Bounds testing approach to analysis of the environment Kuznets curve hypothesis

Energy Economics 44 (2014) 47–62

Contents lists available at ScienceDirect

Energy Economics journal homepage: www.elsevier.com/locate/eneco

Bounds testing approach to analysis of the environment Kuznets curve hypothesis Olugbenga A. Onafowora a, Oluwole Owoye b,⁎ a b

Department of Economics, Susquehanna University, Selinsgrove, PA 17870, USA Department of Social Sciences/Economics, Western Connecticut State University, Danbury, CT 06810, USA

a r t i c l e

i n f o

Article history: Received 16 February 2013 Received in revised form 27 January 2014 Accepted 24 March 2014 Available online 12 April 2014 JEL classification: F18 O13 C23 Q53 C32 Keywords: Environmental Kuznets curve (EKC) ARDL bounds test CO2 emissions Economic growth Energy consumption

a b s t r a c t This paper examines the long-run and the dynamic temporal relationships between economic growth, energy consumption, population density, trade openness, and carbon dioxide (CO2) emissions in Brazil, China, Egypt, Japan, Mexico, Nigeria, South Korea, and South Africa based on the environment Kuznets curve (EKC) hypothesis. We employ the ARDL Bounds test to cointegration and CUSUM and CUSUMSQ tests to ensure cointegration and parameter stability. The estimated results show that the inverted U-shaped EKC hypothesis holds in Japan and South Korea. In the other six countries, the long-run relationship between economic growth and CO2 emissions follows an N-shaped trajectory and the estimated turning points are much higher than the sample mean. In addition, the results indicate that energy consumption Granger-causes both CO2 emissions and economic growth in all the countries. Our results are consistent with previous studies that show that there is no unique relationship between energy consumption, population density, economic growth, trade openness, and the environment across countries. © 2014 Elsevier B.V. All rights reserved.

1. Introduction Achieving higher economic growth at lower intensity of resource use without compromising the quality of life of future generations is a continuing and common concern of governments around the world, one that is exacerbated by the increasing threat of global warming and climate change. An urgent issue for environmental policy makers is to understand and predict how environmental quality will evolve over time with continued increase in economic activity. Environmental economics uses the environmental Kuznets curve (EKC) to model empirically the economic growth–environmental quality trajectory. The EKC postulates an inverted U-shaped relationship between environmental quality and income per capita. Proponents of the EKC claim that in the early stages of industrialization, environmental degradation increases because greater priority is given to increasing material output, and people are more interested in jobs and income than in public properties like environment and its resources. Higher economic activity, however, demands higher inputs of energy and other natural resources and thus higher emissions of pollutants, which ⁎ Corresponding author. Tel.: +1 203 837 8456; fax: +1 203 837 3960. E-mail address: [email protected] (O. Owoye).

http://dx.doi.org/10.1016/j.eneco.2014.03.025 0140-9883/© 2014 Elsevier B.V. All rights reserved.

in turn worsen the environmental conditions. At the later stage of industrialization, and as income increases beyond a threshold (known as the “turning point”), the willingness to pay for a clean environment increases by a greater proportion than income; regulatory institutions become more effective in reducing pollution levels leading to gradual improvement of environmental conditions1 (Panayotou, 1993; Stern, 2004; Dinda, 2004). Following Grossman and Krueger (1991), who first described the EKC and its potentially promising implications for making economic growth sustainable in the future, a plethora of empirical studies have searched for systematic relationships by regressing different measures of air and water quality on various polynomial specifications of income per capita. In general, the EKC hypothesis holds for certain pollutants, including sulfur dioxide (SO2), suspended particulate matters (SPM), nitrogen dioxide (NO2), but less likely for carbon dioxide (CO2). While some studies on CO2 emissions find evidence of an inverted Ushaped path relative to income growth,2 others find a close positive 1 For a thorough survey of theoretical and empirical studies dealing with the EKC please, see Dinda (2004) and Stern (2004). 2 See, for example Grossman and Krueger (1995), Shafik and Bandyopadhyay (1992), Coondoo and Dinda (2002), Apergis and Payne (2010) and Narayan and Narayan (2010).

48

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

relationship between the two variables.3 Others find that the turning point income value needed to start decreasing emissions is very high or is nonexistent; and others even find an N-shaped path, which may be interpreted to imply that rising income initially deteriorates environmental quality and then improves it—the standard EKC result—but then with additional income, emissions increase again.4 The mixed and sometimes inconclusive results from extant empirical studies of the EKC, both for different environmental indicators and also, for different studies looking at the same environmental indicator, is worrisome to environmental decision makers desperately seeking possibilities for simultaneous higher income and improved environmental quality. Most studies of the EKC use cross-section or panel data (under fixed or random effects) analysis, and the implicit assumption is that the coefficients of the inverted U-shaped relationship are country invariant; which implies that the expected shape of the EKC is the same for every country and the predicted turning point income is also the same for every country.5 One of the shortcomings of the cross-section or panel data analysis is the fact that countries exhibit significant differences in development paths, macroeconomic conditions, natural resource endowment, trade orientation, climate, culture, socio-political structures, and institutions. Given the heterogeneous structural and technical characters between countries, different countries exhibit different patterns for their relationships between environmental quality and economic growth. Therefore, pooling all countries together and testing one EKC for all of them are a biased procedure because it implicitly assumes that all countries in the sample share the same experience (Unruh and Moomaw, 1998). Cross-section analysis allows for the likelihood that some important variables which are correlated with income but not common to all countries in the panel could be omitted. Stern and Common (2001) pointed out that omitted important explanatory variables may result in a biased estimate of the EKC in a non-random sample of countries. According to de Bruyn et al. (1998) and Fodha, and Zaghdoud (2010), the conventional panel data estimation techniques have generated spurious EKC estimates because they do not adequately capture the dynamic process involved. Fodha and Zaghdoud (2010) cautioned that EKC results from panel data analysis are unrealistic and dangerous. Vincent (1997, p. 417) argued that an EKC obtained from cross-country regressions “may simply reflect the juxtaposition of a positive relationship between pollution and income in developing countries with a fundamentally different negative one in developed countries, rather than a single relationship that applies to both categories of countries.” Stern et al. (1996) suggested that it would be more appropriate to study the relationship between environment and economic growth, analyzing the experience of individual countries using both econometric and historical analysis. Following the suggestion of Stern et al. (1996), this paper examines the determinants of CO2 emissions in Brazil, China, Egypt, Japan, South Korea, Mexico, Nigeria, and South Africa by using an estimation equation that incorporates energy consumption, population density, and trade openness into the equation of the environmental Kuznets curve. These linkages have not been thoroughly explored to provide useful policy information for environmental decision makers, particularly in developing countries. This paper presents findings to integrate the environment into economic development decisions. This study contributes to the literature confirming the relationship between economic growth and environmental quality, which has 3

See, for example Shafik (1994), Cole et al. (2000), Akbostanci et al. (2009), Ozturk and Acaravci (2010) and Pao and Tsai (2010). 4 See, for example Shafik and Bandyopadhyay (1992), Grossman and Krueger (1995), Moomaw and Unruh (1997), Dinda et al. (2000), Friedl and Getzner (2003), Cole (2004), Martínez-Zarzoso and Bengochea-Morancho (2004) and Lipford and Yandle (2010). 5 According to Dijkgraaf and Vollebergh (1998, p. 3–4): “The question, not answered by the empirical studies is what the intuition behind this implicit assumption is. It seems strange that countries, which are very different in geographical conditions, culture and history, would react identical”.

been empirically modeled for several developed countries. There is little evidence in the context of developing countries using time series data. This underrepresentation is explained by the difficulty of obtaining data of sufficient length for developing countries. Stern et al. (1998) cautioned that this underrepresentation could create bias for the estimated EKC, given the on-going structural changes and specialization in favor of less polluting activities in industrialized countries. To avoid possible bias, it is imperative to include the developing countries, in particular the developing countries of Africa, Asia and South America, in the field of study for better understanding of the evolution of the EKC hypothesis within both developed and developing countries. To this end, this paper tests the EKC hypothesis for a sample of eight countries drawn from industrialized, developing, and emerging market economies. These are countries from different geographical regions and income groups, and who are at different stages of economic development in terms of energy consumption, population growth, income growth, and institutional capacity. The study focuses on the trend of CO2 emissions of each country and analyzes its relationship with respect to GDP per capita growth conditional on specific energy consumption, trade openness, and population density characteristics. Such an approach maps each country's economic growth–environmental quality trajectory on the EKC before and after the turning point. The results indicating their position on the EKC may be useful to formulate policy recommendations directed at conservation, emission reduction, and economic growth and may prove relevant to other countries that have to go through a similar development path. Second, most previous studies of the EKC have been conducted over a relatively short time horizon and have not taken into account structural breaks in the data series. The use of small sample size creates problem with hypothesis tests with low statistical power and higher confidence interval.6 Central limit theory suggests that as the size of the sample becomes large, the sampling distribution of the sample mean approaches a more normal distribution, which calls into question the robustness of the results relative to an analysis over a longer time horizon. Moreover, as noted by Perron (1989) and Enders (2004), the ignorance of structural breaks may bias test for unit root to suggest the presence of unit root even though the data generating process is trend stationary. This study overcomes the small sample deficiencies by employing a reasonably longer sample size of about four decades and controlling for possibility of structural breaks in unit root tests along with performing bounds test for cointegration, variance decompositions analysis, Granger causality tests, and CUSUM and CUSUMSQ tests for stability of the economic growth–environmental quality nexus. These allow for a more in-depth analysis of the interrelationships among the variables and enable the determination of the variables with stronger effects. Third, this study emphasizes identifying turning points of the EKC that have not been adequately looked at, using more up-to-date data. The rest of the paper is organized as follows: Section 2 provides a review of the EKC literature. In Section 3 we discuss the econometric methodology and data used in estimation. Section 4 presents the empirical results and a discussion of the results. Policy implications of the results are presented in Section 5. Our conclusions are reported in Section 6. 2. Literature review The relationship between economic growth and environmental pollution as well as, energy consumption and economic growth, has been one of the most widely investigated topics in the economics literature during the last few decades. Grossmann and Krueger (1991) who investigated the environmental impacts of the North American Free Trade Agreement and discovered that the relationship between the total discharge of various environmental pollutants and economic growth 6 See, Zachariadis (2007) for detailed discussion of the limitations associated with a small sample size in terms of inferences drawn from causality tests.

