Does the environmental Kuznets curve exist?
A panel regression analysisAuthor: Ruben Greven Student number: 10639691 Supervisor: Egle Jakucionyte Date: 03-02-2017
1
Abstract
Ever since the early 90’s, researchers have modelled the relation between income and environmental degradation within the context of the environmental Kuznets curve (EKC) hypothesis. The EKC hypothesis describes an inverted U-shape relation. At early stages of income, environmental degradation increases as income increases. However, at some point environmental degradation starts to decrease and environmental improvements are achieved. This thesis attempts to find empirical evidence for the EKC, using CO2 emissions as indicator. A fixed effects regression is employed using three panels containing approximately 180 countries and a period ranging from 1960 to 2013. Different compositions of countries and years included in the regression are tested. The results support the EKC hypothesis, as turning points are found within the observed data sample. However, the lack of observations beyond the turning points results in much uncertainty about the effects of increases in income on CO2 emissions beyond the turning point.
Statement of Originality
This document is written by Student Ruben Greven who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
2
Introduction
The relation between economic growth and environmental degradation has been a widely discussed topic ever since the Club of Rome presented their research Limits to Growth in 1972. The belief that economic growth and environmental well-being were mutually exclusive became an established belief, which consequently led to much public concern, as it appeared that a collapse of our global biophysical systems might have been inevitable. This rather negative outlook would eventually be somewhat tuned down, when in 1987 the Brundtland Commission published Our Common Future (WCED, 1987). The Brundtland Commission was created by the UN General Assembly and was designated to examine the relation between the global environment and development. It concluded that development did not necessarily result in environmental damage. In fact, it could help mitigate the damage through poverty reduction, as it would increase demand for environmental protection. This finding was conceptualized by the environmental Kuznets curve (EKC) hypothesis. The EKC hypothesis proposes an inverted U-shape relation between GDP per capita and environmental degradation. At early stages of development, when GDP per capita is low, there is little economic activity and thus a low impact on the environment. As a country is pursuing economic growth, more economic activity results in a larger impact on the environment. At some point however, this trend reverses and further economic development results in environmental improvements (Barbier, 1997). The Environmental Kuznets curve (EKC), the curve that plots this relation, is named after the Kuznets curve due to its resemblance. If the relation between GDP per capita and environmental impact is correctly described by the EKC, than economic growth can be considered both the cause and solution to environmental degradation. Consequently, it can be argued that reducing economic activity will not necessarily be the path towards environmental improvements, as is sometimes claimed. The existence of the EKC is of importance to both countries who seek to grow economically without compromising their environmental endowments and countries who already achieved high levels of economic development and now seek to reverse previous damage to the environment.
This study attempts to find empirical evidence for an EKC, using more recent data in the form of a panel containing approximately 180 countries and a period ranging from 1960 to 2013. In contrast to many other studies, this study focusses on only one indicator for environmental degradation and uses three different measures for GPD per capita. For further investigation, the countries are divided into four income groups, which are tested individually. To study the effects of recent global efforts to reduce environmental degradation, the period 1992-2013 is also tested separately. In accordance with recent studies, the issue of non-stationarity of the data is considered as well. The following section will present a literature review, in which the empirical evidence and the theoretical framework is discussed. Next, the methodology and data are discussed. Afterwards, the results are presented. The final section contains the conclusion.
Literature review
This section reviews the literature on the environmental Kuznets curves (EKC) hypothesis in two parts. First, the empirical evidence for and against the EKC will be discussed, the main results found in these studies is whether an inverted U-shape relation is found and at what income levels the turning points are. Next, the theoretical foundation for the EKC is briefly discussed.
3 Empirical evidence
A wide range of indicators for environmental well-being have been employed to test the EKC hypothesis. Most papers focus on the pollution of air and water bodies, but some studies have also used deforestation, energy consumption, biodiversity conservation, production input as well as other indicators for environmental degradation.
The paper by Grossman and Krueger (1991) on the effects of trade liberalization as brought about by NAFTA on pollution in Mexico is often considered the first study of the EKC literature. In contrast to much of the earlier work on the relationship between economic development and environmental degradation, their findings suggested that at high income levels economic development would result in lower pollution, in accordance with what would later become known as the EKC. They used data from the Global Environmental Monitoring System, one of the first databases to collect data on various environmental indicators. They used measurements of SO2, suspended dark matter and the density of suspended particles for various urban areas. For SO2 and suspended dark matter they found turning points somewhere between $4,000 and $5,000 (1985 US dollars). The density of suspended particles appeared to decrease monotonically.
In later research, Grossman and Krueger (1995) extend on their initial research by testing for additional indicators, namely urban air pollution, the state of the oxygen regime in river basins, faecal contamination of river basins and contamination of river basins by several heavy metals. They found that turning points for most pollutants are less than $8,000 (1985 US dollars). Only lead contamination exceeded this amount, instead a turning point was found at $10,500.
Using the same database as Grossman and Krueger, Selden and Song (1994) improved the econometric model and still found support for an EKC. Besides the two air pollutants used by Grossman and Krueger, their research also includes oxides of nitrogen and carbon monoxide. They forecast that for most pollutants a turning point is reached in the second half of the 21st century, considerably later than Grossman and Krueger (1991) suggests. They also hypothesize that urban areas have lower turning points for pollutants, because urban areas have relatively more political representation and reducing pollution in cities might be more cost-effective. Therefore, they include population density as a variable. The coefficient for population density typically has the right sign, but is not always statistically different from zero.
As part of the background study for the 1992 World Development Report Shafik and Bandyopadhyay (1992) estimated an EKC for 10 environmental indicators. The indicators are: lack of clean water, lack of urban sanitation, ambient levels of suspended particulate matter, ambient sulphur oxides, change in forest area during 86, the annual rate of deforestation during 1961-86, dissolved oxygen in rivers, faecal coliforms in rivers, municipal waste per capita, and carbon emissions per capita. They tested three models, a log-linear, log-quadratic and log-cubic, which included a time trend. The two ambient air pollutants appear to be in line with the EKC and have a turning point between $3,000 and $4,000. Lack of clean water and urban sanitation declined at increasing income levels, when income is still low. For most indicators they find that at ‘medium’ levels of income, the indicators show improvements. The exemptions are dissolved oxygen in rivers, municipal waste and carbon dioxide. They also estimate the effects of various policy variables, such as trade, distortions or debt. They argue these policy variables might result in higher levels of pollution. Countries more open for trade might lower environmental standards in order to remain competitive. Energy subsidies result in excessive energy consumption and the associated pollution.
