis and Political, University of Seville - Universidad de Sevilla * Corresponding Author ([email protected]) Abstract The effect of income and extreme temperatures on the residential electricity consumption in Spain is analyzed. An electricity demand function is estimated by means of ordinary least squares- Driscoll and Kraay, feasible generalized least squares, and quantiles panel data techniques, for the total of the sample, and by coastal climatic zones. The results support the energy-environmental Kuznets curve hypothesis for Spain, the continental and Mediterranean zones. An N form is found for the Atlantic zone. The results also reflect that increases in temperature above 22ºC, and decreases below 15ºC, increase electricity consumption, this increase being progressive as temperatures vary. Thus, extreme temperature generates electricity consumption growth. The Mediterranean zone is most sensitive to these temperature changes. The results also show that electricity consumption is more sensitive to cold than to heat. The results suggest an inefficient use of heating and cooling appliances in the areas with the highest electricity consumption. Keywords: residential electricity consumption, temperatures, income, Spanish Provinces, global warming. 2 Income, extreme temperature and residential electricity consumption in the Spanish Provinces 1. Introduction In recent decades, the global average temperature has risen significantly, compared to previous historical periods, a situation that has been related to the increase in greenhouse gas (GHG) emissions. These gases are the result of human activity (Zilio and Recalde 2011), mainly through energy consumption in the transport sector, heating and electricity generation, which represented 73.2% of total emissions in 2016 (Ritchie and Roser 2020). Against this, the UN (UNEP 2019) suggests that, to limit the increase in global warming to below 1.5°C, and to a maximum of 1.8°C up to the year 2100, it is necessary to reduce emissions to 25 GtCO2e by the year 2030, for which, all economic sectors must be part of that change. The transformation of the electricity sector is of special importance. Although this sector has the greatest increase in the use of non-emissions producing renewable energies, according to IEA (2019), the global electricity demand has been greater than the increase in the consumption of renewable energy in the sector, therefore the increase in the use of renewable energy has not been enough to reduce the associated emissions. According to the International Energy Agency (IEA 2019), these emissions increased by 2.5% in 2018. The study of the growth of electricity demand therefore becomes relevant in a context of the need to control emissions. Several relevant issues related to the growth of electricity demand should be noted. Firstly, it is considered that one of the main determinants of said demand is the evolution of income, or economic growth. Since Kraft and Kraft (1978) studied the relationship between both variables, many other researchers have focused on this issue, mostly showing that economic growth positively affects the demand for energy and electricity (Tiba and Omri 2017). However, this relationship is non-linear, and passes through different stages, which is why the 3 analysis of the Environmental Kuznets Curve (EKC), has recently been extended for energy and electricity demand (Dong and Hao 2017). Secondly, it should be noted that the residential sector is particularly decisive in the growth of electricity consumption. The growth of electricity consumption in the sector was 3.9% in 2018, representing 27% of total electricity consumption worldwide (IEA 2019). Thirdly, the report of the International Energy Agency (IEA 2019) indicates that one of the main causes of the increase in electricity demand is global warming itself. Specifically, the aforementioned report indicates that climate changes are responsible for at least one fifth of the increase in global electricity demand. The relevance of this fact increases when considering that, for the coming years, an increase in both average and extreme temperatures is expected (IPCC 2018). This report also points out that the warmer territories will be those most affected, through consuming more electricity in air conditioning (Burke et al. 2015; IEA 2019). Thus, in these territories, the growth of electrical consumption, linked to temperature rises, can be of special relevance. Finally, it should be noted that this effect can also be aggravated in areas of high tourism, or with significant population changes in the summer months, since, as stated in Pablo-Romero et al. (2019a), the number of tourists is directly related to the electricity consumption of certain economic sectors. This can be extended to the consumption of electricity, derived from the population increase that occurs in certain regions, through occupying second homes or renting empty houses, during the vacation months. In Europe, one of the areas especially affected by climate change and global warming is Spain, since it belongs to the Mediterranean area where temperature increases are expected to be higher than the European average (Ministry of the Environment 2006; European Environment Agency 2017). However, not all Spanish Provinces present the same climatic pattern and, basically, there are three different climatic zones: Mediterranean, Continental and Atlantic (IDAE 2011). Therefore, knowing the relationship between temperatures and electricity consumption in Spain, and especially in the Mediterranean climate zone, is of special 4 relevance, particularly in the residential sector. It is important to note that the electricity consumption of the residential sector represented 31.5% of total electricity consumption in Spain in 2018, with the