The empirical studies in this section are composed of a descriptive analysis of Asian countries’ EEKC, their econometric analysis and the case analysis of Uzbekistan.
This study assumes that the energy-use efficiency in each sample country can be represented by the heterogeneity of energy use at the same level of GDP per capita, which is corresponding to the difference in the locations of the EEKC in the descriptive analysis, and the country’s time-invariant fixed effect in the EEKC econometric estimation. There has been the discussion in general on whether energy efficiency is time “invariant” or time “variant” as in Zheng and Heshmati [
25] (This study defines energy use as the kilogram of oil equivalent per capita as in Aruga [
21]. Regarding the concepts of energy efficiency, there have still been a number of debates, for instance, on the choice of single factor efficiency or all-factor energy efficiency (see Zheng and Heshmati [
25])). This study supposes that the heterogeneity of energy use that is not explained on the EEKC trajectories originates from a country-specific time-invariant fixed effect as in, e.g., Akram et al. [
26] and Song and Yu [
27]. This section starts with the EEKC descriptive analysis.
3.1. Descriptive Analysis
Figure 1 displays the EEKCs of selected Asian countries for 1970–2015 (The sample period for the CA countries is 1992–2015). The EEKC is drawn with the vertical axis being the energy use expressed as the kilogram of oil equivalent per capita. The horizontal axis is the gross domestic product (GDP) per capita in terms of US dollars at constant prices in 2015. The data of the energy use is retrieved from the databases of the World Bank Open Data (See the website:
https://data.worldbank.org/accessed on 1 July 2021) and that of GDP per capita from the UNCTAD Stat (See the website:
https://unctadstat.unctad.org/EN/accessed on 1 July 2021). Here, 12 sample countries are selected as follows: Kazakhstan, Kyrgyz, Tajikistan, Turkmenistan, and Uzbekistan in the CA countries, and China, Indonesia, Japan, Korea, Malaysia, Singapore, and Thailand in the other Asian countries. The 12 sample countries here are chosen for easily visualizing the different trajectories of EEKC by removing the countries with similar trends of GDP per capita, while the subsequent econometric analysis targets 23 Asian countries (it will be explained later).
The important finding from
Figure 1 is that the locations of Kazakhstan, Turkmenistan, and Uzbekistan among the CA countries reveal higher positions than the trends commonly seen in the other Asian countries. This finding, together with the simple comparison in
Table A1, suggests that the three CA countries have experienced extraordinary energy uses at their levels of GDP per capita, thereby implying energy-use inefficiency.
The subsequent section examines the countries’ positions of energy use by the country-specific fixed-effect through an econometric approach and investigates the factors contributing to the difference in the fixed effects.
3.2. Econometric Analysis: Methodology and Data
This section focuses on the EEKC econometric analysis for Asian countries and starts with the description of methodology and data.
This study basically follows the original form of the EEKC presented by Suri and Chapman [
12]. Regarding the model specification, this study applies the standard nonlinear model where energy use per capita is explained by GDP per capita and its square, which was shown by Suri and Chapman [
12] as the simple version of their models, and by Aruga [
21] targeting Asia-Pacific countries. This study also adopts the fixed-effect model for panel-data estimation as in Suri and Chapman [
12]. From the statistical perspective, the Hausman test statistic is generally utilized to choose between a fixed-effect model and a random effect one [
28] as in Aruga [
21]. This study, however, emphasizes presenting a country-specific effect on energy use explicitly, and also a time-specific factor such as economic fluctuations due to external shocks such as the Asian financial crises in 1997–1998 and the global financial crises in 2008–2009. In addition, adopting the fixed-effect model contributes to alleviating the endogeneity problem by absorbing unobserved time-invariant heterogeneity among the sample countries. The EEKC analysis is classified into simultaneous equation modeling by Menegaki [
5], thereby requiring a prescription for endogeneity among variables. The fixed-effect model removes the omitted variable bias as a source of endogeneity in panel estimation.
As for the estimation technique, this study applies the ordinary least squares (OLS) estimator for the following Equations (1) and (2) and the Poisson pseudo-maximum likelihood (PPML) estimator for the following Equations (3) and (4) (This study uses EViews 12, as the software for the estimations). The reason for the additional use of the PPML estimator is that the data of energy use might be plagued by the heteroskedasticity problem, in which the OLS estimator leads to a bias and an inconsistency in its estimate. Thus, this study applies both estimators to ensure the robustness of their estimations, following the suggestions as in Santos Silva and Tenreyro [
29] and Head and Mayer [
30]. Suri and Chapman [
12] applied the general least squares (GLS) estimator for addressing the heteroskedasticity problem. This study, however, uses the PPML instead of the GLS considering the property of the nonlinear estimation.
