Environmental Kuznets Curve and the Pollution-Halo/Haven Hypotheses: An Investigation in Brazilian Municipalities

Abstract

There is much discussion on the non-linear relationship between economic growth and carbon dioxide (CO₂) emissions. Additionally, the effects of Foreign Direct Investment (FDI) on the environment are ambiguous, as both beneficial (i.e., pollution-halo) and harmful (i.e., pollution-haven) effects were found. Therefore, the literature presents no consensus on either of these topics. This is especially problematic for developing regions, as these regions represent growing economies interested in receiving foreign investments, and their CO₂-related research is limited. This study aims to understand the impacts of economic growth and FDI on the CO₂ emissions of São Paulo state, Brazil. To perform this study, a unique dataset on regional FDI was built, and 592 municipalities were included. The analyses combine linear and non-linear estimations, and the results suggest a non-linear relationship between Gross Domestic Product (GDP) per capita and CO₂ emissions, along with a negative association between FDI and CO₂. Finally, this study discusses possible policy implications and contributes to the international literature.

1. Introduction

Economic growth plays a vital role in countries’ economic and human development. Over the past 40 years, the world economy has more than tripled [1]. Despite economic growth raising living standards in most countries, it was also responsible for an increase in CO₂ emissions and a depletion of natural resources. The results presented in the IPCC report confirm that global warming is largely attributed to human activities, especially those related to CO₂ emissions [2].

According to Mardani et al. [3], the majority of previous studies for the past two dec-ades have been intensively focused on the study of the nexus between economic growth and energy use and their effects on CO₂ emissions.

Dinda [4] and Liu et al. [5] states that the investigation of the relationship between economic growth and CO₂ emissions mainly focuses on whether this relationship follows the Environmental Kuznets Curve (EKC) hypothesis. According to this hypothesis, the environmental quality deteriorates in the early stage of economic development/growth and improves in a later stage as the economy develops.

Most studies following this approach confirm the existence of an inverted U-shaped (and other curve shapes) for the relationship between growth and emissions (i.e., EKC) in many countries and regions [6,7,8,9]. Still, Al-mulali et al. [10] and Sadik-Zada and Ferrari [11] state that this transition in the GDP–CO₂ relationship depends on the region’s economic development stage and technology adoption, which arguably explains how the EKC theory still generates debate across distinct regions/countries.

Accordingly, pollutant emissions are the most critical environmental problem worldwide, and their reduction is part of the political agenda and international organiza-tions debate, such as the Paris Agreement and the 2030 Agenda for Sustainable Develop-ment [12]. According to the Sustainable Development Goals (SGD), countries and local governments must foster climate action (SGD-13) and global partnerships (SGD-17), as well as inclusive industrialization, innovation, and economic growth (SGD-8 and SGD-9) to achieve a better stage of sustainable development.

Moreover, policies must be adequate to deal with the reduction of CO₂ emissions without affecting economic growth, especially in developing countries, which have been stressing efforts to promote and upgrade their industrial activity [13,14]. Generally, productivity growth is associated with growth in real wages, and ultimately an improve-ment in living standards [14].

In this sense, Foreign Direct Investment (FDI) is a remarkable phenomenon for both developed and developing countries due to its notorious effects on globalization [15] and economic growth [16]. Bakhsh et al. [17] state that FDI contributes to the host economy by stimulating its economic development, by being a source of external finance, and by re-ducing the bridge between domestic savings and target investment. In sum, FDI facilitates local development, which further encourages employment opportunities for skilled and unskilled labor in the host country [17].

However, there is still some debate on the effects of FDI on carbon emissions. In sum, a multitude of studies argues FDI may negatively affect the environment due to its impact on economic growth [7,8,16,18,19]. While FDI inflows may rapidly affect the host economy and bring hope to the deprived populations, they may also increase environmental deg-radation and pollution [16]. In other terms, the promotion of economic growth by inward FDI is not free from environmental costs [20], which is alarming considering that nations, especially developing countries, are eager to attract foreign investments [21]. Indeed, He and Yao [16] assert that emerging countries may trade off FDI’s financial benefits against environmental losses.

Nonetheless, other studies pointed out that FDI may bring green technologies, which could improve the host country’s environmental quality and boost productivity [17,21]. Also, foreign investments might create new investments in the energy sector, promoting alternative energy resources such as wind turbines and photovoltaic cells [22]. These cleaner resources can reduce pollution caused by coal and diesel consumption Moreover, Shahbaz et al. [20] argue that the effect of FDI on the environment depends on the group of analyzed countries. For example, Yi et al. [23] found that green FDI has significantly re-duced China’s carbon emissions.

In addition to the current lack of consensus on the FDI–CO₂ relationship, a major limitation is the lack of regional studies. Research on FDI usually employs country-level data, with a few exceptions. For instance, He and Yao [16] studied FDI in 29 Chinese provinces to avoid significant regional disparities across the country. A similar approach was presented by Hille et al. [24] and their research in Korea. Zhang and Zhou [25] also argue that FDI studies rarely consider regional differences, although these differences may alter results. In fact, the international business literature recognizes the importance of regional or “subnational” differences in FDI activities [26,27], and claim that these subnational differences are particularly relevant for large emerging economies like the BRICS [28]. Specifically, Brazil’s inequality and regional differences are well documented [29].

To fill this gap, our study aims to analyze the impact of inward FDI on CO₂ emissions in Sao Paulo state, Brazil. We employed advanced econometric models and a unique pan-el dataset regarding FDI within 592 municipalities in São Paulo state ranging from 2010 to 2016. This dataset is the first and only regional FDI database available for Brazil, which implies an extensive manual check of approximately 22,000 export records to identify the source of capital of a company in a given municipality. Such data allows an in-depth analysis of FDI in the region and may serve as a base for future research worldwide.