sector consumption growth being 5% more than the global value. The objective of this work is to analyze the evolution of electricity consumption in the residential sector of the 50 Provinces of Spain, and its relationship with temperatures and the evolution of income, taking into account Spain's different climatic zone, which are also affected by the occupation of second residences in the holiday periods. As at 2011 (last available housing census in Spain), more than 60% of second residences were located in coastal zones (IDAE 2011), emphasizing the Mediterranean zones. To make this analysis, using econometric procedures of panel data referring to the period 2000-2016, for which there is sufficient data available, an electricity consumption function is estimated. This function depends non-linearly on temperatures and income for the 50 Spanish Provinces, and for the subsamples of the coastal provinces with a Mediterranean climate, coastal provinces with an Atlantic climate, and, interior provinces with a continental climate. This division, practically coinciding with the climatic division, also makes it possible to take into account that, in the warmer months, there is greater occupation of the Spanish coastal areas and greater unemployment in the inland areas (CaixaBank Research 2020). The provinces of Ceuta and Melilla are not considered, since these are autonomous cities and complete information is not available for the period analyzed. The econometric estimation of the residential electricity demand function is initially performed by using parametric panel data models. Specifically, by the Fixed Effects Ordinary Least Squares method, with Driscoll-Kraay standard errors (FE-OLS-DK), and by the Feasible Generalized Least Squares (FGLS) method. The method by Hasanov et al. (2021) is observed to determine the polynomial order that best defines the nonlinear relationship between income, temperatures, and electricity consumption. This method, previously used in Pablo-Romero et al. (2023), does not specify a priori the non-linear relationship of the variables but establishes 5 a criterion to determine the order of the polynomial to be considered, both for the temperature variables and the income. Secondly, the econometric estimation of the residential electricity demand function is performed by means of the Method of Moments-Quantile Regression (MM-QR). This method allows us to consider the heterogeneity between the Spanish provinces. To our knowledge, there are no previous studies, for the whole of Spain and, for its subsamples, by climatic zones, which analyze the impact of the non-linearity of the effect of temperatures on residential electricity consumption, by verifying the Environmental Kuznets Curve for energy (E-EKC), also adding quantile regression. The results may help to design energy policy measures aimed at achieving Spain's national energy and emissions-related objectives. Likewise, these will also allow a deepening of the knowledge of the effect of global warming on electricity demand in the residential sector, in the different provinces. 2. Literature review The traditional Environmental Kuznets Curve (EKC) reflects the relationship between economic growth and environmental deterioration (Zilio, 2012). Recently, some researchers have adopted the term Energy Environmental Kuznets Curve, or E-EKC, to test whether an inverted U shape can explain the relationship between energy consumption and income (Dong and Hao 2017). Some of these works, such as those by Dong and Hao (2017) and Sbia et al. (2017), support this hypothesis, while others, such as those of Luzzati and Orsini (2009) and Pablo-Romero and De Jesús (2016), reject this. From the perspective of the E-EKC, several investigations have focused on the analysis of energy consumption in the residential sector. These include the studies by Lenzen et al. (2006), for Australia, Brazil, Denmark, India and Japan; the Volland study (2017), for the United Kingdom; those by Pablo-Romero and Sánchez-Braza (2017) and Borozan (2019), 6 referring to the European Union; that by Bouznit et al. (2018), referring to Algeria; that of Pablo-Romero et al. (2019b), referring to the transition economies; the works of Tilov et al. (2019), for Swiss households; the recent studies by Raza et al. (2020), referring to emerging economies; and, Pablo-Romero et al. (2021), on Andalusian municipalities. The econometric techniques used in these studies are diverse, highlighting the recent studies by Borozan (2019), Tilov et al. (2019), and Pablo-Romero et al. (2021), employing quantile regressions. This allows analysis of whether the independent variables have different effects on the dependent variable (that is, residential energy consumption) throughout the residential energy consumption quantiles, providing more precise definitions of the effects of the studied variables. In these studies, those of Borozan (2019) and Pablo-Romero et al. (2021), consider the climatic conditions as determinants of residential energy consumption. Borozan (2019), uses a dummy variable, constructed from the climatic variable known as Heating Degree Days (HDD), to differentiate between climatic zones, whereas the study of Pablo-Romero et al. (2021), uses HDD and also Cooling Degree Days (CDD). In agreement with Fazeli et al. (2016), the HDD and CDD variables have been most used in studies which consider that temperatures affect the energy demand in the residential sector. Usually, studies have modeled the behavior of temperatures in a linear V-shape, with or without a comfort zone, although the authors point out that other studies that consider non-linear temperature behavior are more sensitive. Along these lines, the study by Labandeira et al. (2012), referring to the electricity demand in Spanish households, defines a