Reassessing the environmental Kuznets curve: a multidimensional analysis of air pollution indices and economic development in ASEM

The relationship between economic growth and environmental degradation has long been a central topic in environmental and development economics. A key framework for understanding this relationship is the Environmental Kuznets Curve (EKC) hypothesis, which posits an inverted-U shaped relationship between environmental degradation and income per capita. In the early stages of economic growth, pollution increases as production expands, and industrialization accelerates. However, after surpassing a certain income threshold, further growth leads to improved environmental quality as structural change, technological innovation and stronger environmental regulations take effect (Grossman and Krueger, 1995; Panayotou, 1997; De Bruyn et al., 1998; Dinda, 2004). This hypothesis has significant policy implications: if the EKC holds, economic growth may eventually become compatible with environmental improvement; if not, growth alone may not ensure sustainability (Jaeger et al., 2023).

The EKC hypothesis has been extensively examined in empirical literature, using a wide range of data, pollutants and econometric methods. Many studies found inverted-U shaped relationships between economic growth and environmental degradation, supporting the traditional EKC hypothesis (Al-Mulali et al., 2013; Ahmed et al., 2019; Ridwan et al., 2024; Wang et al., 2024a, 2024b). However, others have found no evidence of EKC, or alternative functional forms such as N-shaped curves, where environmental degradation declines at intermediate income levels but rises again at higher income levels (Culas, 2012; Hao and Liu, 2016; Dong et al., 2018; Ding et al., 2019; Dkhili, 2023). These conflicting findings raise fundamental questions regarding the shape and validity of the EKC across different contexts, pollutants and stages of development (Maneejuk et al., 2020; Bibi and Jamil, 2021).

One important reason for these discrepancies lies in the methodological approaches used to model the EKC. Most studies employed quadratic or cubic functional forms in panel linear regressions to model U- or N-shaped curves (Grossman and Krueger, 1995; Torras and Boyce, 1998; Sarkodie and Strezov, 2019). While straightforward, these approaches impose symmetry on the estimated curves and require the shape to be globally U- or N-shaped. More recent studies have applied threshold models (Hansen, 2017; Twerefou et al., 2017; Dkhili, 2023) and smooth transition regressions (Maneejuk and Yamaka, 2022) to capture nonlinearities. These approaches allow for regime changes at endogenously determined income levels, but they have important limitations: threshold models focus only on changes in levels of the relationship across regimes, without allowing for kinks or slope changes at threshold points, while smooth transition models impose functional smoothness that may not adequately represent abrupt slope shifts (Hansen, 2017; Tarkhamtham and Yamaka, 2019). In other words, neither quadratic nor threshold nor smooth transition models are fully suited to capturing kink-type nonlinearities, where slopes change sharply at certain income levels while maintaining continuity of the regression function.

To address these methodological limitations, this study proposes the use of two-regime and three-regime kink regression models to reexamine the EKC hypothesis. Unlike threshold models, kink models focus on changes in slopes rather than levels, allowing for a more flexible and realistic representation of environmental–economic relationships (Hansen, 2017). Compared to smooth transition models, kink regressions can capture abrupt but continuous changes in the relationship between income and pollution, making them particularly well-suited to identifying U- or N-shaped patterns where slope changes may occur at multiple income levels. By allowing for different marginal effects in low-, medium- and high-income regimes, our model provides a richer framework to investigate whether the EKC holds, and if so, whether its shape is inverted-U, N-shaped or deviates from both.

The analysis focuses on the Asia–Europe Meeting (ASEM) region, which offers a unique setting for revisiting the EKC hypothesis. ASEM is a multilateral forum established in 1996 to strengthen cooperation between Asia and Europe. It has expanded to encompass 53 partners, including 30 European and 21 Asian countries, as well as the European Union and the ASEAN Secretariat. Collectively, ASEM countries account for approximately 59% of global gross domestic product (GDP) (US$50.2tn) and 61% of the world’s population (4.7 billion people). ASEM countries exhibit diverse development trajectories, ranging from highly industrialized European economies to rapidly growing Asian emerging markets. This heterogeneity, combined with shared commitments to sustainable development and intergovernmental cooperation, provides an ideal context to compare and contrast EKC patterns across different development stages within a single integrated framework.

Beyond methodological innovation and regional focus, this study also expands the economic dimension of the EKC analysis. Traditionally, EKC studies have relied primarily on GDP per capita as the sole measure of economic development (Grossman and Krueger, 1995; Panayotou, 1997; Bibi and Jamil, 2021). However, economic development is a multifaceted process involving structural change, financial deepening and urbanization, all of which may influence environmental outcomes in distinct ways (Tamazian et al., 2009; Shahbaz et al., 2016; Hashmi et al., 2020; Ozturk et al., 2024). To account for this, the present study simultaneously examines the roles of GDP per capita, financial development, the share of the tertiary sector, and urbanization in shaping environmental degradation. In addition, foreign direct investment (FDI) and trade openness are included as control variables to capture globalization effects.

