The trade-off between GDP and CO2 emissions in poor countries: A dynamic common correlated effects approach for unbalanced panel data

In recent years, the term "climate change" has been increasingly receiving a lot of attention from scholars and policy makers, adversely affecting the lives of people (mostly of the poor) around the world in the present, and threatening the environment quality in the future. With many concerns about environmental degradation, countries tend to transform economic growth models causing negative impacts on the environment, especially for those in the stage of industrialization and modernization. This study was aimed at investigating the trade-offs between economic development and climate change among poor nations – the most affected by and most likely causing to climate change. By using a dynamic common correlated effects approach for unbalanced panel data which deals with cross-sectional dependency and time-series persistence, the paper showed that GDP is strongly correlated to CO2 emissions both in the short and long run, and one of the reasons is the use of CO2-generating energy sources.


INTRODUCTION
In the development progress, each region and country appears many commonly pressing problems. The more the economy grows, the more deleted fuel and energy sources due to the increasing consumption of non-renewable resources, the more damaged the natural environment, causing catastrophic natural disasters. It is widely supposed that economic growth not keeping pace with environment protection. One of the solutions to this problem is to build a green economy as proposed by The United Nations Environment Programme (2011). A green economy has low emissions, efficient use of resources, and social justice. In a "green economy", growth in income and employment is generated through public and private investments, along with reducing carbon dioxide emissions and pollution, and using energy and resources effectively as well as preventing biodiversity loss. In recent years, the term "climate change" has been increasingly receiving a lot of attention from scholars and policy makers, adversely affecting the lives of people (mostly of the poor) around the world in the present, and threatening the environment quality in the future. With many concerns about environmental degradation, countries tend to transform economic growth models causing negative impacts on the environment, especially for those in the stage of industrialization and modernization.
Developing countries are the most supposedly affected by climate change. The poor are indeed vulnerable to climate change (Rayner and Malone, 2001). According to the Intergovernmental Panel on Climate Change (IPCC) Report, less well-off countries are at risk of environmental pollution and find much difficult to enhance environmental quality under negative impacts from climate change (USGCRP, 2018). For example, it will take them longer (compared to the better-off) to recover the economy after suffering heavy losses from a natural disaster.
Likewise, the United Nations Development Program states that nearly all of the effects of climate change affect developing countries (United Nations Development Programme, 2007). One of the evidences for this argument is that about 32 to 132 million people are being pushed into extremely poor circumstances because of climate change.
Although being the most vulnerable countries affected by climate change, poor countries also is the most likely CO2-causing countries. Accordingly, the trade-offs between economic development and environmental pollution at low-income levels are acknowledged by economists, that is, developing countries tend to discharge more and more in the future in the direction of economic development. Specifically, since 1991, the Environmental Kuznets Curve (EKC) has become a mean of describing the relationship between environmental quality and income per-capita over time (Grossman and Krueger, 1991). Economists used data on the environment as well as per capita income in countries to study this relationship. Additionally, many evidences showed that this nexus also follow an inverted-U-shape, that is, environmental degradation increases in the early stages of development, afterward reaches a peak or a turning point and starts to decline when income levels exceed a certain threshold. Subsequently, numerous later studies also show the relationship between economic growth and environmental pollution in different countries and regions.
The purpose of this study was to investigate the relationship between CO2, GDP, and energy consumption, and to measure their impacts over the long and short term in low-and middle-income countries. Focusing on solving the cross-sectional dependency and time-series persistence, I used the DCCE method developed by Ditzen (2018). The final DCCE estimation model was written in an ECM form; combining PMG method, its long-run estimated coefficients were pooled and its short-run coefficients mean-grouped. To ensure the consistency of estimation, the unit-root test, cointegration test and Hausman test were also adopted.

