Corresponding author: Natalya Ketenci ( nketenci@yeditepe.edu.tr ) © 2019 Non-profit partnership “Voprosy Ekonomiki”.
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits to copy and distribute the article for non-commercial purposes, provided that the article is not altered or modified and the original author and source are credited.
Citation:
Ketenci N (2018) The environmental Kuznets curve in the case of Russia. Russian Journal of Economics 4(3): 249-265. https://doi.org/10.3897/j.ruje.4.28482
|
This study explores the relationships between carbon emissions and their main determinants such as energy consumption, real income, international trade, level of education and level of urbanization in the Russian Federation, employing data for the period 1991–2016. Support for the environmental Kuznets curve hypothesis is found in this study, stating that environment pollution decreases in Russia after income achieves a certain threshold. The ARDL bounds test is employed in order to estimate short-run and long-run relationships in the estimated model. Energy consumption, real income, education and urbanization levels are found to be significant determinants of carbon emissions, while trade openness does not have an impact. The Granger causality test indicates two-way relationships between carbon emissions and energy use, real income and education. Only a single one-way causality runs from carbon emission to trade and no causality was found between carbon emissions and level of urbanization.
CO2 emissions, environment Kuznets curve, cointegration, Russia.
Russia is one of the largest contributors in the world to carbon dioxide (CO2) emissions after China, the US and India in total kilotons (kt). However, in terms of kilogram (kg) of CO2 emissions per 2010 dollars of GDP in 2014, Russia left the US far behind with 0.99 kg compared to 0.33 kg in the US. Contrary to the general tendency of decline in CO2 emissions among developed countries, CO2 emissions in Russia continue to rise. Since 1998, Russia has increased CO2 emissions by 14 percent, making an 8 percent overall increase since 2009. The changes are in favor of the environment, however, if GDP is taken into account as CO2 emissions per 2010 dollars of GDP declined by 45.6 percent in 2014 from 1998 levels and by 7.3 percent from 2009.
In 2015, the Russian Federation declared, as its target of intended nationally determined contribution (INDC), a decrease in the level of greenhouse gas (GHG) emissions by 25–30 percent from the 1990 level by 2030. However, this target is criticized in literature on the grounds that, in 2012, emissions in Russia were already 68 percent of the 1990 level (
New policies need to be designed in order to follow the global continuous declining trend. For example, since 1990, the environmental policy of the European Union (EU) has been focused on the commitment of its countries to decrease greenhouse emissions by 40 percent by 2030. In 2015, 195 countries signed the Paris Agreement, one of the important goals of which is to decrease world greenhouse emissions.
This study investigates factors affecting CO2 emissions in Russia that contribute to climate policies. At the same time, it examines the environmental Kuznets curve hypothesis in the context of that country. Factors of CO2 emissions in Russia have not been investigated empirically to date using an expanded framework. Therefore, the aim of this study is to conduct the first test of the environmental Kuznets curve in Russia by eliminating omitted variables bias, investigating the long-run relationships between CO2 emissions and variables that are responsible for changes, such as energy use, income per capita, trade openness, education and degree of urbanization. The next part reviews empirical findings in literature. The third part introduces the methodology applied in this paper and then empirical results are outlined in the fourth part, followed by a discussion in the conclusion on policy implications.
The problem of worsening environmental quality has been receiving increased attention in literature. Most of the empirical studies have been conducted within the framework of the environmental Kuznets curve (EKC). The EKC hypothesis assumes an inverted U-curve relationship between environmental degradation and income per capita. With an increase in income, pollution initially rises as well; however, after the economic growth reaches a certain threshold, levels of pollution start to decrease. Testing the EKC hypothesis, some studies explain changes in environmental quality purely by economic growth (
However, recently more studies have incorporated the factor of international trade (
In order to solve the problem of omitted variables, in literature, different determinants are incorporated in the EKC framework. For example, urbanization has been found to have a dual effect on the environment (
Some researchers argue that energy prices play an important role in the EKC (
To eliminate the possibility of omitted variable bias, researchers include in the EKC framework analysis such variables as inequality (
The list of major contributors to world CO2 emissions consists of developed as well as developing countries; therefore, numerous studies on the EKC hypothesis include developed (
In support of general findings,
Despite the interest in literature in the EKC for polluting countries, empirical studies for Russia are few in number. Among them are
There is no common conclusion in literature on the long-run relationships between CO2 emissions and income. There is also no expanded research on EKC for Russia. This study thus seeks to fill the gap in literature.