Then, the equations for the estimation are specified as follows:
where the subscripts
i and
t denote sample countries and years, respectively;
eng represents the energy use expressed as the kilogram of oil equivalent per capita;
ypc shows GDP per capita in terms of US dollars at constant prices in 2015;
nrr denotes the natural resource rents (sum of oil, natural gas, and coal rents) expressed as a percentage of GDP;
gov represents the governance indicators;
fi and
ft show a time-invariant country-specific fixed effect and a country-invariant time-specific fixed effect, respectively;
ε denotes a residual error term; α
0…2 and β
0…4 represent estimated coefficients, respectively; ln shows a logarithm form, which is set to avoid scaling issues for the energy use (
eng) and GDP per capita (
ypc); and exp shows an exponential form.
The details of the variables and the sample size for the estimation are shown as follows. The data sources for the energy use (
eng) and GDP per capita (
ypc) are the same as those in
Section 2.1. The data of the natural resources rents (
nrr) are retrieved from the World Bank Open Data database. The governance indicators (
gov) are represented by World Governance Indicators (WGI) of the World Bank (For the data acquisition and their definitions, see the website:
https://info.worldbank.org/governance/wgi/accessed on 1 July 2021). This study, whose analytical concern is the energy policy performances, selects the following four indicators out of a total of six: effectiveness of government (
gve), regulatory quality (
rgq), rule of law (
rol), and control of corruption (
cor). Each index takes the number ranging from −2.5 (weak governance) to 2.5 (strong governance), with the world average being approximately zero. As for the sample size, the estimation targets 23 Asian countries: the 12 countries in
Section 3.1 and an additional 11 countries (Bangladesh, Brunei Darussalam, Cambodia, India, Mongolia, Myanmar, Nepal, Pakistan, the Philippines, Sri Lanka, and Vietnam). The sample period is 1970–2015 for Equations (1) and (3), and 1996–2015 for Equations (2) and (4) due to the data constraints of WGI. The study then constructs a set of panel data of the sample countries and periods. The study winsorizes the data of all the variables at the 0.01st and 99.9th percentile to remove the outliers. The descriptive statistics for the data of all the variables are displayed in
Table A2.
For the subsequent estimation, this study investigates the stationary property of the constructed panel data by employing panel unit root tests: the Levin, Lin, and Chu test (Levin et al. [
31]) as a common unit root test; and the Fisher ADF and Fisher PP tests (Maddala and Wu [
32]; Choi 2001 [
33]) as individual unit root tests. The common unit root test assumes that there is a common unit root process across cross-sections, and the individual unit root test allows for individual unit root processes that vary across cross-sections. These tests are conducted based on the null hypothesis that a level of panel data has a unit root, by including the trend and intercept in the test equations.
Table A3 shows that both of the common and individual unit root tests identify the rejection of the null hypothesis of a unit root at the conventional significance levels in all the variables. Therefore, this study uses the level of panel data for the estimation.
The notes on the specifications of the estimation models are described as follows: Equations (1) and (3) apply a fixed-effect model represented by
fi and
ft, respectively, for panel estimation. The estimation sets China as the benchmark country for country-specific effects because China is located in the middle position in the EEKC descriptive analysis in
Figure 1. The significantly positive coefficient of the country-specific effect would suggest that the country’s energy use is more inefficient than that of China. The ordinary hypothesis of EEKC postulating the inverted-U-shaped path between energy use and GDP per capita would be verified if α
1 > 0 and α
2 < 0 are significant (in Equations (2) and (4), β
1 > 0 and β
2 < 0 are significant).
Equations (2) and (4) replace the country-specific fixed effects above with possible contributors to the fixed effects. This study adopts the natural resource rents (
nrr) and governance indicators (
gov) as possible contributors. This is because the lower performance of energy policies has led to inefficient energy use, in particular in CA countries as shown in Mehta et al. [
22], Dyussembekova [
23], and Gomez et al. [
24], while natural resource abundance is supposed to give less incentive to save energy and use energy efficiently. The energy policy performance is represented by the aforementioned four governance indicators. In Equations (2) and (4), each governance indicator is separately inserted as an independent regressor since the indicators have a multicollinearity problem.
