Sectoral analysis of club convergence in EU countries ’ CO 2 emissions

This paper examines convergence clubs for per capita CO 2 emissions among 28 European countries in two main activity sectors (Industry and Manufacturing) between 1970 and 2018, with a focus on the energy sector. The method used is the Phillips-Sul log t -test using two ordering criteria to run the al-gorithm for the panel countries. The ﬁ rst one is using the last observation and the second one uses the sample average. The results of analyses of data strongly support the existence of convergence clubs, indicating that ﬁ ve groups of European countries are converging to distinct steady states for the aggregate CO 2 emissions. We also ﬁ nd evidence of convergence clubs for industry sectors while manufacturing sector shows clubs convergence only when we use the ﬁ rst criterion while in the second case, we ﬁ nd only a single steady state. © 2021 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Introduction
In recent years there has been increasing public concern about global warming and associated climate change.According to the IPCC [1], 1 the largest increase in global GHG emissions since the 1970s has originated from the global energy supply sector, which is still dominated by fossil fuels.In efforts to mitigate climate change, the European Union (EU) countries adopted a policy framework called the 2030 Framework for Climate and Energy in October 2014.The implementation of the Paris Agreement 2 is also essential to achieve two of the 2030 Agenda for Sustainable Development goals, 3 the Goal 13 which calls for urgent action to combat climate change and its impacts and Goal 7 which ensures affordable and clean energy [2,3].
In addition, the 2008e2020 European Energy and Climate Change Package has set '20-20-20' targets to be reached by 2020 (at least 20% lower GHG emissions than in 1990, with 20% of energy from renewable sources, and a 20% increase in energy efficiency).Further targets are to reduce EU emissions to 40, 60, and 80% of 1990 levels by 2030, 2040 and 2050, respectively [4].
To assist efforts to implement successful policies to achieve these goals, it is crucial to identify the trends in emissions in each European country and examine their evolution in specific economic sectors such as Industry and Manufacturing which focus on energy use and production.This will help to determine where actions should be directed and whether implemented policies have been effective.Both EU institutions and national (or even regional) authorities have crucial roles to play in efforts to meet the above targets.Individual EU members were also required to develop their own national long-term strategies by January 1, 2020, and ensure consistency between these long-term plans and corresponding National Climate and Energy Plans (NECPs).A clear implication of the policies is that groups or 'clubs' of EU countries should converge to the same level of emissions per capita, and all countries eventually, following the roadmap to a low-carbon economy via the mentioned milestone cuts.Thus, the aim of this empirical article is to investigate the process of convergence in CO 2 emissions associated with two sectors (Industry and Manufacturing) from which most of the CO 2 emissions is associated with energy use produced by fossil fuel combustion.We examine club convergence using the methodology introduced by Phillips and Sul [5,6] in 28 EU 4 countries from 1970 to 2018.
We consider this approach the most suitable for the study since it considers the transition path followed by each country in the convergence process, rather than simply static snapshots of convergence.The premise is that CO 2 emissions should converge (eventually) to the same equilibrium, while the foundation of club convergence is that CO 2 emissions of different countries could converge to different equilibria.In this context, it is important to determine whether or not European countries' emissions are converging to different steady states, in order to set appropriate reduction targets for them.Emissions convergence is crucial for policymakers to set the long-term goal of allocating emissions across countries on a per-capita basis.Hence the focus on examining the existence of convergence of CO 2 emissions among European countries is essential so as to clarify whether a common energy and environmental policy is reasonable to be applied to these countries.Several studies published in recent years have tested convergence patterns generally, and particularly club convergence patterns in CO 2 and GHG emissions [7e10].
However, to our knowledge, no previous empirical study has examined the distribution of CO 2 emissions either among European countries for Industry and Manufacturing sectors which are associated with heavy use of fossil fuels and high carbon dioxide emissions.Moreover, the study covers a long time series of data drawn from the [11] database. 5We have found only one published investigation of convergence clubs in Europe, by Morales-Lage et al. [7]; who used a short time series (1990e2012) and different database.Hence, we contribute to the literature on convergence in CO 2 emissions in several important ways.First, we examine convergence in CO 2 disaggregated by European countries and sectors (Industry and Manufacturing).This is important because emissions from such sectors must be reduced by 30%, relative to 2005 levels, by 2030.Second, we use a longer time series (1970e2018) than earlier works on the topic, and a unique database that has not been previously employed for this purpose.Third, although we use the same method, Phillips and Sul [5] algorithm, our work differs from other studies, because we also order the panel members based on the sample average from 1970 to 2018 as in stochastic convergence.
Understanding trends in CO 2 emissions in these specific sectors in 28 EU countries in recent decades could provide robust foundations for debate regarding energy policies, and valuable insights for policy-makers seeking to implement efficient local policies to meet national CO 2 emission reduction targets.The heterogeneous behaviour of the individual EU member states also warrants attention, not only because each country is assigned specific objectives in the climate agreements, but also to set policy measures that provide incentives for abatement adapted to each country's characteristics.
The main results show strong evidence of convergence.We find convergence within several clubs or groups of countries for the Industry sector, but also some differences between sectors when we compare the use of two different criteria to order the panel members.The remainder of the paper is structured as follows.Section 2 briefly discusses the Industry and Manufacturing sectors while Section 3 reviews the literature on club convergence.Section 4 presents the methodology proposed by Phillips and Sul [5,6] to test for convergence and convergence clubs.Section 5 describes the scrutinized data.Section 6 presents and discusses the results.The final section presents conclusions.