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

takes on the shape of an inverted U-shaped curve made the pioneering attempt in this area in the early 1990s. Panayotou (1993) later coined this U-shaped relationship as the environmental Kuznets curve (EKC) because of its similarity to the relationship between per capita income and the distribution level analyzed by Kuznets (1955). According to (Deacon and Norman, 2004, p.6), “The EKC hypothesis is fundamentally a within-country story, that is, a prediction of how a country's pollution will change as it experiences economic growth”. Whether continued increase in national income leads to more environmental degradation is critical for the design of development strategies for an economy (Ang, 2007). Hence, several empirical studies have examined the causal linkages between economic growth and environmental quality. These studies used different measures of environmental degradation as the dependent variable(s) and income as the independent variable. In terms of the approach to measurement, researchers have developed a range of models based on different assumed conditions and different dominant factors, and have used well over 20 different measures of pollution. The four most commonly used measures are CO2 and SO2, emissions (to proxy air pollution) and biochemical oxygen demand (BOD) and the level of dissolved oxygen (to proxy water pollution). Most studies on the EKC hypothesis use panel or cross-country data from a series of countries, especially high-income countries, for their empirical estimations. Earlier studies used reduced form singleequation specifications in which they relate an environmental impact indicator to income per capita. This amounts to assuming that the environment has no feedback effect on income growth. The most common equation form used in these models to arrive at the characteristic Ushaped relationship between income and the environment is a quadratic equation. In contrast, an N-shaped cubic equation indicates that there are many fluctuations in any real-world system. Coondoo and Dinda, (2002) and Dinda (2004), argued that the unidirectional hypothesis, in which the economy influences the environment with no inverse effect, is unrealistic and leads directly to the emergence of what is called the “endogeneity bias” and inaccurate assessments since economic growth itself is an endogenous variable determined by environmental changes and other factors. More emphatically, Panayotou (2003) cautioned that the different shapes of EKC found by these studies only capture the net effects of income on environment, where “income growth is used as an omnibus variable representing a variety of underlying influences, whose separate effects are obscured”, and therefore no clear development policy implications can be directly drawn from the estimated coefficients of the polynomial income terms. A number of the empirical studies used vector autoregression (VAR) and Granger (1969) causality tests to uncover whether income growth causes environmental degradation. With advances in time series econometric techniques, more recent studies have focused on the vector error correction model (VECM) and the cointegration approach. Furthermore, mindful of the problem of omitted variables bias, some authors incorporate factors, such as international trade, population dynamics, prices, environmental policy, income inequality, education, technology, and human development indicators, into the income–environment relation. The inclusion of these additional variables is to decompose the effect of economic growth on environment into the scale, technique, and output composition effects (see, Grossmann and Krueger, 1991; 1995; Cole and Neumayer, 2004). The results from the cross-national and panel studies have been mixed and are not conclusive to offer policy recommendations that can be applied across countries. For instance, Shafik and Bandyopadhyay (1992) estimated log-linear, log quadratic and logarithmic cubic polynomial functional forms of EKC for 10 different indicators of environmental degradation using cross-section data for 149 countries for the period 1960–1990. They found that air pollutants conform to the EKC hypothesis with turning points at income levels between $3000 and $4000. Panayotou (1993) used a translog specification and found similar results for these pollutants with turning points at income levels ranging from $3000 to $5000. Grossman and Krueger (1995) estimated cubic

49

transformation of EKC functions for SO2, dark matter and suspended particulate matter (SMP) using data for 52 cities in 32 counties over the period 1977–1988. For SO2 and dark matter, they found turning points at $4000–$5000 per capita; SMP continually declined at even low-income levels. However, at income levels over $10,000–$15,000 all three pollutants began to increase again. Moomaw and Unruh (1997) and Unruh and Moomaw (1998) examined the CO2 emission trajectories of sixteen OECD countries, and found that most of the countries showed an inverted U-shaped path. Applying the cubic model specification to the 16 countries, the authors determined that the N-shaped curves for all the estimated coefficients were statistically significant. List and Gallet (1999) used US states data between 1988 and 1994 and the seemingly unrelated regression model (SUR) based on Zellner and Theil (1962) and they found statelevel differences in the predicted EKC in states where such relationship exists. The authors pointed out that “Parameter estimates suggest, that previous studies, which restrict cross-sections to undergo identical experiences over time, may be presenting biased results” (List and Gallet (1999), p. 409). Stern and Common (2001) used a common structure for all countries, and they found some support for the EKC hypothesis in OECD but not in Non-OECD countries. Similarly, Apergis and Payne (2009) confirmed the existence of an EKC for a panel of six Central American economies. Cole and Neumayer (2004) considered 86 countries during the period from 1975 to 1998 and found a positive link between CO2 emissions and a set of explanatory variables including population, urbanization rate, energy intensity, and smaller household sizes. Chen (2009) investigated the relationship between industrial sector's development and CO2 emissions using Chinese provincial data, and found that industrial development increases CO2 emissions. Lean and Smyth (2010) arrived at a similar conclusion in their study of the ASEAN countries and Commonwealth of Independent States. In contrast to the vast empirical studies on the EKC for high-income countries, there are only a few studies that have focused on developing countries. Country specific studies as opposed to cross-national studies are also very few. de Bruyn et al. (1998) estimated individual dynamic time series models for the Netherlands, West Germany, UK, and the US. They found that economic growth has had a positive effect on emissions of SO2, NO2, and CO2. Friedl and Getzner (2003) investigated the economic growth-CO2 emissions nexus for Australia between 1960 and 1999, and they found an N-shaped relationship between income and CO2. Akbostanci et al. (2009) tested for the existence of EKC in Turkey using cointegration techniques and both time series and provincial panel data for the periods 1968 to 2003 and 1992 to 2001. They found a monotonically increasing relationship between CO2 emissions and income in the times series analysis, which suggests that the EKC hypothesis does not hold for CO2 emissions. He and Richard (2010) investigated the relationship between CO2 emissions per capita and GDP per capita for Canada between 1948 and 2004. They found little evidence in favor of the EKC. In stark contrast to He and Richard (2010), Ang (2007), Jalil and Mahmud (2009), and Iwata et al. (2010) found evidence supporting the EKC for CO2 emissions in France and China. Copeland and Taylor (2004), Grossman and Krueger, (1995), Machado (2000), Ang, (2009), Halicioglu, (2009), and Jalil and Mahmud, (2009) examined the effects of trade openness on the EKC for various countries. A positive link between trade and carbon dioxide emissions was found by Halicioglu (2009) for Turkey, Machado (2000) for Brazil, and Ang (2009), Jalil and Mahmud (2009) for China. Fodha and Zaghdoud (2010) investigated the existence of the EKC relation for Tunisia for the period 1961 to 2004. They found that CO2 and SO2 are cointegrated with income per capita, but their results for CO2 indicated a monotonically increasing relationship relative to income per capita. The literature reveals that, just like in cross-section panel analysis, the findings from time series country specific studies differ from country to country. The mixed results further confirm that studies based on pooled or cross-section data would provide incorrect inferences

50

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

regarding the real situation of the individual country and, more importantly, they could be grossly misleading in the context of formulating domestic policies for an individual country. 3. Econometric methodology and data 3.1. Basic framework of analysis In the EKC literature, the non-linear relationship between the indicators of environmental pollution and per capita income is usually specified in a semi-reduced form such as: 2

3

EDt ¼ α 0 þ α 1 Y t þ α 1 Y t þ α 1 Y t þ α 4 Z t þ εt

ð1Þ

where ED represents environmental degradation, i.e. the specific pollutant that is used for the estimation, Y is income per capita, and Z are other covariates that are supposed to influence pollution; t denotes a time index and ε is the normally distributed error term. For developing countries, no comprehensive data on environmental indicators and industrial pollution exists due to lack of monitoring of many pollutants and other indicators of the environmental conditions. Given the paucity of suitable data we follow the lead of other researchers, for example Friedl and Getzner (2003), Halicioglu (2009), and He and Richardson (2010), and select carbon dioxide emissions per capita as our measure of environmental degradation.7 With regard to the explanatory variables included in the model, we use gross domestic product (GDP) per capita as our income measure and use population density as the proxy to control for pollution caused by an increasing population. The impact of increases in income growth on environmental conditions is ambiguous. According to Grossman and Krueger (1995), Copeland and Taylor (2004), and Brock and Taylor (2004), the impact depends on which of the three effects—scale effect, technique effect, and composition effect—is stronger and dominates the others. The scale effect refers to the fact that increases in output require more inputs, and, as a by-product, imply more emissions. Economic growth therefore exhibits a scale effect that has a negative impact on the environment (Arrow et al., 1995). The technique effect refers to the invention of and the application of new environmental friendly technologies in production which in turn leads to the reduction of pollutants which degrade the environment (Andreoni and Levinson, 2001). The impact of the technique effect is theoretically positive (de Bruyn, 1997; and Vukina et al., 1999). The composition effect stems from changes in production process caused by specialization as well as the transition from agriculture and/or basic industries to high-tech services. According to Dasgupta et al., (1995), if the sectors with high emission intensities grow faster than sectors with low emission intensities then, ceteris paribus, composition changes will result in increasing pollutant emission. The expected impact of the composition effect is positive deriving from Rostow's (1959) stages of growth postulate. Due to the different nature of these individual effects, the overall impact of increases in income growth on the environment is ambiguous (Grossman and Krueger, 1991; Cole, 2004). The effect of population density on environmental conditions also is ambiguous. On the one hand, if more people live in a given area the effect of individual pollution exacerbates; and thus a high population density may lead to more pollution. On the other hand, Stern (2004) and Selden and Song (1994) argued that a higher population density may lead to lower per capita pollution emissions; as more people are 7 According to the 2007 World Bank report, carbon dioxide emissions are the main greenhouse gas responsible for 58.8% of global warming and climate change (World Bank, 2007a). CO2 emissions are global pollutants whose reduction may suffer from the “freerider” problem. Consequently, some authors have argued that CO2 emissions may not be appropriate for country or regional-specific studies. However, because of the paucity and questionable reliability of time series data for local pollutants such as SO2, many authors resort to using CO2 emissions per capita as the measure of environmental quality.

potentially affected by pollution the benefit of abatement increases. In addition to per capita real GDP and population density, we include energy consumption and trade openness as potential explanatory variables. Higher production and consumption activities demand larger inputs of energy, which in turn creates pollution that is more atmospheric contemporaneously, further degrading the environment. In addition, energy consumption is closely linked to the depletion of natural resources. In view of this, a priori, we expect the impact of increasing energy consumption on the environment to be negative. Regarding the impact of trade openness or trade liberalization on the environment, Antweiler et al. (2001) argued that there are three channels—scale, technique, and composition effects—through which trade openness can result in environmental improvements or deteriorations. The scale effect refers to the fact that trade liberalization increases market size, which presumably increases production and in turn increases emissions. The technique effect is deemed to reduce emissions because of import of efficient and environmental friendly technologies. The composition effect suggests that trade liberalization may reduce or increase emissions depending upon whether the country has comparative advantage in “cleaner” or “dirty” industries. Hence, the composition effect can have both positive and negative impacts. Given the different nature of the individual effects, the overall impact of trade openness on the environment is ambiguous—it depends on which effect is stronger and dominates the others. Based on the preceding discussions, we specify the following equation (in logarithms) to test the validity of the EKC hypothesis for the selected countries8: 2

3

lnC t ¼ β0 þ β1 lnY t þ β2 lnY t þ β3 lnY t þ β4 lnNGt þ β5 lnTRt þ β6 lnPDt þ εt

ð2Þ

where, C is CO2 emissions per capita; Y is per capita real GDP; NG is energy consumption per capita; TR is trade openness; PD is population density; and εt is the standard error term. Eq. (2) allows for testing several kinds of relations between emissions and income: For example, if β1 N 0, β2 b 0, and β3 = 0, the relationship between emissions and income is inverted U-shaped curve that confirms the existence of an EKC, and the monetary value representing the turning point is computed  .   −β 1 . If β1 b 0, β2 N 0, and β3 = 0, the relationship by9 τ ¼ exp 2β2

is U-shaped curve. If β1 N 0, β2 b 0, and β3 N 0 the relationship is an N-shaped curve, which suggests that, a second turning point exists, after which the environmental quality deteriorates again with increasing income. If β1 N 0, β2 b 0, and β3 b 0, the relationship is inverted N-shaped curve, and the turning point is computed10 by τ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  −β2  β22 −3β1 β3 . exp 3β 3

A monotonically increasing or decreasing relationship between income and environment is achieved if only β1 is statistically significant (negative or positive signed, respectively), whereas β2 and β3 remain statistically insignificant11. 3.2. Data Our empirical analysis covers the period from 1970 to 2010 for each country. The data on CO2 emissions were collected from Carbon Dioxide Information Analysis Center (CDIAC) and World Development Indicators (WDI 2012) (World Bank, 2012). The data on gross domestic product (GDP), energy consumption, trade, and population density 8 Several other authors, for example Miliimet et al. (2003), Harbaugh et al. (2002), Panayotou (1997), and Grossman and Krueger (1991) have used cubic models to estimate EKC. 9 See, Dinda (2004). 10 Plassmann and Khanna (2002) elaborate on the algebraic derivation of the turning points and precision in measurement. Also see, Taylor and Mann (1983). 11 For other possible forms of relationship, see Dinda (2004, p. 440–441).

