The Impact of Environmental Protection Expenditures on Reducing Greenhouse Gas Emissions

Abstract

In order to address the adverse impacts of climate change, international initiatives, such as the Sustainable Development Goals, the Paris Agreement and the United Nations Framework Convention on Climate Change, are setting targets to governments and asking them to commit to take action. In pursuit of fulfilling these commitments, countries have set national goals, reflecting their policy choices, local contexts and capabilities, to fight against climate change. Given that governments implement their commitments through their budgets, environmental protection expenditures (EPEs) are the major tool that governments use to fight against climate change. This study investigated the extent to which EPEs are effective in reducing greenhouse gas (GHG) emissions, a major contributor to climate change. Our results provide some evidence that the increase in EPEs has led to a reduction in GHG emissions.

1. Introduction

Climate change is blamed for various extreme weather events such as droughts, water scarcity, wildfires, rising sea levels, destructive storms, hurricanes and typhoons [1] (Caglar and Yavuz 2023). These extreme weather events have important social and economic consequences such as changes in climate change-induced migration patterns, food insecurity, fiscal pressures due to economic losses and damage to infrastructure, as well as heightened health risks for vulnerable populations.

The efforts to fight against the climate change include various intergovernmental frameworks and agreements such as the Sustainable Development Goals (SDGs), the Paris Agreement and the United Nations Framework Convention on Climate Change (UNFCCC) and its bi-yearly Conference of Parties (COP)—the last COP was held very recently in November 2024, commonly referred to as COP29.

All of these efforts ultimately aim to limit the adverse effects of global warming through reducing all emissions [2]. On a global scale, there are various sources of greenhouse gas (GHG) emissions including burning fossil fuels, landfills and agricultural and industrial activities. Although, studies have documented the role of GHG emissions on climate change [3,4,5] (Filonchyk et. al. 2024a; Filonchyk et. al. 2024b; Shah et. al. 2024), the extent of the impact of GHG emissions on livelihoods is not well understood [6] (Johnson et. al. 2024). Governments, through environmental protection spending, try to limit GHG emissions by implementing programs for pollution abatement, protection of the biodiversity and landscape, waste and wastewater management and environmental research and development.

Given that the budget is a government’s central policy document on all social and economic issues, it also reflects governments’ priorities related to the environment and climate change. Environmental protection expenditures (EPEs) in a budget are the major tool that governments use to implement sustainability development policies and fight against climate change. The aim of this study was to determine the extent to which EPEs are effective in reducing greenhouse gas emissions.

The reminder of the paper is organized as follows. The next section reviews the relevant literature on the topic and presents the main findings of the studies in the literature. Section 3 presents information about the empirical model and data. Section 4 discusses the empirical findings. The last section offers concluding remarks and suggestions for further analysis.

2. Literature Review

The multi-disciplinary literature on public spending and its impact on climate change and environment is rich with theoretical formulations and empirical testing of this relationship. There are several studies in the literature that investigated the impact of public spending on various dimensions of environmental degradation. However, both the theoretical formulations and empirical findings were inconclusive on the merits of public spending. Some studies presented a positive impact of public spending on environmental indicators whereas others either could not find a statistically significant relationship or negative impact.

In a pioneer study, Lopez et al. (2011) [7] modeled the impact of fiscal spending on the environment. In their model, they investigated the impact of fiscal spending on the environment through three channels: a scale effect (increasing spending increases GDP but contributes to pollution); composition effect (increasing government spending towards more value-added knowledge products, reducing pollution) and technique effect (increasing government spending contributes to R & D developments that reduce pollution). Their findings suggest that increasing government spending does not necessarily reduce air and water pollution. However, reallocation of government spending to social and public goods had a positive impact on pollution reduction.

Later, Morley (2012) [8] demonstrated a statistically significant negative relationship between environmental taxes, measured as the share of the total government tax revenue, and greenhouse gas emissions using a panel dataset of European Union countries and Norway. Environmental taxes, which finance governments’ environmental spending, have a significant impact on reducing pollution.

In similar studies, Halkos and Paizanos (2013) [9] and Bernauer and Koubi (2013) [10] examined the impact of government spending on the environment directly. Halkos and Paizanos (2013) [9] used a panel dataset of 77 countries for the time period of 1980–2000 to estimate both the direct and indirect effects of government spending on pollution. They concluded that the indirect effects of government spending are more important than the direct effects. Their results indicate that higher levels of public spending as a share of GDP reduced sulfur and nitric oxide emissions but not carbon dioxide emissions. Bernauer and Koubi (2013) [10] investigated the impact of government spending as a percentage of GDP on pollution using a dataset of 42 countries over the period of 1971–1996. Their findings suggested a negative relationship—that is, countries with a larger government tended to suffer more from pollution because of the bureaucratic inefficiency and the influence of special interest groups.

Lopez and Palacios (2014) [11] examined the role of fiscal, trade and environmental policies in 12 European countries using micro data for the period of 1995 to 2008. Their analyses suggested that the increase in the share of fiscal spending based on GDP had a beneficial impact on reducing sulfur dioxide and ozone emissions but not on nitrogen dioxide emissions.

Adewuyi (2016) [12] conducted an extensive global review of the impact of household, firm and government expenditures on emissions in the short and long term during 1990–2015. Overall, the results indicated that the increase in government expenditures increased carbon emissions.

