Productive electricity and non-electricity consumption effects on economic growth: A Latin America analysis

Energy 305 (2024) 132281 Available online 2 July 2024 0360-5442/© 2024 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/bync/4.0/). Productive electricity and non-electricity consumption effects on economic growth: A Latin America analysis María P. Pablo-Romero a , * , Rafael Pozo-Barajas b , Christian Washburn c a Department of Economic Analysis & Political Economics, University of Seville, Spain b Department of Financial Economics & Operations Management, University of Seville, Spain c Faculty of Economic Science, University of Guayaquil, Ecuador ARTICLE INFO Keywords: Electricity use Non-electricity use Economic growth Energy transition Latin America ABSTRACT Electrification is a key pillar in energy transformation to reduce emissions. However, it may also have consequences on production. This study analyzes the effect of using electricity and non-electricity energy sources on production growth. An extended neoclassical production function, with the two uses of energy (electricity and non-electricity), is estimated for 17 Latin American countries from 1990 to 2019 and, for subsamples, based on the level of production per persons employed. The study is performed applying the cross-sectionally augmented, autoregressive distributed lag modeling approach and a non-parametric time-varying coefficients model. The parametric results indicate that the effect of both sources of energy on production are positive and significant. However, the production elasticity with respect to non-electricity consumption is higher than that with respect to electricity consumption. Only in countries with the lowest production levels could electrification be neutral. The non-parametric results show that the effect of non-electricity consumption is decreasing slightly over the period, while the effect of electricity use is increasing slightly, with a tendency towards stabilization at the end of the period. Nevertheless, these changes have been small. 1. Introduction Greenhouse gas (GHG) levels continue to rise at high rates. According to the NOAA report [1], the main contributor to these emissions is carbon dioxide (CO 2 ). It is, therefore, important to achieve carbon neutrality to limit GHG emissions, which requires a profound global energy transformation. According to IRENA [2], this implies a shift from a fossil fuel-based system to one that promotes efficiency and renewable energy. Electrification is a key pillar [3], as it is the only way to increase the share of renewables in total primary energy supply. However, the process of electrification and energy source switching in production processes not only influences emissions, but it can also have consequences on economic growth. Previous literature has found evidence that energy consumption affects economic growth [4]. Furthermore, several authors conclude that it is a key input that determines the growth process [5]. Nevertheless, other authors point out that the relationship between energy consumption and production depends on the type of energy [6]. As a result, there is a growing body of empirical literature on the relationship between different types of energy (including electricity) and economic growth. Majewski et al. [7] conducted a review of the studies which analyzed the impact of electricity on growth. They found at least 19 of them support the hypothesis of its positive influence on economic growth. Two considerations should be made in this regard. First, the evidence suggests that the impact of electricity consumption on production depends on the region/country analyzed [8]. It is appropriate to conduct specific regional analyses as, in this respect, there are few specific studies on Latin America and the Caribbean (LAC) region. These include Al-Mulali et al. [9], Valencia-Herrera and Santillan-Salgado [10] and Solarin and Ozturk [11]. However, this type of analysis is particularly relevant. On the one hand, although the region has undergone a remarkable electrification process [12], there are still 17 million inhabitants without electricity [13]. Thus, the region could have a significant impact on poverty reduction if electrification leads to economic growth [14]. On the other hand, the region is a leader in the use of renewable energy for electricity generation [15]. There is therefore interest in the electrification process as it can have a significant impact on emissions reduction. Second, although there is previous literature on the impact of electricity consumption on economic growth, there are very few studies * Corresponding author. E-mail addresses: [email protected] (M.P. Pablo-Romero), [email protected] (R. Pozo-Barajas), [email protected] (C. Washburn). Contents lists available at ScienceDirect Energy journal homepage: www.elsevier.com/locate/energy https://doi.org/10.1016/j.energy.2024.132281 Received 5 March 2024; Received in revised form 22 May 2024; Accepted 30 June 2024 Energy 305 (2024) 132281 2 which focus on the electrification process [16] and none of these are recent. Nevertheless, this analysis is relevant. The success of electrification to improve sustainability necessarily depends on whether the substitution of electricity for another energy source, does not adversely affect economic growth, especially in developing countries. In this line, there