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

were taken from World Development Indicators (World Bank, 2012). Carbon dioxide emissions (C) are measured as metric tons of carbon per capita. Real GDP (Y) is GDP per capita in constant 2000 dollars. Per capita energy use or consumption (NG) is measured in tons of oil equivalent per capita. Trade (TR) is calculated as the sum of exports and imports of goods and services measured as the percentage of GDP. Population density (PD) is measured in people per square kilometer.

3.3. Estimation methodology Table 1 reports the means and standard deviations of per capita CO2 emissions and per capita real GDP and, Fig. 1 shows their evolution during the period 1970–201012. From Table 1, it can be seen that in all the countries the standard deviation of GDP is higher than that of CO2 emissions and, that the mean values of the series are positive. A visual inspection of the evolution of CO2 emissions and income growth (Fig. 1) reveals that the variables tend to move together over time, which suggests that a long-run cointegration relationship is likely to be present. In addition, we observe some sort of inverted U-shaped curve for most of the countries. The plots for Brazil, Japan, Mexico, Nigeria, and South Africa seem to reveal some type of concavity in their income-pollution trajectory and, those for China, Egypt, and South Korea seem to indicate an increasing tendency to pollute as income rises. However, it is should be noted that even if a country shows increasing pollution level to coincide with rising income, the country may still not contradict the inverted U-shaped EKC proposition, since it may simply be on the rising segment of the curve. Since several other factors besides income growth may affect CO2 emissions, it would be grossly misleading to depend solely on cursory observation of evolution patterns to arrive at conclusions. Therefore, the study proceeds with an in-depth econometric investigation of these relationships. The hypothesized inverted U relationship between economic growth and the environment is a long-run relationship. To date, a number of univariate and multivariate techniques such as the Engle and Granger (1987) residual based approach, the full information maximum likelihood method of Johansen and Juselius (1990), and the fully modified OLS procedures of Phillips and Hansen (1990) have been employed to test long-run relationships. According to Narayan and Smyth (2005), these tests may be inappropriate when the sample size is relatively small. Given the important policy implications that may be drawn from EKC analysis, to avoid drawing incorrect inferences that may result in misleading policies, we use the Autoregressive Distributed Lag (ARDL) Bounds testing approach to cointegration based on unrestricted error correction model (UECM) to analyze the long-run relationships (see Pesaran et al., 2001). The error correction terms from the UECM are used to test for the direction of Granger causality and to conduct generalized variance decomposition analysis13.

12 To conserve on space, we report only the descriptive statistics for CO2 and GDP. Summary statistics for other variables used in estimation can be provided on request. 13 The decisive criterion for our selection of the ARDL technique is two-fold. First, unlike other cointegration techniques (e.g., Engle and Granger (1987) and Johansen and Juselius (1990)) that require all variables in the model to be integrated of the same order, the ARDL procedure can be applied irrespective of the order of integration of the variables (Pesaran and Pesaran, 1997, pp. 302–303). Since the approach does not depend on pretesting the order of integration of the variables, it eliminates the uncertainty associated with pretesting the order of cointegration, and one can concentrate on the more fundamental issue of cointegration and causality. Second, the technique can be applied to studies that employ relatively small or finite sample data sizes such as the present study. As Pesaran and Shin (1999) pointed out, the small sample properties of the ARDL approach are far superior to that of the Johansen and Juselius' (1990) cointegration technique. In addition, the technique can distinguish between dependent vis-à-vis independent variables and perform parameter estimates for both the short-run and long run simultaneously; this eliminates the problems that are normally associated with omitted variables and autocorrelation.

51

Table 1 Descriptive statistics for per capita CO2 emissions and per capita real GDP (in logarithms). Mean

Median

Minimum

Std. dev.

Obs.

0.814 8.459

−0.024 7.595

0.195 0.177

41 41

0.774 5.970

1.834 7.794

−0.059 4.806

0.499 0.939

41 41

0.307 6.987

0.320 7.042

0.951 7.589

−0.505 6.333

0.425 0.372

41 41

Japan per capita CO2 per capita real GDP

2.178 10.290

2.182 10.441

2.288 10.617

1.997 9.720

0.078 0.286

41 41

South Korea per capita CO2 per capita real GDP

1.669 8.745

1.738 8.839

2.477 9.694

0.522 7.598

0.593 0.669

41 41

Mexico per capita CO2 per capita real GDP

1.282 8.511

1.352 8.520

1.459 8.754

0.788 8.157

0.174 0.158

41 41

Nigeria per capita CO2 per capita real GDP

−0.482 24.331

−0.389 24.278

−0.017 25.173

−1.148 23.709

0.310 0.377

41 41

South Africa per capita CO2 per capita real GDP

2.167 8.083

2.184 8.078

2.338 8.242

1.914 7.974

0.113 0.071

41 41

Brazil Per capita CO2 Per capita real GDP

0.443 8.130

0.385 8.165

China per capita CO2 per capita real GDP

0.796 6.097

Egypt per capita CO2 per capita real GDP

Maximum

An unrestricted error correction representation of the ARDL framework of Eq. (2) can be written as:

Δ ln C t ¼ β0 þ

n X

β1i Δ ln C t−i þ þ

i¼1

n X i¼0

β2i Δ ln Y t−i þ

n n X X 3 þ β4i Δ ln ðY t−i Þ þ β5i Δ ln NGt−i þ i¼0 n X

þ

i¼0

n X

2

β3i Δ ln ðY t−i Þ

i¼0 n X

β6i Δ ln TRt−i

i¼0 2

β7i Δ lnPDt−i þ δ1 lnC t−1 þ δ2 ln Y t−1 þδ3 ð ln Y t−1 Þ

i¼0 3

þδ4 ð ln Y t−1 Þ þ δ5 lnNGt−1 þ δ6 ln TRt−1 þ δ7 ln PDt−1 þ ν t ð3Þ where νt is the standard error term. The terms βi, i = 1, 2,....., 7 signify the short-run error correction dynamics while the terms δi, i = 1, 2,....., 7 correspond to the long-run relationship. The ARDL testing procedure starts with the bounds test for the existence of cointegration relationship between the variables in the system. The F-test statistic is used to determine whether the variables are cointegrated by testing the joint significance of the lagged level coefficients. For example, in Eq. (3), where C is the dependent variable, the null hypothesis of no cointegration relationship, H0 : δ1 = δ2 = δ3 = δ4 = δ5 = δ6 = δ7 = 0 is tested against the alternative hypothesis, H1 : δ1 ≠ δ2 ≠ δ3 ≠ δ4 ≠ δ5 ≠ δ6 ≠ δ7 ≠ 0. In the presence of cointegration, one should fail to accept the null hypothesis. Bahmani-Oskooee and Nasir (2004) also suggested that there might be evidence of cointegration when the dependent variable is replaced by the other explanatory variables in the model. To this end, the F-statistics for the joint significance of lagged levels of the variable are calculated when the dependent variable is in turn C, Y, NG, TR, and PD. The test for normalization on the relevant dependent variable can be expressed as FC(C/Y, NG, TR, PD), FY(Y/C, NG, TR, PD), FNG(NG/C, Y, TR, PD), FTR(TR/C, Y, NG, PD) and FPD(PD/C, Y, NG, TR). The F-statistic, obtained by performing the bounds test, has a nonstandard distribution whose asymptotic critical value bounds are provided by Pesaran et al. (2001). Recently, Narayan and Smyth (2005) argued that the critical bounds provided by Pesaran et al. (2001) are inappropriate in small sample size and they regenerated

52

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

BRAZIL

CHINA

2

3

1

2

0 1 -1 0 -2 -1

-3

-4 1970

-2 1975

1980

1985

1990

CO2 per capita

1995

2000

2005

2010

1980

1985

1990

CO2 per capita

1.0

1.0

0.5

0.5

0.0

0.0

-0.5

-0.5

-1.0

-1.0

-1.5

-1.5

-2.0

-2.0

1995

2000

2005

2010

2005

2010

GDP per capita

JAPAN

1.5

1.5

1970

1975

GDP per capita

EGYPT

2.0

1970

-2.5 1975

1980

1985

1990

CO2 per capita

1995

2000

2005

2010

1970

1975

GDP per capita

1980

1985

1990

CO2 per capita

1995

2000

GDP per capita

MEXICO

KOREA 1.5

2

1.0 1 0.5 0

0.0 -0.5

-1

-1.0 -2 -1.5 -2.0 1970

-3 1975

1980

1985

1990

CO2 per capita

1995

2000

2005

2010

1

1

0

0

-1

-1

-2

-2

1980

1985

CO2 per capita

1990

1985

1990

1995

2000

2005

2010

GDP per capita

SOUTH AFRICA

2

1975

1980

3

2

-3 1970

1975

CO2 per capita

NIGERIA

3

1970

GDP per capita

1995

2000

GDP per capita

2005

2010

-3 1970

1975

1980

1985

1990

CO2 per capita

Fig. 1. Evolution of CO2 emissions and real GDP (normalized logarithmic values), 1970–2010.

1995

2000

GDP per capita

2005

2010

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

new sets of critical values for small samples ranging from 30 to 80 observations for the usual levels of significance14. Given the sample size of our study, we compare our computed F-statistics to the critical values of the F-statistics provided in Narayan (2005). If the computed F-statistic is higher than the appropriate upper bound of the critical value, the null hypothesis of no cointegration is rejected; if it is below the appropriate lower bound, the null hypothesis cannot be rejected, and if it lies between the lower and upper bounds, the result is inconclusive. Next, we chose the optimal lag orders of the variables using Akaike Information Criterion (AIC), and estimated the short-run and long-run models based on the selected optimal lag lengths. With the long-run relationship determined and confirmed, the UECM to be estimated is expressed as:

Δ ln C t ¼ ω0 þ

n X

n n X X 2 ω1i Δ lnC t−i þ þ ω2i Δ lnY t−i þ ω3i Δ ln ðY t−i Þ

i¼1 n n i¼0 n i¼0 X X X 3 þ ω4i Δ ln ðY t−i Þ þ ω5i Δ ln NGt−i þ ω6i Δ lnTRt−i ð4Þ i¼0 i¼0 n X þ ω7i Δ lnPDt−i þ φEC t−1 þ μ t

i¼0

i¼0

where, φ is the speed of adjustment parameter, and ECt − 1 is one-period lagged error correction term. All other variables are as defined previously. The coefficient on the error correction term indicates the speed of adjustment back to equilibrium following shock to the system and it should have a statistically significant negative sign. To ensure the goodness of fit of the model we performed diagnostic tests for serial correlation, functional form, normality, and heteroscedasticity. Furthermore, Pesaran et al. (2001) suggested testing for the stability of the estimated model's parameters using the cumulative sum (CUSUM) and cumulative sum of squares (CUSUMSQ) tests of recursive residuals developed by Browns et al. (1975). 4. Empirical results 4.1. Unit root tests The underlying assumption of the ARDL Bounds test is that the variables are either I(1) or I(0). In order to avoid spurious results, we performed unit root tests to ensure that the series are not integrated of order two or higher. We utilized the unit roots test procedure developed by Zivot and Andrews (1992) that allows breaks in the deterministic terms to test for trend break stationarity of the variables. The major advantage of the Zivot–Andrews procedure is that the timing of the structural break is determined endogenously rather than on a priori observation of the data that might introduce pretesting problems. The Zivot–Andrews unit root test results indicate that the variables are level non-stationary, I(1), and difference stationary, I(0).15 Given these results, we tested for cointegration between the variables using the ARDL bounds test approach.16 4.2. Bounds testing for cointegration results Table 2 reports the results of the bounds F-test statistics. The table demonstrates that with CO2 emissions as the dependent variable, the computed F-statistic exceeds the upper bound of the 10%, 5%, and 1% critical values, for all countries. Accordingly, we reject the null hypothesis of no cointegration among the variables and conclude that there is a long-run relationship among CO2 emissions and the determinants. In 14 The critical values tabulated by Pesaran et al. (2001) are based on sample sizes of 500 and 1000 observations respectively. 15 Results of the Zivot–Andrews unit root tests are available upon request. 16 The AIC and SBC suggest that one is the optimal lag length for all the countries in our sample.