non-linear behavior of the HDD and CDD variables, using the squared values of the variables. The use of the HDD and CDD variables to analyze the impact of temperatures on the consumption of residential buildings was reviewed by Li et al. (2012). This review showed that the effect of temperature changes on residential consumption depends on the climatic zone, and concluded that the most significant impact of temperature rise, takes place in zones with warm 7 summers and mild winters. However, in that review, only the study by Lam et al. (2011), in their study on China, specifically analyzed the effect of temperature change on residential and commercial energy consumption by climatic zones. More recently, the study by Li et al. (2018), has insisted on the need to consider the different climatic zones, when studying the impact of temperature changes on electricity consumption in Chinese homes, since substantial differences were found in the use of air conditioning and heating systems, depending on the climatic zone. In the case of Spain, the same conclusion was recently obtained in the study by Zarco-Soto et al. (2020), on the energy consumption of 145 Spanish cities. These studies show that the highest energy consumption is produced in the most extreme climate zones, although these do not take into account other socio-economic variables affecting consumption. Along with the effect of climatic zones, other studies, such as that by Psiloglou et al. (2009), also relate residential electricity demand to the effect of other factors, which may be associated with the weather, and which are partly independent of them, as is the case of vacations. Although this circumstance has not been specifically taken into account in recent studies, it should be noted that the study by Pablo-Romero et al. (2019a), referring to the hospitality sector, pointed out the rise in the number of people associated with tourism. This rise partly explains the increase in electricity consumption in the sector in the Mediterranean provinces, which mainly occurs in the summer months with higher temperatures. The study by Rastogi et al. (2019) shows that population changes affect electricity consumption, so that population changes, throughout the year, may have an influence on electricity consumption. In this study, the E-EKC hypothesis is tested for the residential sector of all the Spanish provinces, for which there are no previous studies. This test is also made taking into account the effect of temperatures, measured through the HDD and CDD variables, in a non-linear manner. This represents an advance within the framework of these studies that allows the sensitivity to the temperature effects to be increased, which is relevant in order to take into 8 account the effect of extreme temperatures, caused by climate change. This study differentiates by climatic and coastal zones, which is an advance because, despite the fact that the climatic zone is considered relevant in considering the effect of the HDD and CDD variables on residential electricity consumption (Li et al. 2012). Likewise, to our knowledge, there is no study that has analyzed the effect of temperature changes on said consumption, as a function of the climatic zone to which it belongs, neither within the E-EKC, nor for the Spanish frameworks. The difference between coastal and non-coastal areas is also incorporated to reflect the population change that occurs in Spain, during the summer months. Finally, the panel data econometric technique is also incorporated through quantile regression, following the recent study by Tilov et al. (2020). 3. Data 3.1. Scope of the study In addition to carrying out the study for the total sample of the 50 Spanish Provinces, this analysis is also carried out for three climatic zones of Spain: Mediterranean, Continental and Atlantic (IDAE 2011), and their sub-divisions into coastal, and non-coastal areas, taking into account that a lot of the Spanish population occupy second homes, at the coast, during vacation periods (CaixaBank Research 2020). The union of both criteria offers an almost perfect crossing, as only three provinces with a Mediterranean climate are inland: Seville, Córdoba and Jaén. For this reason, these were excluded from the group of provinces with a Mediterranean climate, without including them in the continental inland climate zones. This was in order to take into account that the three excluded provinces see their population significantly reduced, during the months of July and August, when temperatures are highest. 9 Thus, the three considered sub-samples are: the group of seven Atlantic coastal climate provinces, the group of 17 Mediterranean coastal climate provinces (excluding the three mentioned above), and the group of 23 continental climate provinces. The Mediterranean coastal climate provinces have the greatest influx of population, during vacation periods, followed by the Atlantic coastal climate provinces. 