Carbon Pricing Impacts on Four Pollutants: A Cross-Country Analysis

: Research on climate change mitigation has increasingly considered carbon pricing, with these efforts concentrating on reductions in carbon dioxide (CO 2 ) emissions. Our comprehensive cross-country analysis extends this focus by quantitatively evaluating the effects of carbon pricing on four major pollutants: CO 2 , nitrous oxide (N 2 O), methane (CH 4 ), and particulate matter (PM). We use regressions and introduce entropy balancing to this research area. Analyzing data from 132 countries from 1992 to 2019, we find that carbon pricing is associated with an average annual reduction in CO 2 emissions by 3 percentage points. A one-unit increase in a coverage-weighted carbon price is associated with reductions in N 2 O emissions by approximately 0.1 percentage points. A shorter panel for 2010–2017 shows a larger impact of 0.3 percentage points for PM. These findings underline the efficacy of carbon pricing not just in curtailing CO 2 but in significantly mitigating other harmful pollutants on a global scale. Reductions in pollutants beyond CO 2 provide further motivation for policymakers to pursue carbon pricing


Introduction
Carbon pricing has the potential to not only reduce carbon dioxide (CO 2 ) emissions and climate change risk, but also to provide co-benefits of reductions in air pollution that result from industrial and other forms of production.More broadly, carbon pricing can enhance a range of outcomes across economic, social, energy, and environmental sustainability categories [1].This paper seeks to quantify emissions outcomes of carbon pricing beyond the impact on CO 2 emissions.This includes assessment of a range of greenhouse gases (GHG) such as nitrous oxide (N 2 O) and methane [2].This paper also considers the effect of carbon pricing on air pollution, which has massive impacts [3], leading to millions of deaths each year [4].Co-benefits have been considered in analyses of key countries, such as the national-level analyses of the co-benefits from various climate policies in China [5][6][7].Our study provides a global perspective through quantifying the carbon pricing co-benefits of emissions reductions for pollutants beyond CO 2 .
Studies of carbon pricing in general are surprisingly scarce, especially at the global scale.The small number of empirical studies on carbon pricing impacts has been described as 'astonishing' [8], relative to the well-known importance of emissions reduction for environmental outcomes.Prior carbon pricing research has tended to focus on country-or regional-level analyses [9].Green [8] lists close to 40 studies on ex-post analysis of carbon pricing impacts on CO 2 (as opposed to non-CO 2 emissions) in the first two tables of that paper, but none of these studies has a global focus.Our study includes countries that account for 93% of the global population.The global focus of our cross-country study is relevant and important due to the research question and prior research context.Greenhouse gases are a global issue, not least because of the global mixing of these gases.
The following section gives a literature review which refers to studies conducting cross-country analysis, including quasi-experimental studies [10].Section 3 gives a concise account of the theoretical basis for assessment of carbon pricing impacts on a range of pollutants.The description of the method and data in Section 4 provides details on the global context covering up to 132 countries accounting for around 93% of the global population and covering various time periods within the range of 1992-2019.The data include both a binary carbon pricing variable for some results, and a coverage-weighted numerical variable for other results.The methods include cross-sectional regressions, the matching method of entropy balancing, and panel regressions.Section 5 has results and explanations, followed by conclusions and policy implications in Section 6.