LITERATURE REVIEW
In-depth research into the relationship between economic development and environmental pollution has been going on for decades. In 1991, Grossman and Krueger first introduced the inverted U-shaped environmental Kuznets curve (EKC) in the study of the North American Free Trade Agreement (NAFTA) and air pollution in Mexico. They found a strong statistical relationship between environmental quality and per capita GDP for a cross-section of countries. This growth-pollution nexus was described as an inverted U-shape, similar to the relationship between inequality and growth proposed by Kuznets in 1955. Specifically, pollution increases in the same direction as economic growth at low levels of income but eventually decreases after reaching a certain level of income. Empirical research in many countries showed that GDP per capita of between $4,000 and $5,000 would reach this threshold.
Several theories could be used to explain the EKC, the first being Panayotou (1993) based on the economic development stages to explain the inverted U-shape. The structural change of a country from rural to urban and agriculture to manufacturing in the first stage of development can lead to environmental degradation. This development resulted in a high amount of greenhouse gas emissions. However, in the next period as the economic structure shifts from manufacturing to services, pollution could be reduced due to the decline in carbon-intensive industries. Additionally, technological advancement would be a reason to reduce emissions once a country reaches a stage of high income. This means that as countries get richer, they have more resources to improve their technology. As a result, obsolete technologies causing pollution are replaced by environmentally friendly ones (Galeotti and Lanza, 2005). Another explanation for EKC was that environmental quality could be considered as a normal or even luxury good (Beckerman, 1992). The income elasticity of demand for environmental quality is greater than zero or even greater than one. In this case, as income increases, the individual's perception of the environment is better, and then the demand for environmental quality is also higher. Furthermore, recent studies also proved the presence of EKC, typically the study of Atici (2009)

in Central and
Eastern European Countries and Apergis and Payne (2010) in Commonwealth of Independent States' countries.
Some studies also found the pollution-growth relationship with an N-shaped curve by estimating in a cubic equation form (Levinson, 2001). This curve illustrated pollution increasing at very low levels of economic growth, then starting to slip down at a certain threshold of income, and eventually rising again at the right end. Likewise, many studies agreed on the nexus between pollution and growth, but fail to support the traditional inverted-U shape. Wang et al. (2011)  While many studies showed that as the economy develops to a certain extent, environmental problems will be improved, some researchers rather pessimistically believed that pollution reduction is still difficult to achieve. Carbon dioxide emissions reduction is far less effective at a local level. A tragedy for the commons is that many people cause pollution but no one cleans up (Hardin, 1968). Thus, this leads to more and more pollution that accumulates and does not decrease despite high income levels, as in the US (Yandle, Vijayaraghavan and Bhattarai, 2002). These arguments clearly do not conform to EKC theory. In addition, Suri and Chapman (1998) also argued that environment improvement may not globally be achieved. Welldeveloped countries tend to export pollution-causing activities to poorer ones undergoing the industrialization process, which means there is nowhere for poor nations to discharge waste. This implies that it is difficult for developing countries to conduct environmental cleanup even if their economies have developed to a turning point in accordance with EKC theory.
Most recent research papers re-confirmed the complex relationship between carbon dioxide emissions and economic growth in different countries. Munir, Lean and Smyth (2020) criticized previous studies for being biased due to cross-sectional dependency, and thus run a new Granger non-causality approach for a dataset of ASEAN-5 between 1980 and 2016. Their findings support EKC and showed a one-way impact from GDP to carbon dioxide emissions in four countries. Similarly in the states of the US over the period 1997-2016, Salari, Javid and Noghanibehambari (2021) affirmed an inverted-U shape nexus following the EKC theory by using both static and dynamic panel data analysis. For analysis of sustainable development in European Union (EU) countries, Gardiner and Hajek (2020) found a long-run equilibrium between growth, energy consumption, carbon dioxide emissions as well as other macroeconomics indicators.
In short, the relationship between economic growth and greenhouse gas emissions remains a subject of controversy. Most of the research did not go into research in the group of poor countries -the most affected by climate change and in the period of industrialization.
Moreover, most of these countries do not pay much attention to environmental issues, as well as little implementation of environmental economic policies, so the data of these countries is not subject to strong interference like developed countries. Another problem is the approach, to be clear, most previous studies have explained the economic relationship and environmental pollution based on the elasticity of environmental perception when income levels change. As analyzed above, such explanation will lead to many complex related issues, such as technology, population, environmental policy, etc. Therefore, in this paper, I considered CO2 emissions as an undesirable product in the process of making economic products (GDP) along with energy consumption. If GDP is positively related to emissions, it reflects a trade-off between environmental quality and economic growth -one of the manifestations of unsustainable development. Another issue to focus on was that the methodology of panel data of previous studies lacks confidence because of cross-sectional dependence and time-serial persistence.