3.2. Data The data on final annual household electricity consumption, expressed in per capita terms (e), comes from the database of the Ministry for Ecological Transition and the Demographic Challenge (Secretary of State for Energy 2022). The figures are expressed in total annual kWh per inhabitant, using population data from the Spanish National Statistics Institute (INE 2022a). Income per capita (y) is measured from the GDP per inhabitant, in constant 2010 euros. This variable was prepared from the data of the INE (2022b), deflating the data of the nominal GDP, for provinces, from the deflator of the GDP, for Autonomous Communities. To consider the effect of temperatures, two indicators are used: Cooling Degree Days (CDD) and Heating Degree Days (HDD), usually used in the literature as, for example, in Blázquez et al. (2013), Serrano et al. (2017), and Pablo-Romero et al. (2019a). These variables were calculated, for each year, from the data provided by the State Meteorological Agency (AEMET 2022), following the expressions below: 16 differences, with constant and with constant and trend. The results in Table 4 show that the null hypothesis of non-stationarity may be rejected for all series. Therefore, it may be considered that these series are I(0) and may be estimated by Ordinary Least Squares (OLS). Table 4. CIPS unit root test in the presence of cross-sectional dependence. Variables Levels First differences Constant Constant and trend Constant Constant and trend e -2.509*** -3.355*** -4.617*** -4.704*** y -2.607*** -3.070*** -3.741*** -3.849*** y2 -4.730*** -4.529*** -3.046*** -2.545* y3 -3.298*** -3.320*** -3.285*** -3.276*** CDD22 -3.319*** -3.633*** -4.959*** -4.860*** CDD222 -3.135*** -3.467*** -4.690*** -4.529*** HDD15 -3.405*** -3.511*** -4.774*** -4.830*** HDD152 -3.358*** -3.451*** -5.069*** -4.946*** d -1.100 -2.625** -3.208*** -3.559*** Note: ***, **, or * indicate significance at the 1%, 5%, and 10% levels, respectively. Lags criterion decision based on F joint test. Finally, it is convenient to analyze the existence of fixed effects, autocorrelation and heteroscedasticity. Table 6 shows the results of the Hausman fixed effects test, which indicate the advantage of estimating by fixed effects. This table also shows the results of the Wald heteroscedasticity test (Greene 2000), which indicate that the null hypothesis of homoscedasticity is rejected. Furthermore, Table 5 includes the results of the autocorrelation test, which indicate that the null hypothesis of no correlation may not be rejected. Table 5. Hausman, heteroscedasticity, and autocorrelation tests. Test Results Hausman chi2 (8) = 67.87*** Heteroscedasticidad chi2 (50) = 5097.39*** F(1,49) 1.068*** Note: *** indicates significance at the 1% level. 17 Taking into account the presence of heteroscedasticity and contemporaneous autocorrelation, and the advantage of estimating by fixed effects, Equation [5] is estimated by using FE-OLS-DK since this method, according to Hoechle et al. (2017), is robust to heteroscedasticity and contemporaneous correlation. Additionally, in order to compare the results, it is also estimated by using FGLS, with the presence of individual dummies, to collect provincial fixed effects. Lastly, the quantile regression model is used to complement the analysis. This model is appropriate when variables may have different effects at distinct points of the conditional distribution of the dependent variable. These effects may not be captured by the OLS, or FGLS model, as these usual regressions refer to the mean (Hübler 2017), but may be captured by quantile regression, which is more efficient in these cases than the OLS method. Thus, in this study, the estimations by using the MM-QR model, developed by Machado and Silva (2019), are also obtained, which allows the presence of fixed effects to be taken into account. This estimation makes it possible to analyze whether income, temperatures and population density, have different effects along the quantiles of residential electricity consumption, allowing a more complete view of the effect of these variables on electricity consumption. 5. Results and discussion 5.1. FE-OLS-DK and FGLS results for the 50 Spanish Provinces Table 6 shows the results of estimating Equation [5] by FE-OLS-DK and by FGLS. Initially, an extended Equation [5] with a quadratic polynomial for income and cubic polynomial for temperatures variables was estimated. Nevertheless, the quadratic coefficient for the income variable and the cubed coefficients for the temperature variables were not significant. Therefore, according to the Hasanov et al. (2021) strategy, Equation [5] has been re-estimated 18 without these coefficients. Column (a) shows the results of the FE-OLS-DK estimation. Once again, since the coefficient associated with the cubed income variable was not significant, the equation has been re-estimated excluding this variable. The results of which are shown in column (b). It can be appreciated that the coefficients of the other variables do not vary substantially. Column (c) shows the results of the FGLS estimation. In this case, the squared coefficient for HDD15 is not significant. Therefore, Equation [5] has been re-estimated without this variable. Column (d) shows these results. Table 6. FE-OLS-DK and FGLS estimate results for the 50 Spanish provinces. Variables (a) (b) (c) (d) y 0.418*** (0.049) 0.443*** (0.031) 0.398*** (0.044) 0.414*** (0.035) y2 -0.369*** (0.053) -0.380*** (0.060) -0.367*** (0.027) -0.319*** (0.069) y3 0.138 (0.110) - 0.182** (0.079) 0.306** (0.118) CDD22 0.040** (0.017) 0.041** (0.017) 0.035*** (0.008) 0.038*** (0.008) CDD222 0.007*** (0.002) 0.007*** (0.002) 0.004** (0.001) 0.007*** (0.001) HDD15 0.118** (0.044) 0.118** (0.044) 0.053** (0.0091) 0.072*** (0.012) HDD152 0.006** (0.002) 0.006** (0.002) 0.001 (0.002) d 0.530*** (0.098) 0.486*** (0.0667) 0.454*** (0.057) 0.459*** (0.052) The estimates in columns (a), (b), (c) and (d) show positive and significant coefficients for income, and negative and significant coefficients for income squared. However, the estimated coefficient for income cubed is positive and not significant in column (a), while it is positive and significant when estimated by FGLS. Thus, the results by FE-OLS-DK are compatible with the Kuznets hypothesis for residential energy, in line with the results obtained for EU countries by Borozan (2019), and the results of Pablo-Romero et al. (2021) and Tilov et 19 al. (2019), in the upper income bracket, as a more flexible form of inverted N is observed by the authors. In the case of the estimate by the FGLS method, the results fit better to the N-shape. Therefore, the