Does Self-Generating Energy Technology Adoption Increase Residential Electricity Consumption Inequality in an Emerging Country? Evidence From Brazil

I. Introduction

Initiatives promoting renewable energy technologies have expanded alternative energy sources in developing nations such as Brazil (Benevit et al., 2023). In Brazil, since the enactment of the Normative Resolution No. 482 in 2012, consumers have been afforded the option to generate and compensate for their level of electricity consumption. The proliferation of renewable generation technologies, alongside rising electricity tariffs, has prompted many consumers to adopt these technologies as a cost-saving measure and for the generation of clean energy.

However, inequalities in energy consumption may arise due to disparities in access to these advanced technologies and the financial ability to invest in them (Carley et al., 2021; Jenkins et al., 2016). In this context, this research aims to contribute to the literature by testing if adopting self-generating energy technology impacts (i) the population’s electricity consumption and (ii) the population’s inequality in electricity consumption in Brazil.

This is the first study to examine the relationship between self-generating energy technology adoption and Brazil’s unconditional distribution of residential electricity consumption. This research innovates by utilizing microdata from the Family Budget Survey (POF/2017-2018) conducted by Brazil’s main research institute, Instituto Brasileiro de Geografia e Estatística (Brazilian Institute of Geography and Statistics - IBGE). The POF sample weight allows inferences for the Brazilian population. We examine the underlying relationship using the Unconditional Quantile Treatment Effect estimator - UQTE (Firpo et al., 2009).

II. Data

The Family Budget Survey analyzes the expenditure structures, income, and part of the wealth variation of families, providing a profile of the population’s living conditions by observing household budgets.^[1] The construction of the variables follows Marques et al. (2024). The outcome variable is the electricity consumed per kilowatts per hour (kW/h). The treatment variable indicates whether the household has alternative energy sources. We created a variable based on the residential electricity tariffs defined by the National Electric Energy Agency (ANEEL) for each federation state to adjust the household’s sensitivity to residential power costs. To control the sensitivity of the home to income, we constructed a variable that considers the total monthly gross income of the family unit. Moreover, the value of the electricity tariff and the household income were deflated to January 15, 2018 values. Finally, we apply the natural logarithm to the three variables.

We performed the analysis for two different samples. We analyzed the full sample (General Sample) and the ordinary consumers sample (Ordinary Consumers). Table 1 presents the descriptive statistics. The study employs four groups of covariates. The first group aims to control the architectural characteristics of the household. The second group controls specific demand characteristics through the number of electronic devices in the home. The third group comprises binary variables aimed at maintaining the status of the property and its classification. The fourth group aims to control the characteristics of the household’s inhabitants.

Table 1 shows that high-intensity consumers tend to have a significantly higher income and more electronic devices. These factors can contribute to greater adoption of self-generating technologies due to the greater return on investment and financial capacity to bear the initial costs, as argued by the energy justice and energy inequality theory (Jenkins et al., 2016).

III. Method

The objective is to test whether adopting self-generating energy technologies impacts: (i) the population’s electricity consumption and (ii) the population’s inequality in electricity consumption. It is possible to model the demand for residential electricity as follows:

$$\begin{align} \ln C_{i s y}&=\alpha_0+\beta T_{i s y}+\alpha_1 \ln P_{i s y}\\ &\quad +\alpha_2 \ln I_{i s y}+\operatorname{Cov}_{i s y}^{\prime} \delta\\ &\quad +\mu_s+\theta_y+u_{i s y} \end{align} \tag{1}$$

The dependent variable $\ln C_{isy}$ is the natural logarithm of electricity consumption at the household level. The variable $T_{i}$ identifies those households that have adopted self-generating energy technologies. The treatment variable assumes the value 1 when using alternative energy sources (generator, solar, and wind board) and zero otherwise. The price and income elasticities of domestic demand for electricity associated with the variables ${lnP}_{isy}$ and ${lnI}_{isy}$ are, respectively, $\alpha_{1}$ and $\alpha_{2}$. The second is the natural logarithm of the residential electricity tariff, and the third is the natural logarithm of the real household income. The indicator $X_{ify}^{\prime}$ represents the vector of covariates. The parameter vector identifies the effects of covariates on electricity consumption. Finally, $u_{isy}$ represents the random perturbation. The terms $\mu_{s}$ and $\theta_{y}$ are the unobserved fixed effects (or fixed confounders). We assume that macroeconomic shocks in the electricity market that have some effect on electricity consumption are controlled by the fixed-year effect ($\theta_{y}$) and that the effects of seasonality and those eventual consumption characteristics and the substitutability of energy goods derived from family preferences due to differences in regional cultures are controlled by the state-fixed effect ($\mu_{s})$.