Finally, this study broadens the environmental dimension of the EKC framework by examining three distinct indicators of environmental degradation: CO₂ emissions, total greenhouse gas (GHG) emissions and PM_2.5 concentrations. Most previous studies focus on a single pollution indicator, usually CO₂, thereby overlooking the multidimensional nature of environmental degradation (Lee et al., 2010; Wang et al., 2018; Bibi and Jamil, 2021; Bekun et al., 2024). By incorporating these atmospheric measures, this study provides a more comprehensive and assessment of the EKC hypothesis.

This study contributes to the EKC literature in several important ways. First, it employs a dynamic three-regime kink regression framework to identify slope-based regime changes in the relationship between economic development and environmental degradation. Unlike conventional polynomial or smooth transition EKC specifications, the kink approach allows for discrete changes in marginal effects at data-driven development levels, providing a transparent characterization of regime shifts rather than smooth turning points. Second, by applying this framework to a heterogeneous group of ASEM economies, the study offers comparative evidence on how EKC dynamics differ across development regimes within a diverse set of countries, highlighting cases where standard inverted-U interpretations fail to capture more complex adjustment patterns. Third, the analysis extends beyond GDP-centric EKC formulations by incorporating multiple dimensions of economic development, including income, financial development, structural transformation toward services, and urbanization, within a unified regime-dependent framework, allowing their marginal environmental impacts to vary across development stages. Finally, by jointly examining global (CO₂, GHG) and local (PM2.5) pollution indicators in a dynamic panel setting, the study shows that regime shifts and persistence effects differ systematically across pollutants with distinct spatial and health characteristics, yielding new policy-relevant insights into the growth–environment nexus.

The remainder of this paper is organized as follows: Section 2 reviews the theoretical foundations and relevant empirical literature. Section 3 outlines the data sources, variable construction and econometric methodology. Section 4 presents the empirical findings and discusses the results. Section 5 concludes the paper with key policy implications and future research directions.

A central approach to understanding sustainable development lies in exploring the relationship between economic growth and environmental quality, most prominently conceptualized by the EKC hypothesis. Originating from the work of Grossman and Krueger (1991) during discussions on the North American Free Trade Agreement, the EKC posits that environmental degradation initially worsens with rising income levels but improves after a critical threshold of economic development is reached. This relationship, graphically represented as an inverted U-shaped curve, reflects the idea that environmental quality deteriorates during the early stages of industrialization and urbanization but improves as economies mature and adopt cleaner technologies (Panayotou, 1997; Kuznets, 1955).

The inverted U-shape of the EKC can be theoretically explained by the interaction of three fundamental mechanisms: the scale effect, the composition effect and the technique effect (Grossman and Krueger, 1995; De Bruyn et al., 1998; Sarkodie and Strezov, 2019). The scale effect implies that as economic activity expands, environmental degradation increases proportionally, since higher production and energy consumption accompany the early stages of industrialization. As economies continue to develop, the composition effect begins to take hold, reflecting structural transformation from pollution-intensive manufacturing sectors toward cleaner, service-based and knowledge-driven industries, thereby reducing overall pollution intensity. Finally, the technique effect captures technological advancement, improved energy efficiency, and the implementation of stricter environmental regulations that become more prevalent at higher income levels, leading to cleaner production methods and reduced emissions. The combined influence of these mechanisms gives rise to the inverted U-shaped EKC: at lower income levels, the scale effect dominates, and pollution rises; beyond a certain threshold, the composition and technique effects surpass the scale effect, resulting in an eventual improvement in environmental quality. However, empirical and theoretical advances have revealed that the relationship between income and pollution may not always follow a single turning point. Instead, in some economies, pollution declines only temporarily before rising again at very high income levels, a pattern referred to as an N-shaped EKC (De Bruyn et al., 1998; Torras and Boyce, 1998; Culas, 2012; Huan et al., 2022). This dynamic may arise from technical obsolescence, consumption-driven rebound effects, or saturation of environmental regulation, whereby sustained economic expansion induces new forms of pollution that offset earlier gains. The N-shaped EKC thus captures the possibility that environmental quality improves only up to a point, after which further economic growth may once again exert pressure on natural resources.

These nonlinear patterns imply that the income–pollution relationship may not be smooth or symmetric, but rather piecewise continuous, with multiple slope changes across different stages of development (Figure 1). Consequently, modeling such dynamics requires econometric techniques capable of identifying kink points, where the marginal effect of income on pollution shifts sharply. The present study extends this theoretical reasoning by employing a two-kink (three-regime) dynamic panel regression model, which allows for abrupt but continuous changes in the slope of the pollution–growth relationship. This approach accommodates interactions among scale, composition, technique and obsolescence effects across development stages, offering a more flexible framework for understanding the environmental consequences of economic growth in ASEM countries.