We choose Sao Paulo for three reasons. First, Brazil has received little attention from the international literature. The absence of empirical studies in Brazil is particularly wor-risome considering the country’s inward FDI stock. The Sao Paulo state is particularly relevant for a more in-depth analysis of FDI in Brazil due to foreign investments in the re-gion. According to the Central Bank of Brazil [30], the state accounted for roughly 39% of foreign investments in the country, at least considering industrial enterprises. Besides, Sao Paulo presented the highest Gross Domestic Product (GDP) in 2016 and is the richest and most economically complex region in Brazil [31]. Second, Brazil is the leading air polluter in Latin America and is a country that experienced significant growth in recent years [32,33], with 441.8 million tons of CO₂ emissions in 2018 [34]. Sao Paulo state is responsi-ble for 71.50 × 106 tCO₂/year [35], which is remarkable to control environmental damage in the country. In addition, Brazilian policymakers do not fully understand the environmental impacts of such investments. Third, there is an absence of studies using subnational data for developing countries.

Accordingly, our study presents three contributions to the literature. First, we present an in-depth CO₂-related analysis in a developing region, which is important for local policymaking, especially for policymakers being pressured to promote economic growth with minimum environmental damage. Policymakers in other parts of Brazil and the world may access our policy recommendations. Second, due to the scarcity of data, we built a unique dataset with information about multinational companies located in Sao Paulo state using an export-oriented FDI measure similar to Moralles and Moreno [36] and Polloni-Silva et al. [37]. Third, our findings reveal the impact of FDI on CO₂ emissions at a municipal level, therefore contributing to the international debate on this issue.

The rest of this paper is organized as follows. The literature review on the impacts of FDI and the possible non-linear relationship between economic growth, FDI, and emis-sions is presented in Section 2. Section 3 focuses on the data gathering process and the es-timation strategy. Section 4 presents the main findings. Section 5 presents the robustness checks. Finally, Section 6 presents the policy implications of our findings, along with a brief conclusion.

2. Literature Review and Hypotheses Development

2.1. Pollution-Halo and Pollution-Haven

According to the literature, FDI may boost the host country’s economic growth while putting the environment at risk [16,20,38]. Although previous research has found incon-clusive results, scholars have attempted to explain the influence of FDI on the environ-ment, as it is often assumed to be connected to economic growth [39,40].

Nonetheless, the literature points to two alternative theories that associate FDI and the environment. Namely, the “pollution-haven” and “pollution-halo” theories. The pol-lution-haven hypothesis defends that inward FDI is attracted to weak regulatory and en-vironmental laws presented by the host country [17,41]. Thus, FDI should negatively affect the environment, especially in developing economies with weaker environmental standards [42]. In fact, several countries are eager to attract FDI [21] and may use their lax environmental regulations to compete in the “FDI tournament” [38]. On the contrary, the pollution-halo hypothesis claims that foreign firms, especially from developed economies, arrive with newer and cleaner technologies and better management practices capable of improving the environmental quality of the host country [17,21]. The pollution-halo hypothesis is especially relevant when domestic firms are considered pollution-intensive when compared to international standards [17].

Concerning the pollution-haven and pollution-halo studies, the results are debatable, once again. Zafar et al. [21] found evidence of FDI bringing advanced technology and in-novative products in the United States, which decreased environmental pollution as the pollution-halo theory suggests. Additionally, Jiang et al. [43] claim that FDI improved air quality in China, especially in populated cities.

By employing cointegration analysis, Nepal et al. [44] found that an increase in FDI inflows may reduce energy use, evidencing that energy-efficient techniques through FDI can reduce carbon emissions, which reinforces the importance of policies that encourage FDI inflow in renewable energy sectors. Evidence of pollution-halo in developing econo-mies is also found by Tiba and Belaid [15] and Yi et al. [23].

However, FDI may present a significant positive impact on environmental degrada-tion, thus supporting (especially in developing economies) the pollution-haven hypothe-sis. In this sense, He and Yao [16] found (at least to some extent) evidence to pollution-havens in China, while similar results are also presented by Jiang [8], Shahbaz et al. [20], and Opoku and Boachie [19]. Furthermore, Kastratović [42] studied 63 different de-veloping countries, including Brazil, and found evidence of pollution-havens. Zakarya et al. [45] studied the BRICS and also found evidence of pollution-havens, as FDI increased CO₂ emissions in the selected countries.

Yet, the results presented by Shao et al. [46] refuted the pollution-haven hypothesis for the BRICS. Thus, in essence, the literature still lacks conclusive results. As Nasir et al. [18] declared, FDI may have either a positive or negative impact on the host’s environ-ment. It depends on the research method, timeframe, and region [16,18,25]. De facto, in-conclusive or mixed results for developing countries were also found by Nadeem et al. [47], Chen and Yang [48], and Xu et al. [49]. Therefore, further research is essential to better comprehend the link between FDI and the environment.

Thus, considering the current literature on FDI, environment, and location choice, we formulate the following hypothesis:

Hypothesis 1 (H1).

The effect of FDI on CO₂ emissions is statistically significant.

Which can be partitioned into:

Hypothesis 1a (H1a).

FDI is negatively associated with CO₂ emissions (pollution-halo behavior);

Hypothesis 1b (H1b).

FDI is positively associated with CO₂ emissions (pollution-haven behavior).

2.2. Non-Linear Relationship between Economic Growth, FDI, and CO₂ Emissions

Global warming is one of the main concerns of world globalization in terms of in-dustrial development. There has been a steady growth in energy demands to sustain the development of countries. At the same time, there are increasing concentrations of green-house gases (GHGs) in the atmosphere, particularly carbon dioxide (CO₂), intensifying the natural greenhouse effect and the heat it generates. According to IPCC [2], human activities are estimated to have caused approximately 1.0 °C of global warming above pre-industrial levels.

Since the 1990s, some authors have come to defend that the interaction between eco-nomic growth and environmental problems has a certain regularity. This empirical evi-dence first emerged by Grossman and Krueger [50,51], in which the relationship between GDP per capita and emission of pollutants takes the form of a “U-inverted” or Environ-mental Kuznets Curve (EKC). Therefore, their work linked economic growth to environ-mental degradation in a non-linear way.