In a comparative analysis, Abid (2017) [13] investigated the impact of public spending on per capita CO₂ emissions in a sample of 58 Middle East and North African (MENA) countries and 41 European Union (EU) countries. The results indicated that public spending contributed to the reduction in GHG emissions in EU countries; however, the impact of public spending on GHG reduction in MENA countries was negligible. His findings also suggested that the interaction between public expenditure and institutional variables play a very important role in reducing GHG emissions. Similar findings for MENA countries were found in Gholipour and Farzanegan (2018) [14]. They analyzed the impact of government expenditures on environmental protection in Middle Eastern countries using data from 1996 to 2015. Their results showed that government expenditures alone did not have any impact on the environment.

Zhang et al. (2017) [15] analyzed the impact of public expenditures on emissions at the subnational level in China. Using city-level data for 106 Chinese cities for the period of 2002–2014, they investigated the direct and indirect impacts of public spending on the emission of sulfur dioxide (SO₂), chemical oxygen demand (COD) and soot production. According to their estimation, the total effects of government expenditure was negative for SO₂ emissions and COD, whereas it was positive for soot production for jurisdictions with a GDP per capita higher than USD 7500.

Postula and Radecka-Moroz (2020) [16] analyzed the impact of the public spending and tax policies of EU countries on environmental protection indicators between 2006 and 2018. Their findings suggested a positive long-term impact of public spending on environmental protection indicators. Their research demonstrated that fiscal instruments, public spending and revenue on and from environmental protection yielded results after several years. However, they argued that policy swings do not help with the fight against climate change; therefore, there needs to be a consensus on climate actions.

Dracea et al. (2020) [17] investigated the impact of environmental expenditures in a single country: Romania. Their main finding was that increasing environmental expenditures by 1 percent of GDP reduced the ecological footprint per capita by 0.01 percent. They used Borucke et al. (2013)’s [18] definition of ecological footprint, which comprises cropland, grazing, fishing, forests, carbon dioxide and infrastructure.

Le Gallo and Ndiaye (2021) [19] investigated the spatial interactions of environmental expenditures in 28 OECD countries for the period of 1995 to 2017. They argued that OECD countries strategically increase their environmental expenditures as a response to increases in their neighbors’ spending. In a deterministic model design study, Barrell et al. (2021) [20] analyzed the efficiency of EPEs in EU countries for the period of 2005–2015 using the data envelope analysis methodology. Their findings suggested diminishing returns: higher EPEs did not necessarily lead to better environmental results. The increase in EU countries’ environmental protection expenditures did not result in proportional improvements in environmental outcomes after a certain level of development.

In more recent studies, Caglar and Yavuz (2023) [1] analyzed the impact of renewable energy consumption and environmental protection expenditures on the load capacity factor (which is a proxy for environmental quality, including biocapacity and ecological footprint) for EU countries for the period of 1995–2018. Their findings suggested that the existing level of environmental protection expenditures alone are insufficient to improve environmental quality. If EU countries want to meet their 2030 and 2050 targets, they need to increase the amount allocated to EPEs. Similarly, Akdag et al. (2024) [21] researched the effectiveness of EPEs in European countries for the period of 1995–2019. Their findings suggested that a one-unit increase in EPE led to a more than two-unit decrease in GHG emissions.

This brief literature review shows that governments intervene through budgetary tools, both in terms of environmental taxes and EPEs, to limit GHG emissions. This paper aims to contribute to the growing literature on the efficacy of public sector interventions. However, our focus was on the expenditure side. The goal was to contribute to the environmental economics literature by investigating the impact of EPEs on GHG levels.

3. Econometric Methodology, Data and Empirical Model

GHG emissions are the major contributor to climate change, affecting all aspects of our environment (Myhrvold and Caldeira, 2012; Zheng et al., 2019; Gallego-Schmid et al., 2020) [22,23,24]. The major greenhouse gases include carbon dioxide (CO₂), methane (CH₄), nitrous oxide (N₂O) and other fluorinated gases, such as hydrofluorocarbons (HFCs), perfluorocarbons (PFCs), sulfur hexafluoride (SF₆) and nitrogen trifluoride (NF₃) (see Figure 1).

This study examined the relationship between public expenditures on environmental protection (EPEs) and environmental outcomes (per capita greenhouse gas (GHG) emissions) using a comprehensive unbalanced panel dataset covering 117 countries from 2000 to 2023. These countries are Afghanistan, Albania, Algeria, Angola, Argentina, Australia, Austria, Azerbaijan, Bahamas, Bahrain, Bangladesh, Belarus, Belgium, Bhutan, Bolivia, Botswana, Brazil, Bulgaria, Burkina Faso, Burundi, Cambodia, Canada, Chile, China, Colombia, the Republic of the Congo, Costa Rica, Cote d’Ivoire, Croatia, Cyprus, Czechia, Denmark, the Dominican Republic, Egypt, El Salvador, Equatorial Guinea, Estonia, Fiji, Finland, France, Georgia, Germany, Greece, Guatemala, Hungary, Iceland, Indonesia, Iran, Ireland, Israel, Italy, Jamaica, Japan, Jordan, Kazakhstan, Kenya, Kiribati, the Kyrgyz Republic, Latvia, Lebanon, Liberia, Lithuania, Luxembourg, Madagascar, Malaysia, the Maldives, Malta, Mauritius, Micronesia, Moldova, Mongolia, Myanmar, Namibia, Nauru, Nepal, the Netherlands, New Zealand, Nicaragua, North Macedonia, Norway, Oman, Palau, Panama, Papua New Guinea, the Philippines, Poland, Portugal, Romania, Russia, Samoa, Senegal, Serbia, Seychelles, Singapore, Slovakia, Slovenia, the Solomon Islands, Somalia, South Africa, South Korea, Sweden, Switzerland, Tanzania, Thailand, Timor-Leste, Tonga, Trinidad and Tobago, Tunisia, Turkey, Uganda, Ukraine, the United Arab Emirates, the United Kingdom, Uruguay, Uzbekistan, Yemen and Zambia. The proposed model is