are previous studies that focus on how other pillars of decarbonization affect economic growth, including Yasmeen et al. [17] and Ivanovski et al. [18]. The authors used the neoclassical production function framework to show the different impact of renewable and non-renewable energy on growth. However, to our knowledge, no author has jointly analyzed the effect of electricity and non-electricity consumption on growth. This paper aims to fill this gap. For this purpose, the model used by Ivanovski et al. [18] for renewable energy is reformulated to examine how the electricity and non-electricity consumption affects growth. Thus, an extended neoclassical production function including the variables of these energies is estimated, using panel data techniques. Moreover, the proposed model considers that not all the energy consumed affects the production process, but only the production energy [19], which is a novelty in this context. The study is first carried out by applying the cross-sectional augmented autoregressive distributive lag model (CS-ARDL) [20]. This model addresses autocorrelation, heteroskedasticity, endogeneity, heterogeneity and cross-sectional dependence (CSD). Secondly, since previous literature suggests that energy effects on growth tend to change over time [21], a non-parametric time-varying coefficient model is implemented to show how the effects of consuming these energies changed over time. The study is conducted for 17 LAC countries, from 1990 to 2019 (and for 3 subsamples based on production levels), extending the literature on the effects of electricity on growth in the region. The main contributions of this paper in the above context are. •Within the framework of the neoclassical production function, a model is formulated that allows the impact of electricity and nonelectricity consumption on economic growth to be jointly analyzed. •Parametric and non-parametric estimates are carried out. The nonparametric estimates allow it to be known how the electricity and non-electricity consumption effects change over time. •It provides an analysis for the LAC region. It distinguishes these effects, according to whether the countries have higher, lower or the lowest levels of production. The results indicate that, although both energy sources have significant positive effects on production, the elasticity, with respect to nonelectricity consumption, is higher than that with respect to electricity consumption. Only in countries with the lowest levels of production per worker could electrification be neutral. The results also show that the effect of non-electricity consumption is decreasing over the period, while the effect of electricity use is increasing. Nevertheless, these changes have been small. The paper is structured as follows. After this introduction, a brief review of the literature is conducted in Section 2. Section 3 describes the model of the study. Section 4 describes the data used and the econometric strategy applied. Section 5 shows the results and provides a discussion of the findings. Finally, the conclusion is given. 2. Literature review Several decarbonization instruments have been put in place to achieve carbon neutrality. In essence, these have been carbon taxes, emissions standards, cap and trade systems, excise taxes on energy sources, policies to promote the development of renewable energy sources and the regulatory mix [22]. Among these instruments, some authors highlight the relevant role of emissions trading systems [23]. Others highlight the increase in the use of renewable energy [24], for which the process of electrification of the economy may be key [25]. However, these last policies can be expected to affect economic growth. This paper focuses on analyzing the impact of using electricity versus alternative energy on economic growth. The following review includes previous literature on energy, specific energy and electrification effects on economic growth. 2.1. Energy and economic growth: the growth hypothesis and the extended production function Since the study by Carter [26], the analysis of the causal relationship between energy and economic growth has been very extensive [4]. These studies are based on four hypotheses. The growth hypothesis states that energy is a factor of production. The conservation hypothesis states that economic growth affects energy consumption. The feedback hypothesis states that there is mutual interdependence, and the neutral hypothesis states that there is no relationship between the variables. The results of these studies are ambiguous [see previous reviews in 4, 27 and 28]. Because of the ambiguity of these results, two lines of research have been developed. On the one hand, one line of research has focused on identifying the effect of economic growth on energy consumption, linking these studies to energy demand analyses. Some authors in this line have emphasized the non-linear effect of income on energy consumption, testing the energy-EKC hypothesis [27]. On the other hand, another line of researchers has analyzed the energy effect on growth in the framework of the neoclassical production function, following Kümmel et al. [28] and Moroney [29]. For these authors, energy is a crucial factor to include in a neoclassical function [5]. A sample of Table 1 Previous studies using the extended production function with energy. Author Year Sample Independent