Literature Review and Novelty
The Carbon Pricing Mechanism (CPM) is often considered as a major policy instrument and an effective means of reducing CO 2 emissions.Many studies suggest the CPM as the most cost-effective way to combat climate change [11][12][13][14].The primary aims of the CPM are to encourage investment in more sustainable sources of energy and to act as an incentive for each country's largest emitters to improve their energy efficiency [15].From a theoretical point of view, the CPM can achieve abatement across emitters at the lowest possible cost by providing a price signal that can equate to marginal abatement costs [16].The introduction of a price signal provides an incentive for emitters to explore abatement opportunities that are cheaper than the carbon price, and therefore leaves subsequent decisions about the quantity of emissions reductions by each emitter to the market.
Arguments in favor of the CPM as an effective climate policy are not unanimously accepted.There are some experts who criticize and downplay the role of the CPM in energy and climate policy and argue that there is little evidence to confirm that the CPM has contributed substantially to decarbonization [17][18][19].They suggest that the CPM addresses climate change as a market failure problem rather than as a more fundamental systemic issue, leading to emphasis being on efficiency as opposed to effectiveness.Critics suggest that observed emissions reductions in some jurisdictions are not attributable to the CPM, but rather that some or most of these cuts have been obtained through regulatory instruments [20][21][22].
Despite conflicting views, as the urgent need for action on climate change mounts, the CPM is likely needed to address climate change [23].Therefore, many countries and jurisdictions around the world have already started to consider the introduction or expansion of carbon pricing initiatives to pursue more ambitious mitigation targets.There are over 70 carbon pricing initiatives either currently in place or scheduled for implementation.These various initiatives include over 30 Emission Trading Schemes (ETS) and over 30 carbon taxes by 2023.The coverage of these combined initiatives is significant and amounts to around 12 gigatons of CO 2 equivalent (GtCO 2 e), which is approximately 23% of global GHG emissions [24].Models suggest substantial potential for carbon pricing, a notable example being a case study in Canada [25].
Although use of the CPM has dominated discussion on climate change policies, the literature lacks extensive evaluations of its actual performance [8,26].The level of emissions is influenced by many different contributing factors, such as economic growth, other energy and climate policy tools, and a range of country-specific factors.Due to the complexity associated with separating out the effects of other contributing factors, there are surprisingly few empirical studies that employ an ex-post quantitative analysis to investigate the actual impact of the CPM on the level of greenhouse gas emissions.Existing studies tend to be country case studies, such as for the Australian electricity sector [27].When also including other country case studies, Green [8] refers to only 37 empirical studies that have assessed the actual performance of the CPM, and these revealed mixed results on the effectiveness of the CPM across different countries [28,29].
A range of quasi-experimental methods have been used in country case studies [10].Appropriate approaches for such studies may include the synthetic control method, regression discontinuity, difference in difference, and instrumental variables [10].Global energy studies often also use other approaches like fixed-effects regression (e.g., [30]), given the challenges in obtaining cross-country data and meeting the various requirements of each quasi-experimental method.A further method in this paper is entropy balancing, where average treatment effects are produced after balancing treatment and control groups in observational studies [31].
Mixed results for country case studies have motivated consideration of the issues at a global scale.Best et al. [32] have evaluated the impact of the CPM on the average annual growth rate of CO 2 in 142 countries by using several econometric modeling approaches over a period of two decades.Their results show that the annual growth rate of emissions has been around 2 percentage points lower in countries that have implemented the CPM compared to countries without the carbon price intervention.
Building on this literature, this paper aims to provide a novel contribution which assesses the impact of carbon pricing on pollutants other than just CO 2 .This includes reductions in N 2 O, methane, and particulate matter.Particulate matter reduction is an important co-benefit of carbon pricing which primarily targets CO 2 emissions.This paper includes a coverage-weighted numerical variable for carbon prices, as opposed to many global energy studies which use binary variables.This paper also incorporates a further method of entropy balancing, which is a novel way to assess robustness of the results in this global carbon pricing context.The authors are not aware of similar prior studies at the global scale.
While the empirical literature on carbon pricing impacts with respect to co-benefits has been very limited, there have been some country case studies using various methods.Most of these case studies of co-benefits have focused on China.For example, Wang et al. [33] considered air quality co-benefits using the coal power industry in China as a case study, Jiang et al. [34] compared co-benefit impacts of different environmental taxes using a computable general equilibrium model, and Li et al. [6] simulated the impact of carbon pricing on air quality co-benefits in China.

Theoretical Insights for the Empirical Context
A fundamental concept related to production is that there are by-products along with the intended outputs of production.For example, production of energy through fossil-fuel combustion involves the emission of pollutants such as CO 2 , particulate matter, methane, and N 2 O.This can be conceptualized in a general production function context where energy production is accompanied by a range of waste products.
The inputs for production vary across economic sectors.This motivates attempts to control for economic structures in empirical assessments that try to explain waste-product outcomes.For example, our models in Section 4 control for industrial-sector production and agricultural-sector production as shares of gross domestic product (GDP) as summary variables for overall production contexts across countries.This is useful as an alternative to detailed data on specific inputs, which is limited by data availability and comparability in cross-country studies.
Another useful way to pursue comprehensiveness is with the Kaya Identity [35].This refers to level-effects relating to population, GDP per capita, energy intensity, and carbon intensity.In the econometric approach in Section 4, the levels of these variables are used to explain the subsequent growth in emissions.
The initial level of energy or emissions variables may be related to subsequent growth, such as through a convergence relationship.For example, countries with higher initial levels for some variables may experience slower growth.Convergence has been found for energy and emissions variables in some contexts but not others [36][37][38].Section 5 explains how fossil fuel shares account for carbon intensity and energy structures, and that these can act as proxies for the initial level of emissions per capita.As the initial fossil fuel shares will be positively correlated with initial emissions per capita, a convergence relationship may be evident if there is a negative relationship between fossil fuel shares and subsequent growth in emissions.
To try to identify the impact of carbon pricing, it is important to attempt to separate its impact from other correlated variables.For instance, countries with carbon pricing might also be more likely to introduce other climate or energy policies.This motivates the inclusion of other policies as control variables.
In this context, the research question is as follows: what is the impact of carbon pricing on a range of pollutants?Our study has a global perspective and considers outcomes beyond just the impact of carbon pricing on CO 2 emissions.
Our research approach is summarized in the flow chart in Figure 1.
relationship may be evident if there is a negative relationship between fossil fuel shares and subsequent growth in emissions.
To try to identify the impact of carbon pricing, it is important to attempt to separate its impact from other correlated variables.For instance, countries with carbon pricing might also be more likely to introduce other climate or energy policies.This motivates the inclusion of other policies as control variables.
In this context, the research question is as follows: what is the impact of carbon pricing on a range of pollutants?Our study has a global perspective and considers outcomes beyond just the impact of carbon pricing on CO2 emissions.
Our research approach is summarized in the flow chart in Figure 1.