Data and expected sign
To assess the relationship between environmental pollution and economic growth in poor countries, this paper used macro data from the World Bank in 79 low-and lower-middle income countries (see Appendix A) in the period of availability. For the clarification of study object, the term "poor countries" is inconsistent with the concept of "developing countries" (including the upper-middle income additionally) or "emerging countries" (countries with economic transformation). To estimate the DCCE model, I used three data variables: CO2 emissions (World Bank, 2020a), GDP constant 2010 (World Bank, 2020b), and Energy consumption (World Bank, 2020c); and then took the log of all as described in Figure 1. Figure 1-a, plotting the relationship between CO2 and GDP, basically showed that there was a positive relationship between these two variables. Moreover, the correlation test between the variables also illustrated that there was little correlation between the two variables GDP and ENG (Figure 1-c).
It could be seen that I did not use GDP per-capita to measure economic development, but GDP constant 2010 because of two reasons. First, the purpose of the study was not to test the EKC theory mainly because of old-fashion and controversy. Second, as mentioned in "literature review", I would explain the relationship between pollution and economic development on the relation to the function of production. In addition, macro indicators such as international investment or trade volume were not included in the model because these factors are strongly correlated with GDP in poor countries. Furthermore, being a determinant of CO2 emissions, energy consumption was also included as a regressor. Using the variable GDP to measure economic development, and adding the variable energy consumption as a determinant of CO2 emissions in this study were similar to recent studies of Munir, Lean and Smyth (2020) in ASEAN+5 countries; Salari, Javid and Noghanibehambari, (2021) in U.S. states; and Gardiner and Hajek (2020) in European Union countries. In the end, this study would not be able to cover fully environmental issues as well as sustainable development, but only consider climate change which is measured by CO2 emissions.  According to Bond (2002), macro data is usually persistent time series, that is, the data in the following years is strongly correlated with of previous years. Therefore, the estimation models for long-time panel data should consider the dynamics of these factors. For large panel data, cross-sectional dependence (CD) should be rigorously considered as mentioned in the study of Munir, Lean and Smyth (2020). One of appropriate reasons for this CD issue lies in the spatial relationship or, more precisely, neighboring countries often interact strongly economically. The similarities in geography, climatic conditions, and natural resources cause effective economic policies to be copied, so there is a spillover effect of economic policy among neighboring countries, including environmental policy. There are two types of cross-sectional dependence defined by Chudik, Pesaran and Tosetti (2011) -strong (based on the unobserved common factor approach) and weak (based on the spatial autocorrelation approach).
Dealing with two these issues, the most recent and reliable approach is Dynamic Common Correlated Estimator (DCCE) model developed by Chudik and Pesaran (2015 (1996) and Common Correlated Effects (CCE) estimation by Pesaran (2006). The CCE model is written in the form of the following equation for a large-N country and large T-year panel data:
In which, CO2, GDP, ENG, and ϵ denote carbon dioxide emissions, gross domestic product, energy consumption, and error term respectively. The presence of cross-section means of CO2 ������ t , GDP ������ t , and ENG ������ t -as known as CCE estimators which relatively eliminate impacts of unobserved common factors (strong CD) generated by the cross-sectional dependency -under strict exogeneity of these variables makes the estimation model consistent (Pesaran, 2006).
However, CCE estimation is effective only for static rather than dynamic panels (Chudik and Pesaran, 2015). Adding into (1) a lagged dependent variable which is not strictly exogenous causes inconsistency in the estimation model, thus Chudik and Pesaran (2015) proposed to add P T =√T 3 lags of cross-section means to solve this problem and possibly reduce weak CD. The equation (1) is re-written as follows: To estimate the short-and long-run relationship, applied to PMG approach which estimates heterogeneous short-run effects and homogeneous long-run effects, the equation (2) is converted into an error correction (EC) model: In which, denotes the EC speed of adjustment towards the equilibrium, and ′ and ′ measure the immediate or short-run impacts while and imply the long-run. To perform this model, we also run Panel Unit-root tests and Cointegration test to check the validity of dynamic panel data. Moreover, because of unbalanced panel data with divergent T among countries, I ran DCCE models with different P T lags, and then performed F-test, CD-test, and Hausman test to choose the best model for this research. For this purpose, F-test determines model fitting with the null hypothesis of R-squared (also known as the coefficient of determination) being zero; Pesaran's CD-test detect cross-section dependency of error terms under the null hypothesis of weakly cross-sectional dependency (Pesaran, 2015); and Hausman test indicates whether coefficients can be considered as equal across countries, more precisely, whether application of pooling coefficients is appropriate (Ditzen, 2018). For this research, I performed DCCE estimation according to the paper of Ditzen (2018), which additionally tests the model consistency including collinearity issues.