The EKC theory asserts that environmental degradation increases in the early stages of economic growth, and decreases when the income surpasses a certain threshold level [16,50]. In other words, the EKC theory suggests that at an early stage of economic devel-opment, people will demand more energy and natural resources, while neglecting the en-vironment. However, at a later stage of development, people may present a different mindset and demand a cleaner environment, energy-efficient products, and the protection of natural resources, as well as the environment, in general [18,21]. In summary, EKC rep-resents the notion that despite the environmental degradation at first, economic growth can improve the environmental quality in the long-run [18]. As FDI is often assumed to be connected to economic growth [39,40], multiple scholars are now analyzing FDI and its effects on the environment from the EKC perspective.

However, results are still inconclusive [18]. Both country-specific and diversified studies have been published. Fosten et al. [52] found significant evidence of EKC in the UK. Esteve and Tamarit [53] studied Spain and found that, in the long-run, economic growth reduces CO₂ emissions. However, Ghosh [54] found no conclusive evidence of this long-run relationship between growth and emissions in India, and similar results were found for Turkey [55], where pollution was not found to be associated with FDI or economic growth. Instead, Haug and Ucal [56] examines the effects of foreign trade and FDI on CO₂ emissions in Turkey and the results show that FDI has no statistically significant long-run effects and CO₂ intensity is not influenced by FDI, but an environmental Kuznets curve is present for CO₂.

More recently, Nasir et al. [18] found evidence of EKC in Asia, although their results were not statistically significant. Contrarily, the authors of [57] validated the EKC phe-nomenon in Malaysia. Furthermore, Onafowora and Owoye [58] analyzed the EKC phe-nomenon in selected economies such as Brazil, Mexico, China, Japan, and South Korea (among others) and only found evidence of EKC in developed economies. Malik et al. [59] studied the case of Pakistan and confirms the EKC hypothesis, and the symmetric results show that economic growth and FDI intensify carbon emission in both the long and short-run and Pakistan is a pollution haven for FDI.

The results of Liu et al. [5] revealed that there is a reverse U-shape relationship be-tween economic growth and carbon dioxide emissions in China, and an inverted N-shaped relationship between FDI and carbon dioxide emissions. However, he suggest-ed re-examining the results by grouping different regions.

Considering the literature, it is possible to infer that the results on the non-linear rela-tionship between economic growth and environmental degradation are still inconclusive [16].

Considering the current literature on economic growth, EKC, and FDI, the following hypotheses were formulated:

Hypothesis 2 (H2).

Economic growth presents a non-linear relationship with CO₂ emissions;

Hypothesis 3 (H3).

FDI presents a non-linear relationship with CO₂ emissions.

3. Data and Method

Our sample includes 592 out of the 645 municipalities in São Paulo state. We excluded 1 municipality for the lack of industry-related data and 52 small-sized municipalities (i.e., averaging 3000 inhabitants each) for the lack of energy-related data. Consequently, the employed sample comprises roughly 99.3% of the state’s population. Also, the data ranges from 2010 to 2016, resulting in a seven-year period and 4144 observations.

3.1. CO₂ Emissions

This study employs CO₂ emissions per capita as the dependent variable. However, a detailed (i.e., municipal-level) CO₂ database is not available at this moment. Previous scholars also faced this issue when studying CO₂ at a more micro-level. For this reason, we manually calculate the emissions from energy consumption for the municipalities of São Paulo, which is a common procedure in CO₂-related studies.

Therefore, we use data on electricity consumption and fossil fuels. Namely, the cal-culation includes gasoline, diesel, liquefied petroleum gas, fuel oil, aviation gasoline, and aviation kerosene. By following this calculation, this study is aligned with previous re-search, employing data from the Brazilian Energy Research Office (EPE) and the National Agency for Petroleum, Natural Gas and Biofuels (ANP). Yet, the Brazilian fuel legislature makes it inappropriate to calculate CO₂ emission directly. Thus, this study followed the IPCC [60] guidelines and subtracted the volume of ethanol added to gasoline C, which is commonly commercialized in the country, to use gasoline A in our calculations. The volume of biodiesel added to diesel oil was also subtracted.

Finally, the CO₂ emissions are calculated using the conversion factors presented in

Table 1. CO₂ conversion factors for various energy sources.

Type of Energy	Conversion Factor	Description
Electricity *	0.0512 (2010); 0.0292 (2011); 0.0653 (2012); 0.096 (2013); 0.1355 (2014); 0.1244 (2015); 0.0817 (2016)	tCO2/Mwh
Automotive gasoline	69.3	tCO2/TJ
Diesel oil	74.1	tCO2/TJ
Liquefied petroleum gas	63.1	tCO2/TJ
Fuel oil	77.4	tCO2/TJ
Aviation gasoline	70	tCO2/TJ
Aviation kerosene	71.5	tCO2/TJ

Source: Ministry of Science, Technology, Innovation, and Communications (MCTIC) and IPCC (2006). * The Brazilian government considers all electricity sources (i.e., hydro, natural gas, oil, biomass, coal, wind, nuclear, and solar) contributing to the national grid to define the CO₂ conversion factor.

.

3.2. Explanatory and Control Variables

This study is interested in better understanding the general effects of the development of São Paulo state on the environment. Accordingly, the regions’ GDP per capita is employed as a measure for economic growth, as done by previous CO₂-studies [7,61,62]. Although alternative variables (e.g., income levels) could provide valuable information on the overall richness of a region, there is a general consensus on the importance of GDP as the primary measure of how an economy or region is performing.

Additionally, in recent years, FDI received considerable recognition in many CO₂-related studies [5,15,62]. However, similar to the CO₂-issue presented earlier, São Paulo state does not present a regional FDI database. De facto, there are no governmental statistics on FDI at the state or city-levels [36], and this lack of data virtually explains the current absence of Brazilian studies on FDI and its regional effects, except for Moralles and Moreno [36] and Polloni-Silva et al. [37].