Method and Data
We assess carbon pricing impacts for various time periods between 1992 and 2019, as specified below.The rationale for this time period includes the following: 1. Start of Carbon Pricing: The early 1990s mark the inception of formal carbon pricing mechanisms, with the first significant implementations beginning around 1990.Selecting 1992 as the start year allows for an analysis shortly after these initial policies were established, providing a clearer view of their long-term effects.2. Data Availability and Quality: The year 1992 also aligns with the availability of robust and comprehensive emissions and economic data, which are crucial for a global analysis.These data, sourced from reliable institutions like the International Energy

Method and Data
We assess carbon pricing impacts for various time periods between 1992 and 2019, as specified below.The rationale for this time period includes the following: 1.
Start of Carbon Pricing: The early 1990s mark the inception of formal carbon pricing mechanisms, with the first significant implementations beginning around 1990.Selecting 1992 as the start year allows for an analysis shortly after these initial policies were established, providing a clearer view of their long-term effects.

2.
Data Availability and Quality: The year 1992 also aligns with the availability of robust and comprehensive emissions and economic data, which are crucial for a global analysis.These data, sourced from reliable institutions like the International Energy Agency and the World Bank, ensure consistency and comparability across the many countries included in our study.

3.
Methodological Consideration for Lagged Variables: This analysis incorporates lagged explanatory variables to account for the delayed effects of carbon pricing and other policy measures on emissions.Starting the analysis in 1992 gives us a sufficient lead time to accurately measure these impacts over subsequent years, enhancing the precision of our econometric models.4.
Exclusion of COVID-19 Pandemic Effects: Ending this study in 2019 strategically avoids the confounding impacts of the COVID-19 pandemic, which started affecting global economies and emissions data from early 2020.This ensures that our findings are not skewed by the atypical economic disruptions caused by the pandemic.
For non-CO 2 GHG, sectoral data are available from the International Energy Agency (IEA) for select years, such as 2010, 2012, and 2015 [39].These emissions relate to energy, agriculture, industry, and other sectors [39].This paper starts with a cross-sectional approach for explaining growth in these non-CO 2 emissions.An ordinary least squares regression model with robust standard errors is given in Equation (1).There are also emissions data relating to total fuel combustion at an annual frequency (but not at a sectoral level), so this paper includes fixed-effects panel analysis for 1992-2019.For particulate matter, there is a panel of data that covers 2010-2017 [40], so fixed-effects panel regressions explain growth in particulate matter exposure.This fixed-effects panel model with year ( I t ) and country ( I c ) fixed effects is shown in Equation (2).
The dependent variable is the annual average growth in emissions (E) calculated as the log of emissions at the end of the growth period minus the log at the start, divided by the length of the growth period in years.This is calculated for each country c.The maximum number of countries in the regression analysis is 132, which is every country for which data are available.These 132 countries cover 93% of the 2012 global population.The time dimension is given by t and the period length is p.The constant is α, the error is ε, and the key coefficient to be estimated for the carbon pricing variable (A) is β.The control variables are described in this section below.
For Equation ( 1), E is for emissions in respect of regressions for total GHG, CO 2 , GHG excluding CO 2 , methane, and N 2 O. Separate regressions are also conducted that restrict the methane and N 2 O variables to energy-sector emissions to focus on this sector, which has generally been the target of carbon pricing schemes.The value of assessing multiple pollutants is evident in Figure 2, as growth in CO 2 and N 2 O do not appear to be closely related.However, Figure 3 shows a stronger positive relationship between these pollutants when N 2 O emissions growth from the energy sector is compared with total CO 2 emissions growth.This is reasonable as total CO 2 emissions largely relate to the energy sector.
The timing of all explanatory variables is at the start of the growth period defined for the dependent variable, which is at a lag of p years.The value of p in Equation ( 1) is either 3 or 5 years.
The carbon pricing variable is denoted as A [41].Robustness tests in Stata code available for Supplementary Material include a proportional variable for carbon pricing as consideration of the fact that some countries change carbon pricing strategies part-way through the growth period.
A coverage-weighted carbon pricing variable is also included.This variable is based on data from the World Bank Carbon Pricing Dashboard.The dashboard has the coverage of carbon pricing schemes as a share of global greenhouse gas emissions.For this paper, this is divided by the share of each country's emissions (independent of carbon pricing) in the global total of greenhouse gas emissions.This gives the share of each country's emissions covered by carbon pricing for each carbon pricing scheme.This is then multiplied by the average price from the dashboard to give a coverage-weighted price in USD per ton of CO 2 equivalent.Since some countries have multiple schemes, the coverage-weighted prices which apply within a country are added to give a country total.This effectively provides a weighted average for each country.A coverage estimate of 40% is used for countries participating in the European Union ETS [24].The dashboard does not have comprehensive data on coverage overlaps between schemes for each country, although the extent of overlap would be minor for the study period.The timing of all explanatory variables is at the start of the growth period defined for the dependent variable, which is at a lag of p years.The value of p in Equation ( 1) is either 3 or 5 years.
The carbon pricing variable is denoted as A [41].Robustness tests in Stata code available for Supplementary Material include a proportional variable for carbon pricing as  The timing of all explanatory variables is at the start of the growth period defined for the dependent variable, which is at a lag of p years.The value of p in Equation ( 1) is either 3 or 5 years.
The carbon pricing variable is denoted as A [41].Robustness tests in Stata code available for Supplementary Material include a proportional variable for carbon pricing as The control set is extensive.O is a vector of socioeconomic and energy variables, with γ representing the coefficient vector, and variables being from the World Bank World Development Indicators [40].These include logs of GDP per capita, population, and energy intensity.The N vector includes fossil fuel shares of total energy supply for coal, natural gas, and oil, based on data from the International Energy Agency [42].The corresponding coefficient vector to be estimated is θ.The S vector includes variables for economic structures, with agricultural and industrial shares of total GDP [40].In this case, the coefficient vector is λ.The L vector has other policy variables for a net gasoline tax [43] and binary variables for policies supporting renewable energy uptake: feed-in tariffs and renewable portfolio standards (RPS) [30,32,44,45].The coefficient vector for these policies is ξ.Some results in Section 5 use the full models, while others exclude some explanatory variables to give larger sample sizes.The error term (ε) is assumed to be normally distributed.Table 1 has variable definitions.Control variables are based on prior literature [8,32].