4
Poorer countries that have considerable debt, may resort to extracting their natural resource base more rapidly than is efficient. This would lower their short term debt burden. However, they found that the results were inconclusive, specific policies seem to only have a significant effect on specific pollutants and the results can therefore not be generalized.
Instead of focussing on a set of indicators, Holtz-Eakin and Selden (1995) only examine the relationship between economic development and carbon dioxide emissions. They employ a quadratic model with fixed country and year effects. Their unbalanced panel of data ranges from 1951 to 1986 for 130 countries. The results from the linear model are in line with the EKC, with a turning point at $35,428. However, this point was outside the range of the observed income, compromising this result to some extent. Using their results, they forecasted CO2 emissions and found that even though an EKC might exist, emissions will continue to increase for the foreseeable future. This is due to the global distribution of wealth and population growth. They argue that low income countries will have the most population and income growth, resulting in an increasing level of global CO2 emissions.
By the mid- 1990s there appeared to be enough consistent findings that for many pollutants it could be concluded that an EKC exists (Yandle et al., 2004). Although most studies remain cautious with respect to interpretation and policy implications, these results seemed to suggest that economic development was indeed the key to overcoming environmental issues. The response to these early studies have been two-fold, efforts to replicate and extent initial findings and a serious discussion of the data, the estimation methods and the extent to which the EKC can be generalized.
With respect to the models and estimation methods used, Stern, Common and Barbier (1996) critically review some of the early studies on the EKC. They identify three problems, namely: simultaneity, data problems and international trade (the issue of international trade will be discussed at a later point). They point out that the models used in previous studies do not allow a feedback from the state of the environment to economic growth, while this feedback does exist. It is therefore inappropriate to estimate a single equation model assuming unidirectional causality from economy to environment. They claim that the data used in early studies contain many missing values, are often of poor quality, are not appropriate as measures for environmental quality and are likely to give rise to heteroskedasticity problems. Taken together they argue that using OLS, which early studies did, results in inconsistent and biased estimates. Finally, they conclude that these early studies are only useful as a descriptive statistic and not for projecting future trends or policy recommendations
In response to Stern, Common and Barbier’s review of the estimation methods typically used, Cole, Rayner and Bates (1997) attempted to overcome several of the weaknesses. They employed fixed effects linear-linear and log-log models for fifteen environmental indicators, which they distinguish between indicators with a direct local environmental impact and indicators with no direct local environmental impact. General least squares (GLS) was utilized to correct for heteroskedasticity and autocorrelation. Income variables were tested for exogeneity and the results indicated that simultaneity was not present. Furthermore, they also calculated standard errors of the turning point to indicate the accuracy of the turning points. For indicators with a direct environmental impact, seven out of eight indicators had turning points at levels of income within the observed range of income, ranging from $5,700 to $18,000, in 1985 US dollars. Turning points for indicators with a direct local environmental impact seemed to be much more robust than indicators with no direct local environmental impact, such as CO2 emissions and total energy use. For most of the indicators with no direct local environmental impact the turning points were well outside the observed income range. For CO2 emissions, the log and linear models estimated turning points at $62,700 and $25,100,
5
in 1985 US dollars, respectively. Only for municipal waste and methane no turning points were estimated. They also included trade intensity as a variable, which has a negative coefficient for all indicators. However, the variable was not statistically significant for any of the indicators. Possibly because the effects were captured by the fixed effects.
More recently, Kaika and Zervas (2013) have collected and summarized studies that empirically tested the EKC for CO2. Most studies use panel data, which differ in terms of countries and years included. The latest year to be included in the panels is 2005. Time-series for individual countries are occasionally employed as well. Although, some studies find an EKC for CO2, mostly for developed countries, the majority of studies find that CO2 emissions rise monotonically with income and therefore do not support the EKC hypothesis. In line with Cole, Rayner and Bates (1997), Kaika and Zervas argue that CO2 does not have a local or direct environmental impact, which could explain the empirical results. The fact that CO2 emissions are strongly related to energy use, which is essential for economic growth, is also offered as an explanation. Among the studies that did find an EKC for CO2 is Dutt (2009). Using robust OLS models and fixed effects models, Turning points were consistently found around 29,000 (constant 2000 international $).
A short essay by Arrow et al. (1995) presents a further argument for caution when interpreting the results of the previously mentioned studies. Taking an ecological approach towards the relation between environmental well-being and economic development, they argued that even though an EKC might exist for a set of specific pollutants, this does not imply a similar relation for environmental quality in general. Their essay reveals a disconnection between measures of pollution, used as indicators by aforementioned studies, and measures of actual environmental quality. They point out that all economic activities ultimately depend on the natural resource base, which includes services provided by ecosystems. If in early stages of economic development resources are depleted and ecosystems are irreversibly damaged, further economic development is no longer supported by natural resources and ecosystem services. A turning point, after which environmental improvements occur, may never be reached. It can therefore not be concluded that economic growth is sufficient for long term environmental well-being.
A final econometric critique is the issue of spurious regression and cointegration (Stern, 2004). The time-series in the panel data may exhibit a stochastic trend which may give a misleading result about the relation between income and environmental well-being. If the variables share a common stochastic trend, they are said to be cointegrated. Different methods are needed to regress cointegrated variables. Like most of the EKC literature in general, the studies that accounted for these concerns have produced mixed results. Perman and Stern (2003) concluded that the EKC does not exist for sulphur emissions. Nayaran and Nayaran (2010) found that for only 15 of 43 developing countries an EKC for CO2 existed. Jaunky (2011) conducted a similar study for 36 ‘high’ income countries. An EKC was only found in five countries. It seems more likely that CO2 emissions stabilize over time, rather than decline. Among the countries that did appear to have an EKC for CO2 were China, Pakistan, India and Malaysia (Jalil and Mahmud, 2009; Shahbaz, Lean and Shabbir, 2012; Tawari, Shahbaz and Hye, 2013; Saboori, Sulaiman and Mohd, 2012).