53

addition, with real GDP (Y) and energy consumption (NG) as the dependent variables, we observe that the calculated F-statistics exceed the upper bound critical values at the 10% level and better for all countries except Egypt and Mexico. In the case of Egypt (energy consumption) and Mexico (real GDP), the computed F-statistics lie between the critical bounds, this suggests that their results are inconclusive. With trade openness (TR) and population density (PD) as the dependent variable, the results indicate the existence of a cointegration relationship for all countries except for the cases of Brazil, Egypt and Mexico (trade openness) and Japan, Korea and Nigeria (population density) where the calculated F-statistics lie between the critical bounds at the 10% level, meaning that their results are inconclusive. For such cases of inconclusiveness, Kremers et al. (1992) suggested that the error correction term is a useful way to establish cointegration. Having found that there is a cointegration relationship between CO2 emissions and the covariates; we estimated the ARDL model within an unrestricted error correction model (UECM) framework to confirm the long-term relationship17. This study selects the optimal number of lags in the UECM-ARDL model using the Akaike Information Criterion (AIC)18. The selected models are reported in the second column of Table 3. The table also reports the long-run parameter estimates, the coefficients of the lagged error correction terms (ECt − 1), as well as, the per capita income turning point(s) for each country19. Additionally, Table 3 shows that the coefficients of the ECt − 1 terms are all significantly negative and smaller than unity in absolute terms. This result provides evidence in favor of a cointegration relationship among the variables established by the bounds testing procedure. As explained earlier, the coefficient of the EC term represents the proportion by which the long-run disequilibrium in the dependent variable is corrected in each short-term period. For example, the coefficient of the EC terms are equal to − 0.665 , − 0.437, and − 0.342 for Egypt, Mexico and Japan respectively, meaning that the deviation from the long-term path of per capita CO2 emissions is corrected by 66.5% in Egypt, 43.7% in Mexico and 34.2% in Japan in each period. Interpretatively, following shock to the system it takes about one year for per capita CO2 emissions in Egypt to get back to the level predicted by its cointegration relationship with Y, NG, TR and PD, but the adjustment process takes about two years in Mexico while it takes about three years in Japan. As can be seen from Table 3, the coefficients of income (Y), income squared (Y2), and income cubed (Y3) in the regressions for Brazil, China, Egypt, Mexico, Nigeria and South Africa are positive, negative and positive respectively; and all are statistically significant at the 10% level or better. The positive–negative–positive coefficient pattern in these six countries suggests an N-shaped path between per capita CO2 emissions and per capita real GDP growth. Interpretatively, as per capita income rises, CO2 emissions also rise, until an income level at which they fall more slowly, after which they begin to rise again. The Nshaped pattern is not consistent with the postulated inverted Ushaped EKC. The estimated per capita income (measured in logarithms) turning point at which CO2 emissions start to decline are 22.083, 17.050, 16.585, 21.344, 32.855, and 22.963 for Brazil, China, Egypt, Mexico, Nigeria and South Africa respectively. Compared to the mean value of per capita real GDP reached by each country during the period of analysis (see Table 1 for the descriptive statistics), the predicted level of per capita income where the turning points occur for these countries are relatively high and out of sample data size. Given these findings, we conclude that the conventional EKC hypothesis does not hold for Brazil, China, Egypt, Mexico, Nigeria, and South Africa over the study period.

17 This study uses annual data. To model the UECM-ARDL model with more explanatory power and simultaneously eliminate autocorrelation of the error terms while keeping reasonable degree of freedom, we set the maximum lags in the model to three (imax = 3). 18 The AIC suggests that one is the optimal lag length for all the countries in our sample. 19 The results of the short-run parameter estimates can be provided upon request.

54

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

Table 2 ARDL bounds test results for cointegration.

FC(C/Y, NG, TR, PD) FY(Y/C, NG, TR, PD) FNG(NG/C, Y, TR, PD) FTR(TR/C, Y, NG, PD) FPD(PD/C, Y, NG, TR)

Brazil

China

Egypt

Japan

South Korea

Mexico

Nigeria

South Africa

7.577* 5.657* 3.399*** 3.312 (a) 4.910**

7.329* 4.103** 4.319** 3.970*** 3.961***

4.271** 4.101** 2.989 (a) 3.387 (a) 3.973***

5.999* 5.963* 4.411** 4.021** 2.991 (a)

7.693* 3.693*** 3.401*** 3.399*** 2.901**

6.844* 2.964 (a) 3.497*** 3.228 (a) 4.991**

4.467** 3.681*** 3.886*** 4.199** 2.477 (a)

8.401* 4.211** 14.721* 4.167** 6.222*

Notes: 1. *, ** and *** are respectively the 1%, 5% and 10% of the significant level. 2. 1% CV [3.967, 5.455], 5% [2.893, 4.000] and 10% [2.427, 3.395] Case II restricted intercept and no trend. 3. (a) denotes the value falls between critical value bands.

In the case of South Korea, the coefficient of Y is negative, that of Y2 is positive and that of Y3 is negative, and each coefficient is statistically significant. The negative–positive–negative coefficient pattern suggests an inverted (reversed) N-shaped trajectory between per capita CO2 emissions and per capita real GDP growth for South Korea. An inverted N-curve can potentially provide empirical support in favor of a bellshaped evolution for CO2 emissions relative to income growth if the estimated per capita income associated with the first and second turning points lie within the sample minimum and maximum income values. As can be seen from Table 3, South Korea's inverted N-shape has the first turning point at per capita income of 8.071 (in logarithms) and the next at per capita income of 8.746 (in logarithms). Both estimated turning points are well within sample size minimum value (7.598) and maximum value (9.694). Given this finding, we conclude that the conventional U-shaped EKC exists for South Korea over the sample period. This result further suggests that the reduction of environmental degradation from economic growth may not be temporary and CO2 emissions will continue to fall at the threshold level of 8.746. For Japan, the coefficient of Y is positive and significant, Y2 is negative and significant, and Y3 is positive but not statistically significant. This result provides weak support in favor of the inverted U-shaped EKC. We interpret this as weak evidence in support of the EKC because the positive coefficient of the Y3 term, even though not statistically

significant, is not zero. To confirm the inverted U relationship the estimated income turning point should be within the sample size. Upon comparing the estimated per capita income turning point value of 10.293 (in logarithms) to the sample size minimum value (9.720) and maximum value (10.617) reached by Japan during the period of analysis we find that the estimated turning point is within sample period data size, thereby confirming that the relationship between economic growth and CO2 emissions in Japan followed an inverted U path over the study period. Energy consumption has a positive and significant effect on CO2 emissions in all the countries. For example, a one percent increase in energy consumption, ceteris paribus, will increase CO2 emissions by 1.28% in Brazil and will increase it by 4.17% in Nigeria. The finding of a positive impact of energy consumption on CO2 emissions is in line with the view that energy consumption is the main source of CO2 emissions (environmental degradation). Our result is consistent with the empirical evidence found by other studies, for instance, Odhiambo (2011), Halicioglu (2009), Liu (2005), Ang (2008), and Jalil and Mahmud (2009). Moving on to trade openness, the results for Mexico, Nigeria, and South Africa indicate a positive and significant impact of trade openness on CO2 emissions. This suggests that increased trade openness contributes to worsening environmental conditions in these three countries.

Table 3 Estimated long-run coefficients for the ARDL model. Dependent variable: ln C [CO2 emissions per capita] AIC-based ARDL ln Y Brazil

(1, 1, 0, 0, 0,0)

China

(1,1, 0, 0, 0,0)

Egypt

(0, 0,0,1,0,0)

Japan

(0,0,0,0,0,1)

South Korea (0, 1, 1, 0, 1,0)

Mexico

(0, 1, 1, 0, 0,0)

Nigeria

(0, 0, 1, 1, 1,0)

South Africa (0, 0, 0, 0,1,1)

(ln Y)2

(ln Y)3

ln NG

ln TR

ln PD

ECt − 1

τ* turning point (in logs)

Diagnostic test statistics

Adj. R2: 0.82; SER: 0.02; DW:2.64; JB: 1.12; RESET(1): 0.28; ARCH (2): 0.20; BG (2): 0.27 3.489 −0.615 0.605 1.187 −0.032 2.724 −0.614 17.050 Adj. R2: 0.84 SER: 0.01; DW: 2.29; JB: 0.08; RESET (1): 0.00 ARCH (2):0.14 (6.227)* (−2.247)** (3.552)* (8.172)* (−1.975)*** (3.319)* (−3.116)* BG(2): 0.09 9.605 −1.710 0.630 0.745 0.414 35.098 −0.655 16.585 Adj. R2: 0.52; SER: 0.07; DW: 2.17; JB: 0.77 RESET (1): 1.61 ARCH(2):2.50 (2.184)** (−2.189)** (2.196)** (2.575)** (1.401) (2.868)** (−2.625)* BG(2):0.35 2.798 −0.601 0.421 0.467 −0.411 0.012 −0.342 10.293 Adj. R2: 0.84; SER: 0.01; DW: 2.14; JB: 0.08 RESET(1):0.91 ARCH(2):1.20 (4.966)* (−4.119)* (1.047) (3.800)* (−2.792)** (4.117)* (−2.791)* BG(2):0.74 −9.121 2.103 −0.012 0.938 −0.026 0.189 −0.659 8.746 and 8.071 Adj.R2: 0.77; SER: 0.03; DW: 2.17; JB: 0.62; RESET(1): 1.41; ARCH (2): 0.20 (−4.431)* (4.381)* (−3.335)* (7.569)* (−0.782) (1.349) (−2.971)** BG (2): 0.09 3.275 −0.535 0.566 0.732 0.044 66.493 −0.437 21.344 Adj.R2: 0.78; SER: 0.44; DW: 2.26; JB: 0.08; RESET (1): 0.14; ARCH (2):0.01; (4.152)* (−2.989)** (2.961)** (4.007)* (2.207)** (2.681)** (−3.526)* BG (2): 0.40 4.728 −0.677 0.405 4.172 0.208 0.176 −0.334 32.855 Adj.R2:0.66; SER:0.59; DW: 2.24; JB: 2.61; RESET(1): 0.91; ARCH (2): 1.20; (4.762)* (−2.416)** (2.389)** (2.916)** (2.989)** (0.923) (−2.905)** BG(2): 0.74 4.895 −0.781 0.659 1.192 1.341 0.401 −0.559 22.963 Adj. R2: 0.84; SER: 0.01; DW: 3.01; JB: 0.08 RESET(1):0.24; ARCH (2): 0.14 (3.772)* (−3.416)* (3.389)** (3.916)** (2.964)** (1.021) (−6.971)** BG(2): 0.05

5.034 (12.225)*

−0.813 (−4.217)*

1.042 (2.210)**

1.277 (6.631)*

−0.037 (−2.557)**

49.923 (4.962)*

−0.434 (−3.173)*

22.083

Notes: (*), (**) and (***) are respectively the 1%, 5% and 10% of the significant level. Asymptotic t-statistics in parentheses. Adj. R2 is adjusted R square. DW is Durbin–Watson statistic. SER is standard error of the regression. JB is Jarque–Bera test for normality of the regression residuals. RESET is Ramsey F-statistic for omitted variables. BG is Breusch–Godfrey F-statistic. ARCH is Autoregressive Conditional Heteroskedasticity F-statistic. In the RESET, BG and AARCH tests, numbers in parentheses are the lag lengths.