Therefore, we build a proxy for the regional FDI presence in each municipality using export-data from the Brazilian Integrated System of Foreign Trade (SISCOMEX). This sys-tem provides information on all exporting companies operating in São Paulo state, in-cluding their names, addresses, and export band value (e.g., up to 1 million BRL, between 1 and 5 million BRL, between 5 and 10 million BRL). With this information, a research group manually verified if these companies presented international origins (i.e., if the headquarters is located in a foreign country). For the whole São Paulo state during the 2010–2016 period, the research group performed approximately 22,000 individual checks. Hence, we follow Moralles and Moreno [36] in building a proxy variable for the intensity of FDI in a region, following Equations (1) and (2) below.

$$FD{I}_{jt}=W_{jt}\times \left(\frac{MN{C}_{jt}}{T{C}_{jt}}\right)$$

(1)

And,

$$W_{jt}=\frac{SF{E}_{jt}}{ST{E}_{jt}}$$

(2)

where FDI is the intensity of FDI for every region j at the year t, W is the adjustment weight regarding the exports from foreign companies, MNC is the number of multinational com-panies (i.e., foreign companies), TC is the total number of exporting companies, k is the number of foreign companies, SFE is the sum of exports from foreign companies, and STE is the sum of total exports (including national and foreign companies). The export band values (i.e., 1 million; 5 million; 10 million; 50 million, 100 million, and 320 million) are used to calculate SFE and STE, similar to Moralles and Moreno [36].

This FDI measure varies from 0 (no foreign companies) to 1 (all exports are depend-ent upon foreign companies). Thus, this measure represents the intensity of FDI for every municipality in every year of the sample and shows the relative importance of FDI for each region. With this information, our study will be the first to evaluate the municipal-level effects of FDI on the environment in a Brazilian region.

It should be noted that this FDI measure is not perfect as it excludes foreign compa-nies commercializing their products exclusively within the country. However, this meas-ure is arguably a valuable proxy for FDI considering that many foreign companies use Brazil as an export station to other Mercosul (Mercado Comum do Sul) countries [63]. As an example, roughly 22% of Paraguay’s imports and 24% of Argentina’s imports come from Brazil, and these products involve a multitude of industrial sectors (e.g., machinery, chemicals, clothing, cars) [64]. As the SISCOMEX system presents data on all exporting companies from all sectors, our FDI measure is arguably capable of including a considerable share of all foreign companies in the country. Additionally, a similar export-oriented proxy for FDI was already employed by Moralles and Moreno [36] and Polloni-Silva et al. [37].

Furthermore, regarding the control variables, previous scholars have employed a multitude of demographic, economic, and technology-related variables in previous stud-ies. For this study, the models include the industrial and service sectors’ share of total GDP, the share of electricity consumption in total consumption, and the population den-sity. Therefore, our model is similar to previous employed models [9,61,62,65].

Table 2. Descriptive statistics and variables description.

Variable	Description	Source	Observations	Mean	S.D.	Min.	Max.
CO2	CO2 per capita (tons)	-	4144	2.055	2.540	0.105	34.413
GDPPC	GDP per capita (2010 values; given in 1000 BRL)	Brazilian Institute of Geography and Statistics (IBGE)	4144	20.197	16.578	4.780	278.840
FDI	Foreign Direct Investment	-	4144	0.046	0.144	0.000	1.000
IND	Industrial sector (value added to total GDP; %)	IBGE	4144	19.365	13.303	2.224	85.966
SERV	Service sector (value added to total GDP; %)	IBGE	4144	39.147	11.086	7.412	75.640
RES	Share of residential electricity consumption in total consumption (%)	Secretary of Energy and Mining of the State of São Paulo	4144	34.679	12.072	0.177	66.417
DENS	Population density (population per square kilometer)	IBGE	4144	342.577	1301.545	3.698	13,534.820

presents the descriptive statistics and the sources used.

3.3. Explanatory and Control Variables

We start the analysis by verifying the linear effects of the explanatory variables on the regions’ CO₂ emissions. Thus, the primary model can be written as:

$$\ln {\mathit{CO}}_{2_{jt}}={\beta }_0+{\beta }_1\ln GDPP{C}_{jt}+{\beta }_{2}FD{I}_{jt}+{\beta }_3\ln {}^{\prime }{\mathbf{X}}_{jt}+a_j+e_{jt}$$

(3)

where CO₂ is the dependent variable for every municipality j at time t, GDPPC is the real GDP per capita, FDI is the aforementioned proxy for regional FDI intensity, X is the set of control variables, $a_j$ is the regional fixed effect, and e is the error term. All variables are employed in their natural-log form, except for FDI as the variable is already in percentage form. In addition to the linear effects of the other explanatory variables, this model formulation allows us to verify if FDI presents a pollution-haven or pollution-halo behavior.

However, this study also aims to test for possible non-linear relationships between economic growth, FDI, and CO₂ emissions. Thus, by following many previous scholars, we add quadratic terms to our model:

$$\ln {\mathit{CO}}_{2_{jt}}={\beta }_0+{\beta }_1\ln GDPP{C}_{jt}+{\beta }_2\ln GDPP{C}_{jt}^2+{\beta }_{3}FD{I}_{jt}+{\beta }_4\ln {}^{\prime }{X}_{jt}+a_j+e_{jt}$$

(4)

And,

$$\ln {\mathit{CO}}_{2_{jt}}={\beta }_0+{\beta }_1\ln GDPP{C}_{jt}+{\beta }_{2}FD{I}_{jt}+{\beta }_{3}FD{I}_{jt}^2+{\beta }_4\ln {}^{\prime }{X}_{jt}+a_j+e_{jt}$$

(5)

with these models, it is possible to test for U or inverted-U curves between GDP, FDI, and the CO₂ emissions, thus verifying the existence of EKC curves and better understanding how future development may affect the environment if São Paulo continues to grow and attract foreign investments.

Furthermore, it is necessary to consider possible non-spherical disturbances that are commonly found in panel datasets. Initially, we verified if our baseline models present collinearity issues (see Appendix A), which is a common issue in environmental studies [66]. Moreover, the Hausman test was employed to examine which specification (i.e., ran-dom or fixed-effects) was better suited for our models. Additionally, the modified Wald test for group-wise heteroscedasticity [67], the Wooldridge test for autocorrelation [68], and the Pesaran test for cross-sectional dependence [69] were used. In sum, these tests’ results show that the fixed-effects model was the preferable specification and that the sample shows heteroscedasticity, autocorrelation, and cross-sectional dependence problems.