Carbon price; binary
A binary variable for countries with carbon pricing at the start of the period.For instance, this assesses the impact of carbon pricing schemes that existed in 2010 on emissions growth in the subsequent 5 years as one example.A robustness test uses a proportional carbon price variable to adjust for countries that had carbon prices for only part of a period.
Carbon price; numerical A coverage-weighted variable using the proportion of emissions covered by carbon pricing and average prices [24].
Emissions trading system (ETS) Similar to the binary carbon price variable, but countries only have a value of one when there is an ETS.
Carbon tax Similar to the binary carbon price variable, but countries only have a value of one when there is a carbon tax.

Log GDP per capita
The natural log of GDP per capita in constant 2017 international dollars in purchasing power parity terms.

Log population
The natural log of population.

Ln energy intensity
The natural log of energy intensity, which is defined as the amount of primary energy in megajoules per unit of GDP in 2017 international dollars in purchasing power parity terms.

Coal share
The coal share of total energy supply [42].

Oil share
The oil share of total energy supply.

Gas share
The natural gas share of total energy supply.
Agricultural share Agricultural value added as a share of total GDP.

Industrial share
Industrial value added as a share of total GDP.

Gasoline tax
The net gasoline tax as calculated by Ross et al. [43]  Despite the comprehensive models, the risk of omitted variables is still possible.For example, technological changes may be relevant.We partly address this through fixed effects.Additionally, some of the explanatory variables included in our models, such as GDP per capita and sectoral outputs, may indirectly reflect technological changes.These variables often correlate with technological advancement, as higher economic outputs and efficiencies typically involve the adoption of newer technologies.While these variables do not provide a direct measure of technological progress, they offer some insight into the economic and industrial conditions that are conducive to innovation.However, these measures might not fully capture the nuanced effects of specific technological innovations or sector-specific dynamics.
Descriptive statistics are presented in Table 2.The growth of emissions tends to be just above zero on average, but there is a substantial range when considering the minimum and maximum values.The results section also considers robustness tests that drop countries with annual emissions growth greater than 10% or less than −10%.In 2012, 28% of countries had carbon pricing.Wide variation is evident for the size and structure of economies.A further method to assess the robustness of the results is entropy balancing.This matching method, which weights a control group to balance covariates like the mean and variance with the treatment group, is presented as an advance on propensity score models [31].The covariates used for balancing match other results except that energy mixes are not included as matching does not occur for countries with diverse energy mixes.Entropy balancing seeks to provide causal estimates of average treatment effects for observational studies [31].In this case, the treatment group is countries with carbon pricing at the start of a growth period, and the control group is countries without carbon pricing.