Taking into account the mixed results in literature on the environmental Kuznets curve to eliminate omitted variables bias, the relationships between CO2 emissions and economic growth following similar methodologies in previous studies of
ct = β 0 + β1et + β2yt + β3yt2 + β4trt + β5edt + ε1t (1)
ct = β 0 + β1et + β2yt + β3yt2 + β4trt + β5ut + ε2t (2)
ct = β 0 + β1et + β2yt + β3yt2 + β4trt + β5edt + β6ut + ε3t (3)
where ct is CO2 emissions per capita; et is commercial energy use per capita; yt represents per capita real income, which is measured as GDP per capita; yt2 is the square of per capita income; trt is trade openness and measured in terms of sum of export and import as a share of GDP; edt is the level of education measured in the number of secondary education pupils; and ut is the urban population as a share of total population.
The reason behind employing three alternative models, where education and urbanization are altered as explanation variables, is the robustness check to eliminate the omitted variables bias. It is discussed in literature that energy consumption is the key determinant in emissions changes; therefore, it is expected that β1 has a positive sign (
The methodology employed in this study allows for I(0), I(1) or fractionally integrated variables; however, it is not designed for I(2) integrated variables. Therefore, in order to check that variables are not integrated of order 2, four alternative unit root tests were conducted. These are the
Due to the relatively low span of data, the
where α1i, α2i, α3i, α4i, α5i, α6i represent the short-run parameters of estimated variables and α7i, α8i, α9i, α10i, α11i, α12i are the corresponding long-run parameters. The procedure is based on the joint F or Wald-statistics, where the null hypothesis of no co-integration in Eq. (4) is H0: α7 = α8 = α9 = α10 = α11 = α12 = 0, against the alternative hypothesis H1: α7 ≠ α8 ≠ α9 ≠ α10 ≠ α11 ≠ α12 ≠ 0. Critical values for the bound test are reported in
After the co-integration has been established, the ECM for Model 1 is estimated in the following form
where ECt–1 is the error correction term and λ1 is the parameter that measures the speed of variables convergence to the equilibrium, which has to be statistically significant with negative sign.
This study employs annual data of carbon emissions and their main determinants such as energy consumption, real income, international trade, level of education and level of urbanization in the Russian Federation for the period 1991–2016. The data are collected from the World Bank’s World Development Indicators database. Carbon emissions are presented by CO2 emissions per capita, metric tonnes; energy consumption is commercial energy use per capita, kg of oil equivalent; real income is represented by GDP per capita, constant 2010 US dollars; international trade is the sum of export and import as a share of GDP; level of education is the number of secondary education pupils; and level of urbanization is expressed by the urban population as a share of total population. All variables are used in natural logarithms.
Descriptive statistics of the sample are presented in Table
Descriptive statistics.
Mean | Max. | Min. | St. dev. | Obs. | |
c | 11.59 | 14.00 | 10.13 | 1.02 | 26 |
e | 4663.75 | 5861.16 | 3981.51 | 441.89 | 26 |
y | 8637.84 | 11615.70 | 5505.63 | 2216.03 | 26 |
tr | 55.25 | 110.58 | 26.26 | 14.31 | 26 |
ed | 12558496 | 15862637 | 9061324 | 2636478 | 26 |
u | 73.54 | 74.10 | 73.34 | 0.26 | 26 |
This study employs annual data for the period 1991-2016 in order to investigate relationships between CO2 emissions and variables that may determine its changes, energy consumption, real income, international trade and levels of education and urbanization in the Russian Federation. The ARDL technique employed in this study allows for variables that are integrated of order 0 or 1, I(0) or I(1); however, the procedure does not allow for integration of order above 1. Therefore, to examine the order of variables integration, four alternative unit root tests are employed in this study, the ADF, the DF-GLS, the PP and the KPSS. The results of the unit root analysis are presented in Table
Unit root tests.
ADF | DF-GLS | PP | KPSS | |
c | –3.47** | –3.55* | –3.47** | 0.29 |
e | –3.48** | –2.82* | –3.43** | 0.35 |
y | –3.08** | –2.14** | –3.09** | 0.28 |
tr | –4.22* | –3.07* | –14.31* | 0.33 |
ed | –4.83* | –3.02* | –1.24 | 0.19 |
u | 0.63 | –0.07 | 0.62 | 0.68** |
After detecting the order of variables integration, the ARDL co-integration approach is employed. The first stage is the selection of the optimum lag length of the unrestricted vector auto regression (VAR). The Akaike Information Criterion (AIC) and the Schwarz Bayes Criterion (SBC) imply that the optimal lag length is 2. The second stage of the analysis involves the investigation of log-run relationships between dependent and independent variables. A bound F-test is employed for equations (4), (6) and (7). The results of the test are reported in Table
Cointegration F test, F(c|e, y, y2, t, ed) .