To deal with these issues, CO₂-studies have been using the Driskoll–Kraay [70] esti-mator [9,61,71], as this technique deals with heteroscedasticity and autocorrelation prob-lems and is robust to general forms of cross-sectional (spatial) and temporal dependence with panels of all sizes, including an N > T dataset [72]. Therefore, the fixed-effects Dris-coll–Kraay (DK) estimator is the main technique employed in the analyses. Yet, the sam-ple is limited as a higher time frame is not available at the moment. Thus, to take ad-vantage of the higher degrees of freedom as the random-effects specification has, the Fea-sible Generalized Least Squares (FGLS) technique was employed to double-check our findings. FGLS is usually a random-effects estimator capable of dealing with heterosce-dasticity, autocorrelation, and cross-sectional correlation [68,73].

4. Main Results

Our main findings are presented in

Table 3. DK and FGLS estimations for the whole sample.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	DK	FGLS	DK	FGLS	DK	FGLS
GDPPC	0.309 ***	1.372 **	0.901 ***	5.518 **	0.308 ***	0.913 ***
	(0.0251)	(0.685)	(0.120)	(2.457)	(0.0252)	(0.306)
GDPPC2			−0.0916 ***	−0.729 **
			(0.0164)	(0.348)
FDI	−0.0614 *	−2.563 **	−0.0553 *	−0.636	−0.00806	−2.278
	(0.0336)	(1.265)	(0.0319)	(1.177)	(0.0975)	(9.335)
FDI2					−0.0551	0.324
					(0.0762)	(9.171)
IND	0.0458 ***	1.340 **	0.0427 **	0.607	0.0459 ***	1.721 ***
	(0.0138)	(0.535)	(0.0182)	(0.477)	(0.0137)	(0.591)
SERV	0.503 ***	2.033 **	0.489 ***	−0.584	0.503 ***	1.662 **
	(0.0872)	(0.859)	(0.0985)	(1.020)	(0.0871)	(0.784)
RES	−0.221 ***	−0.250	−0.218 ***	−0.152	−0.221 ***	1.383
	(0.0381)	(0.364)	(0.0390)	(0.328)	(0.0380)	(1.409)
DENS	0.990 **	−1.376 *	1.048 ***	−0.0203	0.988 **	−0.947
	(0.401)	(0.778)	(0.403)	(0.736)	(0.401)	(0.638)
Constant	−5.800 ***	−7.787 **	−6.909 ***	−7.523 *	−5.792 ***	−13.43 *
	(2.130)	(3.648)	(2.270)	(4.055)	(2.128)	(7.236)
F	228.10 ***		193.48 ***		728.43 ***
Wald Chi2		33.97 ***		27.99 ***		41.45 ***
Hausman	117.64 ***		200.59 ***		114.69 ***
Mod. Wald.	190,000 ***		200,000 ***		190,000 ***
Wooldridge	343.60 ***		328.73 ***		340.18 ***
Pesaran	294.31 ***		290.91 ***		294.45 ***
Observations	4144	4144	4144	4144	4144	4144
Municipalities	592	592	592	592	592	592

*** p < 0.01, ** p < 0.05, * p < 0.1. Standard errors in parentheses. DK: Fixed-effects regression with Driscoll–Kraay standard errors. FGLS: Feasible Generalized Least Squares.

, which shows the linear and quadratic relationships between our explanatory variables and the regions’ CO₂ emissions. As shown, economic growth increases CO₂ emissions, as expected. Moreover, Table 3 shows that a non-linear relationship between GDP per capita and CO₂ emissions exists, as the quadratic term for GDP per capita is significant and negative. Thus, Table 3 suggests that after a certain threshold, the overall growth/richness of a region, here represented by the GDP per capita, helps to decrease the emission level.

Additionally, FDI has a linear and negative relationship with CO₂ emissions, which indicates that FDI presents a pollution-halo behavior. Thus, these results are aligned with some previous studies defending the pollution-halo hypothesis for FDI such as Jiang et al. [74], Zhang and Zhou [25], and Shao et al. [46]. Indeed, FDI seems to be environmentally beneficial to the region. Furthermore, a non-linear relationship between FDI and CO₂ is not found, as the quadratic term is not significant. Yet, before declaring support for any hypothesis, further tests will be made.

Regarding the control variables, Table 3 shows that both the industrial and the ser-vice sectors contribute to higher emission levels. Again, this demonstrates that economic growth is environmentally costly. The results demonstrate that the industrial sector is positive and significant, while FDI is negative and significant. Thus, the overall industrial sector increases the emission levels, while a higher FDI intensity is beneficial.

Likewise, the service sector, which is ignored by many scholars, also increases pollu-tion. This result is expected considering that the CO₂ measure employed by our study is calculated using energy consumption. De facto, the service sector is expanding in the re-gion, and many services, particularly in highly urbanized regions, involve road transpor-tation, which generates higher levels of pollution [75]. In addition, a higher share of resi-dential electricity consumption in the total consumption presents negative coefficients, indicating that higher shares of consumption by other sectors (e.g., industry) result in higher CO₂ than the residential sector. In other words, the impact from the residential sec-tor is lower than other sectors. Also, regions with higher population density levels present higher emissions, as these regions are arguably larger and more populated.

To further evaluate if the results from Table 3 are reliable, we split the sample into sub-samples. First, we estimated the models exclusively with the municipalities with foreign companies operating during the 2010–2016 period, therefore excluding smaller (i.e., less populated/developed) regions. Moreover, we excluded the city of São Paulo and the São Paulo metropolitan region, as these municipalities represent significantly well-developed and populated regions. São Paulo city and its neighborhood are responsible for a significant share of the state’s GDP and CO₂ emissions and therefore represent dissident regions compared to the rest of the state. In sum, results from

Table 4. DK estimations for the sub-samples.