Some of the studies also find that cubic specifications, which result in a N-curve, are significant. However, this might be the result of the stabilization of pollution, once low pollution is achieved, and the inability of the quadratic specification to account for this, rather than an increase in environmental degradation at ‘very high’ income levels (Carson, 2010). Furthermore, the theoretical explanation for a cubic relation is lacking. Another peculiarity is that early papers either
6
exclude additional variables or find that those variables are often statistically insignificant. More recent paper do find statistically significant variables other than income. However, the included additional explanatory variables are likely to be subject to omitted variable bias and it is not clear what we can infer from them (Stern, 2004). In conclusion, it appears that there is sufficient evidence in support of an EKC for indicators that have a local and direct impact on people’s lives. For indicators that have a global and indirect impact on the environment, the results seem at best mixed. Additionally, nearly all studies emphasize that the process of environmental improvements is not automatic, but rather is the result of government policies, social institutions and well-functioning markets.
Theoretical framework
In an effort to explain the EKC, it was recognized that environmental quality is the result of the interaction between emissions and abatement, i.e. reductions in emissions. Next, three determinants of how this interaction is affected by economic development, through the income variable, were identified as: the scale and composition of economic activity as well as the effect of income on the demand and supply of pollution abatement effort (Panayotou, 1997). All else equal, the scale of economic activity positively effects environmental degradation, because a larger scale is associated with increased resource use and waste generation. However, if emissions increase at a similar rate as the population, only total concentrations are affected, whereas emissions per capita are mostly unaffected. The composition of economic activity determines the intensity of resource use and waste generation. At early stages of development, production occurs mostly in the primary sector, generating little waste. At later stages, countries develop a secondary sector comprising mostly of highly polluting industries. Moving towards a service economy is considered the final stage of development and is associated with low levels of pollution. As income increases monotonically through these stages, an inverted U-shaped relation can be found between income and environmental degradation. Contrary to the scale of economic activity, increased income is expected to have a negative effect on the demand for environmental degradation, and thus a positive effect on demand for abatement. At low income levels, people prioritize food and other material needs over environmental well-being. Once a higher level of income is reached and the initial needs are satisfied, people become more concerned with the quality of the environment. Furthermore, higher income levels increase the amount of funds available for the supply of abatement. These factors allow for a decline in pollution.
In addition, Torras and Boyce (1998) point out that at later stages of development, the demand for environmental well-being might be translated into government policy more effectively. They also discuss the effects of technological developments as a reason for environmental improvements. Rising resource costs and consumer preferences can persuade firms to invest in cleaner technologies. However, they suspect that government policies have been most effective in spurring pollution-reducing technological change.
Economists have also attempted to derive theoretical models from these explanations. Using a neoclassical model with more or less standard assumptions that are conventionally used in general equilibrium and growth analyses, which include an aggregator function of capital (K), labour (L) and technology (t), f(K, L, t) as well as internalization of social marginal cost of pollution by firms, Lopez
7
(1994) derived a condition which can describe an inverted U-shape relation between income and pollution. The effect of growth (factor expansion) on the environment depends on two parameters, the elasticity of substitution in production between pollution and non-pollution inputs, and the relative degree at which marginal utility from income decreases. A decrease in environmental degradation is likely to occur when both are high. The elasticity of substitution in production depends to a large extent on the rate of innovation and the ability of technology to allow for the substitution of polluting inputs with non-polluting inputs. The relative degree at which marginal utility decreases is expected to be higher when income is high, and thus is the main driver of the EKC.
Stokey (1998) and Andreoni and Levinson (2001) have developed slightly different models, in which the EKC is driven by technology instead of changing preferences. However, Pasten and Figueroa (2012) show that all three models are special cases of a general model and that changes in both parameters can cause a turning point in pollution, provided that the other parameter is sufficiently high. As income grows, willingness to pay for a better environment increases and pollution becomes expensive. Profit-maximizing firms will change polluting technologies by clean technologies and, consequently, pollution decreases. Or, as income and technological capabilities increase, the elasticity of substitution between polluting and non-polluting increases. If, at some point, people are also willing to pay a positive price for the environment, firms will find it cheaper to use environmentally friendly technologies, which will result in reduced pollution. However, they point out that it seems more consistent to assume that changing preferences are themselves a driver of technological innovation. Changing consumer preferences can affect the state of the environment in multiple ways, and thus create the downward slope of the EKC. They can signal their desire for environmental improvements by purchasing products with a positive or reduced negative effect on the environment. Conditional on the market being efficient and well-defined property rights, firms with a harmful impact on the environment or inefficient production methods will lose market share. Firms that develop technologies to increase efficiency, and reduce their environmental impact, can reduce costs or appeal to consumers directly, with their more environmentally friendly productions methods and as result remain competitive. Another channel through which consumers can affect the quality of the environment, is to create or increase political support for government policies, which in turn can improve the environment directly or induce technological innovation (Torras and Boyce, 1998). Finally, consumers can also alter their behaviour with respect to waste in order to have a positive effect on the environment (recycling, avoiding unnecessary purchases, etc.). However, this has been given little attention, presumably because of its small-scale impact.
There have also been a few drivers put forward that create the illusion of an EKC, but do not reduce global environmental degradation. In particular, the emergence of new pollutants and the off-shoring of pollution (Dasgupta et al., 2002). Most of the empirical studies use specific pollutants as an indicator. If the decline of that pollutant is the result of changing production processes, which also introduce new pollutants, than the analysis would support the EKC hypothesis. However, no actual environmental improvement has been achieved, as one pollutant has been replaced by another. Increased international trade may also have a misleading impact on observed EKCs. Although trade can increase income and thus generate demand for environmental improvements, lower trade barriers can give polluting industries in developing countries, where environmental regulations are less strict, a competitive advantage. Developed countries can respond in two ways (Dinda, 2004). First, developed countries can import more polluting products from developing countries and thereby reducing domestic pollution. The decline of domestic pollution in developed
8
countries may give the illusion that an EKC exists. The reason it does not, is that as developing countries reach a higher level of income and impose stricter environmental regulations, the initial competitive advantage is lost and polluting industries might return. The developing countries cannot off-shore their polluting production, as there are no more countries with loose environmental regulations (Stern, 1996). Second, developed countries can soften their environmental regulations. This can result in a ‘race to the bottom’, at which environmental regulations are virtually non-existent. This does not so much create the illusion of an EKC, but rather contradicts the possibility of an EKC. However, according to Dasgupta et al. (2002), empirical evidence does not support this argument.