Kuznets hypothesis may only be considered to be confirmed when income levels do not reach high levels since, in that case, residential electricity consumption tends to grow again. Regarding the temperature variables, the coefficients of CDD22 and HDD15 are positive and significant in all estimates, although the coefficients are somewhat lower when estimated by FGLS. Therefore, an increase in temperature above 22ºC, and a decrease below 15ºC, increase residential electricity consumption in Spanish Provinces. In addition to this positive relationship, the results also indicate that this effect grows progressively in both cases, since it is observed that the coefficients of both variables squared are positive and significant, except in the estimation by FGLS, where the coefficient associated with HDD15 squared is not significant. It should also be noted that the coefficients related to HDD15 are higher than those related to CDD22. Therefore, more extreme cold has a greater impact on household electricity consumption than extreme heat. This conclusion is in line with the studies for Athens and London by Psiloglou et al. (2009), and the recent study for Portugal by Silva et al. (2020), which show that electricity consumption increases more with heating needs than with cooling needs. This is also in line with the results obtained by Blázquez et al. (2013), for the Spanish economy. Finally, the results also show a positive and significant impact of population density on residential energy consumption. However, it should be noted that this difference is not as pronounced when estimated by FGLS, as when estimated by FE-OLS-DK. Once estimates have been performed, and according to the Hasanov et al. (2021) strategy, it is adequate to analyze the electricity elasticity values respect to the income and temperature variables to study if the turning points are within a reasonable range, and therefore the estimate functions are robust. The elasticity values for each country and year, may be 20 obtained by calculating the derivative of the electricity consumption variable with respect to income and with respect to the temperature variables. Figure 2 shows the elasticity values for each transformed income, CDD22 and HDD15 level in logs, when using column (b) and (d) coefficients. The elasticity of energy consumption with respect to GDPpc is decreasing and reach the zero value, indicating an inverted U shape relationships between variables. Nevertheless, when using the Column d coefficients, the decreasing elasticity Figure 2. Estimated elasticity of energy consumption with respect to GVA per capita, CDD22 and HDD15. Source: Own elaboration The elasticity of energy consumption with respect to GDPpc is decreasing and reaches zero, indicating an inverted U-shaped relationship between the variables. However, using the coefficients in column d, the decreasing elasticity becomes increasing, indicating an inflection 0.5 11.5 elasticity of energy consumption with respect to GVApc 2 2.5 3 3.5 4 GDPpc (in logs) with Column b coefficients with Column d coefficients -.04 -.02 0 .02 .04 .06 elasticity of energy consumption with respect to CDD22 -2 0 2 4 6 CDD22 (in logs) with Column b coefficients with Column d coefficients .06 .08 .1 .12 .14 elasticity of energy consumption with respect to HDD15 5 6 7 8 HDD15 (in logs) with Column b coefficients with Column d coefficients 21 point. This indicates an N-shaped relationship. Since the elasticity never reaches zero, the Kuznets hypothesis cannot be said to hold for any GDPpc interval. The elasticity of energy with respect to CDD22 is increasing at all points, regardless of the coefficients used. This shows a progressively increasing effect of heat on energy consumption. Finally, the energy elasticity with respect to HDD15 is increasing in all its points when using the coefficients of Column b, and constant if using those of Column d. The former shows a progressively increasing effect of cooling on energy consumption, and the latter a constant positive effect. In short, the value of the elasticities shows that extreme heat induces a more than proportional energy consumption, while the same cannot be said, according to all the results, for extreme cold. 5.2. MM-QR results for the 50 Spanish Provinces The above results show that there are some differences in what is happening with the significance of the income and temperature variable coefficients, which may be due to some heterogeneity of the data that previous econometric procedures do not capture. Therefore, Equation [5] is re-estimated by using MM-QR, selecting nine quantiles (10, 20, 30, 40, 50, 60, 70, 80, and 90). The obtained results are presented in Table 7. Table 7. MMQR estimate results for the 50 Spanish provinces. Variables e q(0.1) e q(0.2) e q(0.3) e q(0.4) e q(0.5) e q(0.6) e q(0.7) e q(0.8) e q(0.9) y 0.480*** (0.087) 0.461*** (0.070) 0.448*** (0.059) 0.431*** (0.051) 0.418*** (0.048) 0.406*** (0.050) 0.393*** (0.056) 0.378*** (0.067) 0.354*** (0.087) y2 -0.3951*** (0.095) -0.3873*** (0.076) -0.381*** (0.065) -0.373*** (0.055) -0.3701*** (0.0527) -0.3652*** (0.0545) -0.3596*** (0.0612) -0.3535*** (0.0728) -0.3442*** (0.0954) y3 0.1736 (0.2597) 0.1629 (0.2079) 0.1554 (0.1774) 0.1462 (0.1512) 0.1391 (0.1438) 0.1324 (0.1487) 0.1247 (0.1671) 0.1163 (0.1987) 0.1034 (0.2602) CDD22 0.0501*** (0.0198) 0.0471*** (0.0159) 0.0450*** (0.0135) 0.0424*** (0.0116) 0.04046*** (0.0110) 0.0385*** (0.0113) 0.0364*** (0.0127) 0.0340*** (0.0152) 0.0304*** (0.0199) CDD222 0.0123*** (0.0048) 0.0108*** (0.0038) 0.0098*** (0.0033) 0.0085*** (0.0028) 0.0075*** (0.0026) 0.0066*** (0.0027) 0.0055** (0.0031) 0.0044 (0.0037) 0.0026 (0.0048) HDD15 0.1504*** (0.0423) 0.1405*** (0.0338) 0.1335*** (0.0288) 0.1250*** (0.0246) 0.1184*** (0.0234) 0.1122*** (0.0241) 0.1050*** (0.0271) 