Industry and manufacturing sectors in europe
Industry and manufacturing sectors are significant drivers for the European economy.Industry and manufacturing require energy to operate.They include power generation and heating, combustion for industrial manufacturing and fuel production.In the 2017, Industry was responsible for 24.6% of energy consumed, 20.5 % of Gross Value Added (European Environmental Agency, EEA, 2020).
Carbon dioxide is emitted as a result of the combustion of fuels such as coal, oil, natural gas and biomass for industrial, domestic and transport purposes.CO 2 is the most significant greenhouse gas influencing climate change, thereby posing a threat to public health and the environment.Industrial activities are a source of pressure on the environment in the form of emissions to the atmosphere and water ecosystems, waste generation and resource consumption.Even though releases of pollutants by European industry have generally decreased over the last decade, the impacts and costs of pollution from industry remain high (European Environment Agency (EEA)).
In recent decades, the European industry has shown improvements in terms of environmental performance.The main reasons for this change are: stricter environmental regulations, improvements in energy efficiency, the propensity of European industry to abandon certain types of production, in particular the most polluting ones, and also the participation of companies in voluntary programs aimed at reducing environmental impact.Despite these improvements, the industry sector is still responsible for environmental pollution, such as emissions, and the production of waste.
Fig. 1 shows the development of CO 2 emissions with regards to the two sectors during 1970e2018.The manufacturing sector has had a strong continuously downward trend and lowered emissions by approximately 200%.The industrial sector has experienced an increase in emission from 1970 to 1990.After 1990 the emissions in the sector has decreased to come back to almost the same level as 1970.
Fig. 1.Per capita CO 2 emission trend by sectors for 28 EU countries (1970e2018). 4UK is included in the analysis because in 2008 it was still part of the EU.The UK left the EU on 31 January 2020. 5 The concept of convergence emerges from the neoclassical economic growth model originally proposed by Solow 12.It states that relatively poorer economies with low capital/labor ratioswith the same investment and savings -grow faster than relatively richer ones.This empirical approach of economic growth focused on the so-called beta convergence hypothesis.Several studies confirm the validity of this hypothesis established by Barro and Sala-i-Martin [13] and Mankiw et al. [14].Further studies have used different empirical methods to study the convergence.Some research on convergence focused only on stochastic trends.Models of stochastic convergence (e.g.Ref. [15], adopt an a-theoretical approach and consider the time-series properties of the data (e.g., unit-root or cointegration methods) to describe the dynamic aspect of the economic growth.Another methodology which allows for both deterministic and stochastic trends was proposed by Phillips and Sul [5,6].
The above methodologies were applied to many other different fields than economic growth, among others, energy consumption, energy intensities and greenhouse gases GHGs.In this literature review, we focus mainly on presenting works which test convergence on CO 2 emissions at national or regional level.Numerous studies have investigated convergence patterns of CO 2 emissions, using various datasets, econometric techniques, and concepts of convergence (absolute, conditional, stochastic or club) and the results have been rather ambiguous.For further information on this topic, an extensive survey of empirical convergence on CO 2 was recently published by Payne [16,17].
One of the first papers to explore convergence in CO 2 emissions, by Strazicich and List [18]; examined stochastic and conditional convergence in emissions from a sample of OECD (Organisation for Economic Co-operation and Development) countries during 1960e1997, and found significant evidence that CO 2 emissions have converged.Since then, several studies have explored convergence in various groups of countries and obtained mixed results.For instance, Romero-Avila [19]; Westerlund and Basher [20] and Churchill et al. [21]; among others, have detected convergence across different groups of countries, while Aldy [22]; Nguyen Van [23] and Panopoulou and Pantelidis [24]; among others, have found evidence of divergence.
A complication is that identified trends and conclusions of some previous studies are not comparable because different datasets and analytical methods were used.Clearly, it is essential to use appropriate estimation methods, samples, and tests not only to assess convergence patterns of CO 2 emissions, but also to assess the consistency of obtained results.Thus, here we specifically focus on studies on convergence clubs that used similar methodology to our selected approach.This methodology was initially proposed in two seminal papers, by Phillips and Sul [5,6]; to test for convergence based on an algorithm that also allows countries to be classified into convergence clubs.Using this technique, Panopoulou and Pantelidis [24] detected evidence of convergence of CO 2 emissions in 128 countries between 1960 and 2003, with two clubs of countries moving towards different, gradually converging steady states.
Many subsequent studies have applied the methodology proposed by Phillips and Sul [5,6] (hereafter P&S) to test the existence of convergence clubs among countries and regions.For instance, Herrerias [9] used the convergence clubs technique to assess the hypothesis of environmental convergence in CO 2 emissions, on the basis of energy sources, among a large group of developed and developing countries from 1980 to 2009.Although some countries diverged, he found convergence clubs for a large group of countries.
In a similar study, Camarero et al. [25] assessed convergence in ecoefficiency among 27 European countries during the period 1990e2009, considering CO 2 , NO 2 and CH 4 emissions.They found four to six convergence clubs, depending on the gas considered.In another exploration of convergence clubs, Huang and Meng [26] analyzed per capita CO 2 emissions in China from 1985 to 2008 across spatial areas.They detected convergence towards higher levels of emissions per capita in all the areas considered, and some indications that spatial factors accelerate the rate of convergence in neighboring urban areas.Wu et al. [27] reported that CO 2 emissions per capita of 286 cities in China tended to converge during 2002e2011.Similarly, Wang and Zhang [28] detected convergence in CO 2 emissions per capita in all of six studied sectors (agriculture, industry, construction, transportation, service and residential) across 28 provinces in China from 1996 to 2010.
In contrast, Burnett [29] investigated club convergence in per capita aggregate CO 2 emissions among a panel of US states during 1960e2010 and identified three clubs composed of 23 states in total, and 25 diverging states.Moreover, Apergis and Payne [30] found evidence for multiple equilibria with respect to per capita CO 2 emissions at both aggregate and sectoral levels in an analysis of convergence clubs for per capita CO 2 emissions across US states at both aggregate and sectoral levels.Yu et al. [31] and Liu et al. [32] also found evidence of multiple homogenous clubs in terms of CO 2 emission convergence among 24 industrial sectors and 285 cities in China.
More recently, Morales-Lage et al. [7] explored club convergence in CO 2 emissions in selected sectors of the EU's 28 member countries from 1971 to 2012.They first investigated convergence clubs in three main activity sectors (agriculture, energy and industry) then focused specifically on energy subsectors (power generation and heating, manufacture and construction, transportation, and other minor fuel combustion activities).Their results support the existence of clubs weakly converging towards the group average, with core European countries (France, The Netherlands, Germany and the UK) among the best performing clubs, across sectors and subsectors, but a few of the Central and Eastern European Countries diverging from the average towards higher emissions.