On the theoretical level, the complicated and indirect relation between income and environmental degradation, that is positive at low levels of income and negative at high levels of income, has been compressed into two factors: changing consumer preferences and technological innovation. The effects of international trade and the emergence of new pollutants have often been put forward as arguments against the existence of the EKC.
Method and data
This paper extends on previous empirical analysis by utilizing a simple model with more recent data, as well as taking several econometric concerns into consideration. To test the EKC hypothesis, a country and time fixed effects regression of the following model is used:
Yij = αi + αj + β1*GDPij + β2*GDPij2 + εij
Where Yij is the indicator for environmental degradation of country i and GDPij is the GDP per capita of country i, across years j. αi and αj are country and year fixed effects, respectively. The fixed effects control for time and country invariant characteristics that may have an effect on environmental degradation. Because this study focusses on whether an EKC can be detected in the data, rather than explain the phenomenon, there are no additional explanatory variables included in the regressions. Stern (1996) also point that additional explanatory variables are likely subject to omitted variable bias, making inference complicated. The choice for fixed effects, as opposed to random effects, is supported by a test of overidentifying restrictions for which the Sargan-Hansen statistic is reported in Appendix A (Table A1). The environmental Kuznets curve suggests a positive marginal effect between GDP per capita and environmental degradation for countries with a low GDP per capita and a negative marginal effect for countries with a high GDP per capita. The level of GDP per capita at which a turning points occurs can be calculated as follows: GDP* = -β
1/(2*β2). For inference it is important that the turning point is within the data sample, i.e. GDP* < GDPmax. Whereas in the past, turning points for CO2 emissions were often found outside the observed GDP per capita range. With more recent data and higher maximum GDP per capita values, a turning point could possibly be found within the observed range.
As an indicator for environmental degradation this study uses CO2 emissions (metric tons per capita). Data is obtained from the World Bank Database (World Bank, 2016), which reports CO2 emissions per capita, GDP per capita from 1960 to 2013 and GDP adjusted for PPP from 1990-2013. Additionally, real GDP at chained PPPs from 1960-2013 have been retrieved from the Penn World Tables (Feenstra, Inklaar and Timmer, 2015). From a theoretical point of view, it would be more
9
appropriate to use GDP adjusted for PPP. The decline in CO2 emissions at higher income levels is partly explained by changing preferences. Preferences are unlikely to change due to increases in GDP per capita when purchasing power remains the same. Accounting for purchasing power better indicates how certain needs have been met and new preferences have developed. Using three measures for GDP allows for a comparison, between GDP and GDP adjusted for PPP, and the two sources for GDP adjusted for PPP (World Bank and Penn World Tables). With these data, three unbalanced panels have been constructed, one for each measure of GDP per capita. All panels contain approximately 180 countries, differences are due to data requirements. The World Bank divides countries into four income groups which are: low income, lower middle income, upper middle income and high income. Although there are twice more high income countries than low income countries, all groups are reasonably well represented. The distribution of these income groups in the panels are presented in Table B1 in the appendix B. Summary statistics for the three panels are presented in Table 1.
Table 1. Summary statistics for CO2 and GDP per capita (adjusted for PPP) from World Bank (WB) data for
1990-2013, for CO2 and GDP per capita at chained PPPs from Penn World Tables (PWT) data for 1960-2013,
and for CO2 and GDP per capita from World Bank (WB) data for 1960-2013:
MEAN STD. DEV. MIN MAX OBSERVATIONS
CO2 4.37 5.91 .01 67.45 N = 4227 n = 185 PPP - WB 12332.98 15527.05 239.74 141968.1 T = 24 T-bar = 22.85 CO2 4.35 7.40 -0.02 87.73 N = 7560 n = 165 PPP - PWT 11079.51 17197.27 142.39 245077.8 T = 54 T-bar = 45.82 CO2 4.37 7.11 -0.02 87.73 N = 7702 n = 186 GDP - WB 6295.10 12290.58 37.52 157093.00 T = 54 T-bar = 41.41
There are several advantages of using CO2 emissions per capita as an indicator for environmental degradation. First, CO2 is a by-product from the combustion of fossil fuels, which to a large extent originate from the energy and transportation sector. Because these sectors are local (occur within a country’s border) and are vital to a country’s economy, these sectors cannot be relocated to countries with less strict environmental regulations, thus avoiding the problem that international trade introduces. Furthermore, the processes that emit CO2 cannot easily be replaced by other processes which produce other pollutants. Although nuclear energy produces nuclear waste, which could be considered a ‘new pollutant’, this waste is not discharged into the environment, which means there is no environmental impact. Second, despite the fact that CO2 emissions measure pollution rather than the actual state of the environment, it remains one of the most suitable indicator among indicators that measure pollution. CO2 emissions play an important role in many processes that affect environmental degradation. Processes such as acidification of the oceans, desertification, changing weather patterns and loss of biodiversity. CO2 emissions are not only related to one specific aspect of environmental degradation, like many other indicators. Finally,
10
the theory argues that the EKC is partly explained by changing consumer preferences. Consumers require information about their life and the state of the environment in order to develop such preferences. CO2 emissions are widely reported and discussed as well as relatively easily understood by the general public. Therefore, CO2 may play an important role in determining consumer preferences.
To test the robustness of the results, additional tests are conducted for heteroskedasticity, autocorrelation and cross-sectional dependence. The results for these tests can be found in Appendix A (Table A2) and support the use of Driscoll-Kraay standard errors. These standard errors account for heteroskedasticity, autocorrelation as well as cross-sectional dependence. Additionally, the four income groups will be tested individually. One would expect the low income countries to exhibit a linear relation between CO2 and GDP per capita. For the higher income categories, it is expected that the negative second order effect of GDP per capita on CO2 is more prevalent. Furthermore, regressions will also be performed with data from 1992 until 2013 only. This is to briefly investigate some of the effect of the Earth Summit in 1992, which led to the Rio Declaration on environment and development, and the Kyoto Protocol, which was adopted in 1997. Both addressed the issue of climate change and sought to reduce greenhouse gas emissions, which includes CO2 emissions. The Rio declaration introduced 27 principles, one of which states that countries shall conduct an environmental impact assessment for potentially damaging activities. The Kyoto Protocol introduced a mechanism that allowed for the trading in CO2 as if it were a commodity. This gave CO2 a price and allowed countries and firms to internalize CO2 emissions. An additional benefit is that the panels will become more balanced when only recent periods are considered.