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

Brazil

Brazil

55

China

China

16

1.4

16

1.4

12

1.2

12

1.2

8

1.0

8

0.8

4

1.0 0.8

4

0.6

0.6

0

0 0.4

0.4 -4

-4

0.2

-8 -12

-8

0.0

-0.2

-12

-0.2

-0.4

-16 1985

1990

1995

CUSUM

2000

0.2

0.0

2005

1985

1990

1995

CUSUM of Squares

5%Significance

Egypt

2000

-16 1980

2005

5% Significance

1985

1990 CUSUM

Egypt

1995

2000

2005

-0.4 1980

5% Significance

Japan

1995

2000

2005

5% Significance

Japan

1.4

20

1.4

12

1.2

15

1.2

8

1.0

10

1.0

0.8

1990

CUSUM of Squares

16

4

1985

0.8

5

0.6

0.6 0

0

0.4

0.4 -4

-5

0.2

-8 -12

1990

1995

CUSUM

2000

2005

1985

1990

5%Significance

1995

2000

CUSUM of Squares

Korea

2005

1.4

15

1.2

-0.4 1980 1985 1990 1995 2000 2005 2010

5% Significance

CUSUM

1985

1990

1995

2000

2005

2010

5% Significance

CUSUM of Squares

Mexico 1.4

15

1.2 10 1.0

1.0

10

1980

5% Significance

Mexico

Korea

20

-0.2

-20

-0.4 1985

0.0

-15

-0.2

-16

0.2

-10

0.0

5

0.8

0.8

5 0.6

0.6

0

0

0.4

0.4 -5

-5

0.2

-10

0.2 0.0

0.0 -10

-15

-0.2

-0.2

-20 1980

1985

1990 CUSUM

1995

2000

2005

2010

-0.4 1980

5%Significance

1990

1995

CUSUM of Squares

NIGERIA 15

-15 1985

2000

2005

-0.4 86 88 90 92 94 96 98 00 02 04 06 08 10

2010

CUSUM

5% Significance

Nigeria

86 88 90 92 94 96 98 00 02 04 06 08 10 CUSUM of Squares

5% Significance

South Africa 15

1.4

5% Significance

South Africa 1.4 1.2

1.2 10

10 1.0

1.0 5

5

0.8

0.8 0.6

0.6 0

0 0.4

0.4 -5

0.2

-5

0.2

0.0

0.0 -10

-10 -0.2

-0.2 -0.4

-15 86 88 90 92 94 96 98 00 02 04 06 08 10 CUSUM

5%Significance

-0.4

-15 86 88 90 92 94 96 98 00 02 04 06 08 10 CUSUM of Squares

5% Significance

86 88 90 92 94 96 98 00 02 04 06 08 CUSUM

5% Significance

86 88 90 92 94 96 98 00 02 04 06 08 CUSUM of Squares

5% Significance

Fig. 2. Plots of the CUSUM and CUSUMSQ stability tests statistics.

This finding is in line with the Pollution Haven Hypothesis (PPH) and the works of Copeland and Taylor (2004), Cole (2004), and GamperRabindran and Jha (2004) who indicated that the pollution haven effect could exist in developing countries20. In the case of South Korea, the coefficient of trade openness is negative but not statistically significant. In

20 The pollution haven hypothesis is the hypothesis that a reduction in trade barriers will cause a shift of polluting industries to countries with weaker regulations.

contrast, for Brazil, China, and Japan trade openness has significant negative effects on per capita CO2 emissions, meaning that increased trade liberalization contributes to reducing per capita CO2 emissions, which in turn improves the environment. This finding is consistent with Shahbaz et al. (2012) who found that trade openness improves environmental quality by enhancing the capacity of the country to implement advanced technology to increase domestic production. With respect to population density, the results are mixed. For Brazil, China, Egypt, Japan, and Mexico, increasing population density has a

56

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

positive and significant influence on CO2 emissions, suggesting that more inhabitants per square kilometer leads to more environmental degradation in the long-run. For example, a 1% rise in population density, ceteris paribus, will increase CO2 emissions by 49.9% in Brazil, 2.72% in China, 35.9% in Egypt, and 66.49% in Mexico. In the case of South Korea, Nigeria, and South Africa, population density has a positive but statistically insignificant impact on CO2 emissions. The various specification and diagnostic tests applied to the models, reported in the last column of Table 3, appear significant and robust, indicating that the estimated models fit the data adequately. The RESET (Regression Specification Test) statistics reveal no serious omission of variables, indicating the correct specification of the models. The ARCH (Autoregressive Conditional Heterosckedasticity) tests suggest that the errors are homoscedastic independent of the regressors. The BG (Breusch–Godfrey) tests reveal no significant serial correlation in the disturbances of the error term. The JB (Jarque–Bera) statistics suggest that the disturbances of the regressors are normally distributed. The adjusted R square (Adj. R2) suggest that more than 50% of the variation in per capita CO2 emissions in any country is explained by the explanatory variables. Our sample covers a period over which countries experienced a number of major events that may affect the variables included in the model21. Parameter stability over the sample period is therefore of vital importance to ensure reliability of policy simulations based on the model. To test for parameter constancy, we applied the CUSUM and CUSUMSQ test statistics to the recursive residuals of the models. Plots of the CUSUM and CUSUMSQ test statistics (presented in Fig. 2) reveal no evidence of parameter instability in the models, since the statistics do not cross the 5% critical bounds. Stability of the estimated parameters suggests that the models can be considered stable enough for forecasting and policy analysis. (See Fig. 3.)

4.3. Variance decompositions of CO2 emissions and GDP growth The bounds test procedure established that the variables in the ARDL model of each country are cointegrated and that the estimated parameters are stable. The existence of cointegration among the variables suggests that there must be Granger causality in at least one direction to provide the necessary dynamics. However, detecting Granger causality is strictly restricted to within the sample period and does not allow us to gage the relative strength of the causality among the series beyond the sample period. In order to ascertain the dynamic impacts of the Granger causality relations beyond the sample period we utilized the generalized impulse response approach proposed by Pesaran and Shin (1999) that does not require orthogonalization of shocks and is invariant to the ordering of the variables within the context of a VAR framework (Payne, 2002). The variance decomposition indicates how much of the forecast error variance of each variable can be explained by exogenous shocks (innovations) to the variables in the same VAR model. Innovations to an individual variable can affect both own changes and changes in the other variables in the system. As we are relatively more interested in the relationship between CO2, real GDP, and energy consumption, we shall focus more on the variance decomposition results pertaining to these variables and analyze how much of the forecast error variance of CO2, real GDP and energy consumption is explained by each variable in the model. The results of the variance decomposition analysis presented in Table 4 show that over a 10 year horizon, the relative contribution of innovations in CO2 emissions to GDP forecast error variance dominates 21 These include the oil shocks in 1973 and 1979 and the counter-shock in 1986; the implementation of NAFTA in 1994; the 1994 Marrakech Agreement leading up to the establishment of WTO in 1995; the 1997 Kyoto protocol agenda and subsequent negotiations by the developed and developing countries (including China, Korea, and South Africa) leading up to the 2009 Copenhagen Summit; the 2011 Tsunami and earthquake disasters; the recent global economic recession and financial crisis.

the relative contribution of innovations in GDP to CO2 emissions in five countries: Brazil, Egypt, Japan, Nigeria and South Africa, suggesting that CO2 emissions Granger-cause GDP growth in these five countries, both in the short-run and the long-run. The implication of this finding is that it would not be possible to reduce emissions without undermining long-run economic growth as reduction in CO2 emissions can cause GDP to decline. Our result is consistent with the empirical evidence found by Menyah and Wolde-Rufael (2010) for South Africa, Soytas and Sari (2009) for Turkey and Ang (2008) for Malaysia, but contradicts Odhiambo (2011) for South Africa, and Soytas et al. (2007) for the United States. In contrast, in China and South Korea, the share of CO2 emission variance explained by innovations in GDP is larger than the share of GDP forecast error variance explained by innovations in CO2 emissions; this suggests that GDP growth Granger-causes CO2 emissions in these two countries. Our finding is consistent with those of Zhang and Cheng (2009) for China. An implication of this finding is that economic growth is not compatible with environmental improvements in these two countries. The results for Mexico suggest that CO2 emissions and economic growth are complements, since an almost equal proportion of their forecast error variance is explained by each other. For instance, at the 10-period horizon innovations from economic growth explain around 24% of CO2 variations while the forecast error variance of CO2 emissions explains around 22% of the forecast error variance of economic growth. This finding also suggests that there is a bi-directional causality between CO2 emissions and economic growth. That is, an increase in economic growth leads to an increase in CO2 emissions and vice versa. This bi-directional causal relationship implies that economic growth and CO2 emissions (environmental quality) are interrelated. If this is true, then the design of economic policies directed at fostering growth should include considerations of the direct impact of higher economic growth on the environment and the feedback effects of environmental degradation and resource depletion on long-run output growth. The relative contribution of innovations in energy consumption to GDP forecast error variance dominates the relative contribution of innovations in GDP growth to energy consumption forecast error variance in five countries: Brazil, China, Japan, Egypt and South Africa, which suggests that energy consumption Granger-causes GDP growth. Our result is consistent with the empirical evidence found by Stern (2000), WoldeRufael (2009) for a group of African countries, Wolde-Rufael (2004) for Shanghai, China, Soytas and Sari (2003) for Japan, Germany and Turkey, and Odhiambo (2011) for South Africa. The implication of this finding is that energy consumption and economic growth are causally related, and energy conservation measures may adversely affect economic growth. In contrast, for South Korea, Mexico, and Nigeria, the relative contribution of innovations in GDP growth to energy consumption forecast error variance dominates the relative contribution of innovations in energy consumption to GDP variations. This suggests that real GDP growth Granger-causes energy consumption in these three countries. The Granger causality running from GDP growth to energy consumption indicates that increased economic activity (production and consumption) leads to higher inputs of energy. Menyah and Wolde-Rufael (2010) suggested that this might be an indication of inefficient and excessive energy consumption. They pointed out that “it is not uncommon for some developing countries that are endowed with abundant energy resources to experience inefficiency in their energy use” (Menyah and Wolde-Rufael, 2010, p. 1379). In light of this finding, it is necessary for the decision makers to have an integrated energy policy aimed at increasing energy use efficiency by lowering energy consumption for a given level of economic growth. Table 4 also shows that in Egypt and South Korea, the relative contribution of innovations in energy consumption to CO2 forecast error variance dominates the relative contribution of innovations in CO2 to energy consumption variations, which suggests a unidirectional causal relationship running from energy consumption to CO2 emission. This