variables Kümmel et al. [28] 1985 West German K, L, E Moroney [29] 1992 USA K, L, E Stern [30] 1993 USA K, L, E Stern [31] 2000 USA K, L, E Kummel et al. [32] 2002 USA, Japan, Germany K, L, E Ghali and El-Sakka [33] 2004 Canada K, L, E Oh and Lee [34] 2004 Korea K, L, E Oh and Lee [35] 2004 Korea K, L, E, price Ayres and Warr [36] 2005 USA K, L, E Sari and Soytas [37] 2007 6 developing countries K, L, E Lee and Chang [38] 2008 16 Asian countries K, L, E Apergis and Payne [39] 2009 6 Central American economies K, L, E Apergis and Payne [40] 2009 11 Commonwealth of Independent States K, L, E Lee and Chien [41] 2010 G-7 countries K, L, E Stern [42] 2011 USA K, L, E Ayres and Voudouris [43] 2014 USA, UK, Japan K, L, E Pablo-Romero and S´ anchez-Braza [19] 2015 OECD, BRIC, NAFTA, East Asia, East Europe and EU15 K, L, E, H Streimikiene and Kasperowicz [44] 2016 European Union countries K, L, E Fang and Chen [45] 2016 16 Asian countries K, L, E, H Fang and Chen [46] 2017 China K, L, E, H Hao et al. [47] 2020 China K, E, financial variables, trade Chang et al. [48] 2020 5 Asian countries K, L, E, H Chi et al. [49] 2021 China K, L, E Aghaei and Lin Lawell [50] 2022 Iran K, L, E Zakari and Oluwaseyi Musibau [51] 2024 OECD K, L, E, energy security Warsame et al. [52] 2024 Somalia K, L, E Note: K, L, E and H indicate capital stock, labor or population, energy consumption and human capital, respectively. M.P. Pablo-Romero et al. Energy 305 (2024) 132281 3 previous studies, using this extended production function with total energy consumption, is detailed in Table 1. The independent variables usually included in this extended production function are capital stock, employment or population, and energy consumption. In addition, based on the objective of the study, some studies incorporate additional factors, such as [35,47]. In this line, the study by Fang and Chen [46] also includes human capital and they conclude that it plays a relevant role in economic growth. However, only a few studies have included this factor in the literature, related to the energy-growth nexus, which may be due to the difficulty in finding quality data. Among those studies, it is worth mentioning that by Pablo-Romero and S´ anchez-Braza [19], which uses a translog production function, the study by Fang and Chang [45] for Asia, and the study by Chang et al. [48]. In the latter study, the authors point out that energy seems to play a less important role when human capital is considered. Therefore, the inclusion of this variable is relevant to have an adequate test of the impact of energy on growth. Finally, it is worth noting that previous studies have also shown that the impact of energy on economic growth can vary over time. In this respect, the study by Ren et al. [21] for the provinces of China is noteworthy. 2.2. The extended production function with energy: specific energy sources Recently, the interest in examining the energy impact on economic growth remains in force, raising even more attention [21]. Although there are still studies focusing on total energy effects, such as [51,52], much of the recent literature focuses on a specific source of energy. Among these studies, Ozturk and Al-Mulali [53], and Hassan et al. [54], for example, focused on natural gas, while Rahman et al. [55] focused, alternatively, on coal, gas, and oil. Among these studies, those related to electricity can also be cited. Majewski et al. [7] conducted a review of studies analyzing the relationship between electricity and growth. They found that at least 19 studies support a causal relationship running from electricity to economic growth. In this line, there are also a few studies that use the neoclassical framework to analyze the effect of electricity on output. Among them, Soytas and Sari [56] analyzed the effect of electricity on industrial production in Turkey. Later, Kasperowicz [57] studied the effect of electricity consumption on economic growth in Poland. Shahbaz et al. [58] investigated the effects of electricity consumption, capital formation and financial development on economic growth in Portugal. Similarly, Fang and Chen [46] investigated the relationship between economic growth and various disaggregated alternative energy sources, including electricity, in China. Focusing on Latin American countries, Solarin and Ozturk [11] analyzed the impact of hydropower on growth in seven countries in the region, while Valencia-Herrera and Santillan-Salgado [10] did the same. However, the latter authors only conducted a causality analysis. In addition to these studies, to assess the impact of using clean energy to mitigate the increase in CO 2 emissions, there has been recent interest in determining the impact of renewable energy on economic growth (a review of these studies can be found in Ref. [59]). Furthermore, other authors have focused on determining the joint impact of renewable and non-renewable energy consumption on production growth. These studies make it possible to assess the impact of replacing polluting sources with non-polluting ones, that is, to evaluate the effects of the energy transition. One of the first studies to analyze the joint effect of both renewable and non-renewable energy consumption variables, using the neoclassical production function framework, is the study by Apergis and Payne [60]. The authors find positive elasticities for both energy