Table A4 reports the bivariate correlations and the variance inflation factors (VIF), a method of measuring the level of collinearity between the regressors. It shows a high bivariate correlation (around 0.9) in each combination and high values of VIF that are beyond (or close to) the criteria of collinearity, namely, ten points. The natural resource rents (
nrr) are supposed to have a positive effect on energy use. Each governance indicator is expected to equip a negative coefficient on energy use because better governance leads to more energy-use efficiency.
3.3. Econometric Analysis: Estimation Results
Table 1 reports the results of OLS and PPML estimations in the form of a log-link function. Column (i) displays the outcome of the fixed-effect model of each estimation, and Columns (ii), (iii), (iv), and (v) present the results of the alternative model containing the natural resource rents and the governance indicators instead of the fixed effects. The OLS and PPML estimations show similar results in the sign and significance of each coefficient. Thus, the subsequent description focuses on the result of PPML estimation that adjusts the heteroskedasticity. The findings from the estimation results are summarized as follows.
First, the EEKC hypothesis, which assumes the inverted-U-shaped relationship between energy use and GDP per capita is confirmed in all the estimations from Columns (i) to (v) because the coefficients of GDP per capita are significantly positive and those of its square are significantly negative. The turning points are, however, far beyond the reasonable range of GDP per capita (The turning point is computed by—α
1/2α
2, or—β
1/2β
2. Using the estimated coefficients in
Table 1, the turning points would be beyond 1 million US dollars as GDP per capita). It might come from the observation in
Section 2.1 that most sample countries stay at the increasing trends of their EEKC. This finding leads the research to focus on the locations of the EEKC trajectories rather than the EEKC shapes.
Second, focusing on the fixed-effect model in Column (i), the coefficients of the country-specific dummies are significantly positive in Turkmenistan, Uzbekistan, and Kazakhstan in the CA countries (in common with the OLS estimation), and insignificant or significantly negative in the other sample countries. This means that the energy uses of the three CA countries are inefficient due to their country-specific factors as compared to China, the benchmark country, and this result is consistent with the descriptive analysis in
Section 3.1. The degree of the energy-use inefficiency is shown by the magnitude of the coefficient of the country-specific dummy: exp.(0.926) = 2.524 in Turkmenistan (the energy use of Turkmenistan is 2.524 times larger than that of China), exp.(0.735) = 2.085 in Uzbekistan, and exp.(0.424) = 1.528 in Kazakhstan.
Third, turning to the alternative model containing the natural resource rents and the governance indicators in Columns (ii)–(v), the coefficients of the natural resource rents (nrr) are significantly positive as expected in all the cases. As for the governance indicators, all the coefficients are significantly negative as supposed: the regulatory quality (rgq), the rule of law (rol), the control of corruption (cor) show robust significance at the 99% level (the rule of law is insignificant in the OLS estimation, though), while the government effectiveness (gve) indicate weak significance at the 90% level. This result suggests that energy use is highly correlated with the natural resource abundance and is more importantly affected by policy governance, such as the regulatory quality and the rule of law. The joint estimation outcomes of the country-specific fixed effect and the policy governance effect on energy use lead to the question of the degree of contribution of the policy-governance factors to the country-specific energy use inefficiencies in the CA countries.
The most critical information extracted from
Table 1 is the identification of the energy-use inefficiency with its magnitude in Turkmenistan, Uzbekistan, and Kazakhstan as their country-specific effects in the EEKC framework, and of the statistical linkage of policy governance with energy-use efficiency in the alternative EEKC model.
3.4. Econometric Analysis: Factor Compositions in Energy-Use Inefficiencies
The final step is to examine the contributions of the abundance of natural resources and policy governance to the country-specific energy-use efficiencies in the CA countries (here, also based on the PPML estimation).
Table A5 and
Figure 2 focus only on the three countries (Turkmenistan, Uzbekistan, and Kazakhstan), since these countries only have significantly positive country-specific fixed-effects representing their energy use inefficiencies in the PPML estimation.
Table A5 shows the fixed effects and contributors in the three countries, focusing on two governance indicators: the regulatory quality and the rule of law. Column (a) shows the coefficients of the country-specific fixed-effect dummies; Column (b) presents the period-average natural resource rents (
nrr); Column (c) computes the
nrr deviations from China (the benchmark country); Column (d) obtains the
nrr contributions by multiplying the
nrr deviations by the estimated
nrr coefficient in
Table 1; Column (e) presents the period-average governance indicators (the regulatory quality,
rgq and the rule of law,
rol); Column (f) computes the deviations of
rgq and
rol from China; Column (g) obtains the contributions of
rgq and
rol by multiplying their deviations by the estimated coefficients in
Table 1, respectively; and Column (h) shows the total contributions by summing up each of Columns (d) and (g). In
Figure 2, the line displays the country-specific fixed-effects while the bar graphs indicate the contributions of the natural resource rents and the governance indicators in the three CA countries.