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

57

Mexico, Nigeria, and South Africa show that the relative contribution of innovations in CO2 to energy consumption forecast error variance dominates the relative contribution of innovations in energy consumption to CO2 variations. This indicates a unidirectional causal relationship running from CO2 emission to energy consumption. The direction of this

result is consistent with the view that energy consumption is the primary cause of CO2 emissions (environmental degradation). This finding suggests that reducing energy consumption, especially the consumption of fossil fuels, is a viable option that can help curb CO2 emissions in these countries. In contrast, the results for Brazil, China, Japan,

Brazil Response of C to Y

Response of C to NG

Response of C to PD

Response of C to TR

.08

.08

.08

.08

.04

.04

.04

.04

.00

.00

.00

.00

-.04

-.04

-.04

-.04

-.08

-.08 1

2

3

4

5

6

7

8

9

10

-.08 1

2

3

4

5

6

7

8

9

10

-.08 1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

China Response of C to Y

Response of C to NG

Response of C to TR

Response of C to PD

.10

.10

.10

.10

.05

.05

.05

.05

.00

.00

.00

.00

-.05

-.05

-.05

-.05

-.10

-.10

-.10

-.10

-.15

-.15 1

2

3

4

5

6

7

8

9

10

-.15

-.15 1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

Egypt Response of C to Y

Response of C to NG

Response of C to TR

Response of C to PD

.08

.08

.08

.08

.04

.04

.04

.04

.00

.00

.00

.00

-.04

-.04

-.04

-.04

-.08

-.08 1

2

3

4

5

6

7

8

9

10

-.08

-.08 1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

Japan Response of C to Y

Response of C to NG

Response of C to TR

Response of C to PD

.04

.04

.04

.04

.02

.02

.02

.02

.00

.00

.00

.00

-.02

-.02

-.02

-.02

-.04

-.04

-.04

-.04

-.06

-.06 1

2

3

4

5

6

7

8

9

10

-.06

-.06 1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

Fig. 3. Plots of the generalized impulse response functions. [Response to +/−2 S.D. innovations].

1

2

3

4

5

6

7

8

9

10

58

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

Korea Response of C to NG

Response of C to Y

Response of C to TR

.10

.10

.10

.10

.05

.05

.05

.05

.00

.00

.00

.00

-.05

-.05

-.05

-.05

-.10

-.10

-.10

-.10

-.15

-.15 1

2

3

4

5

6

7

8

9

10

-.15

-.15 1

2

3

4

5

6

7

8

9

10

Response of C to PD

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

Mexico Response of C to NG

Response of C to Y

Response of C to TR

Response of C to PD

.04

.04

.04

.04

.02

.02

.02

.02

.00

.00

.00

.00

-.02

-.02

-.02

-.02

-.04

-.04

-.04 1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

-.04 1

2

3

4

5

6

7

8

9

1

10

2

3

4

5

6

7

8

9

10

Nigeria Response of C to Y

Response of C to NG

Response of C to TR

Response of C to PD

.4

.4

.4

.4

.2

.2

.2

.2

.0

.0

.0

.0

-.2

-.2

-.2

-.2

-.4

-.4 1

2

3

4

5

6

7

8

9

10

-.4 1

2

3

4

5

6

7

8

9

10

-.4 1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

8

9

10

South Africa Response of C to Y

Response of C to PD

Response of C to TR

Response of C to NG

.12

.12

.12

.12

.08

.08

.08

.08

.04

.04

.04

.04

.00

.00

.00

.00

-.04

-.04

-.04

-.04

-.08

-.08 1

2

3

4

5

6

7

8

9

10

-.08 1

2

3

4

5

6

7

8

9

10

-.08 1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

Fig. 3 (continued).

causal relationship implies that energy consumption may not necessarily have to be reduced to mitigate the adverse effects of CO2 emissions on environmental quality. Concerning the contributions of innovations to trade openness and population density, Table 4 shows that in South Africa major shocks to CO2 emissions in any period comes from its own innovations and this is closely followed by innovations to trade openness that reached 19%

in the final period. Concerning GDP growth and energy consumption variations, innovations to these variables in any period is largely accounted for by innovations in trade openness (reaching 25% and 18%, respectively by the 10-period horizon.) Forecast error variance of population density made the least contribution (less than 9% in any period) to the forecast error variance of CO2, GDP growth and energy consumption.

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

In Japan, most of the variance of CO2 emissions in any period is accounted for by its own innovations and by innovations to trade openness and population density (around 20% and 19% respectively). In addition, in Japan and South Korea, perturbation in population density accounted for around 25% and 13% respectively of the forecast error variance of GDP growth in the long-run (5 to 10 years). Noticeably, in Japan, sustained shock to trade openness accounted for about 20% of CO2 emissions forecast error variance and for about 30% of energy consumption variations in the long run. The same happens for Mexico. In Mexico, while innovations to trade openness and population density had relatively little impact in explaining CO2 emissions, GDP growth and energy consumption variations in the short-run (1 to 5 years), their weight picked up considerably in the long-run. In Egypt, innovations to trade openness accounted for the second largest share of energy consumption variations in the long run. In Nigeria, innovations to population density made the least contribution to the forecast error variance of CO2, GDP growth and energy consumption in the long run. 5. Policy implications The results of our analysis showed that except for Japan and South Korea, CO2 emissions–economic growth trajectory in all the sampled countries follows an N-shaped curve and the estimated turning points are nowhere near the income range associated with maximum pollution on the conventional EKC. A literal interpretation of these findings would imply substantial increases in pollution during the next several decades. Policy makers however should be mindful that a persistent decline in environmental quality might exert negative externalities on the economy through a reduction in human health and thereby reduce productivity and economic growth in the long-term (Ang, 2008). Indeed, empirical research suggests that pollution costs are already quite high in these countries. For example, the World Bank (2007b) estimated that the total health cost associated with outdoor air pollution in urban areas of China in 2003 was between 157 and 520 billion Chinese yuan, accounting for 1.2–3.3% of China's gross domestic product. Recognizing the high costs of the damage to human health caused by emissions, adopting explicit national policies for environmental protection are matters of high urgency. Given the likelihood that some of these countries will not reach the estimated per capita income turning points for several decades to come, it is even more imperative that the governments should not see economic growth as a solution for their environmental problems. In other words, it would be naïve of these policy makers to take the EKC postulate at face value and bet on reaching a turning point that might only be a statistical artifact. The adoption of national policy initiatives to mitigate environmental degradation and resource depletion in the earlier stages of economic development may be justified on purely economic grounds. In the same vein, current prevention may be far less costly than plans to clean up accumulated damages. The results of our analysis provide hardly any evidence that the environmental pollution problem will solve itself any time soon; thus tighter and concentrated environmental policy regimes may be needed in order to direct the environment–economic growth nexus toward a downward trend. The finding of a long-run positive relationship between energy consumption and CO2 emissions in all of these countries, and the evidence that CO2 emissions and energy consumption Granger-cause GDP growth, creates a policy quandary for the decision makers: how to coordinate energy consumption, carbon emissions, and economic growth. Viewed differently, the positive relationship between energy consumption and CO2 emissions could imply that the growth (industrialization) processes in these countries are highly pollution intensive. Hence, to ensure environmental sustainability without limiting growth, there may still be a need to implement a wide range of environmental policies that would induce industries to adopt new technologies that could help mitigate environmental pollution without sacrificing economic growth. Importantly, our finding that energy consumption Granger-

59

causes economic growth (Brazil, China, Egypt, Japan, and South Africa) does not necessarily mean that energy conservation will hinder economic growth. In fact, a reduction in energy consumption due to improvements in energy efficiency may raise productivity, which in turn may spur economic growth. Therefore, the implementation of energy efficient policies that transform energy consumption from lowefficiency and high-pollution pattern to high-efficiency and lowpollution pattern may stimulate rather than hinder economic growth (Belke et al., 2011; Costantini and Martini, 2010). The significantly positive relationship between population density and environmental degradation observed in some of these countries suggests that policy alone on demographic control may not solve the environmental problem. In combination with sound economic policy, it can have far-reaching effects on the demand for a higher environmental quality. The mixed evidence on the effects of trade openness on environmental quality is consistent with a priori expectation that the overall effect of increased trade liberalization on national welfare can be ambiguous, but with the proper amount of environmental policy intervention as trade is further liberalized, there can be a net gain from trade. 6. Conclusion The objective of this study has been to examine the determinants of per capita CO2 emissions in Brazil, China, Egypt, Japan, South Korea, Mexico, Nigeria and South Africa over the period 1970–2010 using a model that incorporates energy consumption, trade openness and population density into the equation of the environmental Kuznets curve (EKC). The empirical analysis was carried out using two methodologies: the Autoregressive Distributed Lag (ARDL) bounds test to cointegration, which is valid regardless of whether a series is I(0) or I(1), or integrated of any arbitrary order; and the generalized variance decomposition analysis that is invariant to the ordering of the variables in the VAR system. The bounds F-test confirmed cointegration relationships among the variables for all the selected countries based on the results of the estimated coefficients of the lagged error correction terms. CUSUM and CUSUMSQ parameter constancy tests confirmed that the models are stable over the period of analysis. However, the significance and signs of variables in the cointegration vector space was diverse. Our results indicated that income growth and energy consumption are statistically significant in all the countries, providing evidence that income growth and energy consumption are main factors in increasing CO2 emission in the long-run. For trade openness and population density, the results are mixed. Our results also indicated that the signs of the estimated long-run coefficients of income, squared income, and income cubed satisfy the inverted U-shaped EKC for only two countries: Japan and South Korea. In all the other six countries, the long-run relationship between economic growth and CO2 emissions follows an N-shaped trajectory and, the estimated tuning points are out of the sample data size. The results from our variance decomposition analysis indicated unidirectional causality running from CO2 emission to output growth in Brazil, Japan, Egypt, Nigeria, and South Africa; unidirectional causality running from output growth to CO2 emissions in China and South Korea; and bidirectional causality between economic growth and CO2 emissions in Mexico. The results also indicated that sustained shocks to energy consumption have long lasting impacts on output growth in most of the sampled countries. Given these findings, it would be ill advised for the policy decision makers to adopt the EKC postulate as the conceptual basis for policies favoring economic growth unconditionally. In other words, given the probability of high environmental damage, costs to human health, productivity and national output, the high cost of improving the environment after damage has occurred, and, the likelihood that irreversible environmental damage may have been caused even before the implied turning points are reached, it is imperative that the governments enact

60

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

Table 4 Generalized forecast error variance decomposition at various horizons. Δ ln Yt

Δ ln NGt

Δ ln TRt

Δ ln PDt

China

Δ ln Ct

Horizon 1 2 5 8 10

Dependent variable Δ ln Ct 100.00 0.00 94.79 2.95 75.90 9.31 61.60 22.86 57.28 26.14