$$\begin{gathered} \mathrm{l}\mathrm{n}\left({GHG emissions}_{i, t}\right)={\beta }_0+{\beta }_1 environmental protection expenditures \left(EPE\right)+ \\ {\beta }_2 \mathrm{l}\mathrm{n}\left({GDP per capita}_{i, t}\right)+{\beta }_3{ln \left({GDP per capita}_{i, t}\right)}^2+{\beta }_4 { urban population}_{i, t}+{\beta }_5 {trade openness}_{i, t}+{\epsilon }_{i, t} \end{gathered}$$

where subscripts i and t represent the country and time, respectively—i is the cross section and t represents the time period (2000 to 2023). Some of the variables in the model were transformed into the natural logarithm form because the others were already in percentage form.

3.1. Data

The dependent variable in the analysis was greenhouse gas (GHG) emissions, expressed in CO₂ equivalent (CO₂e). GHG emissions account for the global warming potentials of various gases such as methane (CH₄) and nitrous oxide (N₂O)—it provides a standardized metric for cross-country comparisons. The emissions data were sourced from Our World in Data (https://ourworldindata.org/), which compiles information from the Global Carbon Project and United Nations Framework Convention on Climate Change (UNFCCC). The dependent variable was in natural logarithm form. The main explanatory variable was the share of EPEs in Gross Domestic Product (GDP).

To control for population differences, the dependent variable was per capita GHG emissions. Per capita GHG emissions is the main environmental metric in the literature that provide insights into both aggregate emissions and the efficiency of resource use. Empirical studies have demonstrated that fiscal policies can significantly influence environmental outcomes. For instance, Sharma (2011), Morley (2012) and Miceikienė et al. (2021) [8,25,26] showed that public spending on environmental services can reduce air pollutants such as nitrogen dioxide (NO₂), sulfur dioxide (SO₂) and carbon monoxide.

The variable of interest in this analysis was environmental protection expenditures (EPEs), which were measured as the share of GDP used for environmental protection expenditures. EPEs are the government expenditures deployed for the reduction, prevention or elimination of environmental degradation according to the Classification of the Functions of Government (COFOG). EPEs include expenditures on activities for pollution abatement, protection of the biodiversity landscape, and waste and wastewater management. The hypothesis was EPE is negatively correlated with GHG emissions.

Hypothesis 1:

There is a significant negative relationship between emissions and EPEs.

H0: 

There is no significant relationship between emissions and EPEs.

The second important variable was the development level of the countries, expressed as the Gross Domestic Product (GDP) per capita. Similar to the estimations by Morley (2012), Halkos and Paizanos (2013), Abid (2017), Gholipour et al. (2018), Le Gallo and Ndiaye (2021) and Caglar and Yavuz (2023) [1,8,9,13,14,19], the GDP per capita (expressed in 2015 USD) variable was included into the model to serve as a proxy for the economic development level and to reflect the availability of fiscal resources. This variable was used in the natural logarithm form. Higher-income countries may have a greater financial capacity to invest in environmental protection, but they may also exhibit higher emissions due to more energy-intensive economic activities.

Hypothesis 2:

There is a significant positive relationship between emissions and GDP per capita.

H0: 

There is no significant relationship between emissions and GDP per capita.

In the early stages of development, the economic activities of countries may cause environmental degradation. However, after reaching a certain level of per capita income, economic growth could lead to environmental improvements (Grossman and Krueger 1995) [27]. This is known as the environmental Kuznet curve (EKC). The EKC implies that GHG emissions per capita is an inverted U-shaped function of per capita income. Therefore, to test the EKC hypothesis, the squared GDP variable in the natural logarithm form was included into the model.

Hypothesis 3:

There is a significant negative relationship between emissions and (GDP per capita)².

H0: 

There is no significant relationship between emissions and (GDP per capita)².

Our two control variables were urban population and trade openness. Urban population was expected to be positively correlated with GHG emissions because of the environmental degradation problems associated with urbanization—pollution, traffic congestion and loss of green spaces. However, urban population could be negatively correlated with the environmental protection spending as well due to the higher levels of income in urban areas. In jurisdictions with higher incomes, there might be higher levels of willingness to spend on environmental protection (Le Gallo and Ndiaye 2021). The second control variable was trade openness, which is the share of exports and imports in GDP. Similar to Lopez et al. (2011), Halkos and Paizanos (2013), Abid (2017), Gholipour et al. (2018) and Le Gallo and Ndiaye (2021) [7,9,13,14,19], we investigated whether international trade affects the level of emissions. Our hypothesis was that greater trade openness leads to a lowering of GHG emissions. However, there is no consensus in the literature on the impact of trade openness on the environment. Dogan and Seker (2016), Destek et al. (2018), Park et al. (2018), Alola et al. (2019) and Leitao and Lorente (2020) [28,29,30,31,32] found a positive impact of trade on reducing GHG emissions, whereas Kasman and Duman (2015), Balsalobre-Lorente et al. (2018) and Tachie et al. (2020) [33,34,35] found a negative impact. Caglar and Yavuz (2023) [1] argued that trade openness can influence the environment through technology, scale and composition channels and the ultimate effect depends on which channel prevails.