variables. In the same vein, Gozgor et al. [61] and Rahman and Velayutham [5] found a positive effect of both types of energy. However, different effects were observed by Chen et al. [62], where a positive effect of non-renewable energy on economic growth was found, with the positive effect of renewable energy only observed above a threshold. Yasmeen et al. [17], found that the impact of conventional energy on output is stronger than that of clean energy in Pakistan. Similarly, Ivanovski et al. [18] found a positive effect of non-renewable energy and an insignificant effect of renewable energy on output in OECD countries. Some of the most recent studies are those by Mujtaba et al. [63], Espoir et al. [64], and Ali et al. [65]. Finally, focusing on Latin American countries, it is worth mentioning the study by Al-Mulali et al. [9], which concludes that the consumption of renewable electricity has driven economic growth more than the consumption of non-renewable electricity. 2.3. Electrification and economic growth Basically, the papers that jointly analyze the impact of renewable and non-renewable energy on growth, aim to determine the extent to which sustainable goals, related to the use of renewable energy, can affect economic growth. However, up to now, renewable energy use has been mainly linked to electricity consumption. Thus, electrification is part of the energy transition strategy. It is viewed as an opportunity to decarbonize the economy in many regions, including Latin America [66]. This fact has led to a growing interest in the study of the effect of electrification on the decrease of emissions. There is a rich literature in this area, including [67,68]. Nevertheless, despite the interest in knowing the impact of energy transition on economic production [69,70], this has not been accompanied by others analyzing the impact on economic growth. Instead, studies on electrification and growth have mainly focused on access to electricity and energy poverty (referring to less developed regions [71] and rural areas [72]), and on company productivity [73]. Few papers have analyzed the impact of electrification on production, with the understanding that the use of electricity could increase the technological progress of companies [42]. Moreover, only the study by Jung and Lee [16] focused on fuel substitution. The research results show that switching from fossil fuels to electricity could lead to a decrease in companies’ productivity in the short run, but an increase in the long run. In fact, studies focusing on how electrification and switching energy sources affect economic growth are scarce. To our knowledge, from a macroeconomic point of view, there are no studies analyzing the effect of electricity versus non-electricity sources on economic growth, integrating both energy sources in the same production function, as is done with renewable and non-renewable energy sources. Thus, this study will fill this gap. Several considerations can be drawn from this review. First, at least one stream of research believes energy affects economic growth, and its effect can be analyzed using neoclassical production functions. Second, not all energy sources affect production and economic growth in the same way. Third, there is interest in knowing how the energies, promoted contribute to sustainable growth, affect economic growth. Fourth, increasing the use of electricity over other types of final energy consumption may be advisable to achieve sustainable goals but, the effects of this substitution of energy sources on economic growth are not well studied and there is no prior evidence from a macroeconomic point of view. Fifth, there is little evidence on the impact of electricity on economic growth in Latin America and the Caribbean. Finally, no previous studies have analyzed the effect of the use of electricity and nonelectricity sources on production, at least in the previous context. Therefore, this paper examines whether the use of electricity and non-electricity productive energy have the same effect on output growth. For this purpose, an extended neoclassical production function is estimated. Electricity and non-electricity consumption are included as additional factors of production to physical capital, labor, and human capital. M.P. Pablo-Romero et al. Energy 305 (2024) 132281 4 3. The model Energy-augmented Cobb-Douglas production functions have been used to study the effect of energy on economic growth (see Table 1). According to Oryani et al. [74], this function presents certain advantages. It is simple to estimate the output elasticities by using the logarithmic form, which includes the effect of the technology used, and the estimated coefficients are close to factor productivity. This function may be expressed as follows: Y=AL α Kβ1Єγ[1] where Y is the output of the economy, A is the technology, K is the capital stock, L is the persons engaged and Є denotes the energy use. The elasticities of the capital stock, energy use, and persons engaged, are α , β1and γ. Two considerations can be drawn from the above expression. First, the impact of human capital on production is well known [75]. However, few studies have included human capital as an additional factor in the empirical estimation of energy-extended production functions (see Section 2.1 and Table 1). These studies conclude that both factors are relevant for production [19,46]. Therefore, the inclusion of human capital as an additional factor in the above expression is appropriate. Second, as detailed in the literature section (see Section 2.2), it is worth considering that not all types of energy contribute equally to economic growth. In this sense, for example, Azam et al. [76] and Luqman et al. [77] point out that nuclear energy does not have the same impact on output as renewable energy. Following this approach, other authors, such as Ivanovski et al. [18] and, recently, Espoir et al. [64], differentiate the impact of renewable and non-renewable energy on production. In these analyses, both energy variables are simultaneously included in a neoclassical production function. Following the methodology used in these analyses, and to analyze the different effect of electricity and non-electricity consumption on growth, both different types of final energy are included in Equation [1]. This differentiation of energy sources is better suited to the production decisions of companies than the inclusion of renewable and nonrenewable sources, since companies, cannot decide on the renewable or non-renewable origin of the energy they consume. In addition, the differentiation of energy sources proposed in this study could guide the policies, recently adopted by many countries, which tend to electrify the economy to use more renewable energies and thus reduce emissions. The production function is then expressed as: Y=AL α Kβ1Eβ2NEβ3Hδ[2] where E, NE and H represent productive electricity use, productive nonelectricity use and human capital, respectively. β2,β3 and δ indicate the output elasticities of productive electricity and productive nonelectricity consumption, and human capital, respectively. Assuming constant returns to scale within the capital stock, engaged persons and use of energies, given the human capital ( α +β1+β2+ β3=1), Equation [2] can be expressed in terms of output per engaged persons as: y=Akβ1eβ2neβ3Hδ[3] where y, k, e and ne represent GDP, capital stock, productive electricity consumption, and productive non-electricity consumption in per engaged term. 4. Problem Statement, data and econometric strategy 4.1. Problem statement This paper analyzes whether productive electricity consumption and productive non-electricity consumption (in per engaged terms) have different effects on production in per engaged term (GDPpc), in the LAC region. In addition, the study is extended to three subsamples: LAC countries with higher GDPpc, LAC countries with lower GDPpc and those with the lowest GDPpc values. To determine these impacts, the production elasticities of productive electricity and productive nonelectricity consumption are estimated and analyzed for each sample. To estimate these elasticities, the augmented Cobb-Douglas production function is linearized by taking natural logarithms on both sides of Equation [3]. Estimates of this log-linear form of the extended Cobb-Douglas production function give direct elasticities of variables. The equation to be estimated is expressed as: ln yit = α i+γt+β1lnkit +β2lneit +β3lnneit +δlnHit + ε it [4] where ln indicates natural log, i and t indicate country and year, α i indicates the individual effects, γ t indicates the time effects accounting for time varying omitted variables, and ε denotes the random error. The β2 and β3 estimated coefficients inform us of the productive electricity consumption and the productive non-electricity consumption effect on economic growth, which allows their analysis and comparison. 4.2. Data The data are from the Penn World Table version 10.01 (PWT 10.01) database [78] and the International Energy Agency [79]. The sample covers 17 LAC countries (Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, El Salvador, Guatemala, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru, and Uruguay), for the period 1990 to 2019. The three subsamples are: 1. Countries with higher GDPpc which are Argentina, Brazil, Chile, Colombia, Costa Rica, D. Republic, Mexico, Panama, and Uruguay; 2. Countries with lower Table 2 Descriptive statistics. Variables Mean Std. Dev. Min. Max. Observations GDP per engaged in logs. overall 10.125 0.472 9.251 11.123 N =510 between 0.468 9.398 10.796 n =17 within 0.127 9.689 10.467 T =30 Capital stock per engaged in logs overall 11.439 0.532 10.166 12.518 N =510 between 0.529 10.316 12.183 n =17 within 0. 38 11.011 12.123 T =30 Productive electricity use per engaged in logs overall 8.705 0. 699 6.959 10.199 N =510 between 0. 643 7.765 9.885 n =17 within 0. 314 7.172 9.404 T =30 Productive non-electricity use per engaged in logs overall 10.626 0. 