To operationalize these theoretical mechanisms within an empirical framework, it is necessary to explicitly link the EKC channels to observable dimensions of economic development. While income per capita provides a composite measure capturing the joint influence of scale, composition and technique effects, additional variables are required to disentangle these mechanisms more precisely. In this study, each development indicator is interpreted as a proxy for a specific channel of environmental impact. Financial development, measured by credit to the private sector, reflects both scale and technique effects: it facilitates capital accumulation and industrial expansion in early stages, while also enabling investment in cleaner technologies and environmental innovation at higher levels of development. At advanced stages, however, financial deepening may also generate rebound or obsolescence effects by stimulating consumption and carbon-intensive activities.

The share of the tertiary sector is employed as a direct proxy for the composition effect, capturing the structural transition from manufacturing-based to service-oriented economies, which is typically associated with lower emission intensity. Urbanization, in turn, reflects the interaction between scale effects and agglomeration economies. Early-stage urban expansion tends to increase environmental pressure through higher energy demand and infrastructure needs, whereas mature urban systems may achieve efficiency gains through improved public transport, technological adoption and spatial optimization.

Numerous empirical studies have sought to validate the EKC hypothesis across different countries and environmental indicators, often producing mixed results. In its simplest form, this relationship has been estimated through quadratic functional specifications using time series or panel data. Studies such as Leitão (2010), who analyzed 94 countries, and Culas (2012), focusing on nine Latin American economies, confirmed the existence of an inverted U-shaped EKC. Their findings suggest that economic growth initially deteriorates environmental quality but ultimately improves it once a threshold income level is surpassed, implying that sustainable development can be achieved through continued economic progress.

Several studies focusing on CO₂ emissions support this pattern. Ahmed et al. (2019) identified an inverted U-shaped relationship between income and CO₂ emissions in Indonesia, while Gierałtowska et al. (2022) found similar evidence across 163 countries from 2000 to 2016. Likewise, studies on airborne pollutants such as PM_2.5 concentrations also align with the EKC hypothesis. For instance, Gupta et al. (2022) found an inverted U-shaped relationship between PM_2.5 exposure and income levels in Bangladesh during 1990–2016. These results highlight that the EKC relationship may depend on regional, structural and institutional differences. Furthermore, Al-Mulali et al. (2013) confirmed the inverted U-shaped EKC between urbanization and CO₂ emissions in middle east and north Africa (MENA) countries using the Pedroni cointegration and dynamic ordinary least squares (DOLS) techniques, while Shahbaz et al. (2016) identified a U-shaped (rather than inverted U-shaped) relationship in Malaysia using an autoregressive distributed lag (ARDL) framework. Despite methodological variations, most of these studies rely on quadratic or cubic polynomial regressions that impose symmetry and global curvature, potentially oversimplifying the true nonlinear relationship between growth and pollution (Torras and Boyce, 1998; Ridwan et al., 2024).

A growing body of research, however, challenges the conventional inverted U-shaped EKC, proposing an N-shaped relationship between income and environmental degradation. This hypothesis suggests that while environmental quality initially deteriorates and later improves with income, it may again worsen beyond a second turning point, reflecting technical obsolescence, rebound effects or the saturation of environmental regulations (De Bruyn et al., 1998; Torras and Boyce, 1998). Early observations of this N-shaped pattern were made by Grossman and Krueger (1995) and Panayotou (1997), particularly for sulfur dioxide (SO₂) emissions. Friedl and Getzner (2003) documented a similar relationship for Austria, while Alvarez-Herranz and Balsalobre-Lorente (2015) confirmed the N-shaped EKC among organisation for economic co-operation and development (OECD) countries. More recent studies have found comparable results in emerging economies: Shehzad et al. (2022) reported an N-shaped EKC in Algeria using a cubic panel regression model, and Huan et al. (2022) identified an N-shaped association between industrial development and environmental degradation in China from 1972 to 2020 using the ARDL bounds approach.

Although these studies provide valuable insights into the nonlinear dynamics of the pollution–growth nexus, most continue to rely on cubic polynomial regressions, which are highly sensitive to functional specification and assume smooth, continuous curvature across regimes. Such an assumption often oversimplifies the true nature of environmental responses to economic expansion, leading to potential misidentification of local turning points and an inability to capture abrupt slope changes that typically characterize real-world pollution–growth relationships (Maneejuk et al., 2020). Moreover, conventional static models, such as ARDL or standard panel estimators, tend to overlook critical econometric challenges, including endogeneity, unobserved heterogeneity and dynamic persistence, which can bias long-run inferences regarding the shape and stability of the EKC (Ganda, 2019; Dkhili, 2023). To address these limitations, the present study adopts a dynamic panel kink model, which effectively accounts for these econometric complexities while allowing for structural changes in the marginal effects of economic development on environmental degradation. This framework enables the identification of one and two kink points, delineating two and three distinct regimes in the pollution–growth relationship, and thereby captures both inverted U-shaped and N-shaped patterns within a unified, data-driven structure.