0.00 1.12 7.26 8.80 8.15

0.00 0.50 0.55 0.48 2.12

0.00 0.64 6.98 6.27 6.30

8.35 5.88 13.51 11.54 14.06

1 2 5 8 10

Dependent variable Δ ln Yt 32.86 65.80 1.33 24.20 62.78 9.43 17.11 35.53 43.59 14.99 25.47 53.57 13.77 22.37 58.04

0.00 2.58 3.49 5.31 5.24

0.00 0.01 0.27 0.67 0.58

1.79 1.18 8.99 14.18 14.85

0.00 7.77 16.84 15.82 19.28

1 2 5 8 10

Dependent variable Δ ln NGt 60.02 0.00 39.98 63.70 4.54 29.93 70.48 9.01 10.19 56.64 22.53 9.24 43.95 22.40 18.52

0.00 0.72 1.35 2.11 6.27

0.00 1.12 8.97 9.48 8.85

Dependent variable Δ ln Ct 100.00 0.00 0.00 83.26 1.24 12.28 60.63 7.02 22.01 54.71 6.42 26.86 49.67 6.81 31.86

0.00 0.37 3.99 6.49 6.64

0.00 2.85 6.34 5.52 5.02

Japan 1 2 5 8 10

Dependent variable Δ ln Ct 100.00 0.00 81.34 4.51 55.34 5.41 51.92 5.93 52.35 5.95

0.00 0.83 1.54 3.01 3.05

0.00 0.14 19.59 20.28 20.09

0.00 13.18 18.13 18.87 18.56

1 2 5 8 10

Dependent variable Δ ln Yt 5.55 94.46 0.00 9.40 71.28 13.11 23.47 44.06 14.17 16.18 30.54 35.34 15.35 28.58 39.58

0.00 6.02 15.37 15.36 14.09

0.00 0.20 2.94 2.58 2.39

1 2 5 8 10

Dependent variable Δ ln Yt 60.72 22.44 53.17 30.37 43.84 27.58 40.24 25.65 38.94 22.12

0.02 1.98 7.79 7.85 7.53

8.93 5.90 6.35 7.36 6.63

7.90 8.57 14.43 18.90 24.77

1 2 5 8 10

Dependent variable Δ ln NGt 5.50 0.00 94.50 4.53 0.82 91.48 5.27 1.23 74.09 8.38 0.94 65.99 7.47 0.74 68.64

0.00 1.72 10.65 14.63 14.86

0.00 1.45 8.75 10.06 8.30

1 2 5 8 10

Dependent variable Δ ln NGt 72.25 22.43 27.68 50.80 30.37 20.23 35.34 27.58 14.15 34.62 25.65 14.12 39.36 22.12 12.93

0.00 8.15 28.72 30.48 27.80

0.00 3.42 4.71 5.27 4.76

South Korea 1 2 5 8 10

Dependent variable Δ ln Ct 100.00 0.00 0.00 93.03 0.16 5.56 76.09 4.44 14.81 52.49 22.49 18.73 40.72 34.86 16.53

0.00 1.07 3.92 4.45 3.07

0.00 0.18 0.74 1.85 4.81

Mexico 1 2 5 8 10

Dependent variable Δ ln Ct 100.00 0.00 0.00 61.49 30.12 6.77 59.08 17.29 20.98 51.92 19.24 22.72 48.71 24.25 20.98

0.00 0.06 0.44 3.73 3.63

0.00 1.56 2.21 2.39 2.43

1 2 5 8 10

Dependent variable Δ ln Yt 68.07 30.57 39.30 50.50 19.91 61.38 9.30 74.55 6.04 77.12

1.01 3.29 2.52 0.94 0.84

0.00 5.56 6.64 4.12 2.85

0.35 1.35 9.56 11.09 13.15

1 2 5 8 10

Dependent variable Δ ln Yt 39.51 59.87 0.00 25.20 72.79 0.02 23.72 53.16 10.40 22.72 51.41 12.48 22.32 49.51 14.13

0.00 0.69 9.30 9.04 8.36

0.62 1.32 3.43 4.54 5.68

1 2 5 8 10

Dependent variable Δ ln NGt 72.50 0.00 27.50 58.86 8.24 25.21 32.91 31.81 29.55 13.49 58.13 20.75 8.16 67.09 16.31

0.00 4.84 3.01 2.93 1.70

0.00 2.84 2.71 4.71 6.73

1 2 5 8 10

Dependent variable Δ ln NGt 36.19 21.60 27.49 29.10 49.74 13.25 33.12 35.86 22.66 30.29 33.07 24.94 28.57 36.52 23.20

12.72 6.12 4.70 7.48 7.27

2.01 1.80 3.67 4.22 4.44

Nigeria 1 2 5 8 10

Dependent variable Δ ln Ct 100.00 0.00 0.00 95.15 3.75 0.16 73.70 8.16 10.56 73.69 8.35 10.64 72.65 9.02 10.46

0.00 0.11 4.45 4.26 4.78

0.00 0.84 3.14 3.06 3.10

South Africa 1 2 5 8 10

Dependent variable Δ ln Ct 100.00 0.00 0.00 95.24 2.52 0.42 80.60 4.83 6.06 58.95 8.26 12.45 48.44 10.47 14.20

0.00 1.53 6.04 14.31 18.73

0.00 0.28 2.46 6.03 8.16

1 2 5 8 10

Dependent variable Δ ln Yt 68.07 30.57 39.30 50.50 19.91 61.38 9.30 74.55 6.04 77.12

0.00 5.56 6.64 4.12 2.85

0.35 1.35 9.56 11.09 13.15

1 2 5 8 10

Dependent variable Δ ln Yt 3.64 50.60 16.45 2.98 52.70 20.15 9.00 21.18 22.72 29.29 15.97 17.14 39.68 13.65 13.96

25.03 21.87 37.05 28.71 24.81

4.28 2.31 10.05 8.89 7.90

Brazil

Δ ln Ct

Horizon 1 2 5 8 10

Dependent variable Δ ln Ct 100.00 0.00 0.00 86.26 1.85 7.10 66.47 2.41 13.80 56.75 2.85 24.14 54.68 5.13 23.69

0.00 4.77 16.22 14.14 14.05

0.00 0.02 1.10 2.12 2.46

1 2 5 8 10

Dependent variable Δ ln Yt 50.30 36.33 4.86 56.69 26.72 10.45 49.55 16.08 19.88 36.65 13.61 27.93 33.69 15.06 26.71

0.16 0.26 0.97 10.27 10.48

1 2 5 8 10

Dependent variable Δ ln NGt 59.93 0.00 38.28 48.62 5.44 36.98 34.81 4.09 35.28 26.96 4.95 38.08 24.19 5.71 35.97

Egypt 1 2 5 8 10

1.01 3.29 2.52 0.94 0.84

Δ ln Yt

Δ ln NGt

Δ ln TRt

Δ ln PDt

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

61

Table 4 (continued) Brazil

Δ ln Ct

Nigeria 1 2 5 8 10

Dependent variable Δ ln NGt 39.24 2.27 43.52 49.05 1.81 32.72 42.05 6.77 33.57 49.60 5.77 27.45 54.31 6.20 24.15

Δ ln Yt

Δ ln NGt

Δ ln TRt 14.97 10.04 11.41 8.46 7.73

Δ ln PDt 0.00 6.38 6.21 8.72 7.62

conscious national policies for mitigating environmental degradation and resource depletion rather than rely on increasing economic growth to solve their environmental problems later. A wide range of policy initiatives that would induce increased demand for better environment quality and its sustainability should be explored in tandem with measures to spur economic growth. References Akbostanci, E., Turut-Asik, S., Tunc, G., 2009. The relationship between income and environment in Turkey: is there an environmental Kuznets curve? Energy Policy 37, 861–867. Andreoni, J., Levinson, A., 2001. The simple analytics of the environmental Kuznets curve. J. Public Econ. 80, 269–286. Ang, J.B., 2007. CO2 emissions, energy consumption, and output in France. Energy Policy 35, 4772–4778. Ang, J.B., 2008. Economic development, pollutant emissions and energy consumption in Malaysia. J. Policy Model 30, 271–278. Ang, J.B., 2009. CO2 emissions, research, and technology transfer in China. Ecol. Econ. 68, 2658–2665. Antweiler, W., Copeland, B.R., Taylor, M.S., 2001. Is free trade good for the environment? Am. Econ. Rev. 91 (4), 877–908. Apergis, N., Payne, J.E., 2009. Energy consumption and economic growth in Central America: evidence from a panel cointegration and error correction model. Energy Econ. 31, 211–216. Apergis, N., Payne, J.E., 2010. The emissions, energy consumption, and growth nexus: evidence from the commonwealth of independent states. Energy Policy 38, 650–655. Arrow, K., Bolin, B., Costanza, R., Dasgupta, P., Folke, C., Holling, C.S., Jansson, B.-O., Levin, S. , Mäler, K.G., Perrings, C., Pimental, D., 1995. Economic growth, carrying capacity and the environment. Science 268, 520–521 (Policy Forum). Bahmani-Oskooee, M., Nasir, A., 2004. ARDL approach to test the productivity bias hypothesis. Rev. Dev. Econ. 8, 483–488. Belke, A., Dobnik, F., Dreger, C., 2011. Energy consumption and economic growth: new insights into the cointegration relationship. Energy Econ. 33 (5), 782–789. Brock, W.A., Taylor, M.S., 2004. Economic growth and the environment: a review of theory and empirics. In: Durlauf, S., Aghion, P. (Eds.), Handbook of Economic Growth (North Holland). Browns, R.L., Durbin, J., Ewans, J.M., 1975. Techniques for testing the constancy of regression relations overtime. J. R. Stat. Soc. 37, 149–172. Chen, S., 2009. Engine or drag: can high-energy consumption and CO2 emission derive the sustainable development of Chinese industry. Front. Econ. China 4, 548–571. Cole, M.A., 2004. Trade, the pollution haven hypothesis and the environmental Kuznets curve: Examining the linkage. Ecol. Econ. 48, 71–81. Cole, M.A., Neumayer, E., 2004. Examining the impact of demographic factors on air pollution. Popul. Dev. Rev. 2 (1), 5–21. Cole, M.A., Elliott, R., and Azhar, A.K., (2000). The determinants of trade in pollution intensive industries: north–south evidence. University of Birmingham, UK, Mimeo. Coondoo, D., Dinda, S., 2002. Causality between income and emission: a country-group specific econometric analysis. Ecol. Econ. 40, 351–367. Copeland, B., Taylor, M.S., 2004. Trade, growth, and the environment. J. Econ. Lit. 42 (1), 7–71. Costantini, V., Martini, C., 2010. The causality between energy consumption and economic growth: a multi-sectorial analysis using non-stationary cointegrated panel data. Energy Econ. 32, 591–603. Dasgupta, S., Mody, A., Roy, S., Wheeler, D., 1995. Environmental regulation and development: a cross-country empirical analysis. Working Papers. Policy Research. WPS 1448. de Bruyn, S.M., 1997. Explaining the environmental Kuznets curve: structural change and international agreements in reducing sulphur emissions. Environ. Dev. Econ. 2, 485–503. de Bruyn, S.M., ven den Bergh, J.C.J.M., Opschoor, J.B., 1998. Economic growth and emissions: reconsidering the empirical basis of environmental Kuznets curves. Ecological Economics. 25, 161–175. Deacon, R., Norman, C.S., 2004. Does the environmental Kuznets curve describe how individual countries behave? Departmental Working Papers. Paper 05-04Department Department of Economics, UCSB. Dijkgraaf, E., Vollebergh, H., 1998. Growth and/or environment: is there a Kuznets curve for carbon emissions? Paper presented at the 2nd biennial meeting of the European Society for Ecological Economics, Geneva, 4–7th March. Dinda, S., 2004. Environmental Kuznets curve hypothesis: a survey. Ecol. Econ. 49, 431–455.