Variables	Municipalities with FDI during the 2010–16 Period			Excluding São Paulo City			Excluding São Paulo Metropolitan Region
	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)	(15)
GDPPC	0.182 ***	1.714 ***	0.180 ***	0.182 ***	1.714 ***	0.179 ***	0.210 ***	2.048 ***	0.207 ***
	(0.0477)	(0.125)	(0.0465)	(0.0477)	(0.125)	(0.0464)	(0.0524)	(0.280)	(0.0506)
GDPPC2		−0.214 ***			−0.214 ***			−0.254 ***
		(0.0229)			(0.0230)			(0.0418)
FDI	−0.120 ***	−0.123 ***	0.0884	−0.120 ***	−0.123 ***	0.0886	−0.120 ***	−0.125 ***	0.102
	(0.0324)	(0.0305)	(0.0841)	(0.0324)	(0.0305)	(0.0844)	(0.0324)	(0.0310)	(0.0892)
FDI2			−0.211 ***			−0.211 ***			−0.221 ***
			(0.0740)			(0.0742)			(0.0841)
IND	0.00849	−0.00715	0.00931	0.00822	−0.00755	0.00907	0.0127	0.000710	0.0149
	(0.0342)	(0.0420)	(0.0329)	(0.0348)	(0.0427)	(0.0335)	(0.0384)	(0.0474)	(0.0377)
SERV	0.234 ***	0.218 ***	0.228 ***	0.234 ***	0.218 ***	0.229 ***	0.309 ***	0.307 ***	0.302 ***
	(0.0616)	(0.0478)	(0.0632)	(0.0617)	(0.0484)	(0.0633)	(0.0772)	(0.0471)	(0.0791)
RES	−0.0748 ***	−0.0720 ***	−0.0752 ***	−0.0744 ***	−0.0717 ***	−0.0748 ***	−0.0669 ***	−0.0648 ***	−0.0667 ***
	(0.0148)	(0.0145)	(0.0145)	(0.0149)	(0.0145)	(0.0146)	(0.0171)	(0.0170)	(0.0169)
DENS	0.726	0.782	0.726	0.722	0.779	0.722	0.565	0.623	0.560
	(0.657)	(0.605)	(0.654)	(0.657)	(0.605)	(0.653)	(0.682)	(0.613)	(0.676)
Constant	−4.993	−7.883 **	−4.985	−4.951	−7.840 **	−4.942	−4.074	−7.569 **	−4.041
	(3.939)	(3.741)	(3.910)	(3.928)	(3.729)	(3.898)	(3.684)	(3.703)	(3.647)
F	51.36 ***	1189.08 ***	14.29 ***	50.86 ***	1189.18 ***	55.28 ***	23.90 ***	1329.63 ***	10.30 ***
Hausman	63.40 ***	37.74 ***	78.87 ***	62.50 ***	32.85 ***	77.97 ***	43.59 ***	93.89 ***	61.90 ***
Mod. Wald.	4516 ***	18185 ***	4584 ***	4503 ***	18120 ***	4576 ***	4067 ***	9411 ***	4344 ***
Wooldridge	58.98 ***	47.52 ***	58.76 ***	58.67 ***	47.29 ***	58.45 ***	56.56 ***	43.60 ***	56.77 ***
Pesaran	128.97 ***	120.66 ***	128.38 ***	127.32 ***	119.04 ***	126.74 ***	94.41 ***	87.71 ***	93.73 ***
Observations	1001	1001	1001	994	994	994	784	784	784
Municipalities	143	143	143	142	142	142	112	112	112

*** p < 0.01, ** p < 0.05, * p < 0.1. Standard errors in parentheses. DK: Fixed-effects regression with Driscoll-Kraay standard errors. Sub-sample 1: only municipalities with FDI during the whole time period. Sub-sample 2: São Paulo city excluded. Sub-sample 3: São Paulo metropolitan region excluded.

confirm the effects of GDP and FDI on the environment and are aligned with Table 3.

The results from Table 4 confirm the findings in Table 3. In sum, GDP seems to pre-sent a non-linear relationship with CO₂. An inverted U-shaped relationship between GDP and CO₂ seems to represent the Brazilian reality similar to the original Simon Kuznets [76] curve. Moreover, FDI is, again, linear and negatively associated with emissions, thus being environmentally beneficial. Once more, before declaring support for the hypotheses, we perform robustness checks.

5. Robustness Checks

The robustness checking section is twofold. First, we verify if endogeneity is an issue. Second, we evaluate if a non-linear relationship is significant using the threshold regres-sion technique, as some scholars presented criticism towards the use of quadratic varia-bles to test for non-linearity.

The main explanatory variables (GDP per capita and FDI) were tested for endogenei-ty. Sustainability studies showed that economic growth and environmental degradation may suffer from simultaneity problems. Additionally, the international business literature argues that foreign companies do not choose their subsidiaries’ locations randomly. For-eign firms may be attracted to some of the host’s characteristics [26]. If foreign firms are choosing to invest in more developed regions, and these regions generate more pollution, then the model may suffer from reverse causality.

Therefore, the C-statistic test (also known as the difference-in-Sargan test) was em-ployed to verify endogeneity issues [77] and we employed the lagged values as instru-ments, following Yue et al. [78] and Song [79]. Previous scholars argue that finding in-struments to employ in CO₂-related studies is difficult, but defend that past outputs should not impact present emissions. Therefore, one-year lagged values of GDP per capita and FDI were used. Results (see Appendix B) show that FDI is exogenous, but endogeneity is significant for GDP per capita.

To mitigate this issue, we employed the Limited Information Maximum Likelihood (LIML) estimation technique. Previous studies commented that the LIML technique is bet-ter than the traditional Two-Stages Least Squares (2SLS) technique, and more capable of ensuring consistency and asymptotic normality [80]. LIML presents good results even in smaller samples [81] and reduces the bias of the estimations [82].

The results presented in

Table 5. LIML estimations.