Firpo, Fortin, and Lemieux (2009) introduced a new type of regression, a particular case of recentered influence functions regression. According to Firpo and Pinto (2016), under the hypotheses of un-confoundedness and common support, treated groups and controls are comparable in terms of covariates $X_{i}$. The first hypothesis assumes that the vector of observable variables $X_{i}$ contains all the information about the potential outcome. The second hypothesis assumes that each household in the treatment group is comparable to those in the control group. Thus, the inequality treatment effects are identifiable from the data.

We applied the UQTE using the propensity score framework analysis for the mean, the quantile Q90, and the inequality measures: inter-quantile ratios Q90/Q01, Q90/05, Q95/Q10; Atkin inequality index, Gini inequality index, and Lorenz inequality index. All regressions fit the survey sampling plan.

IV. Results

Table 2 presents the balances of the covariates after the weighting by the propensity scores. All covariates are statistically comparable between the treated and control groups.

Table 3 presents the results. The first column shows how adopting self-generating energy technology affects the unconditional average. The second column shows the effect on the 90^th quantile. We present inter-quantile residential electricity consumption from columns (3) to (5). From column (6) to column (9), we present the results for the inequality measures of Gini, Absolute Gini, Atkinson, General Entropy, and Lorenz.

Panel A applies the analysis to the entire sample. In the first column, the results show that an increase in self-generating energy technology adoption by a percentage point does not affect the average unconditional consumption distribution. We corroborate the hypothesis that adopting self-generating renewable energy technologies does not affect the population’s electricity consumption. In the second column, the 1% increase in adoption generates an increase of approximately 0.34% in residential energy consumption at the 90th quantile. This result indicates an increase in consumption in the upper tail of the unconditional distribution. Columns (3) to (5) show an increase in the gap between the tails of the distribution of unconditional energy consumption. Finally, all the inequality measures show that adopting self-generating energy technologies increases the concentration of residential electricity consumption. Thus, we corroborate the hypothesis that adopting self-generating energy technologies in Brazil generated inequality in residential electricity consumption.

In Panel B, the one percentage point increase in the adoption of self-generating residential energy technologies generates an increase of approximately 0.14%. In economic terms, it is possible to argue that households with higher electricity consumption adopt modern self-generating energy technologies as a strategy to increase electricity consumption. On the other hand, adopting self-generating technologies for households with lower consumption may indicate replacing the electricity source by maintaining the same level of consumption.

In the second column of Panel B, the 1% increase in adoption generates an increase of approximately 0.28% in residential energy consumption at the 90th quantile. Columns (3) to (5) follow the same pattern of increasing the distance between the distribution’s tails presented in Panel A. All concentration measures generally follow the same pattern as in the previous panel.

V. Conclusion

This research pioneered the study of the impact of renewable energy technology adoption on the unconditional distribution of residential electricity consumption in Brazil. The study was innovative, using microdata from the 2017-2018 POF/IBGE Survey and methodologies that had not previously been used in the literature. The results provided reliable estimates of the adoption of self-generating energy technologies on the residential electricity demand of the Brazilian population. In strict terms, the results indicated that adopting self-generating energy technologies increased the average energy consumption of ordinary consumers. However, it was consumers with higher consumption that shifted the average. The results suggested that adopting energy transition technologies can increase the inequality of residential electricity consumption. Public managers must pay particular attention to the potential mechanisms for generating electricity inequality and income concentration within the current socio-economic context.

Funding

Funding support for this article was provided by the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) of Brazil (309836/2021-2).