China

Δ ln Ct

South Africa 1 2 5 8 10

Dependent variable Δ ln NGt 85.34 0.00 14.66 85.60 1.65 11.35 73.79 4.90 10.92 58.52 8.84 14.70 48.78 10.26 15.70

Δ ln Yt

Δ ln NGt

Δ ln TRt 0.00 0.88 8.91 13.62 18.05

Δ ln PDt 0.00 0.52 1.48 4.33 7.21

Dinda, S., Coondoo, D., Pal, M., 2000. Air quality and economic growth: an empirical study. Ecol. Econ. 34, 409–423. Enders, W., 2004. Applied Econometric Time Series, 2nd ed. John Wiley and Sons. Engle, R.F., Granger, C.W., 1987. Co-integration and error correction: representation, estimation, and testing. Econometrica 55 (2), 251–276. Fodha, M., Zaghdoud, O., 2010. Economic growth and pollutant emissions in Tunisia: an empirical analysis of the environmental Kuznets curve. Energy Policy 38, 1150–1156. Friedl, B., Getzner, M., 2003. Determinants of CO2 emissions in a small open economy. Ecol. Econ. 45, 133–148. Gamper-Rabindran, S., Jha, S., 2004. Environmental impact of India's trade liberalization. (Accessed at: http://unpan1.un.org/intradoc/groups/public/documents/APCITY/ UNPAN024230.pdf). Granger, C.W., 1969. Investigating causal relations by econometric models and crossspectral methods. Econometrica 37 (3), 424–438. Grossman, G.M., Krueger, A., 1991. Environmental impacts of a North American free trade agreement. National Bureau of Economics Research Working Paper, No. 3194, NBERWorld Bank, Cambridge. Grossman, G.M., Krueger, A., 1995. Economic growth and the environment. Q. J. Econ. 110, 353–377. Halicioglu, F., 2009. An econometric study of CO2 emissions, energy consumption, income and foreign trade in Turkey. Energy Policy 37, 1156–1164. Harbaugh, W.T., Levinson, A., Wilson, D., 2002. Reexamining the empirical evidence for an environmental Kuznets curve. Rev. Econ. Stat. 84 (3), 541–551. He, J., Richard, P., 2010. Environmental Kuznets curve for CO2 in Canada. Ecol. Econ. 69, 1083–1093. Iwata, H., Okada, K., Samreth, S., 2010. Empirical study on the environmental Kuznets curve for CO 2 in France: the role of nuclear energy. Energy Policy 38, 4057–4063. Jalil, A., Mahmud, S., 2009. Environment Kuznets curve for CO2 emissions: a cointegration analysis for China. Energy Policy 37, 5167–5172. Johansen, S., Juselius, K., 1990. Maximum likelihood estimation and inference on cointegration, with applications to the demand for money. Oxf. Bull. Econ. Stat. 52, 169–210. Kremers, J., Ericsson, N., Dolado, J., 1992. The power of cointegration tests. Oxf. Bull. Econ. Stat. 54, 325–348. Kuznets, S., 1955. Economic growth and income inequality. Am. Econ. Rev. 45, 1–28. Lean, H.H., Smyth, R., 2010. CO2 emissions, electricity consumption, and output in ASEAN. Appl. Energy 87, 1858–1864. Lipford, J.W., Yandle, B., 2010. Environmental Kuznets curves, carbon emissions, and public choice. Environ. Dev. Econ. 15, 417–438. List, J.A., Gallet, C.A., 1999. The environmental Kuznets curve: does one size fit all? Ecol. Econ. 31, 409–423. Liu, X., 2005. Explaining the relationship between CO2 emissions and national income— the role of energy consumption. Econ. Lett. 87, 325–328. Machado, G.V., 2000. Energy use, CO2 emissions, and foreign trade: an IO approach applied to the Brazilian case. Thirteenth International Conference on Input–output Techniques, Macerata, Italy (August 21–25, 2000). Martínez-Zarzoso, I., Bengochea-Morancho, A., 2004. Testing for an environmental Kuznets curve in Latin-American countries. Rev. Econ. Anal. 18, 3–26. Menyah, K., Wolde-Rufael, Y., 2010. Energy consumption, pollutant emissions, and economic growth in South Africa. Energy Econ. 32, 1374–1382. Miliimet, D., List, J., Stengos, T., 2003. The environmental Kuznets curve: real progress or misspecified models? Rev. Econ. Stat. 85 (4), 1038–1047. Moomaw, W.R., Unruh, G.C., 1997. Are environmental Kuznets curves misleading us? The case of CO2 emissions. Environ. Dev. Econ. 2 (4), 451–463. Narayan, P.K., 2005. The saving and investment nexus for China: evidence from cointegration tests. Appl. Econ. 37, 1979–1990. Narayan, P.K., Narayan, S., 2010. Carbon dioxide emissions and economic growth: panel data evidence from developing countries. Energy Policy 38, 661–666. Narayan, P.K., Smyth, R., 2005. Electricity consumption, employment, and real income in Australia: evidence from multivariate Granger causality tests. Energy Policy 33, 1109–1116. Odhiambo, N.M., 2011. Economic growth and carbon emissions in South Africa: an empirical investigation. Int. Bus. Econ. Res. J. 10 (7), 75–84. Ozturk, I., Acaravci, A., 2010. CO2 emissions, energy consumption and economic growth in Turkey. Renew. Sust. Energ. Rev. 14, 3220–3225. Panayotou, T., 1993. Empirical tests and policy analysis of environmental degradation at different stages of economic development. Working Paper WP238, Technology and Employment Programme. International Labor Office, Geneva. Panayotou, T., 1997. Demystifying the environmental Kuznets curve: turning a black box into a policy tool. Environ. Dev. Econ. 2, 465–484. Panayotou, T., 2003. Economic growth and the environment. Spring Seminar of the United Nations Economic Commission for Europe (Geneva).

62

O.A. Onafowora, O. Owoye / Energy Economics 44 (2014) 47–62

Pao, H.T., Tsai, C.M., 2010. CO2 emissions, energy consumption, and economic growth in BRIC countries. Energy Policy 38, 7850–7860. Payne, J.E., 2002. Inflationary dynamics of a transition economy: the Croatian experience. J. Policy Model. 34, 219–230. Perron, P., 1989. The great crash, the oil price shock, and the unit root hypothesis. Econometrica 57, 1361–1401. Pesaran, M.H., Pesaran, B., 1997. Working with Microfit 4.0: Interactive Economteric Analysis. Oxford University Press, London. Pesaran, M.H., Shin, Y., 1999. An autoregressive distributed lag-modeling approach to cointegration analysis. In: Strom, S. (Ed.), Econometrics and Economic Theory in the 20th Century: The Ragnar Frisch Centennial Symposium. Cambridge University Press, Cambridge (Chapter 11). Pesaran, M.H., Shin, Y., Smith, R.J., 2001. Bounds testing approaches to the analysis of level relationships. J. Appl. Econ. 16, 289–326. Phillips, P.C., Hansen, B., 1990. Statistical inference in instrumental variables regression with I(1) processes. Rev. Econ. Stud. 57, 99–125. Plassmann, F., Khanna, N., 2002. Assessing the precision of turning point estimates in polynomial regression functions. Working Paper 0213, Department of EconomicsBinghamton University, NY. Rostow, W., 1959. The stages of economic growth. Econ. Hist. Rev. XII (1), 1–17. Selden, T., Song, D., 1994. Economic growth and environmental quality: is there a Kuznets curve for air pollution emissions? J. Environ. Econ. Manag. 27, 147–162. Shafik, N., 1994. Economic development and environmental quality: an econometric analysis. Oxf. Econ. Pap. 46, 757–773. Shafik, N., Bandyopadhyay, S., 1992. Economic growth and environmental quality: time series and cross-country evidence. Background paper for the World Development Report 1992, Working Paper No 904The World Bank, Washington DC. Shahbaz, M., Lean, H.H., Shabbir, M.S., 2012. Environmental Kuznets curve hypothesis in Pakistan: cointegration and Granger causality. Renew. Sust. Energ. Rev. 16, 2947–2953. Soytas, U., Sari, R., 2003. Energy consumption and GDP: causality relationship in G-7 countries and emerging markets. Energy Econ. 25, 33–37. Soytas, U., Sari, R., 2009. Energy consumption, economic growth, and carbon emissions: challenges faced by an EU candidate member. Ecol. Econ. 68, 1667–1675. Soytas, U., Sari, U., Ewing, B.T., 2007. Energy consumption, income, and carbon emissions in the United States. Ecol. Econ. 62, 482–489. Stern, D.I., 2000. A multivariate cointegration analysis of the role of energy in the US macroeconomy. Energy Econ. 22, 267–283. Stern, D.I., 2004. The rise and fall of the environmental Kuznets curve. World Dev. 32, 1419–1439.

Stern, D.I., Common, M.S., 2001. Is there an environmental Kuznets curve for sulfur? J. Environ. Econ. Manag. 41, 162–178. Stern, D.I., Common, M.S., Barbier, E.B., 1996. Economic growth and environmental degradation: the environmental Kuznets curve and sustainable development. World Dev. 24, 1151–1160. Stern, J.E., Collier, T.K., Reichert, W.L., Caillas, E., Hom, T., Varanasi, U., 1998. Bioindicators of contaminant exposure and sublethal effects: Studies with benthic fish in Puget Sound, Washington. Environmental Toxicology and Chemistry 11 (5), 581–728. Taylor, A.E., Mann, W.R., 1983. Advanced Calculus, 3rd ed. John Wiley and Sons, Inc., New York. Unruh, G.C., Moomaw, W.R., 1998. An alternative analysis of apparent EKC-type transitions. Ecol. Econ. 25, 221–229. Vincent, J.R., 1997. Testing for environmental Kuznets curves within a developing country. Environ. Dev. Econ. 2, 417–432. Vukina, T., Beghin, J.C., Solakoglu, E.G., 1999. Transition to markets and the environment: effects of the change in the composition of manufacturing output. Environ. Dev. Econ. 4 (4), 582–598. Wolde-Rufael, Y., 2004. Disaggregated industrial energy consumption and GDP: the case of Shanghai, 1952–1999. Energy Econ. 26, 69–75. Wolde-Rufael, Y., 2009. Energy consumption and economic growth: the African experience revisited. Energy Econ. 31, 217–224. World Bank, 2007a. Growth and CO2 emissions: how do different countries fare. Environment Department, Washington, DC. World Bank, 2007b. Cost of Pollution in China. World Bank, Washington, DC. World Bank, 2012. World Development Indicators. (Accessed at: http://www.worldbank. org/data/onlinedatabases.html). Zachariadis, T., 2007. Exploring the relationship between energy use and economic growth with bivariate models: new evidence from G-7 countries. Energy Econ. 29, 1233–1253. Zellner, A., Theil, H., 1962. Three-stage least squares: simultaneous estimation of simultaneous equations. Econometrica 30, 54–78. Zhang, X.P., Cheng, X.M., 2009. Energy consumption, carbon emissions, and economic growth in China. Ecol. Econ. 68, 2706–2712. Zivot, E., Andrews, D., 1992. Further evidence on the great crash, the oil-price shock, and the unit-root hypothesis. J. Bus. Econ. Stat. 10 (3), 251–270.