0.0972*** (0.0323) 0.0852*** (0.0424) HDD152 0.0109*** (0.0036) 0.0095*** (0.0028) 0.0086*** (0.0024) 0.0075*** (0.0021) 0.0066*** (0.0019) 0.0058*** (0.0020) 0.0048*** (0.0023) 0.0038 (0.0027) 0.0022 (0.0036) 22 d 0.7305*** (0.1540) 0.6690*** (0.1226) 0.6259*** (0.1047) 0.5734*** (0.0898) 0.5323*** (0.0851) 0.4939*** (0.0879) 0.4498*** (0.0987) 0.4013*** (0.1174) 0.3271*** (0.1543) Once again, the absence of significance of the income cubed variable can be observed, highlighting that the model with these three variables may not be the most appropriate to represent the data for the Spanish Provinces. The elimination of this variable does not substantially modify the value and significance of the rest of the coefficients, so it is not considered necessary to present the estimation data omitting this variable. In relation to the coefficients of income and income squared, both are significant in all quantiles, in all cases presenting a positive sign for income, and a negative sign for income squared. These results again show the compatibility of the data with the E-EKC hypothesis, in this case, in all quantiles of the sample. However, the values of the coefficients are not the same for all quantiles, but for both variables these tend to decrease in absolute terms. This reflects that as the quantiles increase, which implies higher residential electricity consumption, the Kuznets curve tends to be less steep, that is, it tends to flatten out, making the effect of increasing income translate into a smaller reduction in electricity consumption, in the higher electricity consumption quantiles. This fact may evidence the lack of energy efficiency in those locations, and a reduction in electricity consumption cannot be guaranteed, based on increasing household income alone. On the contrary, for lower per capita electricity consumption areas (first quantiles), the Kuznets curve is steeper. Thus, once a given income threshold is exceeded, small increases in income are accompanied by significant reductions in per capita electricity consumption. This leads to the consideration that significant reductions in electricity consumption may be achieved, if accompanied by appropriate measures. Regarding temperature, similar values to those obtained in previous estimations are observed. In all cases, positive and significant coefficients are obtained for the four variables related to temperature, although the lack of significance of the squared variables is observed 23 for the higher quantiles. Therefore, these results suggest a positive effect on electricity consumption for both an increase in temperature above 22ºC, and a decrease below 15ºC. Moreover, these also suggest a progressively increasing relationship of this effect, as temperatures increase or decrease, except for the higher quantiles. Nevertheless, it is curious to note how this temperature effect decreases as the quantile increases, which may suggest inefficient use of heating and cooling appliances in areas of higher electricity consumption, where sensitivity to temperature changes is lost. This fact may be due, for example, to always having the heating on, regardless of how cold it is. In this sense, it can be indicated that the provinces with lower electricity consumption use their devices more efficiently. Several reasons could induce this behavior, including less use of devices through income considerations, or a greater awareness of energy saving in certain provinces. It is also worth noting that the coefficients associated to HDD15 are again higher than those related to CDD22, so those provinces with lower temperatures tend to have higher electricity consumption. These results are in line with previous studies for European countries, such as those by Borozan (2018, 2019), and with the results presented by Blázquez et al. (2013), for Spanish provinces. The results of this study show a higher sensitivity of electricity consumption to low temperatures, than to high temperatures, which the authors attribute to the fact that households in Spain are provided with more heating, than with air conditioning. Likewise, the results are also in line with the study by Pablo-Romero et al. (2019a) on the hospitality sector in the Mediterranean provinces in Spain, where a greater effect of low temperatures on the sector’s electricity consumption, is also evident. Finally, the results also reveal that population density has a positive impact on household electricity consumption. In this case, all the coefficients by quantiles are positive, significant and decreasing, which implies that a percentage increase in population density has a greater positive impact in the provinces, with lower electricity consumption. 24 5.3. FE-OLS-DK and FGLS results by climatic zones The differences observed in the previous estimates may also be associated with the sample heterogeneity, related to the differences in climate and location of the provinces. Thus, in order to better specify the effect of income and temperatures on residential electricity consumption in Spain, estimates of Equation [5] by FE-OLS-DK and FGLS are performed for each subsample of provinces considered: coastal Mediterranean, interior and Atlantic. Initially, the complete Equation [5] was estimated for each subsample and, subsequently, it was re-estimated by eliminating the non-significant variables. However, it is observed that the coefficients do not vary substantially, with respect to the full estimates. Table 8 shows the estimates for each subsample, with only the significant variables. Table 8. FE-OLS-DK and FGLS estimate results for the three climatic zones. Variables FE-OLS-DK: Mediterranean FGLS: Mediterranean FE-OLS-DK: Interior FGLS: Interior FE-OLS-DK: Atlantic FGLS: Atlantic Y 0.479*** (0.048) 0.526*** (0.001) 0.420*** (0.045) 