The GDP-CO2 Relationship Revised
Testing for long-term correlation and asymmetry with a heterogeneous panel estimator
Abstract
This article revisits the relationship between country-level variations in economic development and carbon dioxide (CO2) emissions totake advantage of new analytical tools andto explore the dynamism between these variables. The articleapplies Chudik and Pesaran's (2015) «DynamicCommon Correlated Effects Mean Group» estimatorto analyze data for a wide selection of countries over the time-period 1966-2013, which relaxes several assumptions that have been problematic in previous research on this topic. The analysis finds a thresholdat$13,640 gross domestic product (GDP) in purchasing parity per capita, where increasing GDP per capitais not associated with further increases in CO2 emissions per capita. To underline the policy implications and social relevance of this analysis, the article compares the obtainedGDP-CO2relationship withtrajectories in the scenario database that isappliedinthe Intergovernmental Panel on Climate Change's (IPCC) Fifth Assessment Report (AR5)(Analysis, 2015; Clarke et al., 2014).Interestingly,only 6 of 1,184 scenarios in AR5 are compatible with the results in this article, whichsuggests that the predominant understanding of future climate change is in sharp contrast with historical evidence about the relationship between economic development and CO2 emissions.
Introduction
The overarching research question in this article is not a new one: How is economic development related to country-level carbon dioxide (CO2) emissions? The keywords in this question return 386,000 hits in Google Scholar, with several articles in the range of 100-3000 citations. Attention is drawn to this topic because economic development is considered as an important driver of land-use, CO2- and other greenhouse gas emissions that are typically argued to determine the extent of climate change outcomes (e.g. sea level- and temperature rise) (Blanco et al., 2014). To illustrate, the Intergovernmental Panel on Climate Change's (IPPC) fifth assessment report (AR5)relieson a set ofclimate change scenarios where GDP- and CO2 per capita correlate on average at .88(Analysis, 2015).[1]It is therefor worrying thatenvironmental economicresearch has not accumulated to a consensual understanding of how economic development affects CO2 emissions(Carson, 2010; Pasten & Figueroa, 2012; Zapata & Paudel, 2009).This article highlights the potentially long-term and asymmetric nature of the GDP-CO2relationship as elements in previous research that can be improved, and presents an empirical application to test this claim.[2]The issues with long-termand asymmetrical effects are introduced in the following paragraphs and these topics are discussed more extensively in the literature review and methods section.
A symmetrical relationship implies that increasing and decreasing values of an independent variable have opposite effects on the dependent variable, and an asymmetrical relationship is one that deviates from this principal. An asymmetrical GDP-CO2 relationship can for example occur is if infrastructure for renewable energy productionis improved during a period of economic growthand itis still applied at later time points when the economy is stagnating.[3]This dynamic is incompatible with the typical interpretation of regression coefficients (i.e. «an increase in X leads to [insert the coefficient-value] increases in Y»), which is based on an assumption about symmetrical relation(Schorderet, 2003; Shin, Yu, & Greenwood-Nimmo, 2014). Nevertheless,most studies of economic development and CO2 emissions apply the standard interpretation of regression coefficients without testing if the symmetry assumption holds, which implies that previous research on this topiccan be suffering from serious Type I- or II error. Thisarticle builds on a handful of studiesthat have examined the degree of symmetry between economic development and CO2 emissions(i.e. Doda, 2013; Sheldon, 2013; York, 2012), and suggests that this issue can be understood better with new analytical tools.