491 9.634 11.513 N =510 between 0.483 9.904 11.376 n =17 within 0.145 10.220 11.141 T =30 Human capital index in logs overall 0.851 0.162 0.408 1.146 N =510 between 0.142 0.524 1.055 n =17 within 0.086 0.578 1.166 T =30 M.P. Pablo-Romero et al. Energy 305 (2024) 132281 5 GDPpc; and 3. The four countries with the lowest GDPpc, El Salvador, Guatemala, Honduras, and Nicaragua. The subdivisions have been made by considering where a country lies in relation to the GDPpc median spline. A country is in the first subsample when it is over the GDPpc median spline, while the second and third subsamples are below it. The variables used are real Gross Domestic Product (GDP), capital stock, human capital, engaged persons and two energy variables: the productive electricity consumption and the productive non-electricity consumption. The energy variables come from Ref. [79] and the rest of them from Ref. [78]. GDP and capital stock are expressed at constant 2017 national prices, in million 2017US$. Persons engaged are expressed in millions. Human capital is an index calculated from years of schooling and returns to education. These variables have been expressed in terms of logs. GDP and capital stock have been converted into per engaged terms, according to the proposed model explained below. As in Pablo-Romero and S´ anchez-Braza [19], to suitably assess the relationship between energy and production, productive energy use has been considered instead of total energy consumption. Productive energy is defined as the energy use in the productive processes. Therefore, this figure does not include domestic energy use. In this paper, the productive energy is determined as the sum of energy consumption, minus total residential energy consumption. Once the productive energy has been calculated, this variable is divided into productive electricity consumption (electricity consumption, minus residential electricity consumption) and productive non-electricity consumption (productive energy, minus productive electricity consumption). The variables are in tons of oil equivalent (toe) per engaged terms (in logs), according to the proposed model explained below. 4.3. Descriptive analysis The main descriptive statistics are displayed in Table 2. As the ‘within’ standard deviation is smaller than the ‘between’ standard deviation, for all variables, these deviations are higher across countries than across time. Fig. 1. Productive non-electricity and electricity consumption per persons employed: 1990–2019. Fig. 2. Scatter plots: Productive non-electricity and electricity consumption per persons employed and GDP per persons employed (GDPpc). M.P. Pablo-Romero et al. Energy 305 (2024) 132281 6 The lefthand graph in Fig. 1 presents the evolution of productive non-electricity consumption in the LAC countries. The median spline shows quite a constant trend for the region, with a slight growth in the latter years. Nevertheless, there are differences between the countries that should be noted. The most non-electricity consumption countries, by persons employed, are Brazil, Chile, Mexico, and Argentina. The less non-electricity consumption countries, by persons employed, are El Salvador, Honduras, Nicaragua, and Guatemala. In addition, it is also worth noting that Peru, Bolivia, and Uruguay meaningfully increase their non-electricity consumption over time. The rest of the countries (which do not appear in the graph) present values around the median spline line. The righthand graph in Fig. 1 presents the productive electricity consumption in the sample, over the analyzed period. All countries show a growing trend. Once again, Argentina, Brazil, Chile, and Mexico show the highest values, while Honduras, Guatemala, and Nicaragua show the lowest. Additionally, it is observed that Uruguay and Bolivia have noticeable growing trends, the latter being the country with the lowest productive electricity value in the region. It is also worth noting that electricity values are lower than non-electricity values in all countries. Finally, scatter plots of data, between productive non-electricity and electricity consumption per persons employed and GDPpc levels, are provided in Fig. 2. A smooth line (LOWESS) is added through the scatter plot to see the relationship between variables. In both cases, there is an initial positive relationship between the variables. 4.4. Econometric strategy Various econometric strategies have been used to study the impact of long-term impact of exogenous variables on GDPpc. Nevertheless, prior to estimating Equation [4], the slope heterogeneity and the cross-sectional dependence (CSD) of the variables have been studied. Subsequently, the order of integration was discovered. Regarding the heterogeneity of the potential slope, the Swamey’s test for slope homogeneity by Pesaran and Yamagata [80] was performed. The results obtained from counting for CSD, and for autocorrelation in the residual [81], shows that the model suffers from a heterogeneity problem. This indicates that slopes are heterogeneous, varying across countries (Table 3). To analyze the CSD, the weak test by Pesaran [82] was performed, finding strong cross-section dependence for all variables (Table 4). To avoid CSD issues, the unit root test (CIPS) by Pesaran [83] was performed in levels and first differences. The results indicate that all series are I(1) (Table 5). Next, panel cointegration tests were performed employing the Pedroni and Westerlund tests [84–86], as both tests consider CSD. The Pedroni test [84,85] was performed considering the demeaning process to mitigate the CSD impact. All statistical results reject the null hypothesis (Table 6). The Westerlund test [86] is an efficient cointegration estimator that addresses CSD. According to the statistical results shown in Table 6, the null hypothesis can be rejected. Considering the above results, and for the purposes of robustness, several econometric procedures have been carried out. Initially, the dynamic ordinary least square (DOLS) method was performed, which also addresses potential endogeneity. In this sense, previous reviews on the energy–growth nexus reveal that there is evidence