Empirical methodology
In this section, we present the P&S approach, proposed by Phillips and Sul [5,6]; to identify convergence patterns in per capita CO 2 emissions across European countries at the aggregate and sectoral levels.The novel aspect of this methodology is the application of a nonlinear model with a growth component and a time varying factor that allows for transitional dynamics and captures heterogeneity across countries and over time.The method also makes it possible to classify convergence clusters endogenously.Furthermore, it is not sensitive to the stationarity properties of the time series studied and hence not affected by the small sample properties of conventional stationarity tests such as unit root and cointegration tests.Their methodology expresses the time-varying common-factor for the observable series X it ;of country i at time t as follows: where X it is the log value of CO 2 emissions; q i represents the countries' characteristic component; m t is a common component that may follow either a non-stationary stochastic trend with drift or a trend-stationary process; and ε it is the error term.In the specification above, the per capita CO 2 emissions can be further decomposed into a common trend component, m t and an individual element, d it such that: Since it is impossible to estimate d it from equation (2) due to over-parameterization, Phillips and Sul [5,6] construct the convergence and long run equilibrium of the series based on a relative measure of the loading coefficient as: Hence, h it is the relative transition path of country i to the panel average at time t, which enables tracing of its individual trajectory relative to the panel average.
If the factor loadings d it converge to d i ; the relative transition paths governed by h it converge to 1 for all i as t/ ∞: Therefore, the cross-sectional variance h it given by H it ¼ P N t¼1 ðh it À 1Þ 2 approaches zero as t/∞: To test the null hypothesis of convergence, Phillips and Sul [5,6] propose the following semiparametric form for the loading coefficient h it : where d i is fixed, s i is an idiosyncratic scale parameter, x it ~iidð0; 1Þ, L(t) is a slow varying function of time and a denotes the speed of convergence.This representation ensures that d it converges to d i for all values of a !0. The null hypothesis of convergence can be written as: We can test H 0 by estimating the following log t regression model: where _ where a _ is the ordinary least squares estimate of ain H 0 .The null hypothesis of convergence can be tested by applying a conventional one-sided t-test for the slope coefficient g _ constructed using heteroskedasticity and autocorrelation consistent (HAC) standard errors.At the 5% probability level, the null hypothesis of convergence is rejected if t g _ < À 1:65: Note that the regression starts at some point t ¼ ½rT; where ½rT is the integer part of rT: Phillips and Sul [5,6] recommend r ¼ 1/3 as a satisfactory choice in terms of both size and power for small or moderate T 50: However, the rejection of full convergence does not imply the absence of convergence in subgroups of the panel.