To The focus on ASEM economies is motivated by their unique combination of advanced European countries and rapidly developing Asian economies, providing a broad spectrum of income levels, institutional frameworks, and environmental policy regimes within a unified analytical setting. This heterogeneity makes ASEM particularly suitable for examining nonlinear and regime-dependent relationships, as it captures different stages of structural transformation and environmental transition. In addition, the comparison between European and Asian economies allows for a clearer assessment of how regulatory capacity, technological development and policy timing influence environmental outcomes.

The study period (2008–2023) is selected to capture the post-global financial crisis era, during which significant economic restructuring, technological progress and environmental policy developments have taken place. This period also benefits from improved data availability and comparability across countries. However, it is important to note that the inclusion of the COVID-19 pandemic may introduce short-term fluctuations in pollution levels, particularly due to temporary reductions in economic activity. While such effects may influence short-run dynamics, the empirical framework focuses on medium- to long-run relationships, and the main findings are therefore unlikely to be driven by pandemic-specific shocks. This data set provides a suitable balance between cross-country heterogeneity and temporal consistency, allowing for a comprehensive assessment of EKC dynamics across different stages of economic development (Grossman and Krueger, 1995; Panayotou, 1997; De Bruyn et al., 1998; Dinda, 2004; Jaeger et al., 2023).

The study employs three indicators of environmental degradation to provide a multidimensional perspective: CO₂ emissions, total GHG emissions and PM_2.5 concentrations. Data on CO₂ and GHG emissions are obtained from Our World in Data (Friedlingstein et al., 2023; Jones et al., 2023). CO₂ emissions reflect fossil fuel combustion and industrial processes, while GHG emissions include methane and nitrous oxide, providing a broader view of atmospheric degradation (Ahmed et al., 2019; Gierałtowska et al., 2022; Ridwan et al., 2024; Wang et al., 2024a, 2024b). These indicators are widely used in EKC analyses for their direct links to energy use and industrialization (Akbostanci et al., 2007; Mert and Bölük, 2016; Sarkodie and Strezov, 2019).

PM2.5 exposure data are obtained from the World Development Indicators (WDI) of the World Bank. The indicator measures the population-weighted annual mean exposure to ambient particulate matter with an aerodynamic diameter of less than 2.5 micrometers (PM2.5), expressed in micrograms per cubic meter. The estimates are derived from the Global Burden of Disease (GBD) air pollution exposure data set produced by the Institute for Health Metrics and Evaluation (IHME), which integrates satellite observations, chemical transport models, and available ground monitoring data, while excluding natural dust sources. Importantly, the GBD framework applies a consistent measurement methodology across all years, ensuring intertemporal comparability in panel analyses. This PM2.5 indicator has been widely used in empirical studies on environmental quality, urbanization and the Environmental Kuznets Curve (Hao and Liu, 2016; Dong et al., 2018; Ding et al., 2019; Le et al., 2021; Gupta et al., 2022; Ciarlantini et al., 2023). Note that although most explanatory variables are available through 2023, PM2.5 exposure data from the World Development Indicators are available only up to 2020. Accordingly, all regressions in which PM2.5 serves as the dependent variable are estimated over the 2008–2020 period.

The explanatory variables include economic development, financial development, industrial structure and urbanization, along with control variables such as FDI and trade openness. Economic development is measured by real GDP per capita, the canonical EKC variable capturing scale, composition and technique effects (Grossman and Krueger, 1995; Kuznets, 1955; De Bruyn et al., 1998; Dkhili, 2023; Mahmood et al., 2023). Financial development is proxied by domestic credit to the private sector (% of GDP), reflecting the role of financial systems in influencing environmental outcomes through investment channels (Tamazian et al., 2009; Apergis and Ozturk, 2015; Ganda, 2019; Ozturk et al., 2024). Industrial structure is represented by the share of the tertiary sector in GDP, which captures structural shifts from manufacturing to services as economies mature, often associated with reduced emission intensities (Hashmi et al., 2020; Mujtaba and Shahzad, 2021; Huan et al., 2022).

Urbanization is measured by the urban population share, which can exert both positive and negative pressures on environmental quality depending on infrastructure, energy use and policy context (Al-Mulali et al., 2013; Shahbaz et al., 2016; Wang et al., 2018; Ahmed et al., 2019). FDI (% of GDP) and trade openness (trade as % of GDP) are included to control for the effects of globalization, reflecting potential pollution haven (Akbostanci et al., 2007; Kellenberg, 2009; Bekun et al., 2024) or pollution halo effects (Baek, 2016; Sarkodie and Strezov, 2019; Shehzad et al., 2022).

To prepare the data for analysis, all variables are transformed into their natural logarithmic form. However, FDI is treated as a special case. Since net FDI inflows, measured as a percentage of GDP, may include zero or negative values in some years for certain countries, a standard logarithmic transformation cannot be directly applied. To address this issue, we adopt a shift-log transformation, as $FDIit*=ln⁡(FDIit-min⁡(FDI) + 1)$⁠. This transformation shifts the distribution to ensure positivity, making the logarithmic operation well-defined while preserving relative differences across observations. This approach is widely used in empirical studies when variables can take both positive and negative values, ensuring that no observations are lost due to negative FDI (Kellenberg, 2009).