Variables	Whole Sample			Municipalities with FDI during the 2010–16 Period			Excluding São Paulo City			Excluding São Paulo Metropolitan Region
	(16)	(17)	(18)	(19)	(20)	(21)	(22)	(23)	(24)	(25)	(26)	(27)
GDPPC †	0.526 ***	1.356 **	0.527 ***	0.696 **	0.649	0.696 **	0.695 **	0.649	0.696 **	1.002 **	0.785	1.002 **
	(0.105)	(0.592)	(0.106)	(0.289)	(0.552)	(0.290)	(0.290)	(0.552)	(0.290)	(0.504)	(0.596)	(0.505)
GDPPC2 †		−0.127			0.00735			0.00733			0.0253
		(0.115)			(0.118)			(0.118)			(0.106)
FDI	−0.0989 **	−0.130 **	−0.169	−0.139 *	−0.136 *	−0.0115	−0.139 *	−0.135 *	−0.0106	−0.168 *	−0.151 *	−0.118
	(0.0451)	(0.0567)	(0.153)	(0.0783)	(0.0821)	(0.207)	(0.0782)	(0.0821)	(0.207)	(0.0899)	(0.0841)	(0.244)
FDI2			0.0712			−0.128			−0.128			−0.0485
			(0.144)			(0.183)			(0.183)			(0.207)
IND	−0.00756	−0.0952	−0.00779	0.0155	0.0128	0.0170	0.0154	0.0126	0.0168	0.0517	0.0342	0.0524
	(0.0240)	(0.0826)	(0.0241)	(0.0746)	(0.0680)	(0.0748)	(0.0749)	(0.0683)	(0.0751)	(0.106)	(0.0811)	(0.106)
SERV	0.551 ***	0.980 ***	0.552 ***	0.454 *	0.433	0.452 *	0.454 *	0.433	0.453 *	0.759 *	0.636	0.758 *
	(0.0796)	(0.320)	(0.0799)	(0.255)	(0.304)	(0.256)	(0.256)	(0.305)	(0.257)	(0.404)	(0.406)	(0.406)
RES	−0.237 ***	−0.254 ***	−0.237 ***	−0.0940 ***	−0.0928 ***	−0.0939 ***	−0.0934 ***	−0.0922 ***	−0.0933 ***	−0.0952 ***	−0.0893 ***	−0.0951 ***
	(0.0706)	(0.0773)	(0.0706)	(0.0256)	(0.0280)	(0.0257)	(0.0256)	(0.0280)	(0.0256)	(0.0277)	(0.0277)	(0.0278)
DENS	0.450 ***	0.0114	0.452 ***	0.246	0.245	0.245	0.240	0.239	0.239	−0.0767	−0.0434	−0.0781
	(0.164)	(0.416)	(0.165)	(0.339)	(0.327)	(0.340)	(0.340)	(0.327)	(0.340)	(0.453)	(0.407)	(0.453)
F	17.59 ***	10.56 ***	15.17 ***	3.19 ***	4.28 ***	2.98 ***	3.17 ***	4.26 ***	2.97 ***	2.24 **	3.17 ***	2.09 **
Observations	3552	3552	3552	858	858	858	852	852	852	672	672	672
Municipalities	592	592	592	143	143	143	142	142	142	112	112	112

*** p < 0.01, ** p < 0.05, * p < 0.1. Standard errors in parentheses. ^† Predicted using one-year lagged values as instruments.

demonstrate that our linear results are consistent, as they are aligned with the previous estimations in Table 3. GDP per capita is positively as-sociated with CO₂ emissions. Conversely, FDI is negatively associated with emissions.

Furthermore, there is the issue of possible non-linear relationships between our ex-planatory variables and CO₂ emissions. Previously, we employed the quadratic forms of GDP and FDI in the models, as it is done by the majority of EKC studies [7,15,62]. Recently, however, scholars have commented on the limitations of this method. Girma [83] argues that, although the quadratic specification tends to be more informative than the linear model, it establishes a very specific functional form on the non-linearity that may not be a good representation of the pattern in the data. He and Yao [16] pointed out that, although EKC is traditionally measured in quadratic or even cubic forms, this specification is arbitrary and subjective. For these reasons, scholars are now using more novel methods such as nonparametric regressions and the Non-linear Autoregressive Distributed Lag (NARDL) in non-linear studies [84]. Particularly, the threshold regression is being em-ployed by CO₂-related studies [16,85,86].

Thus, this study also investigates the non-linear behavior of the variables using the fixed-effect panel threshold regression proposed by Hansen [87] as presented in Equations (6) and (7).

$$\ln {\mathit{CO}}_{2_{jt}}={\beta }_0+{\beta }_1\ln GDPP{C}_{jt} I\left(GDPP{C}_{jt}\le \mathsf{\lambda }\right)+{\beta }_2\ln GDPP{C}_{jt} I(GDPP{C}_{jt}>\mathsf{\lambda })+{\beta }_{3}FD{I}_{jt}+{\beta }_4\ln {}^{\prime }{X}_{jt}+a_j+e_{jt}$$

(6)

And,

$$\ln {\mathit{CO}}_{2_{jt}}={\beta }_0+{\beta }_1\ln FD{I}_{jt} I\left(FD{I}_{jt}\le \mathsf{\lambda }\right)+{\beta }_2\ln FD{I}_{jt} I(FD{I}_{jt}>\mathsf{\lambda })+{\beta }_{3}GDPP{C}_{jt}+{\beta }_4\ln {}^{\prime }{X}_{jt}+a_j+e_{jt}$$

(7)

where I(·) is an indicator function. We follow Girma [83] and employ a set of quantiles (1%, 1.25%, 1.50%, …, 98.75%; 99%) to estimate the threshold parameter (λ). Then, we test the null hypothesis of no non-linearity (H₀: ${\beta }_1={\beta }_2$) using the likelihood ratio test statistics and their bootstrapped p-values.

Table 6. Test results for the threshold effects.

Variables	Threshold	RSS	MSE	F	p-Value
GDPPC	Single (λ1)	154.769	0.037	20.300	0.077
	Double (λ2)	154.374	0.037	10.580	0.340
	Triple (λ3)	154.037	0.037	9.050	0.573
GDPPC †	Single (δ1)	102.709	0.029	24.930	0.054
	Double (δ2)	102.410	0.028	10.350	0.636
	Triple (δ3)	102.186	0.028	7.780	0.858
FDI	Single (ϕ1)	155.377	0.038	4.040	0.773
	Double (ϕ2)	155.250	0.038	3.370	0.760
	Triple (ϕ3)	155.178	0.038	1.920	0.937

^† Predicted using one-year lagged values as instruments.

shows that a non-linear relationship is only significant for the GDP–CO₂ relationship, with one significant threshold point.