F-statistics | 95% LB | 95% UB | 90% LB | 90% UB | 99% LB | 99% UB |
F(c|e, y, y2, t, ed) | ||||||
5.25 | 2.39 | 3.38 | 2.08 | 3 | 3.06 | 4.15 |
F(c|e, y, y2, t, u) | ||||||
7.69 | 2.39 | 3.38 | 2.08 | 3 | 3.06 | 4.15 |
F(c|e, y, y2, t, ed, u) | ||||||
5.08 | 2.27 | 3.28 | 1.99 | 2.94 | 2.88 | 3.99 |
After the long-run relationships were detected, the short-run and long-run coefficients were estimated. The short-run results and diagnostic test statistics are presented in Table
ARDL short run results.
F(c|e, y, y2, t, ed) | F(c|e, y, y2, t, u) | F(c|e, y, y2, t, ed, u) | |||||||||
Regressors | β | t-ratios | Regressors | β | t-ratios | Regressors | β | t-ratios | |||
∆e | 1.19 | 26.57* | ∆e | 1.06 | 22.63* | ∆e | 1.17 | 24.99* | |||
∆y | 0.01 | 6.03* | ∆y | 0.01 | 5.05* | ∆y | 0.01 | 5.77* | |||
∆y2 | –0.27 | –6.39* | ∆y2 | –0.23 | –5.45* | ∆y2 | –0.27 | –5.91* | |||
∆tr | –0.02 | –1.99 | ∆tr | 0.01 | 1.37 | ∆tr | –0.02 | –1.86 | |||
∆ed | 0.09 | 1.12 | ∆u | –3.11 | –1.62 | ∆ed | 0.09 | 1.17 | |||
∆u | –1.16 | –0.56 | |||||||||
ECMt–1 | –0.96 | –6.86* | ECMt–1 | –0.98 | –5.39* | ECMt–1 | –0.95 | –6.09* | |||
Diagnostic test statistics | |||||||||||
R 2 | 0.72 | R 2 | 0.92 | R 2 | 0.74 | ||||||
DW-statistic | 2.13 | DW-statistic | 2.43 | DW-statistic | 2.28 | ||||||
F-statistic | 4.72 | F-statistic | 14.42 | F-statistic | 4.63 | ||||||
RSS | 0.01 | RSS | 0.01 | RSS | 0.01 |
The long-run results of the ARDL estimations are presented in Table
ARDL long run results ARDL (1, 0, 0, 0, 0, 2).
F(c|e, y, y2, t, ed) — ARDL (1, 0, 0, 0, 0, 2) | F(c|e, y, y2, t, u) — ARDL (1, 0, 2, 2, 0, 0) | F(c|e, y, y2, t, ed, u) — ARDL (1, 0, 0, 0, 0, 2, 0) | ||||||||||
Regressors | β | t-ratios | Regressors | β | t-ratios | Regressors | β | t-ratios | ||||
e | 1.27 | 39.09* | e | 1.14 | 23.64* | e | 1.26 | 32.08* | ||||
y | 0.01 | 4.62* | y | 0.01 | 0.76 | y | 0.01 | 4.13* | ||||
y 2 | –0.28 | –5.97* | y 2 | –0.08 | –1.49 | y 2 | –0.27 | –4.79* | ||||
t | –0.02 | –1.01 | t | 0.02 | 0.97 | t | –0.02 | –0.98 | ||||
ed | 0.09 | 2.57** | u | –3.27 | –2.15** | ed | 0.09 | 2.23* | ||||
c | –5.18 | –6.34* | c | 8.08 | 1.14 | u | –0.84 | –0.41 | ||||
c | –1.59 | –0.18 | ||||||||||
Diagnostic test statistics | ||||||||||||
t-ratios | p-values | t-ratios | p-values | t-ratios | p-values | |||||||
χ 2 SC | 1.72 | 0.22 | 2.11 | 0.16 | 2.49 | 0.32 | ||||||
χ 2 FF | 0.09 | 0.92 | 0.04 | 0.96 | 0.24 | 0.81 | ||||||
χ 2 N | 3.78 | 0.15 | 0.09 | 0.95 | 3.56 | 0.16 | ||||||
χ 2 H | 0.84 | 0.59 | 1.29 | 0.32 | 0.80 | 0.62 |
The magnitude of the negative impact of real income squared is slightly higher when compared to the positive impact of real income on CO2 emissions, implying that the speed of environmental improvement is greater when an economy reaches a certain income threshold compared to the initial degradation speed. Education level has a positive effect on emissions, being significant only in the long run, suggesting that rising education level exposes the population to wider knowledge for the consumption of technology-intensive products and that possible environmental awareness, which depends on education level, does not takes place in Russia. On the other hand, the urbanization level in the Russian Federation has a strong negative impact on CO2 emissions, implying better organization for a clean environment in urban areas. Similar results have been found for India by
To examine the direction of the impact of series, the pairwise Granger causality test was conducted. The results of the causality test are presented in Table
Pairwise Granger causality test.