We used three sources of data: (i) Our World in Data (https://ourworldindata.org/); (ii) the International Monetary Fund Government Finance Statistics (https://data.imf.org/?sk=a0867067-d23c-4ebc-ad23-d3b015045405, accessed on 2 February 2025); and World Bank World Development Indicators (https://databank.worldbank.org/source/world-development-indicators, accessed on 2 February 2025). After collecting data from these sources, we deployed data quality assessment methods, such as data validation, data auditing and data cleaning, focusing on ensuring the accuracy, completeness, consistency and validity of the data. A detailed description of the dependent and explanatory variables, as well as the data sources are presented in

Table A1. Variable Definition and Sources.

Variable	Abbreviation	Definition
Per Capita GHG Emissions	GHG emissions	Total greenhouse gas emissions including carbon dioxide, methane and nitrous oxide emissions from all sources, including land-use change per capita (measured in metric tons of CO2 equivalent). Unit: Per capita metric tons of Co2e Data source: CO2 and Greenhouse Gas Emissions—Our World in Data (available at https://ourworldindata.org/co2-and-greenhouse-gas-emissions) Accessed on 2 February 2025.
Share of Environmental Protection Expenditures in GDP	EPE	Government expenditure dedicated to environmental protection efforts. Unit: Percentage of GDP Data Source: International Monetary Fund (IMF), Statistics Department, 2021. Government Finance Statistics (GFS) Database (available at https://data.imf.org/?sk=a0867067-d23c-4ebc-ad23-d3b015045405) Accessed on 2 February 2025.
Share of Urban Population in Total	urban population	Percentage of urban population in total population. Unit: Percentage of total population Data Source: Food and Agriculture Organization and World Bank population estimates—World Development Indicators (available at https://databank.worldbank.org/source/world-development-indicators) Accessed on 2 February 2025.
Per Capita Gross Domestic Product	GDP per capita	Represents the Gross Domestic Product per capita, measured in constant 2015 USD. Unit: USD Data Source: World Bank national accounts data and OECD National Accounts data files—World Development Indicators (available at https://databank.worldbank.org/source/world-development-indicators) Accessed on 2 February 2025.
Trade Openness	trade openness	Sum of exports and imports of goods and services divided by gross domestic product (expressed as a percentage). Unit: Percentage of GDP Data Source: World Bank national accounts data and OECD National Accounts data files—World Development Indicators (available at https://databank.worldbank.org/source/world-development-indicators) Accessed on 2 February 2025.
Electricity Generation from Coal	Coal electricity	Electricity generation from coal. Unit: Measured in terawatt-hours. Data Source: Data on Energy by Our World in Data (available at https://github.com/owid/energy-data/blob/master/owid-energy-codebook.csv#L21) Accessed on 2 February 2025
Coal Production	Coal production	Coal production in a year. Unit: Measured in terawatt-hours. Data Source: Data on Energy by Our World in Data (available at https://github.com/owid/energy-data/blob/master/owid-energy-codebook.csv#L21) Accessed on 2 February 2025
Share of Electricity Generated by Coal	Coal share of electricity	Share of electricity generated using coal. Unit: Measured as a percentage of total electricity. Data Source: Data on Energy by Our World in Data (available at https://github.com/owid/energy-data/blob/master/owid-energy-codebook.csv#L21) Accessed on 2 February 2025
Total Electricity Demand per Person	Per capita electricity demand	Total electricity demand per person. Unit: Measured in kilowatt-hours per person. Data Source: Data on Energy by Our World in Data (available at https://github.com/owid/energy-data/blob/master/owid-energy-codebook.csv#L21) Accessed on 2 February 2025
Fossil Fuel Consumption per Capita	Per capita energy production from fossil fuels	Fossil fuel consumption per capita. Unit: Measured in kilowatt-hours per person. Data Source: Data on Energy by Our World in Data (available at https://github.com/owid/energy-data/blob/master/owid-energy-codebook.csv#L21) Accessed on 2 February 2025
Emissions from Electricity Generation	GHG emissions from electricity production	Emissions from electricity generation. Unit: Measured in megatonnes of CO2 equivalent. Data Source: Data on Energy by Our World in Data (available at https://github.com/owid/energy-data/blob/master/owid-energy-codebook.csv#L21) Accessed on 2 February 2025

.

3.2. Econometric Issues and Estimation

In this model, the error term, ${\epsilon }_{i, t}$, is assumed to be distributed independently in all time periods of a country and ${\beta }_0$ is the country-specific effect, which is also assumed to be distributed independently and constantly across the country. This means that no matter which value we choose for X, the error term ε must not show any systematic pattern and must have a mean of 0. However, any external factor not captured by the model, such as international environmental policy coordination, are collected in the error term. This might introduce an estimation bias. In a panel data environment, the fixed-effects model allows unobserved effects to be correlated with explanatory variables. On the other hand, the random-effects model assumes that the unobserved effects are not correlated with the explanatory variables.