Therefore, the threshold tests confirm the linear relationship between FDI and CO₂, and FDI seems to be negatively associated with emissions. Therefore, the results support H1a and reject H1b.

Table 7. Panel threshold estimations.

Variables	(28)	(29)
GDPPC (≤λ1)	0.351 ***
	(0.0543)
GDPPC (>λ1)	0.307 ***
	(0.0508)
GDPPC † (≤δ1)		0.636 ***
		(0.117)
GDPPC † (>δ1)		0.574 ***
		(0.108)
FDI	−0.0585	−0.0874
	(0.0792)	(0.0581)
IND	0.0456 *	0.0536 **
	(0.0247)	(0.0209)
SERV	0.493 ***	0.279 ***
	(0.0655)	(0.0483)
RES	−0.218 **	−0.222 **
	(0.0952)	(0.104)
DENS	1.022 ***	0.753 ***
	(0.211)	(0.197)
Constant	−6.018 ***	−4.925 ***
	(0.711)	(0.718)
F	23.47 ***	13.33 ***
Observations	4144	3552
Municipalities	592	592

*** p < 0.01, ** p < 0.05, * p < 0.1. Robust standard errors in parentheses. ^† Predicted using one-year lagged values as instruments.

shows the estimated coefficients for GDP following two distinct regimes: lower and higher levels of GDP. These results suggest that, indeed, there is a non-linear relationship between GDP and CO₂. After the threshold point, the impact of GDP on the environment is lower, meaning that richer regions (i.e., more developed) tend to have a smaller environmental impact, which is beneficial to the region.

Nonetheless, the threshold regression does not support the EKC hypothesis, as GDP still presents positive effects in both regimes. Hence, Table 7 shows support for H2 and rejects H3. São Paulo state does not present the inverted-U curve that European [88] and Asian [5,7,62] regions present. In sum, the GDP–CO₂ relationship is not linear or quadratic. The policy implications of these findings will be discussed in the next section.

6. Conclusions and Policy Implications

The findings show that the relationship between economic growth and emissions is environmentally costly and non-linear. To some extent, GDP increases CO₂ emissions, which corroborates with previous studies’ understanding [21,89]. Other studies proposed an inverted U-shape curve between economic growth and emissions [5,8,15,19]. Our study contributes by showing that GDP is non-linear with CO₂ emissions, although this relationship did not present the U-shape form previously proposed. In this sense, we found that GDP becomes less environmentally damaging after a certain threshold point. This result supports hypothesis 2 (H2: Economic growth presents a non-linear relationship with CO₂ emissions).

Additionally, the results reveal a negative impact of FDI on CO₂ emissions, therefore supporting hypothesis 1 (H1a: FDI is negatively associated with CO₂ emissions). This result corroborates with previous studies that advocate in favor of FDI to reduce pollutant emissions [6,21,23,78]. In contrast, our main finding presents evidence against studies arguing that foreign investments increase environmental damage [7,8,16,19,62,90]. The dataset we employed shows that the majority of foreign companies operating in São Paulo state have origins in developed economies (e.g., United States, Germany, and Japan), and are operating in medium and high-tech sectors (e.g., cars, energy, mobile technologies, and electronics). Therefore, the negative impact of FDI on CO₂ emissions might result from multinational companies with green practices and higher levels of productivity. This is relevant because public authorities in developing regions must create specific policies to attract international enterprises to generate green growth conditions. We argue that FDI must be one of the major drivers to promote economic growth along with sustainable development.

There are two main useful contributions for policymakers. First, the non-linear effect of GDP on emissions might encourage public authorities to find alternative economic growth strategies. Arguably, local policymakers have an essential role in boosting eco-nomic growth while reducing power consumption, as energy policies should be managed at the local level [91]. Second, FDI is an essential policy tool to create green development paths in Sao Paulo state. In this sense, local policies must focus on strategies to attract green technologies [92], clean energy sources [93], green jobs and products [94], and higher productivity [95]. Note that FDI is crucial to promote economic growth because it guarantees access to foreign firms’ knowledge and technologies to transform the economic structure. For example, the industry is a relevant sector to the Sao Paulo region. Public authorities could reduce taxes to national and multinational companies that emit less CO₂. A tax policy could also lead green companies to regions that are already presenting higher levels of CO₂, such as the Sao Paulo city and densely populated regions [96]. New firms might also create local new jobs, increase the local income, and generate green economic growth paths. This new cycle would generate development with lower CO₂ emissions. Besides, alternative transport policies are required in metropolitan areas, such as efficient public transport for passengers, vehicle utilization improvements, and fuel efficiency to reduce pollutant emissions [75,97]. The public sector might invest in environmental projects to develop national green products, such as photovoltaic cells, smart grids, bioplastics, among others [98]. Also, the vast regions in Sao Paulo countryside could focus on alternative energy resources [99], such as the electric energy from sugarcane mills or by building photovoltaic power plants.

Noteworthy, Sao Paulo is the most developed state in Brazil. It means that Sao Paulo has adequate infrastructure (roads, railways, ports, and airports) to connect the region with the entire world. Also, Sao Paulo presents one of the highest human development in Brazil, which is crucial to create human capital (education, health, income, among others). These conditions favor the attraction of FDI to Sao Paulo state. Other emerging regions may use the policy recommendations presented here to facilitate their development.

Finally, some limitations of this study are expected to open multiple avenues for future research. First, this study measures CO₂ emissions through consumption. Future studies are encouraged to examine the impact of FDI and GDP on other types of emissions (e.g., sulfur dioxide emissions (SO₂), air quality, or ecological footprint). Second, this study did not verify how renewable and non-renewable energy sources can individually affect São Paulo state’s emissions. Third, we did not control the effect of economic structure and the type of FDI on pollution.

Despite these limitations, this study contributes to the literature on growth, foreign investments, and CO₂ emissions. As previously described, there is no consensus on the effects of growth and FDI on the environment, particularly in emerging economies. To the best of our knowledge, this is the first study to apply Brazilian regional FDI data to better comprehend its environmental consequences. To perform such an analysis, a handmade and unique dataset was built, and more novel econometric techniques were employed. Our findings have important policy implications and are arguably capable of guiding future research in developing regions.