To examine the existence of kink effects while accounting for the dynamic behavior of environmental outcomes, this study applies a dynamic panel kink regression model. This approach incorporates lagged dependent variables and allows for changes in marginal effects at endogenously determined kink points, making it well suited to capture nonlinear relationships between economic development and environmental degradation. Following the empirical strategies of Halkos (2003), Twerefou et al. (2017) and Ganda (2019), the analysis employs a dynamic panel framework that includes lagged pollution terms to account for temporal persistence. Consistent with Hansen’s (2017) kink regression, the model is parameterized using min–max transformations, which ensure continuity at kink points and allow regime-specific slopes to be interpreted directly. The modified EKC model is formally expressed as:

where $Yit$ represents environmental indicators (e.g. $CO2it$⁠, $TGHGit$⁠, $PMit$⁠). For each development indicator $Dmit∈{ln⁡GDPCit,lnCRit,lnTIit,lnURBit}$⁠, the kink points $γm,1$ and $γm,2$(with $γm,1<γm,2$⁠) divide the support of that indicator into three regions and allow its marginal effect to differ across regions, while maintaining continuity of the regression function. Country-specific fixed effects $ci$capture time-invariant heterogeneity across countries, $λt$ denotes year fixed effects that control common global shocks and time-specific influences, and $εit$ is the idiosyncratic error term.

Within this framework, the “regimes” are variable-specific: each development indicator has its own low-, middle- and high-level regions defined by its own kink points. Therefore, the marginal effect of each indicator can be read directly from the corresponding slope coefficients in each region. To assess EKC-type patterns, the signs of the estimated regime-specific slopes are examined. For example, an N-shaped relationship with respect to $Dmit$is supported when its estimated slopes satisfy $βm,1>0$⁠, $βm,2<0$ and $βm,3>0$⁠. In particular, when $β1,1β1,2β1,3$ satisfies this pattern, the results indicate an N-shaped EKC between income per capita and environmental degradation. Analogously, the same sign pattern for $βm,1βm,2βm,3$ provides evidence of EKC-type nonlinearities with respect to alternative development dimensions such as credit development, tertiary industry development, and urbanization.

By contrast, if the estimated slope coefficients exhibit the pattern $βm,1>0$and $βm,2<0$ (with no statistically significant third regime), the relationship is characterized as inverted-U shaped. This case corresponds to the traditional EKC hypothesis, in which environmental degradation initially increases with development but declines once the development indicator surpasses a critical level. Conversely, if the estimated coefficients are positive across all regions $βm,1>0,βm,2>0,βm,3>0$⁠, the relationship is monotonically increasing, implying that higher levels of development are consistently associated with greater environmental pressure. Similarly, if all estimated coefficients are negative $βm,1<0,βm,2<0,βm,3<0$⁠, the relationship is monotonically decreasing, indicating that environmental degradation declines steadily as development progresses

In the first stage, kink (threshold) values are identified for each development indicator separately using a grid search procedure. For each candidate pair of kink points $γm,1γm,2$⁠, the dynamic panel kink regression model is estimated while maintaining linearity in all other covariates. The optimal kink points are selected by minimizing the corresponding generalized method of moments (GMM) objective function. The statistical significance of kink effects is assessed using bootstrap-based tests to account for the nonstandard distribution of the threshold estimators.

It is important to note that the kink points are estimated as common thresholds across the panel, rather than country-specific values. This approach follows the standard dynamic panel kink regression framework (Hansen, 2017), where a single set of threshold parameters is identified by minimizing the GMM objective function over the pooled sample. As such, the estimated kink points should be interpreted as representative regime boundaries for the average development trajectory across ASEM economies, rather than exact turning points for any individual country. Given the heterogeneity of countries in the sample, these common thresholds implicitly reflect an aggregation of country-specific dynamics. Therefore, while they provide a useful benchmark for identifying broad development regimes, some cross-country variation in the precise location of turning points is naturally averaged out in the estimation process.

In the second stage, the estimated kink points are used to construct kinked regressors via min–max transformations and are incorporated into a dynamic panel model estimated by the System GMM estimator, following Arellano and Bond (1991) and Blundell and Bond (1998). To ensure valid inference in the presence of estimated kink points (generated regressors), a country-cluster bootstrap procedure is employed in which both the kink points and the System GMM model are reestimated within each bootstrap replication. Confidence intervals and hypothesis tests for regime-specific coefficients are therefore constructed from the bootstrap distributions, allowing kink estimation uncertainty to be fully propagated into the dynamic panel inference.

For notational convenience, define the vector of kinked regressors as:

where each element corresponds to the min–max kink representation of a development indicator and captures variable-specific nonlinear effects rather than common development regimes.