Therefore, when estimating the model, the unobserved heterogeneity across countries needs to be accounted for. The standard approach is to use the fixed- (FE) or random- (RE) effects model formulations. The choice of the model depends on the diagnostic test result of the correlation between the cross-section specific error term and the explanatory variables. This study employed a step-by-step approach, estimating the pooled ordinary least square (POLS) first, and then the FE and RE panel models. Diagnostic tests were then applied and the three models were compared in order to select the most appropriate one [36].

The results of the POLS, FE and RE estimations are presented in

Table 1. Results for POLS, FE and RE models.

Variable	POLS GHG Emissions	FE GHG Emissions	RE GHG Emissions
Environmental protection expenditures (EPEs)	−0.265 * (0.036)	−0.033 * (0.019)	−0.044 ** (0.022)
GDP per capita (2015 USD)	−0.050 (0.024)	2.243 *** (0.157)	1.926 *** (0.166)
GDP per capita2	0.028 ** (0.012)	−0.112 *** (0.008)	−0.098 *** (0.008)
Trade openness	0.000 * (0.000)	−0.000 (0.000)	−0.000 (0.000)
Urban population	0.000 *** (0.001)	−0.028 *** (0.002)	−0.009 *** (0.002)
Constant	−0.522 (1.109)	−7.234 (0.735)	−6.837 *** (0.784)
Observations	1648	1648	1648
R-squared	0.50	0.30	0.17

Standard errors are shown in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

. The first column presents the results for the POLS method of estimation with the hypothesis that there are no differences between the data matrices of the cross-sectional dimension. In order to select between the POLS and FE models, an F test was conducted. The F-statistic was computed by comparing the residual sum of squares (RSS) from the POLS model to the RSS from the FE model. The null hypothesis was that all the constants are homogenous and therefore the POLS model is applicable. In

Table 2. Cross test results for model specification.

Model Specification
Cross Test	Test Statistics		Decision for Null Hypothesis
POLS vs. FE	F	68.39 (0.000)	Rejected
POLS vs. RE	ALM	8016.66 (0.000)	Rejected
FE vs. RE	Hausman	370.01 (0.000)	Rejected

Note: p-values are shown in parentheses.

, the F-statistic results are presented. The high F-statistic value that was significant at the 1 percent level (p-value = 0.000) indicates that the FE model explained significantly more variation in the data. Therefore, we preferred the FE panel model over the POLS model.

In order to compare the POLS model with the RE model, the Adjusted Lagrange Multiplier test (ALM) was applied. The null hypothesis was that the variance of the random effect is zero (the random effects are insignificant), and the POLS model is appropriate for the alternative hypothesis that the variance of the random effect is larger than zero. According to the result of the ALM test (p-value = 0.000), the null hypothesis was rejected; this means the random-effects model is more appropriate than the POLS model. The ALM test provided evidence of significant differences across countries; therefore, we decided to not use the POLS model.

The Hausman (1978) [37] test was used to choose between the RE panel model and the FE panel model. The null hypothesis was that the difference in coefficients is not structural. In Table 2, the results of the Hausman test (p-value = 0.000) indicated that the fixed-effects model was superior to the random-effects model. The fixed-effects model allows for different constants for each group. Using the fixed-effects estimator gave consistent results even when the estimators were correlated with the individual effects. The fixed-effects model captured all the effects that were specific to individual observations that did not vary over time.

Our step-by-step diagnostic approach indicated that the fixed-effects model was more reliable than the other two options. Before interpreting the results, we ran diagnostic tests on the FE estimations. We first started with the heteroscedasticity test to see if the standard error estimates were biased, which can lead to incorrect interpretations of the results. A modified Wald test statistic for groupwise heteroscedasticity in the fixed-effects model was used to detect if there is heteroskedasticity in the data (Greene, 2000) [38]. The modified Wald test statistic values led us reject the null hypothesis, indicating presence of heteroscedasticity (see

Table 3. Modified Wald test for groupwise heteroskedasticity in fixed-effects regression model.

H0: sigma(i)^2 = sigma^2 for all i chi2 (117) = 1,014,780.01 Prob > chi2 = 0.0000

).

In order to avoid potential heteroscedasticity problems, we used the cross-sectional time-series feasible generalized least square (FGLS) model, which is a more robust estimation method and can account for correlated errors and non-constant variances.

Table 4. Cross-sectional time-series FGLS regression.

Coefficients: generalized least squares Panels: heteroskedastic Correlation: no autocorrelation Estimated covariances = 117 Estimated autocorrelations = 0 Estimated coefficients = 6	Number of obs = 1648 Number of groups = 117 Obs per group: min = 1 avg = 14.08547 max = 22 Wald chi2(5) = 4448.58 Prob > chi2 = 0.0000
Variable
Environmental protection expenditures (EPEs)	−0.156 *** (0.021)
GDP per capita (2015 USD)	0.564 *** (0.139)
GDP per capita2	−0.010 (0.139)
Trade openness	−0.000 * (0.000)
Urban population	0.011 *** (0.000)
Constant	−3.178 *** (0.651)
Observations	1648
R-squared	0.50

Standard errors are shown in parentheses. *** p < 0.01, * p < 0.1.

presents the results for the FGLS estimators for the FE model.