Initial Results: Carbon Pricing Impact on GHG
Table 3 shows results with dependent variables being the annual change over the five years to 2015 for each pollutant.The binary carbon pricing variable has negative and significant coefficients at the 1% or 5% levels in the columns for (total) GHG, CO 2 , and methane.For countries with carbon pricing in 2010, which is the start of the fiveyear growth period, subsequent annual growth of GHG is lower by around 3 percentage points, compared to countries without carbon pricing in 2010.The average impact on CO 2 is greater at around 4 percentage points per annum.There is a smaller point estimate for methane of around 2 percentage points per annum.Smaller proportional impacts on methane are reasonable given that methane emissions also often come from sectors which are not covered by carbon prices.In robustness tests in Stata code that is available for Supplementary Material, there are similar negative and significant coefficients if the binary carbon pricing variable is adjusted to incorporate the introduction or removal of carbon pricing during the five years to 2015.Carbon pricing results are also similar when controlling for the initial level of emissions.Some of the controls also reveal interesting impacts in Table 3.For example, the economic structure of economies appears to be important.There are positive and significant impacts of the agricultural share of GDP in 2010 on the subsequent growth of pollutants in each case, except for the CO 2 column where the coefficient is insignificant.Positive and significant coefficients for the agricultural share are reasonable for GHG such as methane and N 2 O in the final two columns, as these pollutants are major by-products from agricultural production processes [2].
Entropy balancing results in Table 4 show a similar theme of carbon pricing lowering pollution.More specifically, countries with carbon pricing tend to have lower emissions growth of approximately 2 percentage points for non-carbon-dioxide pollutants and around 3 percentage points for CO 2 .These results tend to show slightly larger impacts on non-CO 2 pollutants compared to Table 3, but slightly lower impacts on CO 2 .The results use the same time period as Table 3. Results are similar if six countries are dropped from the sample to remove the impact of carbon price introductions during the five years ending in 2015.Notes: *** and ** show statistical significance at the 1% and 5% levels respectively.ATE is the average treatment effect.The lower and upper bounds show 95% confidence interval boundaries.There are 131 observations for the first two average treatment effects and 127 for the final three.The covariates match Table 3 except that energy mixes are not included as matching will not occur for countries with diverse energy mixes.