supporting bidirectional, unidirectional and no causality, between GDP and energy use [4,87,88]. The DOLS method of estimation proposes a parametric correction for the effect of potential endogeneity of integrated regressors, in cointegrating time series and panel data regression models, through a parametric method and lags and leads. In addition, the generalized method of moments (GMM) has been performed to check for robustness. The GMM has been performed in the presence of heteroskedasticity and CSD. Lags of the capital stock, electricity and non-electricity energy variables and human capital are employed as instruments. Additionally, clustering has been used to account for CSD. Additionally, as recent literature indicates that the DOLS method may yield suboptimal results, as it cannot address the CSD appropriately [89], the CS-ARDL technique was also performed to calculate the long run effects between variables. This approach addresses autocorrelation, heteroscedasticity, endogeneity, heterogeneity and CSD, by performing the dynamic common correlated effects (CCE) estimator [89]. According to Kuo et al. [90], this approach has several advantages: it tackles the CSD by determining the logs and averages of all CSD units; and, Table 3 Slope heterogeneity test. Delta 10.421*** Adj. 11.953*** Note: *** indicates significance at the 1 % level with average bandwidth 3, Cross Sectional Averaged Variables: y k H e ne. Table 4 Cross-sectional dependence test. Variable CD-test lny 39.51*** lnk 16.58*** lne 52.96*** lnne 3.81*** lnH 61.70*** Note: *** indicates significance at the 1 % level. Table 5 CIPS unit root test. Variables Levels First differences Constant Constant and trend Constant Constant and trend lny −2.198* −2.612 −4.772*** −5.171*** lnk −1.614 −2.536 −4.438*** −4.751*** lne −1.981 −2.457 −4.765*** −5.088*** lnne −1.688 −2.234 −5.260*** −5.356*** lnH −1.142 −1.941 −2.268** −3.775*** Note: ***, **, or * indicate significance at the 1 %, 5 %, and 10 % levels, respectively. Lags criterion decision based on F joint test. Table 6 Pedroni and Westerlund test results. Dependent variable: lny Independent variables: lnk lne lnne lnH Pedroni Test H0: No cointegration. Ha: All panels are cointegrated. Modified Phillips–Perron t 2.5962*** Phillips–Perron t −2.3785*** Augmented Dickey–Fuller t −1.8118** Westerlund test H0: No cointegration. Ha: Some panels are cointegrated Variance ratio −2.0497** H0: No cointegration. Ha: All panels are cointegrated Variance ratio −2.3799*** Note: Test regression fitted with constant and trend. ***. ** indicates significance at the 1 % and 5 % level, respectively. M.P. Pablo-Romero et al. Energy 305 (2024) 132281 7 heterogeneous slopes are used to calculate dynamic CCE and can handle small sample sizes. The model is expressed as Δlnyit = α i+∑ p j=1 βijΔlnyi,t−j+∑ p j=1 β ´ ijIV,i,t−j+∑ 1 j=0 β ´ ijΔCSi,t−j+ ε it [5] where IV is used for the independent variables and CS for the average of cross sections for respective variables. Finally, as previous literature suggests that energy effects on growth tend to change with time [21], a non-parametric estimate was performed to show how productive elasticity, with respect to productive electricity and non-electricity consumption, has changed over time. The LLDVE by Li et al. (2011) was performed. This model is expressed as: ln yit =fi(t) + α i+β1(t)lnkit +β2(t)lneit +β3(t)lnneit +δ(t)lnHit +uit [6] where fi(t) = fi(t/T)across all i is the common trend function, α i represents the individual effects, βj(t)are the time-varying estimates and uit is stationary for each i. It is supposed that ∑N i=1ai=0, where N is the cross-section size. 5. Results and discussion 5.1. Parametric estimates: whole sample results Columns 2 and 3 in Table 7 show the estimated results of [4] by the DOLS and GMM approaches. The GMM was performed by using instrumental variables. Physical and human capital, and the energy variables have been considered endogenous. Up to 3 period delayed values of these variables have been used as instruments. The estimations have been performed accounting for the heteroscedasticity and CSD. The underidentification test by Kleibergen and Paap [91], the overidentification test of all instruments by Hansen [92] and the endogeneity test have been performed to evaluate this estimate (see the last rows in column 3 in Table 7). All estimate coefficients are significant and positive, while their values are plausible. Focusing on the energy coefficients, these results indicate that the effect of productive electricity and non-electricity consumption on GDP is different. In both estimates, the elasticity of the GDPpc, with respect to the non-electricity consumption, is higher than that with respect to electricity consumption. Additionally, the CS-ARDL method has been performed (Equation [5]). It accounts for heterogeneous slope coefficients by using the MG estimator. Table 8 shows the long-run estimate results. The CD statistics by Pesaran [82] are relatively low, indicating that the CSD has been taken out of the model correctly. It is worth noting that human capital has a non-significant effect on GDPpc. Therefore, Equation [5] has been re-estimated, eliminating the human capital variable. This elimination does not remarkably alter the energy coefficients. The estimate results, using CS-ARDL, are quite similar to those obtained before, showing the estimate’s