Next, we tested for serial correlation. The correlation of the error term for each cross-sectional unit with the error term in the following period is called serial correlation (Bhargava et al. 1982) [39]. To detect the presence of serial correlation in the dataset, we used the Durbin–Watson test. The test statistic was less than 2, confirming the presence of serial correlation. Therefore, we ran the same model with robust standard errors.

Table 5. Fixed-effects model with robust standard errors.

	Number of obs = 1648 Number of groups = 117		Number of obs = 1640 Number of groups = 117
	R-squared: Within = 0.1831 Between = 0.3493 Overall = 0.3006 corr(u_i, Xb) = −0.82	Obs per group: min = 1 avg = 14.1 max = 22 F(5, 116) = 10.43 Prob > F = 0.000	R-squared: Within = 0.1932 Between = 0.3668 Overall = 0.3136 corr(u_i, Xb) = −0.83	Obs per group: min = 1 avg = 14.0 max = 22 F(5, 116) = 10.49 Prob > F = 0.000
Variable
Environmental protection expenditures (EPEs)	−0.033 (0.046)
Environmental protection expenditures (EPEs)(t−1)			−0.032 (0.045)
GDP per capita (2015 USD)	2.243 *** (0.472)		2.165 *** (0.495)
GDP per capita2	−0.112 *** (0.023)		−0.109 *** (0.024)
Trade openness	−0.000 (0.000)		−0.000 (0.000)
Urban population	−0.028 *** (0.007)		−0.027 *** (0.007)
Constant	−7.235 *** (0.007)		−6.816 *** (0.007)
Observations	1648		1640
R-squared	0.30		0.31

Robust standard errors are shown in parentheses. *** p < 0.01

presents the results of the FE estimators with robust standard errors. The first column in Table 5 presents the results for the proposed model. However, an important issue is time delay in the manifestation of the effects of EPEs on GHG emission outcomes. In order to examine whether the effects of environmental protection expenditures manifest with a delay, we estimated the fixed-effects model with robust standard errors in two ways: without a time lag and with a one-year time lag. The first column in Table 5 presents the results without a time lag for EPEs whereas the second column introduces a one-year time lag for the EPE variable.

In addition to examining the role of time delay, we also wanted to examine the role of income levels of the countries. In order to examine the income effect in the relationship between GHG emissions and EPEs, we conducted the same analysis without including low-income countries.

Table 6. Fixed-effects model with robust standard errors without low-income countries.

	Number of obs = 1291 Number of groups = 81		Number of obs = 1286 Number of groups = 81
	R-squared: Within = 0.3362 Between = 0.2730 Overall = 0.1622 corr(u_i, Xb) = −0.72	Obs per group: min = 2 avg = 15.9 max = 22 F(5, 80) = 16.29 Prob > F = 0.000	R-squared: Within = 0.3389 Between = 0.2655 Overall = 0.1535 corr(u_i, Xb) = −0.71	Obs per group: min = 2 avg = 15.9 max = 22 F(5, 80) = 11.71 Prob > F = 0.000
Variable
Environmental protection expenditures (EPEs)	−0.040 (0.045)
Environmental protection expenditures (EPEs)(t−1)			−0.042 (0.047)
GDP per capita (2015 USD)	4.501 *** (0.566)		4.417 *** (0.683)
GDP per capita2	−0.224 *** (0.027)		−0.219 *** (0.033)
Trade openness	−0.000 (0.000)		−0.000 (0.000)
Urban population	−0.018 *** (0.006)		−0.017 *** (0.005)
Constant	−18.859 *** (2.716)		−18.555 *** (3.363)
Observations	1291		1286
R-squared	0.16		0.15

Robust standard errors are shown in parentheses. *** p < 0.01

presents the results for 81 countries which had a GDP per capita greater than USD 10,000 in 2015. These countries were Albania, Algeria, Argentina, Armenia, Australia, Austria, Azerbaijan, Bahamas, Bahrain, Belarus, Belgium, Botswana, Brazil, Bulgaria, Canada, Chile, China, Colombia, Costa Rica, Croatia, Cyprus, Czechia, Denmark, the Dominican Republic, Ecuador, Egypt, Equatorial Guinea, Estonia, Fiji, Finland, France, Gabon, Georgia, Germany, Greece, Hungary, Iceland, India, Indonesia, Iran, Ireland, Israel, Italy, Japan, Kazakhstan, Kosovo, Kuwait, Latvia, Lebanon, Libya, Lithuania, Luxembourg, Malaysia, the Maldives, Malta, Mauritius, Mexico, Mongolia, Montenegro, Namibia, the Netherlands, New Zealand, North Macedonia, Norway, Oman, Panama, Paraguay, Peru, Poland, Portugal, Qatar, Romania, Russia, Saudi Arabia, Serbia, Seychelles, Singapore, Slovakia, Slovenia, South Africa, South Korea, Spain, Sri Lanka, Sweden, Switzerland, Thailand, Trinidad and Tobago, Tunisia, Turkey, Turkmenistan, Ukraine, the United Arab Emirates, the United Kingdom, the United States and Uruguay.