0.358*** (0.024) 0.277** (0.099) 0.393*** (0.083) y2 -0.513*** (0.133) -0.212*** (0.002) -0.400*** (0.041) -0.377*** (0.029) -0.518** (0.173) -0.361*** (0.088) y3 - - - - 1.986** (0.830) 1.122*** (0.350) CDD22 0.258*** (0.017) 0.114*** (0.000) 0.020 (0.019) 0.012*** (0.004) - - CDD222 -0.105* (0.048) - 0.008* (0.005) 0.006*** (0.001) - - HDD15 0.146*** (0.020) 0.187*** (0.000) 0.044* (0.026) 0.037*** (0.005) - 0.056* (0.032) HDD152 0.009*** (0.001) 0.012*** (0.000) - - - - D 0.883*** (0.139) 0.542*** (0.002) 0.289*** (0.096) 0.240*** (0.027) - - 25 Columns 2 and 3 show the results for the coastal Mediterranean provinces. These results confirm the E-EKC hypothesis, with coefficients for income somewhat higher than in the estimates, for all provinces. The coefficients associated with CDD22 are positive and significant in both cases, although the coefficient associated with CDD22 squared is not significant, when estimated by FGLS, and negative and barely significant for FE-OLS-DK. Therefore, it cannot be affirmed that there is a progressively increasing relationship between high temperatures and electricity consumption, as obtained for the country as a whole. Regarding the coefficients associated with HDD15, these are positive and significant, showing, as at the national level, a positive and progressively increasing effect on electricity consumption as temperatures fall. The values of these coefficients are also similar to those for the total sample. Finally, the high value of the population density coefficient stands out, almost doubling the coefficient obtained at the national level. In this sense, it should be noted that the Mediterranean area is notable in Spain for its high population density, compared to the inland areas, except for Madrid and some bordering areas. Columns 4 and 5 show the results for the interior provinces, again confirming the E- EKC hypothesis, with coefficients similar to the national provinces, and somewhat lower than those of the Mediterranean zone. The results for coefficients associated with CDD22 are positive, showing a progressively increasing relationship, although it should be noted that the it is not significant when estimating by the FE-OLS-DK method, and the values of the coefficients are lower than those of the national sample and the Mediterranean provinces. Thus, a positive and progressive effect of high temperatures on electricity consumption may be observed, although this effect is smaller and weaker than in the Mediterranean areas. This may be due to the holiday effect, whereby part of the population of the interior provinces tends to leave during the hottest months. Regarding the coefficients associated with HDD15, it is observed that their values and levels of significance are lower than those observed in the 32 Bouznit M, Pablo-Romero MP, Sánchez-Braza A. 2018. Residential electricity consumption and economic growth in Algeria. Energies 11(7):1-18. Boyd GA. 2014. Estimating the changes in the distribution of energy efficiency in the U.S. automobile assembly industry. Energy Economics 42:81-87. Burke M, Hsiang SM, Miguel E. 2015. Global non-linear effect of temperature on economic production. Nature 527(7577):235−239. CaixaBank Research. 2020. Sector Report. Real Estate 2020 (1st semester). Caixabank Research: Barcelona. Domene E, Saurí D. 2006. Urbanisation and water consumption: influencing factors in the metropolitan region of Barcelona. Urban Studies 43(9):1605-1623. Dong XY, Hao Y. 2017. Would income inequality affect electricity consumption? Evidence from China. Energy 142:215-227. European Environment Agency. 2017. Climate Change, Impacts and Vulnerability in Europe 2016, EEA Report No 1/2017. European Environment Agency, Publications Office of the EU: Luxembourg. Fan S, Hyndman RJ. 2011. The price elasticity of electricity demand in South Australia. Energy Policy 39(6):3709-3719. Fazeli R, Davidsdottir B, Hallgrimsson JH. 2016. Residential energy demand fors pace heating in the Nordic countries: Accounting for interfuel substitution. Renewable and Sustainable Energy Reviews 57:1210-1226. Greene W. 2000. Econometric Analysis, 4th edition. Prentice Hall: Upper Saddle River, New Jersey. Hoechle D, Ruenzi S, Schaub N, Schmid M. 2017. The impact of financial advice on trade performance and behavioral biases. Review of Finance 21(2):871-910. Hübler M. 2017. The inequality-emissions nexus in the context of trade and development: a quantile regression approach. Ecological Economics 134:174-185. IDEA. 2011. Análisis del Consumo Energético del Sector Residencial en España. Informe SECH-SPAHOUSEC. Instituto para la Diversificación y Ahorro de la Energía, Ministerio de Industria, Energía y Turismo: Madrid. https://www.idae.es/uploads/documentos/documentos_Informe_SPAHOUSEC_ACC_f68 291a3.pdf IEA. 2019. Global Energy & CO2 Status Report 2019. International Energy Agency: París. https://www.iea.org/reports/global-energy-co2-status-report-2019 33 INE. 2022a. INEbase/Estadísticas/Demografía y Población/Cifras de población y censos demográficos; Datos de Superficie por Provicias. Instituto Nacional de Estadística: Madrid. https://www.ine.es INE. 2022b. INEbase/Estadísticas/Economía/Cuentas Económicas. Instituto Nacional de Estadística: Madrid. https://www.ine.es INE. 2022c. Métodos y proyectos/Nuevos proyectos/Estadística experimental/Atlas de distribución de renta de los hogares. Instituto Nacional de Estadística: Madrid. https://www.ine.es IPCC. 2018. Global Warming of 1.5°C. Intergovernmental Panel on Climate Change, Cambridge University Press: Cambridge, UK and New York. https://www.ipcc.ch/sr15/ Kraft J, Kraft A. 1978. On the relationship between energy and GNP. Journal of Energy and Development 3(2):401-403. Labandeira X, Labeaga JM, López-Otero X. 2012. Estimation of elasticity price of