The clustering algorithm
To identify convergence clubs in a panel of countries, Phillips and Sul [5] developed a data-driven algorithm.The version of this algorithm applied here incorporates minor adjustments advocated by Schnurbus et al. [17], which has the following five steps: Step 1. Ordering: the panel members according to the last observation.
Step 2. Core Group Formation: identification of a core club of countries by sequential log(t) regressions, based on the k highest members (Step 1) with 2 k N: The size of the group is determined by seeking the maximum convergence t-statistic, t k , with t k > À 1:65.
Step 3. Data Sieving for Club Membership: addition of countries to the core group (Step 2), one at a time, if the t-statistic associated with each new country is greater than zero.
Step 4. Recursion and Stopping: formation of a second convergence club by application of log t regression to the group of countries not selected in Step 3, if they converge.If not, steps 1 to 3 are repeated, to detect sub-convergence clusters.If no core group is found in step 2, these countries display divergent behavior.
Step 5: Club Merger: the final step, merger of clubs fulfilling the convergence, according to log (t) regression, of all pairs of subsequent clubs and across formed clubs, until no further mergers meet the criteria.
We also run Phillips and Sul [5] algorithm and ordering the panel members based on the sample average over the period covered as in stochastic convergence.

Data and descriptive statistics
The data used in the study are drawn from the EDGAR (Emission Database for Global Atmospheric Research) database (EDG-ARv5.0_FT2018).We specifically used information on fossil CO 2 emissions during the period 1970e2018 from the 28 European countries that were members of the EU between 2013 and January 2020 (Appendix A).
The variable of interest for the analysis is annual per capita emissions of fossil CO 2 , measured in tons per year, originating from five main activity sectors: Industry and Manufacturing.
These are described in more detail in Appendix B. No short-cycle carbon CO 2 emissions are included for any sector.
In Table 1, we present means and standard deviations (SD) of overall (not differentiated by sectors) per capita CO 2 emission for each country during the whole period 1970e2018.The records show clear variation in both the means and standard deviations of each country's emissions.Emissions of 16 countries (e.g., Sweden) fell during the whole period, while those of 12 countries (e.g., Austria) increased.
It is relevant to mention that Luxembourg has the largest CO 2 emissions per capita, while Portugal has the lowest.

Results
In this section, we present the results on convergence of CO 2 emissions in EU countries using the methodology described above, involving use of the P&S algorithm.We used STATA 16 software for the econometric convergence test and club clustering and ran all our models using the package and code developed by Du [33].
We start our analysis describing the transition path of whole 28 countries.Then, we first present results for the full sample (1970e2018), then specifically focus on each of the five sectors (Industry and Manufacturing).Next, we examine the dynamics of per capita CO 2 emissions by repeating the analysis for all samples and for each of the two sectors using the P&S algorithm with two different ordering criteria as first step of the procedure.
Using Equation ( 3) we construct and plot the relative transition path, h it (see Fig. 2) for 28 EU countries.In such a way, we can remove the common steady-state trend, and trace the trajectory of each country i in relation to the panel average.Convergent is evident when h it for all countries approaches 1.As we can see from the Fig. 2 the relative transition path shows the relative departure of each country from the common growth path.We notice that our panel of countries have different departures and different transition paths.We highlight four countries: Luxembourg, Portugal and Malta and Estonia.Luxembourg starts the transition from high initial state, as we expected because it has EU's highest level of CO 2 emissions per capita while Portugal and Malta start from a low initial state.Estonia instead has a point of departure above 1 and its transition tend to diverge from the group.
Tables 2a, 3a, and 4a show results of the algorithm when the panel members are ordering according to the last observation, while in Tables 2b, 3b and 4b, countries are ordered based on the sample average over the period covered as in stochastic convergence.