To be able to interpret the results, it may be useful to discuss a particular feature of the functional form that will influence the results. Because both the upward and the downward part of the EKC are estimated by a single function, the functional form implies a mathematical relation between the upward and the downward part of the EKC. In particular, the variable GDP2 implies that the curve is symmetrical about the vertical line that goes through the turning point. This would mean that the rate of environmental degradation before the turning points is the same as the rate of environmental improvements beyond the turning point. In reality however, there is no reason to assume that this is true. For the discussion of the results it is important to determine whether there are actually environmental improvements beyond the turning point, rather than a stabilization of CO2 emissions per capita.
Finally, in order to use the results to cautiously forecast future emissions, we must make sure that the panel does not contain a unit root or account for it. An Im-Pesaran-Shin test, based on the augmented Dickey-Fuller (ADF) test, is conducted to test for a unit root. The results for these tests can be found in Appendix A (Table A3) and indicate that GDP and GDP squared contain a unit root, whereas CO2 emissions do not seem to contain a unit root. However, CO2 emissions do appear to be trending. Because the model accounts for the effect of every single year by including dummies for n-1 individual years, the trend in CO2 emissions is already controlled for. Because GDP and GDP squared are non-stationary, first differences of GDP and GDP squared are also used as explanatory variables. Additionally, if the variables also contain a common stochastic trend, which can be tested using an EG-ADF test, different estimation methods should be used, such as dynamic OLS. However, this is outside the scope of this thesis.
11
Results
The section presents and discusses the results from the regressions as described in the previous section as well as the turning points that can be calculated from these results. The regression results for the three datasets are presented in Table 2. The variables GDP and GDP2 are included as well as the first differences, ΔGDP and ΔGDP2, to account for a unit root. 1 and 2 correspond to the dataset with GDP per capita from the World Bank (1960-2013). 3 and 4 correspond to the dataset with real GDP per capita at chained PPPs from the Penn World Tables (1960-2013). Finally, 5 and 6 correspond to the dataset with GDP per capita adjusted for PPP from the World Bank (1990-2013). In the previous section it was concluded that CO2 does not contain a unit root, but both GDP and GDP2 did contain a unit root. Therefore, the models that account for a unit root (2, 4 and 6) are considered to be the more appropriate models.
For all three models a positive coefficient is found for ΔGDP and a negative coefficient for ΔGDP2. This indicates an inverted U-shape and allows a turning point to be calculated. Of the models that do not account for a unit root (1, 3 and 5), the coefficients of models 1 and 5 also indicate an inverted U-shape. Model 3 does not, instead it suggests a U-shape relation between CO2 emissions and GDP per capita. However, the coefficient for GDP in model 3 is not statistically significant. Regression results for a linear relation between CO2 and GDP (Table B2, Appendix B) also suggest that in model 3 CO2 emissions rise monotonically as GDP per capita increases.
Table 2. Regressions results. (1,2) is GDP from World Bank, (3,4) is GDP at chained PPPs from PWT and (5,6)
is GDP adjusted for PPP from World Bank.
(1) (2) (3) (4) (5) (6)
GDP 7.26e-5** 3.14e-5 1.82e-4***
(3.61e-5) (2.63e-5) (0.24e-4)
GDP2 -12.6e-10*** 5.72e-10*** -1.59e-9***
(3.26e-10) (1.69e-10) (0.28e-9)
ΔGDP 4.71e-4*** 4.03e-4*** 1.99e-4***
(1.43e-4) (1.15e-4) (0.52e-4)
ΔGDP2 -2.90e-9*** -1.46e-9** -1.27e-9**
(1.06e-9) (6.87e-10) (0.52e-9)
Prob > F 0.000 0.000 0.000 0.000 0.000 0.000 within R2 0.07 0.06 0.19 0.05 0.11 0.03 N 7702 7516 7560 7395 4227 4042 * Statistically significant at 10% ** Statistically significant at 5% *** Statistically significant at 1%
Turning points for models 1, 2, 4, 5 and 6 are as follows: (1) 28,809.52 (2016 US$)
(2) 81,206.90 (2016 US$) (4) 138,013.70 (2011 US$)
12 (5) 57,232.70 (2016 international $) (6) 78,346.46 (2016 international $)
Comparing these results with the summary statistics in Table 1, we see that turning point for models 2, 4 and 6 are considerably higher than the mean GDP per capita. This is also true when compared to the mean GDP per capita levels for 2013. Yet, the turning points are within the data sample. The turning point of model 1 does not account for a unit root and is roughly consistent with findings from Dutt (2009). There are considerably fewer measurements of GDP per capita beyond the turning point, then there are before the turning point. As mentioned before, the functional form implies a symmetric EKC. Because most of the observations of GDP per capita are below the turning point, the upward part of the EKC is estimated more accurately than the downward part. Because of the symmetry, the downward part of the EKC might be a necessary consequence of the functional form, rather than a feature of the data. Therefore, it could also be that beyond the turning point the curve flattens, instead of declines. Especially the high turning point of model 4 raises the question of how certain we can be of a decline in CO2 emissions, instead of stabilization. If consumer preferences do not change significantly or are not translated into effective environmental regulation and firms can afford to ignore changes in demand, it could be the case CO2 emissions per capita become stable at some point or increase only at a small rate, but a decline is never realised. Based on these results it is difficult to distinguish between the two development paths. Although these results seem to support the EKC hypothesis, it is worth to further investigate the declining part of the curve. Also note that accounting for a unit root results in higher turning points in both cases (2 and 6).
Table 3 presents regression results of all three datasets for the period 1992-2013 only.