Null Hypothesis | F-statistics | Probability |
e does not Granger cause c | 6.695 | 0.006 |
c does not Granger cause e | 12.055 | 0.001 |
y does not Granger cause c | 4.224 | 0.030 |
c does not Granger cause y | 3.402 | 0.055 |
y 2 does not Granger cause c | 5.046 | 0.018 |
c does not Granger cause y2 | 5.796 | 0.011 |
t does not Granger cause c | 1.448 | 0.259 |
c does not Granger cause t | 15.052 | 0.001 |
ed does not Granger cause c | 4.236 | 0.030 |
c does not Granger cause ed | 4.766 | 0.021 |
u does not Granger cause c | 2.048 | 0.157 |
c does not Granger cause u | 0.367 | 0.697 |
The final stage of the ARDL approach is to estimate the stability of the model by employing the cumulative sum (CUSUM) and the cumulative sum of squares (CUSUMSQ) stability tests of
As the results of the CUSUM and the CUSUMSQ test may not always produce similar figures, to check the robustness of the results, the Chow forecast test of
Chow forecast test.
F-statistics | Likelihood ratio | |
Model 1 | 1.46 (0.26) | 12.62 (0.05) |
Model 2 | 0.48 (0.69) | 2.14 (0.54) |
Model 3 | 6.80 (0.01) | 56.11 (0.00) |
The main aim of this study is to test the EKC hypothesis in the Russian Federation for the period 1991–2016. In order to avoid the possible problem of omitted variables in addition to trade, education and urbanization, variables are introduced to the model. Three models are estimated in this study. The first model employs energy use per capita, real income per capita, square of real income, trade and education as independent variables. The second model employs the variable of urbanization instead of education. Finally, the third model employs the variables of both education and urbanization in addition to main independent variables. In all three models, carbon dioxide emission per capita is employed as a dependent variable, a measure of environmental quality. The results provide enough evidence to conclude that the EKC hypothesis is valid for Russia, indicating an increase in carbon dioxide emissions with economic growth; however, after income reaches a certain threshold, CO2 declines. In all three models, the error correction term is significant with an expected negative sign implying that deviations from long-run equilibrium are adjusted by estimated variables in about one year for all models. The results of Granger’s causality test demonstrate bidirectional Granger causality relationships between the pairs of CO2 and energy use, CO2 and real income, CO2 and the square of real income, and CO2 and education. A one-way causality runs through CO2 to trade and there is no causality between CO2 and urbanization. In order to measure the stability of model parameters, the CUSUM and CUSUMSQ techniques and the Chow forecast test were employed. The results provide evidence of parameter stability only for the second model, where the results of tests are inconclusive for the first model and reject the stability of parameters in the third model. Therefore, only the results of the second model may be employed for policy in Russia. From the results of the second model estimations, it can be concluded that economic growth has a short-run impact on CO2, energy use determines CO2 in both the short and long runs, while urbanization has significant high long-run impact on emissions.
Based on the results of the estimations, a number of policy implications can be derived for the Russian Federation. The results support the EKC hypothesis, which states that the relationships of CO2 emissions and economic growth have an inverted U-shape. Therefore, policy-makers should continue to implement policies to sustain economic development that leads to the use of cleaner technologies for lower carbon dioxide emissions. There is strong evidence that energy use has a destructive impact, while the level of urbanization has a refining effect on environmental quality. Therefore, policy-makers should focus on decreasing energy intensity, maintaining environmental policies in urban areas and implementing efficient environmental policies in rural areas to decrease the difference between urban and rural areas. Finally, education in Russia is found to have a damaging effect on environmental quality; therefore, new policies should be directed to increase environmental awareness among the population.