The empirical approach in this article builds on Liddle (2015)who applies Pesaran's (2006) «Common Correlated Mean Group» (CCMG) estimator to examine how economic development is related to CO2 emissions. Liddle demonstrates how the CCMG estimator can be applied to examine curvilinear relationships in the face of non-stationary and cointegrated data, which has been a major source of error in previous environmental economic research (Müller-Fürstenberger & Wagner, 2007; Wagner, 2008). The present article adds toLiddle's work by taking advantage of Chudik and Pesaran's(2015)recent extensionofthe CCMG estimator thatallows for dynamic estimation. Dynamic estimation enables the researcher to examine long-term effects, which are interesting in the context of economic development and CO2 emissions because the relationship between these variables is expected to extend over several time points(Andreoni & Levinson, 2001; Brock & Taylor, 2003, 2005, 2010; Copeland & Taylor, 2003; López, 1994; Panayotou, 1997; Stokey, 1998; Vukina, Beghin, & Solakoglu, 1999).[4]A change in economic development is likely to have an instantaneous effect on CO2 emissions through its impact on the scale of production, but at later time points it may also affectinvestments and development ofinfrastructurethat is related to energy- and carbon intensity. Static models are bounded by two time points and unlikely to account for more than a part of this process, which may cause seriously misguiding inferences. The present article therefor contributes byexaminingforthe potentially long-term nature of economic development's effect on CO2 emissions, while also accounting for non-stationary processes and non-linearity.
The remaining article is structured in the following way. First, a literature review is presented to elaborate on how thestudy contributes to existing research about economic development and environmental outcomes.[5] Second, the data and methods are described, and third, theresults are presented. Fourth, the results are discussed in terms of theoretical and policy relevance, and lastly, the article ends with a concluding section.
Literature review
The relationship between economic development and environmental- and climatic quality has been debated intensively since The Limits to Growth was issued in the 1970's (Meadows, Meadows, Randers and Behrens III 1972). Scholars in this field typically argue that economic growth causes the scale of production to increase, which functions as a multiplier for all polluting side effects of production. However, a number of articles also claim that persistent economic growth causes the composition and technology of production to change in ways that are favorable for the climate and environment(Andreoni & Levinson, 2001; Brock & Taylor, 2003, 2005, 2010; Copeland & Taylor, 2003; López, 1994; Stokey, 1998). By this it is meant that economies tend to change from agricultural and industrial sectors towards service and knowledge production as the economy grows, and economic surplus is often invested in cost-, carbon- and energy efficient means of production in order to maintain growth and/or improve environmental outcomes. Economic progresscan in other words have desirable long-term impacts on environmental outcomes, even if the short-term impactis adverse. The question is whether long-term impacts outweigh short-term effects.[6]
It is often hypothesized that increasing values of economic development has a positive and linear effect on CO2 emissions, which impliesthat the adverse short-term effect outweighs desirable long-term effects. This hypothesis is associated with a model of environmental outcomes (or impact) that is called the «IPAT» (Impact=Population*Affluence*Technology).[7] The IPAT treats population as an independent parameter and the role of economic development (affluence) is expected to weigh heavier than technological progress (Commoner, Corr, & Stamler, 1971; Ehrlich & Holdren, 1971). Meanwhile, it is also common to hypothesize that economic progress causesemissions to curb along an inverse U-shaped slope,which implies that desirable long-term effects eventually outweigh the adverse short-term effect.The curvilinear slope istypicallylabeled as the «environmental Kuznets curve» (EKC), due to its resemblance to Simon Kuznets (1955) prediction of economic inequality, and the EKC has become a brand for hypotheses that predict desirable long-term effects. Panayotou(1994)was the first to use this label.
Both the EKC and the IPAT influence present policy debates and academic literature. An example of this is the AR5, which refers frequently to these perspectives(Blanco et al. 2014).Still, the perspectives are contrasting and bear different implications. To illustrate, the IPAT suggests more severe climate change than the EKC, as well as correspondingly stringent policy measures, if we assume thatthat economic progress will prevail in the future. Misguided trust in one, or a combination of the two, can therefor lead to a false understanding of climate change and suboptimal policy recommendations. Numerous studies have been carried out to clarify whether one should trust either the IPAT or EKC, but there is still no consensus in literature on this topic (Carson, 2010; Pasten & Figueroa, 2012; Zapata & Paudel, 2009). Consequently, there are important social, political and academic reasons to re-examine the GDP-CO2 relationship. The present article addresses arguments about an IPAT and an EKC by examining the following hypotheses:
H1: Increasing GDP per capita is associated with increasing values of CO2 emissions per capita.
H2: Increasing GDP and population are associated with increasing values of CO2 emissions.