electricity with incomplete information. Energy Economics 34(3):627-633. Lam JC, Wan KKW, Pan W, Li DHW. 2011. Long-Term Trends of Degree-Days in the 20th and 21st Centuries and Heating and Cooling Energy Use Implications. Research report. Department of Civil and Architectural Engineering: City University of Hong Kong. Lenzen M, Wier M, Cohen C, Hayamid H, Pachauri S, Schaeffe R. 2006. A comparative multivariate analysis of household energy requirements in Australia, Brazil, Denmark, India and Japan. Energy 31(2-3):181-207. Li DHW, Yang L, Lam JC. 2012. Impact of climate change on energy use in the built environment in different climate zones-A review. Energy 42(1):103-112. Li M, Cao J, Xiong M, Li J, Feng X, Meng F. 2018. Different responses of cooling energy consumption in office buildings to climatic change in major climate zones of China. Energy and Buildings 173:38-44. Luzzati T, Orsini M. 2009. Investigating the energy-environmental Kuznets curve. Energy 34(3):291-300. Machado JAF, Silva JMCS. 2019. Quantiles via moments. Journal of Econometrics 213(1):145- 173. Ministry of the Environment. 2006. Plan Nacional de Adaptación al Cambio Climático (PNACC). Ministerio de Medio Ambiente: Madrid. Nguyen-Van P. 2010. Energy consumption and income: A semiparametric panel data analysis. Energy Economics 32(3):557-563. 34 Pablo-Romero MP, De Jesús J. 2016. Economic growth and energy consumption: The energyenvironmental Kuznets curve for Latin America and the Caribbean. Renewable and Sustainable Energy Reviews 60:1343-1350. Pablo-Romero MP, Sánchez-Braza A. 2017. Residential energy environmental Kuznets curve in the EU-28. Energy 125:44-54. Pablo-Romero M, Pozo-Barajas R, Sánchez-Rivas J. 2019a. Tourism and temperature effects on the electricity consumption of the hospitality sector. Journal of Cleaner Production 240: 118168,11pp. Pablo-Romero MP, Sánchez-Braza A, Galyan A. 2019b. Relationship between economic growth and residencial energy use in transition economies. Climate and Development 11(4):338-354. Pablo-Romero MP, Pozo-Barajas R, Molleda-Jimena G. 2021. Residential energy environmental Kuznets curve extended with non-linear temperature effects: A quantile regression for Andalusian (Spain) municipalities. Environmental Science and Pollution Research 28:48984-48999. Pesaran MH. 2004. General diagnostic tests for cross section dependence in panels. Cambridge Working Papers WP0435, Faculty of Economics, University of Cambridge: Cambridge. http://www.econ.cam.ac.uk/research-files/repec/cam/pdf/cwpe0435.pdf Pesaran MH. 2007. A simple panel unit root test in the presence of cross-section dependence. Journal of Applied Econometrics 22(2):265-312. Psiloglou BE, Giannakopoulos C, Majithia S, Petrakis M. 2009. Factors affecting electricity demand in Athens, Greece and London, UK: A comparative assessment. Energy 34(11):1855-1863. Rastogi D, Holladay JS, Evans KJ, Preston BL, Ashfaq M. 2019. Shift in seasonal climate patterns likely to impact residential energy consumption in the United States. Environmental Research Letters 14(7):074006,10pp. Raza SA, Shah N, Khan KA. 2020. Residential energy environmental Kuznets curve in emerging economies: The role of economic growth, renewable energy consumption, and financial development. Environmental Science and Pollution Research 27(5):5620-5629. Ritchie H, Roser M. 2020. CO₂ and Greenhouse Gas Emissions. Published online at OurWorldInData.org. https://ourworldindata.org/co2-and-other-greenhouse-gas- emissions Sbia R, Shahbaz M, Ozturk I. 2017. Economic growth,financial development, urbanisation and electricity consumption nexus in UA. Economic Research-Ekonomska Istraživanja 30(1):527-549. 35 Secretary of State for Energy. 2022. Energía/Estadísticas y Balances Energéticos/ Publicaciones estadísticas/Estadísticas Eléctricas Anuales. Secretaría de Estado de Energía, Ministerio para la Transición Ecológica y el Reto Demográfico: Madrid. https://energia.gob.es Serrano S, Ürge-Vorsatz D, Barreneche C, Palacios A, Cabeza LF. 2017. Heating and cooling energy trends and drivers in Europe. Energy 119:425-434. Silva S, Soares I, Pinho C. 2020. Climate change impacts on electricity demand: The case of a Southern European country. Utilities Policy 67:101115,6pp. Suri V, Chapman D. 1998. Economic growth, trade and energy: Implications for the environmental Kuznets curve. Ecological Economics 25(2):195-208. Tiba S, Omri A. 2017. Literature survey on the relationships between energy, environment and economic growth. Renewable and Sustainable Energy Reviews 69:1129–1146. Tilov I, Farsi M, Volland B. 2019. Interactions in Swiss households’ energy demand: A holistic approach. Energy Policy 128:136-149. Tilov I, Farsi M,Volland B. 2020. From frugal Jane to wasteful John: A quantile regression analysis of Swiss households’ electricity demand. Energy Police 138:111246,11pp. UNEP. 2019. Emissions Gap Report 2019 - Global Progress Report on Climate Action. United Nations Environment Programme, UNEP: Nairobi. https://www.unep.org/resources/emissions-gap-report-2019 Volland B. 2017. The role of risk and trust attitudes in explaining residential energy demand: Evidence from the United Kingdom. Ecological Economics 132:14-30. Yin J, Zheng M, Chen J. 2015. The effects of environmental regulation and technical progress on CO2 Kuznets curve: An evidence from China. Energy Policy 77:97-108. Zarco-Soto IM, Zarco-Periñán PJ, Sánchez-Durán R. 2020. Influence of climate on energy consumption and CO2 emissions: The case of Spain. Environmental Science and Pollution Research 27:15645-15662. Zilio M. 2012. Curva de Kuznets ambiental: La validez de sus fundamentos. Cuadernos de Economía 35:43-54. Zilio M, Recalde M. 2011. GDP and environment pressure: The role of energy in Latin America and the Caribbean. Energy Policy 36(12):7941-7949.