Let $θ$ denote the vector of slope parameters and let:

be the full regressor vector. The dynamic panel kink regression model can then be written compactly as:

This compact representation facilitates the implementation of System GMM estimation using both the difference and level equations. The inclusion of the lagged dependent variable in the dynamic panel kink regression model introduces endogeneity due to its correlation with the error term. To address this issue, the model is estimated using the System GMM estimator, following Arellano and Bond (1991) and Blundell and Bond (1998), which combines equations in first differences and in levels and exploits suitable lagged values of the endogenous variables as internal instruments.

The lagged dependent variable and the key development indicators, income per capita, financial development, tertiary industry share, and urbanization, are treated as endogenous variables to account for potential simultaneity and reverse causality. For instance, while these variables may influence environmental outcomes, they may also be affected by economic conditions, environmental policies and structural changes. Accordingly, their lagged levels dated (t – 2) and earlier are used as instruments for the differenced equations, while their lagged first differences are employed as instruments for the level equations. The choice of lag (t – 2) reflects the need to ensure instrument validity, as first lags may remain correlated with the error term in the presence of serial correlation. Trade openness and FDI are treated as predetermined variables, as they may be influenced by past shocks but are less likely to be contemporaneously correlated with the current error term. These variables are therefore instrumented using their second lags. The remaining control variables are assumed to be strictly exogenous. Time fixed effects are included in all specifications and treated as exogenous.

To mitigate instrument proliferation, the instrument matrix is collapsed, and the maximum lag depth is restricted. This parsimonious instrument strategy helps preserve the power of the Hansen test and avoids overfitting concerns commonly associated with System GMM estimation. Estimation is conducted using the two-step System GMM procedure, with the finite-sample correction to standard errors proposed by Windmeijer (2005). Model validity is assessed using the Arellano–Bond tests for serial correlation and the Hansen test of overidentifying restrictions.

In the presence of kink regressors, the estimated kink points are treated as fixed when constructing the min–max transformed variables, and the dynamic panel model is estimated conditional on these kink points. This approach preserves linearity in parameters while allowing for variable-specific nonlinearities across regimes.

After estimating the model, we examine whether kink effects are statistically significant, following Hansen (1999) and the extension by Seo and Shin (2016) for dynamic panel models. The null hypothesis assumes linearity, meaning the slope coefficients are equal across regimes. The alternative hypothesis allows for regime-dependent slopes.

Because the threshold parameter is unidentified under the null, standard asymptotic distributions do not apply. Therefore, we use a nonstandard F-test with bootstrap p-values to assess the existence of kink effects. This involves computing the test statistic based on the difference between the linear and kink model fits, then generating the empirical distribution of the statistic through bootstrap resampling, as recommended by Hansen (1999). Rejection of the null hypothesis provides evidence of nonlinear, regime-dependent effects in the relationship between development and environmental outcomes.

To provide a comprehensive overview of the data set, Table 2 presents the descriptive statistics for the key variables used in this study. The table summarizes the central tendency, dispersion and distribution characteristics, including the mean, standard deviation, minimum and maximum values.

The results reported in Table 3 present the outcomes of the Levin, Lin and Chu (LLC) and Im, Pesaran and Shin (IPS) panel unit root tests. The findings consistently reject the null hypothesis of non-stationarity across all variables, confirming that the data series are stationary. This evidence indicates that the variables are suitable for further empirical investigation, thereby validating the use of dynamic panel estimation to explore the relationship between economic development and environmental pollution.

Before implementing the panel kink regression model, it is crucial to verify whether a nonlinear relationship exists between each economic development indicator and its corresponding environmental pollution indicator. Table 4 present the results of the two-regime (one kink point) and three-regime (two kink points) tests.

As shown in Table 4, the F-test statistics reported in Table 4 provide strong empirical evidence of nonlinear linkages between economic development and environmental degradation. The rejection of the linear null hypothesis in favor of the two-regime specification at the 1% level across all cases implies that economic growth exerts differential marginal effects on pollution across distinct stages of development. This pattern aligns with the theoretical expectation that structural, technological and policy shifts alter the income–environment elasticity as economies evolve. Furthermore, the three-regime test results suggest that, for certain indicators, an additional threshold may exist, reflecting a multiphase environmental response to development. In these cases, pollution intensity initially rises with income, subsequently declines, and may increase again beyond a higher level of development, indicating a potential N-shaped EKC rather than the conventional inverted U-shaped EKC. However, the detection of a three-regime kink does not, by itself, confirm the existence of either an N-shaped or U-shaped EKC. To test this hypothesis, the study proceeds with the estimation of a dynamic panel kink regression model, which enables the identification of regime-dependent marginal effects and the formal verification of U-shaped and N-shaped nonlinearities within a unified analytical framework.