Further Results: More Policies and Time Periods
For Table 5, the carbon pricing variable is split into separate binary variables for ETSs and carbon taxes.The robustness of the results is further assessed by adding extra policy controls.This is useful for gaining some understanding about whether the carbon pricing coefficients in Table 3 were being driven by other related policies.Table 5 includes policies covering net gasoline taxes and renewable energy policies.The downside of adding extra policy controls is that the sample size reduces because data for these extra policy controls are unavailable for some countries.The ETS binary-variable coefficients are negative and significant in four of five columns in Table 5, while the carbon tax coefficients are insignificant.A smaller number of countries had introduced carbon taxes by 2010, possibly creating a challenge for more precise estimation of the carbon tax coefficients.For the ETS variable, the coefficients are significant at the 1% level for GHG in total and for CO 2 .Results are similar when excluding outliers with emissions growth greater than 10% or less than −10% per annum (these results are available through the Stata code).The ETS coefficient in the methane column is significant at the 5% level.
The other policy controls in Table 5 show some expected impacts.There are negative and significant coefficients for the gasoline tax and feed-in tariff variables in the column for CO 2 .This is reasonable as both these policies support the transition from fossil fuels to renewables, so should be expected to lower subsequent growth in CO 2 emissions.Also, these policies are aimed at energy usage (either gasoline or the promotion of renewable energy as a substitute for fossil fuels), and so can be expected to have less of an impact on other GHG that relate more to non-energy sectors.
Table 6 switches to a shorter time period, which is useful to reduce the probability of the results being influenced by intra-period changes that are not accounted for by the explanatory variables that give measures at the start of the growth period.There are similar results between Table 6 with the 3-year growth period and Table 3 with the 5-year growth period.The carbon pricing coefficient is negative and significant in explaining GHG and the CO 2 component.The magnitudes of the annual average changes are similar to Table 3. Table 6 also focuses more specifically on GHG in the energy sector.The dependent variables in the final two columns are now only for the energy sector, as opposed to total methane and N 2 O emissions growth in the prior tables.The carbon pricing variable has a negative and significant coefficient in explaining both methane and N 2 O emissions growth from the energy sector.In both cases, the magnitude of the impact at around 3 percentage points lower emission growth for countries with carbon pricing in 2012 is similar to the case for GHG overall.Larger carbon pricing impacts on N 2 O and methane in the energy sector compared to the non-energy sector are reasonable as carbon prices should simultaneously reduce multiple pollutants which are joint by-products from energy production.In contrast, N 2 O and methane in other sectors which are not covered by carbon pricing should not be directly impacted by carbon pricing.The occurrence of convergence patterns in the energy sector is suggested in Table 6 through negative and significant coefficients for the energy explanatory variables in the CO 2 column.In addition, there are also negative and significant coefficients for natural gas shares in explaining methane and N 2 O growth and for the oil share in explaining the N 2 O growth.The agricultural share is no longer a significant explanatory variable in the final two columns, which is consistent with the focus on the energy sector.
The impact of carbon pricing on greenhouse gas emissions can also be assessed using a longer-term perspective with fixed-effects panel analysis as shown in Table 7.There is greater data availability covering more years for greenhouse gas emissions from fuel combustion (without splits for sectors), allowing for the following panel analysis.Notes: All explanatory variables are at the start of the growth period (i.e., lagged 1 year).*, **, *** represent statistical significance at the 10%, 5%, or 1% levels, respectively.Robust standard errors are in brackets below the coefficients.Energy intensity is not included based on missing data in the early years.
Table 7 shows a similar negative relationship as that observed with the cross-sectional analysis in Tables 3-6.The carbon pricing coefficients are statistically significant at the 1% level for CO 2 , N 2 O, and total greenhouse gas emissions.The magnitudes are again similar, with carbon pricing leading to lower annual emissions growth by around 3 percentage points.The carbon pricing coefficient for explaining growth in methane emissions is also negative but is not significant.There are also some negative and significant coefficients for the feed-in tariff variable, although the impact and significance are lower.There are strong negative relationships from some fossil fuel shares to the subsequent growth in emissions, which is expected given that fossil fuel shares at the start of each year have a major influence on the subsequent emissions in the directly following year.It is important to remember that the dependent variables are measuring emissions growth (which can be negatively related to the starting emissions) and not levels.
There are corresponding negative coefficients for the numerical carbon price variable in Table 8.There is statistical significance at the 1% level for CO 2 and total greenhouse gas emissions.The N 2 O column has significance at the 5% level.A one-unit increase in the weighted carbon price variable is associated with a decrease of 0.2 percentage points in CO 2 emissions growth and 0.1 percentage points in N 2 O growth.The control variables match Table 7 but are not shown to save space.

Carbon Pricing Impact on Air Pollution
Table 9 also gives results from the fixed-effects panel analysis but now investigates the impact of carbon pricing on the growth of particulate matter exposure.There are negative and significant coefficients for carbon pricing in both columns.Carbon pricing is associated with lower subsequent growth of exposure to particulate matter by up to 3 percentage points.Column (1) has a larger sample size as it includes less variables.Column (2) drops some observations when adding extra policy controls, which are not available for some years of the sample.There are also negative and significant coefficients for other policy variables in explaining subsequent growth in particulate matter.In particular, the RPS variable has a negative and significant coefficient.The gasoline tax variable also has a negative coefficient, which is statistically significant at the 10% level.It is reasonable that these other policies that motivate moving from fossil fuels toward renewable energy would also have impacts on pollutants such as particulate matter that are jointly released with other pollutants.The magnitude of the RPS and carbon pricing coefficients are roughly similar.
The impact of the numerical carbon pricing variable is again negative and significant in Table 10 when explaining growth of particulate matter exposure.The statistical significance is at the 1% level in both columns.A one-unit increase in the weighted carbon price variable is linked to particulate matter growth being lower by around 0.3 percentage points with the larger sample in column (1).

Conclusions and Policy Implications 6.1. Magnitudes Compared to Prior Studies
While our study aims to be a novel contribution by quantifying the impacts of carbon pricing on non-CO 2 emissions using a global sample of countries, the magnitudes for the results are still comparable with prior studies.This is possible because the analysis still includes impacts on CO 2 , in addition to the focus on non-CO 2 emissions.Also, the use of annual changes in emissions in percentage-point terms is useful for giving a common unit of measurement, as opposed to levels of pollutants which will have different scales.
Our finding of carbon pricing impacting annual emissions growth of CO 2 by 3 to 4 percentage points is slightly larger in absolute-value terms but is still within the range of prior studies.Green [8] notes that many studies find impacts of around 1.5 percentage points on average.Best et al. [32] found an impact of around 2 to 3 percentage points on CO 2 emissions for a cross-country study.Our finding of a larger carbon pricing impact incorporates all channels of carbon pricing influence.In contrast, the study by Best et al. [32] focused on changes through energy efficiency or carbon intensity by controlling for contemporaneous population and economic growth.Our finding is lower in comparison to the study for Sweden by Andersson [14], which uses the synthetic control method to show a carbon pricing impact of around 6 percentage points for CO 2 emissions in the Swedish transport sector.
The magnitudes of the carbon pricing impacts on the various pollutants in our study are internally reasonable.Our findings show that carbon pricing has larger point estimates of coefficients (in absolute-value terms) for the impact on CO 2 , rather than on other pollutants.This is consistent with carbon pricing historically targeting energy sectors where CO 2 emissions are pronounced, as opposed to agricultural sectors where pollutants such as methane and N 2 O are more important.