Table 3. Regressions results for the period 1992-2013. Labels are the same as in Table 2
(1) (2) (3) (4) (5) (6)
GDP 4.6e-5** 2.10e-4*** 1.86e-4***
(2.04e-5) (0.23e-4) (0.25e-4)
GDP2 -7.89e-10*** -1.76e-09*** -1.57e-9***
(2.05e-10) (9.45e-11) (0.27e-9)
ΔGDP 1.48e-4*** 2.03e-4* 1.96e-4***
(0.34e-4) (1.11e-4) (0.53e-4)
ΔGDP2 -9.12e-10** -1.23e-9 -1.24e-9**
(3.73e-10) (0.91e-9) (0.53e-9)
Prob > F 0.000 0.000 0.000 0.000 0.000 0.000 within R2 0.05 0.02 0.16 0.02 0.11 0.03 N 3941 3908 3584 3560 3933 3897 * Statistically significant at 10% ** Statistically significant at 5% *** Statistically significant at 1%
Because the period is the same for all datasets, a comparison can be made between the three different measures for GDP per capita. Furthermore, it allows a closer look at the declining part of the curve. The coefficients in all models are now in line with the EKC hypothesis. However, model 4
13
lacks statistical significance. The turning points for the models are: (1) 29,150.82 (2016 US$) (2) 81,140.35 (2016 US$) (3) 59,659.09 (2011 US$) (4) 82,520.33 (2011 US$) (5) 59,235.67 (2016 international $) (6) 79,032.26 (2016 international $)
All turning points remain inside the data sample. Models 3 and 4 are the only ones that seem to be affected by dropping all measurements before 1992. Estimates from model 3 are now in line with the EKC hypothesis and a turning point can be calculated. The turning point for model 4 has decreased considerably. Results from models 5 and 6 are nearly identical compared to the results for the period 1960-2013. This is to be expected, because roughly the same period is covered. However, despite the shorter period, the turning points for models 1 and 2 are remarkably similar to the ones from Table 2. Because much of the reduction in CO2 emissions is thought to have happened in the period 1992-2013, it was expected that the negative impact on CO2 emissions would be more prevalent and produce a lower turning point. This could either mean that the results are consistent over time or that dropping the earlier measurements had two opposing effects, which cancel each other out. The size of the coefficients in absolute terms for model 2 suggest that both periods may produce similar turning points, but the curve itself is different. Both coefficients for model 2 are smaller (in absolute terms) for the period 1992-2013 than for the period 1960-2013, i.e. both the positive first order effect and the negative second order effect are smaller. This means that the upward and downward part of the slope are less steep, i.e. the curve is flatter. Not only are the turning points similar for both periods, the rate of environmental improvements beyond the turning point is lower for the period 1992-2013 as well. This is also true for model 4, except that the turning points is considerably lower. Because there are now relatively more measurements beyond the turning points than for the period 1960-2013, the downward part of the EKC is more accurately estimated for the period 1992-2013. It is likely that the higher rate of environmental improvements beyond the turning point, as derived from Table 2, is due to the imposed symmetry of the EKC rather than a feature of the data.
Turning points for all measures of GDP are now relatively close to each other as well. The shape of the curves of models 3 and 4 are also very similar, suggesting no significant differences between the two measures of GDP adjusted for PPP. Again, accounting for a unit root results in higher turning points in all cases. Stern (2004) concluded that using more appropriate econometric techniques generally results in higher turning points. The results from Table 2 and 3 are in line with that statement.
Finally, to investigate the different income groups separately, only one dataset is used, with the GDP variable from the Penn World Tables. As mentioned above, GDP adjusted for PPP is preferred, also this dataset ranges from 1960 to 2013. The results are presented in Table 4. The results for low income countries (1), lower middle income countries (2) and upper middle income countries (3) indicate a U-shape relation between CO2 emissions and GDP per capita. However, the coefficients in models 1 and 2 are not statistically significant. The U-shape implies that CO2 emissions will decline first, before rising again. The theory predicts that lower income countries show a positive relation between CO2 emissions and GDP per capita. Therefore, the initial decline is a somewhat surprising result, but once again, this could be due to the imposed symmetry.
14
Moving from the upper middle income countries (3) to the high income countries (4) shows an interesting development, the signs of both coefficients flip. Whereas the results from model 3 imply a positive relation with increasing marginal effects, the results from model 4 show negative marginal effects and a turning point at 136,129.03 (2011 US$). Once again, this turning point is too high to determine whether the curve actually declines or flattens out. Splitting the data for high income countries into the periods 1960-1992 (5) and 1992-2013 (6) may provide further insight. Because the period 1960-1992 is associated with lower level of development than the period 1992-2013, it is expected that the period 1960-1992 shows a positive relation between CO2 emissions and GDP per capita and that the period 1992-2013 provides stronger evidence for a negative relation. Model 5 shows that a turning point exists at 219,724.77 (2011 US$). Only one country reports values of GDP per capita beyond the turning point. Therefore, the path of environmental improvements beyond the turning point is very uncertain. Once again, it seems likely that the turning point is the result of the imposed symmetry and that the data only indicates a flattening of the curve. However, model 6 also finds evidence for an inverted U-shape relation between CO2 emissions and GDP per capita for the period 1992-2013. The turning point is 88,148.15 (2011 US$), which is somewhat comparable to the turning points derived from Table 3. Compared to model 5, model 6 does indeed indicate that the rate of environmental improvement is higher, and thus provides support for the EKC hypothesis. However, both coefficients for GDP2 are not statistically significant. Furthermore, even for model 6, the amount of observations beyond the turning point are still low, giving rise to uncertainty about the declining part of the EKC.
Table 4. Regressions results of the income groups for real GDP per capita at chained PPPs (PWT). 1,2,3 and 4
correspond to low income, lower middle income, upper middle income and high income, respectively. 5 and 6 also correspond to high income, but for the periods 1960-1992 and 1992-2013, respectively.