Table 5 reports the estimated kink points for each economic indicator across different environmental pollution measures, derived from the dynamic panel kink regression framework. These thresholds identify the levels at which the marginal relationship between economic variables and environmental outcomes changes, thereby partitioning the sample into distinct segments.

Focusing on income, the results reveal two statistically significant kink points for CO₂ and total GHG emissions at ln(GDP per capita) ≈ 8.0–8.1 and 10.4–10.8, corresponding to approximately US$3,000–3,200 and US$34,000–47,000, respectively. These values are broadly consistent with the EKC literature, which documents nonlinear relationships between income and environmental outcomes. Compared with previous studies, the upper kink points identified in this study are relatively high. For example, Özbay, 2025 reports a turning point of approximately US$19,800, while Muratoğlu et al. (2024) estimate a threshold around US$29,250. However, more recent evidence suggests that environmental transitions may occur at higher income levels and may involve multiple thresholds. In particular, Wang et al. (2024a, 2024b) document turning points at approximately US$45,000 and above. In this context, the upper kink identified in this study falls within the range reported in recent literature and is therefore economically plausible. Importantly, this upper threshold is not driven by a single outlier country but reflects observations from multiple high-income economies within the sample, including OECD countries with GDP per capita exceeding US$40,000. For PM₂.₅, a single kink is identified at ln(GDP per capita) ≈ 8.0 (approximately US$3,000), which is consistent with existing evidence showing that local pollutants tend to reach turning points at lower income levels compared to global emissions.

Regarding financial development, the estimated kink points occur at approximately 40%–66% and above 100% of GDP, depending on the pollution indicator. These thresholds lie well within the empirical distribution of the sample and correspond to levels of financial depth observed across both emerging and advanced economies. Prior studies have shown that credit-to-GDP ratios exceeding 100% are common in highly developed financial systems (Arcand et al., 2015). Notably, the higher threshold (around 140% of GDP for GHG emissions) is supported by the data rather than driven by a single outlier. The sample includes multiple highly financialized economies, such as Japan, the UK, Spain and Denmark, as well as rapidly expanding credit markets including China and Thailand, where credit-to-GDP ratios exceed this level. As such, the estimated kink reflects the upper tail of the distribution and should be interpreted as capturing a high-financialization regime rather than a universal benchmark.

For the share of tertiary industry value added, the estimated kink points range from approximately 40% to 70% of GDP. These values align closely with the observed distribution of structural transformation across countries. In particular, lower values are characteristic of economies where agriculture and manufacturing remain dominant, whereas higher values correspond to more service-oriented economic structures commonly observed in advanced economies. This pattern is consistent with the environmental economics literature, which highlights the role of structural change and the composition effect in shaping environmental outcomes (Grossman and Krueger, 1995), as well as recent empirical evidence showing that economic restructuring toward services contributes to improved environmental performance (Dong et al., 2023).

Finally, for urbanization, the estimated kink points range from approximately 30% to over 80%. These thresholds align with the empirical distribution of urbanization levels, where developing economies are typically below 50%, while advanced economies often exceed 70%–80%. This result is also broadly consistent with Fang et al. (2021), who report turning points within a comparable range.

To examine whether the EKC follows an N-shaped rather than an inverted U-shaped pattern across different dimensions of economic development, this study first evaluates the slope coefficients of each regime obtained from the dynamic panel kink regression model. The results are reported in Table 6.

The results show no evidence of an N-shaped EKC for income (GDP per capita). Instead, the relationship between economic development and environmental degradation is predominantly inverted-U-shaped or monotonic. For carbon-based pollutants, CO₂ and total GHG emissions, the estimated marginal effects remain positive across all regimes, although their magnitude gradually declines as income increases. This suggests that most ASEM economies have not yet reached the income threshold at which economic growth translates into net environmental improvement. These findings support a weak-form EKC interpretation, characterized by an attenuation of income effects rather than a full turning point, where scale effects continue to dominate while composition and technique effects have begun to emerge but remain insufficient to reverse the overall trend (Grossman and Krueger, 1995; Panayotou, 1997).

The persistence of rising CO₂ and GHG emissions even at advanced stages of development is consistent with prior studies emphasizing the difficulty of decoupling global pollutants from economic growth, given their deep integration into production and consumption systems (Friedl and Getzner, 2003; Apergis and Ozturk, 2015; Bibi and Jamil, 2021; Jaeger et al., 2023). In contrast, PM₂.₅ exhibits clearer evidence of stabilization or decline, reflecting a weak inverted-U pattern. This suggests that local pollutants respond more rapidly to policy enforcement, urban infrastructure development and technological abatement, whereas global emissions display delayed or incomplete turning points due to slower adoption of clean energy and the presence of rebound effects in consumption.