Table 2a shows the club convergence results (estimated g values, together with the corresponding t-statistics and speed of convergence parameter a _ ).For the full sample, the results indicate that the null hypothesis of full panel convergence should be rejected, since t g _ ¼ À 20:752 < À 1:65, suggesting divergence of the full group of 28 countries.
The result is in line with previous findings by Panopoulou and Pantelidis [24]; Nguyen- Van [23]; Stegman [34] and Aldy [22]; of divergence among large groups of developed and developing countries, as well as the tests for convergence club among EU countries reported by Ref. [7].
We can formally test for convergence among the initial six clubs to check whether they can be merged to form larger convergence clubs.For example, in order to test whether club 1 and club 2 can be merged to form a larger convergence club, we can implement the log t-test with a panel that contains all members of both clubs.If the estimated value of the convergence parameter g (and the corresponding t-statistic) indicates convergence, we conclude that we can merge the two clubs.Otherwise, the two clubs cannot be merged, and they represent two different convergence clubs.We first test for convergence between two consecutive clubs.As shown in the middle column of Table 2, results of subsequent merger tests of convergence (step 3 in the algorithm) between consecutive clubs form a larger convergence club, but none of the other pairs.Thus, there is evidence of convergence within subgroups of countries to different steady states with different convergence speeds.The final results of the analysis, presented in the right column of Table 2a, indicate the existence of four convergence clubs (with 10, 7, 6, 2, and divergence of Sweden).
Note that although the point estimate of g is negative for one of the fourth club, the t-statistic indicates that the estimate is not statistically different from zero, suggesting convergence among members of the club.Generally, clubs 1 and 2 include countries with high per capita CO 2 emissions, while clubs 3 and 4 mostly include countries with low per capita CO 2 emissions.Club 4 (Latvia and Romania) has its convergence speed, a _ , faster than that of members of Clubs 1e2, while for club 3 the log t parameter is clearly negative, so we rule out convergence towards the average.Table 2b also shows similar results as Table 2a.The final classification also shows 4 clubs which have some differences in the composition of the clubs than in Table 2a.The divergence club is made up of four countries: Latvia, Lithuania, Luxembourg and Romania.
We sketch in Fig. 3 the relative transition paths for the four final clubs.The trend of the club describes a pattern converging to 1 for the first period (1970e1990) and diverging in four clubs the second period (1990e2018) as expected.
Next, we examine the club convergence of per capita CO 2 emissions for each of the two sectors.For the overall period, in the case of Industry (see Table 3a), we reject the hypothesis that per capita emissions converge, as under the null hypothesis the log t parameter should be equal to or larger than zero and the test is onesided (À6.818<À1.65).However, application of the algorithm identifies two convergence clubs, and after testing the larger convergence club, we reject the hypothesis that this group of countries forms a club and we get the initial clubs 1 and 2 as final classification.Our results support Morales-Lage et al. [7] findings, although we have no diverging countries for this sector.Moreover, previous authors consider industry and power generation heating in two separated sectors, while we consider it as one.
Results in Table 3b shows instead relevant difference in the Industry sector when we use different criteria to order the countries in the S&P algorithm.The final classification indicates that there are five different convergence clubs in comparison to only two clubs illustrated in Table 3a.Our results show that when we use different criteria to sort countries, we get different numbers of clubs and consequently multiple equilibria in the Industry sector.This result also means that EU member are less homogeneous in CO 2 emissions within the Industry sector.
In the case of Manufacturing (see Table 4a), after rejecting the null hypothesis of convergence, we test the three clubs and identify two convergence clubs which clearly are not converging (À21.822<À1.65 for the first club and À2.913<À1.65 for the second club).We then, identify three initial clubs as the final classification.Our results concerning this sector differ from findings by Morales-Lage et al. [7] of three convergence clubs and nine non-converging countries, although in the cited study the manufacturing sector included construction.Contrary to the industry sector, in the