H3: Increasing GDP per capita is associated with increasing values of CO2 emissions per capita until a threshold is reached, where after increasing GDP per capita is eventually associated with decreasing emissions.
One potential reason why environmental economic studies have notaccumulated to consensual understanding of the GDP-CO2 relationship is the lacking and/or inappropriate attention to long-term effects in previous research. The plausibility of a long-term relationship between economic development and environmental outcomes can be explained as a chain reaction that starts with positive economic growth. This growth is naturally associated with increased income levels, which is often thought to increase the demands for goods and services(e.g. Solow, 1957). Businesses are likely to scale up their production to capitalize on this development, and more pollution is emitted as the scale of production increases, unless production is shifted towards greener sectors and more energy- and CO2 efficient technologies(Brock & Taylor, 2005). Changes in the composition and technology of production are however unlikely to proceed at the same pace as the scale. Compositional and technological changes requireentrepreneurship, competence and infrastructure, which are not likely to emerge over night - even if the economy is suddenly booming(Malecki, 1997; Nelson & Phelps, 1966). Consequently, it is important to be aware that the impact of economic development on environmental outcomes is not only instantaneous, but only a select few studies examine the long-term relationship between economic development and CO2 emissions in an appropriate manner(Berenguer-Rico, 2011; Berenguer-Rico & Gonzalo, 2014; Müller-Fürstenberger & Wagner, 2007; Wagner, 2008).
Müller-Fürstenberger & Wagner(2007)present an important critique of the environmental economic literature, as they argue that it is inappropriate toestimate curvilinear relationshipswith polynomials if the data is non-stationary. The asymptotic theory of stationarity of by first-difference does not apply to higher order terms and Wagner's(2008) therefor suggests that one should apply defactoring procedures to estimate the relationship between GDP- and CO2 per capita.[8]Berenguer-Rico(2011) adds to Wagner (2008) by pointing out that curvilinear relationships can also be estimated with non-stationary data if summability procedures are applied.By examiningall peer-reviewed articles that citeWagner (2008) orBerenguer-Rico (2011),I find that no one else has applied defactoring or summability procedures to analyze CO2 emissions. Liddle (2015) has however demonstrated that Pesaran's (2006) CCMG-estimator can be used to overcome the issue with non-stationary data and non-linearity, and with Chudik and Pesaran's (2015) dynamic extension of the CCMG estimator it is also possible to calculate long-term multipliers (LRMs).[9] As the methods section explains, the CCMG and DCCMG approaches enable the researcher to examine non-linear relationships with a post-estimation procedure. To my knowledge, the DCCMG estimator has not yet been applied in studies of economic development and CO2 emissions, and the article therefor contributes to the small literature that accounts for curvilinearity and non-stationarity in an appropriate manner by examining the following hypothesis:[10]
H4: A significant proportion of the association between GDP per capita and CO2 emissions per capita is not instantaneous and it takes more than 1 year before half of the effect diminishes.
Asymmetry isanother potential problem inthesocial scientific environmental literature. The presence of significant asymmetry precludes the standard interpretations of regression coefficients, which may explain why previous research in economic development and CO2 emissions has not accumulated to a consensual understanding of the GDP-CO2 relationship(Doda, 2013; Sheldon, 2013; York, 2012). A«strictly symmetrical» GDP-CO2 relationship implies that a 1-dollar increase in GDP has the exact opposite effect on CO2 emissions as a 1-dollar decrease. The magnitude of effect is in other words the same, but positive and negative growth causes emissions to change in different directions. One can also think of «relative symmetry» as the absence of significant asymmetry, which is more usefulin this article. Regression analyses typically imply relative symmetry because the reported coefficients should be interpretable in both directions. For example, a 0.5 coefficient for the effect of GDP on CO2 emissions can be interpreted both as "a 1-dollar increase in GDP causes CO2 emissions to increase with 0.5 metric tons"and "a 1-dollar decrease in GDP causes CO2 emissions to decrease with 0.5 metric tons". However, the 0.5-coefficientcan be severely misguidingif the reported relationship is asymmetrical. Imagine for example that we have a sample with an equal distribution of increasing and decreasing values of GDP and that decreasing values of GDP have no effect on emissions, while increasing values havesignificant effect.In this case, a 1-dollar increase is actually GDP associated with 0.75 metric ton increase in CO2 emissions, not 0.5. Not accounting for asymmetry can therefore lead to severe Type-I or II error, and the following paragraphs describe an emerging literature that explains whyit is plausible that the GDP-CO2 relationship is asymmetrical.