RESULTS AND DISCUSSION
Firstly, I ran panel unit root tests to check the stationarity of variables at level and first difference without time trend and drift term. Because of unbalanced panel data, it is appropriate to conduct augmented Dickey-Fuller (ADF) tests (Fisher-type) under the null hypothesis of all country groups containing unit roots. The values of statistic are calculated based on the modified inverse chi-squared transformation developed by Choi (2001), at three different significance levels of 1% (***), 5% (**) and 10% (*), when both T and N are assumed to tend to infinity.
Lagged differences indicate the number of lags used in the ADF regression model. Moreover, I also specified that these tests subtract the cross-sectional averages to reduce undesirable effects of CD according to the study of Levin, Lin and James Chu (2002). Figure 2 shows results of panel unit root tests, which three variables are all stationary at first difference. there is no significance to reject that EQ02 is preferred over EQ07. In short, EQ02 is the most preferred model, and is described as Figure 3. Combined with the DCCE model, the equation (3) is simply converted into (4) with , indicating the residuals (including cross-sectional averages). Specifically, the left side denotes the short-run relationship, the right side the long-run relationship and the EC speed of adjustment of -0.43. In short-run, both energy consumption and GDP statistically significantly cause the increase of carbon dioxide emissions, but the former to a greater extent. In long-run, the estimation result shows that there is a trade-off between past carbon emissions and GDP at the rate of 1.03/1.00 (over three times greater than in short-run 0.3/1.00

CONCLUSION
Economically, sustainable development is referred the economy uncorrelated with environmental degradation, especially when climate change is becomingly worse and worse at global and local level. Accordingly, the less well-off are the most vulnerable and should be paid attention to. To compound this puzzle, the introduction of EKC implies that pursuing CO2intensive growth models is one of the main causes of environmental pollution in poor countries.
This study investigated the short-and long-run relationship between economic development (GDP) and climate change (carbon dioxide emissions) in 79 low-and lower-middle income countries. Because of the presence of cross-country dependence and time-series persistence, I performed a DCCE model to gain consistency of estimation.
The estimation results showed that GDP has impacts on CO2 in both the short and long run. Specifically, it represented a trade-off between economic development and air pollution at a rate of 1.03/1 in the long run and 0.3/1 in the short term. This study also pointed out that one of the major reasons for this problem is the use of high-polluting energy sources in poor countries, when a unit of energy consumption generates 0.93 unit of carbon dioxide emissions on average.
All of the above showed that the growth models in these countries are not sustainable, and with huge trade-offs for the nature environment.
The topic of research on sustainable development is indeed not novel, but very essential, and needs to be fully evaluated in many aspects. In this study, I only evaluated air pollution without investigating other types of pollution such as soil and water. Although CO2 is the main cause of climate change and air pollution, other kinds of waste should also be considered (SO2 and NO2, for instance). Furthermore, in order to recommend effective measures against pollution problems, it is necessary to conduct a number of more in-depth studies of the main causes, typically outdated technology in FDI projects. Finally, the trend of using and developing renewable energy sources is one of the key measures to solve environmental problems in general and climate change in particular. Such studies on environmental protection related to the development of new energy sources should also be noted in the coming time.

Availability of data and materials
The datasets used and/or analysed during the current study are obtained from World