Lastly, we examined the role of other confounding factors in environmental degradation. We introduced six new variables that reflect the energy structure of countries in the model: electricity generation from coal, coal production, share of electricity generated by coal, total electricity demand per person, per capita fossil fuel consumption and GHG emissions from electricity generation.

Table 7. Fixed-effects model including energy structure variables with robust standard errors.

	Number of obs = 1291 Number of groups = 81		Number of obs = 1090 Number of groups = 79		Number of obs = 1291 Number of groups = 81
	R-squared: Within = 0.3460 Between = 0.2535 Overall = 0.1553 corr(u_i, Xb) = −0.74	Obs per group: min = 2 avg = 15.9 max = 22 F(5, 80) = 16.90 Prob > F = 0.000	R-squared: Within = 0.2968 Between = 0.2422 Overall = 0.1514 corr(u_i, Xb) = −0.75	Obs per group: min = 1 avg = 13.8 max = 22 F(5, 80) = 18.31 Prob > F = 0.000	R-squared: Within = 0.4092 Between = 0.0724 Overall = 0.0297 corr(u_i, Xb) = −0.61	Obs per group: min = 2 avg = 15.9 max = 22 F(5, 80) = 26.69 Prob > F = 0.000
Variable
Environmental protection expenditures (EPEs)	−0.036 (0.044)		−0.053 (0.051)		−0.020 (0.033)
GDP per capita (2015 USD)	4.380 *** (0.562)		4.354 *** (0.685)		3.841 *** (0.540)
GDP per capita2	−0.218 *** (0.027)		−0.218 *** (0.033)		−0.189 *** (0.026)
Trade openness	−0.000 (0.006)		−0.000 (0.006)		−0.000 (0.00)
Urban population	−0.020 *** (0.006)		−0.021 *** (0.007)		0.015 *** (0.005)
Electricity from coal	0.000 *** (0.000)
Coal production			0.000 ** (0.000)
Coal’s share of electricity					0.008 *** (0.001)
Constant	−18.089 *** (2.706)		−17.841 *** (3.253)		−16.210 *** (2.627)
Observations	1291		1090		1291
R-squared	0.16		0.15		0.03

Robust standard errors are shown in parentheses. *** p < 0.01

and

Table 8. Fixed-effects model including energy structure variables with robust standard errors.

	Number of obs = 1291 Number of groups = 81		Number of obs = 1007 Number of groups = 60		Number of obs = 1291 Number of groups = 81
	R-squared: Within = 0.3523 Between = 0.1943 Overall = 0.0913 corr(u_i, Xb) = −0.64	Obs per group: min = 2 avg = 15.9 max = 22 F(5, 80) = 14.60 Prob > F = 0.000	R-squared: Within = 0.4894 Between = 0.3887 Overall = 0.1861 corr(u_i, Xb) = −0.22	Obs per group: min = 1 avg = 16.8 max = 22 F(5, 80) = 23.02 Prob > F = 0.000	R-squared: Within = 0.3472 Between = 0.2503 Overall = 0.1531 corr(u_i, Xb) = −0.61	Obs per group: min = 2 avg = 15.9 max = 22 F(5, 80) = 16.27 Prob > F = 0.000
Variable
Environmental protection expenditures (EPEs)	−0.040 (0.045)		−0.064 (0.062)		−0.036 (0.044)
GDP per capita (2015 USD)	4.451 *** (0.569)		3.632 *** (0.913)		4.371 *** (0.0.562)
GDP per capita2	−0.223 *** (0.027)		−0.183 *** (0.045)		−0.218 *** (0.027)
Trade openness	−0.000 (0.006)		0.001 * (0.000)		−0.000 (0.00)
Urban population	−0.017 *** (0.006)		−0.013 * (0.007)		0.020 *** (0.006)
Per capita electricity demand	0.000 * (0.000)
Per capita energy production from fossil fuels			0.000 *** (0.000)
GHG emissions from electricity production					0.000 *** (0.000)
Constant	−18.662 *** (2.720)		−15.273 *** (4.442)		−18.042 *** (2.711)
Observations	1291		1007		1291
R-squared	0.09		0.19		0.15

Robust standard errors are shown in parentheses. *** p < 0.01, * p < 0.1.

present the results for the models with these variables.

4. Results

Our results provide some evidence that there is a negative correlation between EPEs and GHG emissions. Our EPE variable consistently demonstrated a negative sign across the different estimation models, with varying degrees of statistical significance (Hypothesis 1). However, it was not statistically significant in the last model with FE estimators. In the FE estimators with robust standard errors models, although the EPE variable had a negative sign, the p-value was higher than commonly acceptable levels in models with and without a one-year time lag. Therefore, it is fair to say there was no significant time delay effect in the relationship between GHG emissions and EPEs. In both models, the magnitudes of the coefficient of the EPE variable were close and in both models, they were not statistically significant. Similarly, the results for higher-income countries in Table 6 were very similar to the results for the full set of countries in Table 5. Table 6 presents the results for the same models for only 81 countries (as opposed to all 117 countries). The EPE variable had a negative sign without statistical significance in the models with and without the time lag. In that sense, there seems to be no income effect in the relationship between GHG emissions and EPEs. In both the developed and developing country groups, EPEs were negatively correlated with GHG emissions but without statistical significance. In Table 7, we present the results for the same 81 countries but introducing six energy structure variables—one at a time to avoid multicollinearity issues. The results with the various energy structure variables in Table 7 were very similar to those of the other models: the EPE variable had a negative sign with no statistical significance. However, in all six models, the energy structure variable was statistically significantly positively correlated with GHG emissions. Increases in electricity generation from coal, coal production, the share of electricity generated by coal, total electricity demand per person, per capita fossil fuel consumption and GHG emissions from electricity generation caused increases in GHG emissions.