(1) (2) (3) (4) (5) (6)
ΔGDP -3.42e-5 -8.25e-4* -3.95e-4*** 4.22e-4*** 4.79e-4** 2.38e-4* (8.71e-5) (4.18e-4) (1.40e-4) (1.26e-4) (1.95e-4) (1.20e-4) ΔGDP2 2.53e-8 9.26e-8 1.98e-8*** -1.55e-9** -1.09e-9 -1.35e-9
(1.90e-8) (5.73e-8) (0.54e-8) (0.73e-9) (0.81e-9) (0.93e-9) Prob > F 0.000 0.000 0.000 0.000 0.000 0.000 within R2 0.07 0.21 0.26 0.09 0.13 0.03 N 1363 1686 1981 2365 1252 1204 * Statistically significant at 10% ** Statistically significant at 5% *** Statistically significant at 1% Conclusion
For over 25 years, economists have used the EKC hypothesis to model the relation between income and environmental degradation. Evidence from early papers for a variety of environmental indicators with a local and direct impact seemed to support the EKC hypothesis. However, until recently it appeared that for CO2 emissions the EKC did not exist. The results in this thesis are generally
15
consistent with the EKC hypothesis, as most of the coefficients have the correct signs. Estimated turning points range from roughly 79,000 to 89,000 for the period 1992-2013 and are around 135,000 for the period 1960-2013 (in 2011 US$).
The differences between both measures of GDP adjusted for PPP seem to be minimal. For the period 1992-2013, both measures produces similar curves and turning points. For this period, GDP per capita also produces a similar turning point, but the rate of environmental degradation before the turning point and the rate of environmental improvements beyond the turning point are higher. Related to this, is the issue of symmetry, as imposed by the functional form. Because of the symmetry and the lack of observations beyond the turning points, a flattening of the curve cannot be excluded. Results for the period 1992-2013 and the different income groups alone, does provide additional evidence for the EKC, but the lack of observations beyond the turning point remains an issue.
All turning points considered, this thesis supports Holtz-Eakin and Selden’s (1995) conclusion that global CO2 emissions will continue to increase for a considerable amount of time, as mean levels of GDP per capita are well below the turning points. In the near future, increases in GDP per capita will most likely still result in higher CO2 emissions. On the other hand, there is also reason for a little more optimism, as efforts to reduce CO2 emissions are also expected to increase in the future. In 2015 the Paris Agreement was adopted by 195 countries and as of December 2016, 127 countries have ratified the treaty. The main goal of this agreement is to prevent the global average temperature to rise more than 2°C above pre-industrial levels (UNFCCC, 2015). As the main contributor to the greenhouse gas effect, CO2 emissions per capita will have to decrease in order to achieve this goal.
These types of studies assume that countries have homogenous development paths, an assumption that is unlikely to hold in reality. Therefore, time-series and historical analysis of individual countries remain relevant. This study used a rather crude test for detecting a unit root in the panels. Tests for a unit root in single time-series regressions may find more evidence of a unit root for CO2 emissions. Also, there are no standard errors presented for the turning points. As a consequence, no statements can be made about the accuracy of the turning points. The turning points also do not give the total level of emissions, which is relevant information if an ecological threshold exists, as we would not want to exceed that threshold. A drawback of testing a reduced-form EKC is that it cannot be used for the development of environment policies, because it does not estimate the factors that directly affect CO2 emissions. Therefore, it should be emphasized that the process of environmental improvements is not automatic, but rather the result of government policies, social institutions and well-functioning markets.
16
References
Andreoni, J., & A. Levinson. (2001). The Simple Analytics of the Environmental Kuznets Curve. Journal of Public Economics, 80, 269–286.
Arrow, K., Bolin, B., Costanza, R., Dasgupta, P., Folke, C., Holling, C. S., et al. (1995). Economic growth, carrying capacity, and the environment. Science, 268, 520–521.
Barbier, B.E. (1997). Introduction to the Environmental Kuznets Curve Special Issue, Environment and Development Economics, 2, 369-381.
Cole, M. A., Rayner, A. J., & Bates, J. M. (1997). The environmental Kuznets curve: An empirical analysis. Environment and Development Economics, 2(4), 401–416.
Dinda, S. (2004). Environmental Kuznets Curve Hypothesis: A Survey. Ecological Economics, 49, 431–455.
Dasgupta, S., Laplante, B., Wang, H. & Wheeler, D. (2002). Confronting the Environmental Kuznets Curve. Journal of Economic Perspectives, 16(1), 147-168.
Dutt, K. (2009). Governance, institutions and the environment-income relationship: a cross-country study. Environment, Development and Sustainability, 11(4), 705–723.
Feenstra, R.C., Inklaar, R., & Timmer M.P. (2015). The Next Generation of the Penn World Table. American Economic Review, 105(10), 3150-3182, available for download at www.ggdc.net/pwt. Grossman, G. M., & Krueger, A. B. (1991). Environmental impacts of a North American Free Trade Agreement. NBER Working Papers, 3914.
Grossman, G. M., & Krueger, A. B. (1995). Economic growth and the environment. Quarterly Journal of Economics, 110, 353–377
Holtz-Eakin, D., & Selden, T. M. (1995). Stoking the fires. CO2 emissions and economic growth. Journal of Public Economics, 57, 85–101.
Jalil, A., & Mahmud, S.F. (2009).Environment Kuznets curve for CO2 emissions: A cointegration analysis for China. Energy Policy, 37(12), 5167-5172.
Jaunky, V.C. (2011). The CO2 emissions-income nexus: Evidence from rich countries. Energy Policy, 39(3), 1228-1240.
Kaika, D., & Zervas, E. (2013). The Environmental Kuznets Curve (EKC) theory—Part A: Concept, causes and the CO2 emissions case. Energy Policy, 62, 1392-1402.
Lopez, R. (1994). The Environment as A Factor of Production: The Effects of Economic Growth and Trade Liberalization. Journal of Environmental Economics and Management, 27, 163–184.
Narayan, P.K., & Narayan, S. (2010). Carbon dioxide emissions and economic growth: Panel data evidence from developing countries. Energy Policy, 38(1), 661-666.
Panayotou, T. (1997). Demystifying the Environmental Kuznets Curve: Turning A Black Box into a Policy Tool. Environment and Development Economics, 2, 465–484.
17
Pasten, R., & Figueroa, E.,B. (2012). The Environmental Kuznets Curve: A Survey of the Theoretical Literature. International Review of Environmental and Resource Economics, 6, 195-224.
Perman, R., & Stern, D. I. (2003). Evidence from panel unit root and cointegration tests that the environmental Kuznets curve does not exist. Australian Journal of Agricultural and Resource Economics, 47, 325–347.