Table 2a
Convergence club classification of overall CO 2 emissions during 1970e2018, with numbers of countries in the clubs shown in parentheses (the last observation is used for ordering).

Initial classification Club merger tests Final classification
Full sample [28] g

Table 2b
Convergence club classification of overall CO 2 emissions during 1970e2018, with numbers of countries in the clubs shown in parentheses (the sample average is used for ordering).

Initial classification Club merger tests Final classification
Full sample [28] g

Table 3a
Convergence club classification of CO 2 emissions for Industry (1970e2018), with numbers of countries in the clubs shown in parentheses (the last observation is used for ordering).

Initial classification Club merger tests Final classification
Full sample [28] g manufacturing sector, using as ordering criterion the sample average over the period covered, we did not find any convergence clubs (See Table 4b).This means that the trend of the CO 2 emissions in the Manufacturing sector suggest a single steady state for all panel of countries.
Another interesting issue probed is whether there is evidence of transitioning between one convergence group and convergence club groups, or vice versa, when we apply the two criteria of ordering the panel members.This evidence of transitioning between the two convergence clubs suggesting either a slow convergence between the two clubs or a tendency for some countries to change club.It is worth noting that manufacturing sector is converging toward different clubs in only when we use the traditional way of ordering the panel members.This means that the level of CO 2 emissions from this sector are still very different across European countries over time.