In terms of Hypothesis 2, there was strong evidence that there was a significant positive relationship between GHG emissions and GDP per capita. The GDP variable was positively correlated with the GHG emission levels, with a very low p-value in almost all of the models. Clearly, countries produce GHG emissions in the initial stages of development. However, as they continue along the development path, their negative impact on the environment tapers out. This phenomenon is known as the environmental Kuznet curve (Hypothesis 3). We found strong evidence of the presence of the EKC. In the FE and RE models as well as the FE models with robust standard errors, the results showed a highly significant negative relationship between GDP per capita squared and GHG emissions. This finding presents evidence of an inverted U-shaped relationship between economic growth and environmental degradation. Although there is no consensus in the literature about the existence of the EKC, our findings add another piece of evidence in support of the presence of the EKC.

As far as the control variables were concerned, the findings across the various models were not consistent. The trade openness variable was not correlated with GHG emissions in a statistically significant manner in the different models. Only in the POLS, FGLS and FE with robust standard errors (with per capita energy production from fossil fuels variable) models, it was significant at the 10 percent level, but the signs were in different directions in the different models. For the urban population variable, although initially, the FGLS model presented a statistically significant positive relationship at the one percent level, the following estimation models presented a strong negative relationship. In the FE and RE models, there was a highly significant but negative relationship. Moreover, various FE with robust standard error models presented a very strong negative relationship between urbanization and GHG emissions in Table 5, Table 6 and Table 7.

5. Conclusions

This study investigated the efficacy of environmental protection expenditures in reducing GHG emissions. Environmental protection expenditures are proxies for efforts to prevent and reduce environmental degradation resulting from economic and social activities. They represent the monetary value of efforts directly aimed at preventing and reducing environmentally harmful activities.

Our empirical findings present a negative correlation between EPEs and GHG emission, which points towards EPEs’ effectiveness in the fight against climate change. EPEs have strategic value, which allow for an evaluation of climate change policies. However, the link between EPEs and environmental outcomes needs to be established more strongly at a granular micro level to guide policymakers in all countries so that budgetary resources are deployed to achieve the best outcome. Although our findings paint a positive picture of the impact of EPEs on GHG emissions at a macro level, the environmental economics research community needs better information on climate change adaptation and mitigation expenditures to deepen the analysis.

The Classification of the Functions of Government (COFOG) is the only available public expenditure data source to analyze EPEs for both developed and developing countries. It is a standard used to classify government expenditures by purpose. It was developed in its current version in 1999 by the Organization for Economic Co-operation and Development (OECD) and is published by the United Nations Statistical Division. It has three levels of details: divisions, groups and classes. Environmental protection expenditure is one of ten divisions (broad government objective), with six groups (sub-items), including waste management, water waste management, pollution abatement, protection of biodiversity and landscape and research and development related to environmental protection and environmental protection.

Since the adaption of the System of Environmental and Economic Accounting Central Framework in 2012, there have been efforts to revise the COFOG to better breakdown public expenditures at a granular level. These efforts focus on integrating biodiversity and climate change and broader sustainability concerns into expenditure decisions. Currently, the definition of environmental protection expenditures is loosely defined and the determination of whether an expenditure item contributes to the efforts against climate change is left to individual countries. More importantly, estimates of EPEs are self-reported by countries. The international community needs to develop standards for the identification and reporting of public expenditures for climate action. The recent efforts to establish internationally accepted reporting standards for the public sector on climate expenditures are encouraging. Recently, the International Public Sector Accounting Standards Board (IPSASB) issued a draft of groundbreaking climate-related disclosure standards specifically for the public sector. This will establish climate-related disclosure standards for the public sector, aiming to help governments report on their environmental impacts and actions to combat climate change. The challenge will be the implementation of these standards by governments. There needs to be concentrated efforts, especially by international organizations, to help countries to adopt and implement green budgeting principles as well as climate-related disclosure standards.

In terms of the relationship between GHG emissions and income levels, we found evidence of a positive relationship. The increase in GDP per capita increased GHG emissions to a certain point, after which, the level of emissions started to decline. This finding of the presence of the EKC suggests that the economic development of countries comes at the expense of environment. The detrimental impact of economic development on the environment, however, slows down after a certain income level. The presence of the EKC suggests that as countries reach a certain level of income, they manage to shift to a more environmentally friendly economic production structure.

The presence of the EKC has important global policy implications: higher-income countries relocate dirty industries to lower-income countries with less resources and capacity to tackle the climate change challenge. We need to find ways to support lower-income countries to increase the proportion of renewables in energy generation, which research has shown to have a positive impact on environmental quality (Destek et. al. 2018) [29]. Moreover, we need to investigate the EKC hypothesis further in order to pinpoint the turnaround level. At what income level do countries shift to a greener economic structure? What is the income level when economic growth stops contributing to environmental degradation? The answers to these questions will help the international community devise programs to support low-income countries in tackling climate change.