Saboori, B., Sulaiman, J., & Mohd, S. (2012). Economic growth and CO2 emissions in Malaysia: A cointegration analysis of the Environmental Kuznets Curve. Energy Policy, 51, 184-191.
Selden, T. M., & D. Song. (1994). Environmental Quality and Development: Is There a Kuznets Curve for Air Pollution Emissions? Journal of Environmental Economics and Management 27, 147-162. Shafik, N., & S. Bandyopadhyay (1992). Economic growth and use and the U.S. economy.
Background Paper for the World Development WCED, Our Common Future (Oxford: Oxford University Report 1992 (Washington, DC: The World Bank, 1992).
Shahbaz, M., Lean, H.H., & Shabbir, M.S. (2012). Environmental Kuznets Curve hypothesis in Pakistan: Cointegration and Granger causality. Renewable and Sustainable Energy Reviews, 16(5), 2947-2953.
Stern, D.I., Common, M.S., & Barbier, E.B. (1996). Economic Growth and Environmental Degradation: The Environmental Kuznets Curve and Sustainable Development. World Development, 24(7), 1151-1160.
Stern, D.I., (2004). The rise and fall of the environmental Kuznets curve. World Development, 32, 1419–1439.
Stokey, N. L. (1998). Are There Limits to Growth? International Economic Review, 39(1), 1–31. Tiwari, A.K., Shahbaz, M., & Hye, Q.M.A. (2013). The environmental Kuznets curve and the role of coal consumption in India: Cointegration and causality analysis in an open economy. Renewable and Sustainable Energy Reviews, 18, 519-527.
Torras, M., & Boyce, J. K. (1998). Income, inequality, and pollution: A reassessment of the environmental Kuznets curve. Ecological Economics, 25, 147–160.
UNFCCC, (2015). Adoption of the Paris Agreement. Report No. FCCC/CP/2015/L.9/Rev.1. http://unfccc.int/resource/docs/2015/cop21/eng/l09r01.pdf.
World Bank, national account data (2016). CO2 per capita, GDP per capita, GDP adjusted PPP per
capita. Retrieved November 28, 2016 from: http://databank.worldbank.org.
World Commission on Environment and Development (1987). Our common future. Oxford: Oxford University Press.
Yandle, B., Bhattarai, M., & Vijayaraghavan, M. (2004). Environmental Kuznets Curves: a Review of Findings, Methods, and Policy Implications. PERC, 2, 1–38.
18
Appendix A
Table A1. Fixed vs random effects.
GDP - WB PPP - PWT PPP - WB
SARGAN-HANSEN STATISTIC 3.8e+07 9.2e+07 4.9e+05
P-VALUE 0.0000 0.0000 0.0000
Table A2. Heteroskedasticity, autocorrelation and cross-sectional dependence.
GDP - WB PPP - PWT PPP - WB HETEROSKEDASTICITY LR CHI2 28174.45 27754.57 13107.94 PROB>CHI2 0.0000 0.0000 0.0000 DF 185 164 184 FIRST-ORDER AUTO-CORRELATION F 83.683 45.871 195.456 PROB>F 0.0000 0.0000 0.0000 DF 185 164 184 CROSS-SECTIONAL DEPENCE PESARAN N.A. 125.948 49.887 P-VALUE N.A. 0.0000 0.0000
Table A3. Results for Im-Pesaran-Shin test for a unit root
COTWO1 COTWO2 COTWO3 GDP -WB PPP - PWT PPP - WB
LAGS(AIC) W-T-BAR STATISTIC -0.2678 0.7162 2.0291 36.3655 27.3214 33.5532 P-VALUE 0.3944 0.7631 0.9788 1.0000 1.0000 1.0000 + TREND W-T-BAR STATISTIC -2.9376 -1.7372 -3.8953 16.2780 11.9354 4.3779 P-VALUE 0.0017 0.0412 0.0000 1.0000 1.0000 1.0000 + DEMEAN W-T-BAR STATISTIC -7.5575 -6.9141 -5.5277 22.0877 12.3025 22.0933 P-VALUE 0.0000 0.0000 0.0000 1.0000 1.0000 1.0000 + TREND AND DEMEAN
W-T-BAR STATISTIC -2.9881 -3.0618 -5.5884 4.0258 6.7667 -0.6453 P-VALUE 0.0014 0.0011 0.0000 1.0000 1.0000 0.2594 - Cotwo1, cotwo2 and cotwo3 refer to the variables for CO2 emissions in the three panels; GDP – WB, PPP –
PWT and PPP – WB, respectively. They are all obtained from the World Bank. However, the different panels have a different composition of countries and years included.
19 Table A3-continued. GDP2 -WB PPP2 - PWT PPP2 - WB LAGS(AIC) W-T-BAR STATISTIC 49.1895 38.6630 49.8465 P-VALUE 1.0000 1.0000 1.0000 + TREND W-T-BAR STATISTIC 33.8754 21.6608 17.4945 P-VALUE 1.0000 1.0000 1.0000 + DEMEAN W-T-BAR STATISTIC 25.9919 5.8676 39.1786 P-VALUE 1.0000 1.0000 1.0000
+ TREND AND DEMEAN
W-T-BAR STATISTIC 10.7526 8.0694 7.1069
20
Appendix B
Table B1. fractions of income groups present in the panels.
PPP - WB PPP - PWT GDP - WB
LOW INCOME 0.15 0.18 0.16
LOWER MIDDLE INCOME 0.26 0.23 0.25
UPPER MIDDLE INCOME 0.29 0.27 0.27
HIGH INCOME 0.31 0.32 0.33
Table B2. Regression results for a linear relation between CO2 and GDP. Labels are the same as in Table 2.
(1) (2) (3) (4) (5) (6)
GDP -2.08e-5 1.29e-4*** 8.55e-6
(1.76e-5) (0.36e-4) (9.90e-6)
ΔGDP 1.60e-4** 1.12e-4* 5.87e-5
(0.68e-4)T (0.64e-4) (6.49e-5)
Prob > F 0.000 0.000 0.000 0.000 0.000 0.000 within R2 0.05 0.06 0.16 0.04 0.02 0.02 N 7702 7516 7560 7395 4227 4042 * Statistically significant at 10% ** Statistically significant at 5% *** Statistically significant at 1%