Implications
The findings from our study have significant implications for policymakers and practitioners focused on environmental sustainability and public health.By showing varied impacts of carbon pricing on different pollutants, our research underscores the necessity for nuanced policy frameworks that address specific environmental challenges.Policymakers could consider socio-economic factors in the design of these mechanisms, assessing the economic impact on different industries and communities and providing support where the impact of pricing might be disproportionately high.Additionally, developing integrated policy frameworks that combine carbon pricing with other regulatory and voluntary measures could enhance the effectiveness of environmental policies, ensuring reductions in greenhouse gas emissions contribute to broader sustainability goals.Increasing efforts to raise public awareness about the benefits of carbon pricing beyond CO 2 reduction and engaging the public and stakeholders in policy formulation can lead to more widely accepted and successful implementations.

Limitations and Potential for Future Research
Our study, while comprehensive, faces limitations that suggest avenues for future research.The availability and granularity of data on certain pollutants were limited, which may affect the generalizability of our findings, and future studies could benefit from more detailed sector-specific and pollutant-specific data.Our primary focus was on data up to 2019, not covering subsequent developments in carbon pricing or environmental policies that could influence the observed relationships.Future research could conduct longitudinal studies to track the impacts of carbon pricing over longer periods, especially in response to evolving economic and environmental conditions, and investigate the interaction effects between different pollutants under varying carbon pricing schemes to provide deeper insights into their synergistic or antagonistic effects.While our study is global, regionspecific analyses could uncover nuanced dynamics that global studies might overlook, suggesting detailed analyses at the regional or even country level could be beneficial.Additionally, more research is needed to assess the economic impacts of carbon pricing, particularly on low-income populations and developing economies, including studying the effectiveness of mitigation strategies like rebates or subsidies.
More specifically, the econometric analysis in this paper is constrained by data limitations.For example, the International Energy Agency [39] displayed sectoral data for non-CO 2 greenhouse gas emissions for the years 2010, 2012, and 2015 at the time of writing.For particulate matter from the World Development Indicators [40], a panel with a continuous time series was available from 2010 to 2017.This paper also uses a panel for emissions from fuel combustion for 1992-2019.As data availability improves over time, there will be great potential for future research on the impacts of carbon pricing.
A further limitation relates to the complexity of the context.It is possible that carbon pricing in one country has impacts on other countries in several ways.For example, offset schemes can allow for emissions reduction by one country to be satisfied through payment for emissions reduction in another country.There is also potential for leakage of emissions from countries with high carbon prices to countries with low or zero carbon prices.While prior studies mention leakage, there has so far been an absence of systematic analysis of the issue in a cross-country context [8,32].There is an opportunity for future research to assess these issues with more intricate modeling approaches, as more data become available over time.It is also possible that leakage will partly be addressed by upcoming plans to introduce border carbon adjustments, such as in the European Union.
The highlights of our study, which can motivate further related research, include the following: • The unique global scope of our analysis, assessing the impact of carbon pricing on multiple pollutants beyond CO 2 across 132 countries.• The introduction of entropy balancing as a novel methodological approach in this field, enhancing the robustness of our findings.

•
Our significant empirical contributions to understanding the broader environmental benefits of carbon pricing, particularly in reducing energy-sector emissions of N 2 O and methane.
Future studies can increasingly seek to compare policy approaches as data availability improves.While it has been difficult in the past to separate the impacts of carbon pricing from other policies, there may be greater potential going forward.This can involve more precise definitions of components of green industrial policy and empirical assessment of the relative value of each policy component.Increased ambition for a range of policies, including but not limited to carbon pricing, has wide support despite apparent conflicts on the relative merit of various components of policy portfolios.Policymakers may be more likely to pursue more ambitious policies that can improve sustainability prospects if there is more evidence through additional detailed empirical studies in the future.Investigating the interactions between carbon pricing and other policy measures, including technological incentives, can provide insights into optimizing combined policy portfolios for greater environmental benefit.
Future research should also incorporate more comprehensive international trade and investment data to better understand how carbon pricing in one region may influence production patterns and emissions in others.Additionally, further studies should focus on disentangling the various channels of impact including technological advancement, possibly by incorporating data on energy efficiency improvements and renewable energy adoption rates.
Future studies can also link the carbon pricing literature to broader assessments of socioeconomic and environmental issues.For example, carbon pricing could be integrated with recent studies which have begun to explore the interlinkages between economic growth and green energy adoption [47].Carbon pricing could also be investigated alongside environmental innovation and green energy deployment.Studies have shown that advancements in technology and the adoption of green energy sources significantly contribute to environmental sustainability [48,49].