The analytical results from
Table A5 and
Figure 2 are summarized as follows. First, the contributions of the natural resource rents (
nrr) to the country fixed effects account for 37.6–89.5% of the country fixed effects in all the cases. Second, the regulatory quality has contribution rates to the country fixed effects by 46.0% in Uzbekistan, 43.5% in Turkmenistan, and 5.7% in Kazakhstan. Third, the rule of law has contribution rates of 23.3% in Uzbekistan, 22.5% in Turkmenistan, and 17.8% in Kazakhstan. The results suggest that the lack of policy governance is an influential factor that explains energy-use inefficiency in Uzbekistan and Turkmenistan. This outcome is consistent with the argument which points out the lower performance of energy policies in Mehta et al. [
22], Dyussembekova [
23], and Gomez et al. [
24]. Thus, there would be enough room for Uzbekistan and Turkmenistan to improve their energy-use efficiencies by enhancing their performance of energy policies. From the global perspective, the CA countries are considered to be some of the most energy-consuming entities with the greatest loss of energy resources in the world as Kaliakparova et al. [
3] argued, while the SDGs require a doubling of the global rate of improvement in energy efficiency by 2030. Therefore, this study’s finding demonstrates the significance of enhancing the CA countries’ energy policy governance from the world-wide context as well as the regional aspect.
3.5. The Case of Uzbekistan
It was in Uzbekistan that the policy governance mattered to the largest degree in explaining its energy-use inefficiency as shown in the previous subsection. Thus, this subsection picked up the case of Uzbekistan as a sample for describing the detailed energy situation and policies.
The growth of population and the development of the economy in Uzbekistan have induced an increasing demand for energy. The population in Uzbekistan has risen from 20.8 million in 1991 to 33.7 million in 2020. Regarding the economy, the agricultural and industrial sectors are considered the dominant energy users. Thus, Uzbekistan’s energy sector has played a significant role; it accounts for 7% of GDP and 72% of the government investment program. Moreover, primary energy demand in Uzbekistan is forecast to increase with an annual growth rate of 1.7% by 2025 (Yuldasheva et al. [
34]). Although there are several mines from which gas-oil, coal, and uranium are extracted to produce energy, there is a shortage of energy generation and transmission. Uzbekistan has even resumed imports of electricity from neighboring Central Asian countries, such as Kazakhstan, in 2019, and Turkmenistan, in February 2021, to meet the rising energy demands, especially during the peak load of the winter period.
The question of the factors that caused the shortage of energy generation is crucial in Uzbekistan. First, the technology and management utilized in energy production are outdated and inefficient (e.g., Gomez et al. [
6]). Although the energy use intensity in Uzbekistan has decreased by approximately 45% in the last 15 years, it remains 35% higher than that of Kazakhstan and three times higher than that of Germany (See the website:
https://www.worldbank.org/en/news/press-release/2018/01/30/industrial-enterprises-to-become-more-energy-efficient-reducing-overall-energy-consumption-in-uzbekistan accessed on 1 July 2021). This is because most of the main energy generators located in a thermal power station that generates 56.5 million kilowatts of electricity, were built fifty years ago and are less productive under an inefficient management system. At the same time, the country’s industrial sector, which largely utilizes inefficient and obsolete technology in its production processes, accounts for about 40% of total energy consumption. The government is now on its way to modernizing the energy generation system with the installation of high-end technologies. Second, Uzbekistan still depends on traditional energy resources such as natural gas. Looking at the share of energy sources of electricity generation in 2018, traditional energies such as natural gas account for 90%, whereas hydropower as a renewable energy source accounts for only 10% (See the website:
https://www.iea.org/reports/uzbekistan-energy-profile accessed on 1 July 2021). In Uzbekistan, more than 300 days in a year are sunny, and there is considerable room to develop the harvesting of renewable energy resources. In fact, the solar energy potential is almost four times the country’s primary energy consumption. Thus, potential solar energy is enough to meet the rising energy demands and a wide range of industrial purposes (See the website:
https://www.iea.org/reports/uzbekistan-energy-profile/sustainable-development accessed on 1 July 2021). Its production is expected to be cost-saving and sustainable, thereby contributing to energy-use intensity in the country.