Sheldon(2013)claims that the short-term relationship between economic development and CO2 emissions is asymmetrical because power and industrial sectors decrease their application of energy-inefficient capital at a higher pace during recession than they invest and increase their application of energy-efficient capital during periods of economic progress. The power and industrial sectors are likely to behave this way becauseit is inexpensive to retire an excessive coal-fired power plant, while an upgrade to renewable energy is costly. Moreover, thelogic of Sheldon's argument extends easily to households and individuals. A family that owns two cars, one electricity driven and another that runs on diesel fuel, is more likely to leave the diesel car in the garage to cut expenditures if income decreases. If income increases with equal magnitude, it is however not as likely that the family will replace the diesel with another electric car,seeing as thatwouldbe a costly investment. An asymmetrical short-term relationship emerges between economic development and emissions where the negative effect of recession is significantly stronger than the positive effect of economic progression, if this logic applies to a sufficiently large proportion of CO2- emitting sources.[11]This particular form ofasymmetry leads to Type I-error because regression coefficients are typically interpreted as «an increase in GDP per capita leads to [insert coefficient value] increases in CO2 emissions per capita» even though the theoretical argument predicts that negative economic growth drives the relationship. The present article examines the following hypothesis to test Sheldon's argument about short-term asymmetry:
H5: Increasing GDP per capita has less effect on CO2 emissions in the short-run than decreasing GDP per capita.
York (2012: 763)argues that the development of infrastructure and other hardware is hard to reverse, which may explain why the long-term relationship between economic development and emissions could be asymmetrical:[12]
Asymmetry is probably due to the fact that economic growth produces durable goods, such as cars and energy-intensive homes, and infrastructure, such as manufacturing facilities and transportation networks, that are not removed by economic decline and that continue to contribute to CO2emissions even after growth is curtailed.
This type of asymmetry implies thatthat positive economic growth has stronger impact on CO2 emissions than negative growth, which deflates the relationship between the variables and causes Type II-error.York extends his argumentation with an empirical analysis, but the estimates in his article are focused onthe short-termGDP-CO2relationship. The hypothesis about long-term asymmetry is therefor not addressed directly, and York (2012: 763) discusses this as a potential caveat: "It remains to be determined whether the effect on emissions of short-term (year to year) trends in economic growth or decline, which I have analyzed here, is the same as the consequences of longer-term trends in growth and decline". Doda (2013) replicates York's analysis with slightly different methods and data, and finds that the GDP-CO2 relationship is symmetrical - contrary to York's results. However, Doda neither examines the long-term relationship, which is necessary in order to test the relevant theoretical argument.[13]The presentarticle thereforeadds to this debate by testing the following hypothesis:
H6: Increasing GDP per capita has more effect on CO2 emissions than decreasing GDP per capita.
Methods and data
This section is organized in the following manner: First, the applied data is described in terms of sources and scales of measurement; Second, descriptive statistics and pre-estimation diagnostics are presented to motivate the analytical design; Third, the DCCMG estimator is described and discussed, and Fourth; control variables are presented.
The article appliesindicators of national fossil-fuel CO2 emissions from Boden, Marland and Andres(2015) and Olivier, Janssens-Maenhout, Muntean, and Peters (2015), as well as the indicators of GDP per capita from James, Gubbins, Murray, and Gakidou (2012). James et al. (2012) have merged six of the most used measures of GDP per capita to create a time-series cross-sectional measure of GDP per capita that cover 210 countries from 1950 to 2015 without gaps. The present article adapts James et al.'s (2012) measure by excluding imputed observations where no real measures exist. The applied measure of GDP per capita is in other words the average of "clean" time-series extensions, and it is measured in constant 2005 US dollars.