Figure 3 .
Figure 3. Energy-sector N2O growth per annum and CO2 growth per annum for 2012-2015.Based on data from the IEA [39].

Figure 2 . 19 Figure 2 .
Figure 2. N 2 O growth per annum and CO 2 growth per annum for 2012-2015.Based on data from the IEA [39].

Figure 3 .
Figure 3. Energy-sector N2O growth per annum and CO2 growth per annum for 2012-2015.Based on data from the IEA [39].

Figure 3 .
Figure 3. Energy-sector N 2 O growth per annum and CO 2 growth per annum for 2012-2015.Based on data from the IEA [39].
GHG excluding CO 2 emissions growth Annual percentage emissions growth of GHG excluding CO 2 .Methane emissions growthAnnual percentage growth of methane emissions.We use total, energy-sector only, and fuel-combustion methane emissions in separate regressions, as identified in each case.N 2 O emissions growth Annual percentage growth of N 2 O.We use total, energy-sector only, and fuel-combustion nitrous oxide emissions in separate regressions, as identified in each case.Particulate matter growthAnnual growth of mean annual exposure of particulate matter of size 2.5 micrograms per cubic meter or smaller.As data are available as a panel for 2010-2017, 1-year growth periods are used.
in 2015 USD per liter.Subsidies can produce negative values.
[32]-in tariffsA binary variable for feed-in tariffs, as in the study by Best et al.[32]and updated using table R10 in the 2021 REN21 [46] Global Status Report.RPS A binary variable for RPS, as in the study by Best et al. [32].The 2018 REN21 [45] Global Status Report shows no change in the number of RPS policies over 2016 and 2017.

Table 2 .
Descriptive statistics.Notes: These descriptive statistics give dependent variables in 2015 and explanatory variables in 2012, which aligns with some results below.There are 132 observations for most variables; however, emissions data are available for 129 countries for non-CO 2 pollutants, 130 countries for particulate matter, and the net gasoline tax is available for 126 countries.The maximum value of CO 2 growth per annum of 0.35 is not shown in the figures because data are unavailable for the N 2 O variable for this country.Data availability therefore leads to the dropping of an outlier value.An outlier value for methane from the energy sector is also not shown in the Graphical Abstract.

Table 3 .
Carbon pricing and greenhouse gas emissions growth per annum: 5 years to 2015.

Table 4 .
Entropy balancing for treatment effects of carbon pricing on annual pollutant growth for the five years to 2015.

Table 5 .
Carbon pricing and GHG emissions growth: 5 years to 2015, extra controls.
Notes: All explanatory variables are at the start of the growth period (i.e., lagged 5 years, meaning that they are values from 2010).*, **, *** represent statistical significance at the 10%, 5%, or 1% levels, respectively.Robust standard errors are in brackets below the coefficients.

Table 6 .
Carbon pricing impacts on greenhouse gas growth: 3 years to 2015.All explanatory variables are at the start of the growth period (i.e., lagged 3 years, meaning that they are values from 2012).Methane and N 2 O are only from the energy sector in Table6.*, **, *** represent statistical significance at the 10%, 5%, or 1% levels, respectively.Robust standard errors are in brackets below the coefficients. Notes:

Table 8 .
Results with the numerical carbon price variable explaining 1-year growth in fuel-combustion emissions, 1992-2019.:Allexplanatory variables are at the start of the growth period (i.e., lagged 1 year).**and*** represent statistical significance at the 5% or 1% levels, respectively.Robust standard errors are in brackets below the coefficients.Controls match Table7.Energy intensity is not included based on missing data in the early years. Notes

Table 9 .
Fixed-effects panel results for the 1-year growth in particulate matter, 2010−2017.All explanatory variables are at the start of the growth period (i.e., lagged 1 year).*, **, *** represent statistical significance at the 10%, 5%, or 1% levels, respectively.Robust standard errors are in brackets below the coefficients. Notes:

Table 10 .
Results with the numerical carbon price variable explaining 1-year growth in particulate matter, 2010-2017.: All explanatory variables are at the start of the growth period (i.e., lagged 1 year).*** represents statistical significance at the 1% level.Robust standard errors are in brackets below the coefficients.Controls are not shown but match Notes