Conclusions
We examined the CO 2 convergence across 28 European countries using methodology introduced by Phillips and Sul [5,6]; considering two sectors: Industry and Manufacturing.We use Du [33] algorithm applying two different criteria to order the panel countries: the first one is based on the last observation and the second one sample average over the period covered.
The results provide several insights.First, we find evidence of convergence clubs in per capita CO 2 emissions across European countries for the whole period considered (1970e2018).When we use the first criterion to order the panel countries, we identify 5 groups of countries that converge to different equilibria in aggregated per capita CO 2 emissions and one diverging country (Sweden).The first club comprises 17 countries, which are the largest emitters.The second group includes 6 countries (Belgium, Hungary, Czechia, Germany and Greece).The third group comprises Hungary and Lithuania, and the fourth group consists of Latvia and Romania.While we use the second criterion, we found four groups.The not convergent group comprises Latvia, Lituania, Luxembourg and Romania.Thus, the club convergence results highlight the importance of tailoring emission abatement policies to convergence clubs comprising clusters of countries with distinct characteristics, and hence mitigation requirements.
The analysis of the two sectors also yields some important insights.Three convergence clubs were detected for the manufacturing sector and two convergence clubs for the Industry sector when we order the countries according to the last observation.
When we order countries based on the average over the period considered, we get five convergence clubs for the Industry and only one steady state for the manufacturing sector.
These findings have policy implications for emissions targets to mitigate climate change, as recognition of the differences in convergence patterns across European countries should help efforts to design and implement policies tailored to avoid adverse sectoral effects in the respective countries.Given these findings, we argue that European countries may consider adopting new technologies and renewable energies, making older equipment more energy efficient and changing consumer behavior to achieve longrun convergence among member states.
In general, although all sectors have decreased their CO 2 emissions, they are still at a high level.In particular in the industrial sector because most of the CO 2 emissions are associated with energy use produced by fossil fuel combustion.This is not in line with the sustainable development goals.
It is particularly important that right policies are introduced at  the EU level so that incentives for technological development and innovations, for energy efficient production and distribution of products are created.Furthermore, our results have policy implications with respect to countries that display divergences.Diverging countries (such as Sweden, Portugal, Ireland, Estonia, Greece, Hungary Latvia, Lithuania, Romania and Slovakia) require tailored policies that may not be convergent in the short term, because they have highly specific emission trends (e.g., as low emitters), or policies that strongly promote convergence.In the industry and manufacturing sectors direct CO 2 emissions are produced by burning fuel for power or heat, through chemical reactions, and from leaks from industrial processes or equipment.Most direct emissions come from the consumption of fossil fuels for energy.A smaller amount of direct emissions come from leaks from natural gas and petroleum systems, the use of fuels in production (e.g., petroleum products used to make plastics), and chemical reactions during the production of chemicals, iron and steel, and cement.Indirect emissions are produced by burning fossil fuel at a power plant to make electricity, which is then used by an industrial facility to power industrial buildings and machinery.Technological differences or the type of industry and manufacturing (with lower emissions in the nonconverging countries) may explain our results.
Finally, the findings should be extended (and rigorously tested) in further empirical work.This should ideally include both broader and more detailed examination of the factors that contribute to emission convergence or divergence among European countries in per capita CO 2 emissions at aggregate and sectoral levels during both the focal and other periods.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Table 1
Means and standard deviations (SD) of annual overall CO 2 emissions per capita in each of the panel of 28 countries (1970e2018).

Table 4b
Convergence club classification of CO 2 emissions for Manufacturing (1970e2018), with numbers of countries in the clubs shown in parentheses (the sample average is used for ordering).Full sample [27] does not include Malta because of missing data.Initial classification: Club 1 [AT, BE, CZ, FI, DE, LU, NL, SK, ES]; Club 2 [BG, HR, CY, DK, EE, FR, GR, HU, IE, IT, LV, LT, PL, PT, RO, SI, SE, UK].Final classification: Club Full sample [28] For a key to the two-letter codes, see Appendix A.