robustness. These results show that both electricity and non-electricity consumption have a significant positive effect on GDPpc. Thus, as previously suggested by Sharma [93] and Pasten [94], energy use plays a fundamental role in the Latin American production growth. Therefore, at the regional level, energy conservation policies cannot be applied without this affecting production growth. These results are also coherent with studies that analyze the renewable and non-renewable energy use effects, on production growth in the region, which show positive and significant effects on production [9,95]. The CS-ARDL estimate results again show that there are notable differences in the coefficient values for the electricity and the nonelectricity variables. Additionally, it can also be detected that the difference between these values is now greater than that shown in Table 7, with the effect of electricity on productivity growth being relatively low. Therefore, on average, the productive non-electricity use has a greater effect on the per employment production than electricity consumption. Table 7 Estimates results by DOLS and GMM. Variables DOLS IV-GMM lnk 0.381*** 0. 400*** (0.038) (0.020) lnH 0.207** 0.153** (0.023) (0.080) lne 0.171*** 0.105*** (0.019) (0.018) lnne 0.243*** 0.171** (0.032) (0.018) Underidentification test Kleibergen-Paap-LM-Chi-sq(8) 81.438*** Overidentification test of all instruments Hansen J Chi-sq(7) 12.853 Endogeneity test Chi-sq(4) 27.858*** Note: Standard errors in brackets; ***, ** and * indicate significance at the 1 %, 5 % and 10 % level, respectively. Table 8 Long-run estimate results by CS-ARDL. Variables CS-ARDL CS-ARDL lnk 0.402*** 0.381*** (0.112) (0.097) lnH −0.058 (0.150) lne 0.098* 0.076** (0.058) (0.039) lnne 0.148** 0.150** (0.060) (0.048) CD Statistic −1.77* −1.49 Note: Standard errors in brackets; ***, ** and * indicate significance at the 1 %, 5 % and 10 % level, respectively. Table 9 Long-run estimate results for sub-samples by CS-ARDL. Variables CS-ARDL Higher GDPpc Sub-sample1 CS-ARDL Higher GDPpcy Sub-sample1 CS-ARDL Lower GDPpc Sub-sample2 CS-ARDL Lower GDPpc Sub-sample2 CS-ARDL Lowest GDPpc CS-ARDL Lowest GDPpc lnk 0.309*** 0.270** 0.384** 0.352** 0.480*** 0.393*** (0.116) (0.108) (0.202) (0.191) (0.207) (0.097) lnH 0.672 0.206 0.112 (0.855) (0.456) (0.180) lne 0.112* 0.079* 0.094** 0.077* 0.172** 0.184** (0.063) (0.060) (0.042) (0.049) (0.067) (0.073) lnne 0.163** 0.195*** 0.149*** 0.183*** 0.161*** 0.191** (0.051) (0.065) (0.053) (0.051) (0.061) (0.025) CD Statistic −1.36 −1.87* −1.80* −1.31 −1.78* −1.63 Note: Standard errors in brackets; ***, ** and * indicate significance at the 1 %, 5 % and 10 % level, respectively. M.P. Pablo-Romero et al. Energy 305 (2024) 132281 8 Therefore, in the region, the substitution of one source for another may not be neutral, at least if the productive structure remains unchanged. These results are in line with those obtained by Jung and Lee [16] for Korean companies in the short term, and with those obtained by Khan et al. [69], who observed negative effects of energy transition on production growth. Likewise, the results are in line with those shown in Refs. [17,18] for renewable and non-renewable energy effects. 5.2. Parametric estimates: subsample results Equation [5] has been re-estimated for two subsamples: for those with higher and lower GDPpc. Table 9 shows the results when the CS-ARDL method is used. As before, Equation [5] has been re-estimated, by omitting the human capital variables, as these are not significant. Additionally, Equation [5] has been re-estimated for the poorest countries, by only including the four poorest countries of subsample 2, which are also those who consume much less energy per persons employed. The results for the subsamples 1 and 2 are quite similar. Nevertheless, the estimated non-electricity coefficient values are slightly lower in subsample 2. All estimates show that the productive non-electricity consumption coefficient is higher than that for electricity. Therefore, the results are in consonance with those previously obtained for the whole sample. Regarding the estimated capital stock coefficients, those for subsample 1 are lower than those for subsample 2. Therefore, in countries with less GDPpc, the effect of capital stock on GDPpc is higher. Finally, focusing on those countries with the lowest GDPpc, the estimated results differ from the previous results. Now, the estimated productive electricity coefficients are still lower than the non-electricity ones, but their values are much closer than before. Therefore, there is evidence that, in the lower GDPpc areas of the region, the effect of greater electrification could be neutral, in the short term. The countries in this sample are the poorest in the region, which also have the lowest Fig. 3. Non-parametric LLDVE estimates of common trends and coefficients: full sample.Note: The solid red lines show the estimated coefficients of Equation [6] for the common trends and the coefficients, over time. Their confidence intervals are shown with gray shading. M.P. Pablo-Romero et al. Energy 305 (2024) 132281 9 Fig. 4. Non-parametric LLDVE estimates: subsamples.Note: The solid red lines show the estimated coefficients of Equation [6] for the common trends and the coefficients over time. Their confidence intervals are shown with gray shading. M.P. Pablo-Romero et al.