The results for the alternative threshold variables further illuminate the mechanisms underlying these nonlinear dynamics. Financial development exhibits an N-shaped relationship with GHG emissions: emissions initially decline as credit expansion supports cleaner investment and environmental financing but subsequently increase beyond a critical threshold when financial deepening channels resources toward energy-intensive and consumption-driven sectors (Tamazian et al., 2009; Ganda, 2019; Ozturk et al., 2024). Industrial structure, by contrast, displays a sustained negative marginal effect beyond the second kink, consistent with the composition effect, whereby the transition from manufacturing to service- and technology-oriented sectors contributes to long-term reductions in environmental degradation (Hashmi et al., 2020; Maneejuk and Yamaka, 2022).

Urbanization demonstrates an inverted-N relationship with CO₂ and GHG emissions, reflecting a multistage development process. In the early stage, emissions increase due to the scale effect associated with infrastructure expansion and rising energy demand. In the intermediate stage, improvements in efficiency, structural transformation and technological adoption contribute to emission reductions. At more advanced stages, emissions tend to stabilize, reflecting the balance between agglomeration benefits and residual consumption-driven pressures. This highlights the dual role of cities as both sources of emissions and centers of environmental efficiency (Ahmed et al., 2019; Ridwan et al., 2024).

Overall, the findings suggest that the environmental consequences of development in ASEM economies are predominantly single-peaked rather than cyclical. While the EKC hypothesis holds for certain pollutants, particularly local emissions, the absence of an N-shaped income–pollution relationship indicates that environmental improvement remains incomplete and heterogeneous across pollutants and drivers. Economic growth alone is therefore insufficient to ensure environmental sustainability, and complementary mechanisms, particularly green finance, industrial upgrading and effective urban governance, are essential to achieve sustained decoupling and to prevent future rebound effects (Panayotou, 1997; De Bruyn et al., 1998; Jaeger et al., 2023).

To provide a clearer interpretation of the empirical results, the regime-specific coefficients reported in Table 6 are translated into the corresponding EKC forms summarized in Table 7 based on the sign and sequence of marginal effects across regimes. A monotonic relationship is identified when marginal effects retain the same sign across all regimes. An inverted-U pattern is observed when the marginal effect is positive in the initial regime and becomes negative in subsequent regimes, while a U-shaped pattern reflects the opposite sequence. More complex forms, such as N-shaped or inverted-N relationships, arise when marginal effects change sign multiple times across regimes.

To ensure the reliability of the estimated nonlinear relationships and mitigate potential econometric biases, a series of robustness checks are conducted using alternative estimation techniques and model specifications. The baseline results are obtained from the dynamic panel kink regression framework, which is well suited to address endogeneity, unobserved heterogeneity and dynamic persistence in the pollution–development relationship (see  Appendix).

Figure 2 presents representative examples of the estimated EKCs derived from the panel kink regression model. The blue lines indicate fitted piecewise relationships, while the red dashed vertical lines denote the estimated kink points at which the marginal effect of economic development on environmental degradation changes. The selected examples display five nonlinear forms, namely, inverted N, monotonic increase, N, inverted U and U shapes, which capturing the diversity of development–pollution dynamics across the ASEM economies.

To explicitly account for regional heterogeneity in environmental policy frameworks and development trajectories, this section conducts a disaggregated analysis for European and Asian economies in addition to the full ASEM sample. The motivation for this split lies in the markedly different timing and intensity of environmental regulation across the two regions. European countries have generally implemented carbon pricing mechanisms, emissions trading schemes, and stringent environmental standards at earlier stages of development, whereas many Asian economies have experienced rapid industrialization with comparatively later adoption of such policies.

The objective of this analysis is therefore not to test parameter homogeneity, but to examine whether the nonlinear income–environment relationships and regime-dependent effects identified at the aggregate level differ systematically across regions with distinct regulatory and structural characteristics. Tables 8 and 9 report the estimated income thresholds defining the development regimes for each pollutant, while Tables 10 and 11 present the full dynamic panel kink regression results for the European and Asian subsamples, respectively. This framework enables a direct comparison of how income growth, financial development, structural transformation toward services, and urbanization affect environmental outcomes across regions, highlighting differences in both the location of regime shifts and the magnitude of marginal effects. In doing so, the analysis provides insight into how earlier environmental policy intervention and economic maturity in Europe contrast with the development–environment dynamics observed in fast-growing Asian economies.

For the European economies, the results provide partial evidence of nonlinear EKC behavior rather than a consistently pronounced N-shaped pattern. CO₂, total GHG emissions and PM₂.₅ generally exhibit weak N-shaped trajectories, emissions increase at low-income levels, decline around the mid-income threshold, and show a mild rebound at the highest income regime. This pattern can be interpreted as reflecting a multistage development process in which early structural adjustments and environmental policies reduce emissions, followed by a temporary rebound driven by consumption patterns, technological obsolescence or saturation of regulatory effects at higher income levels. In contrast, financial development displays an inverted-N relationship with CO₂ and GHG emissions, indicating that early financial expansion supports cleaner investments, while intermediate-stage credit deepening may finance energy-intensive activities, before more mature financial systems reallocate resources toward greener sectors. Industrial transformation demonstrates a similar inverted-N pattern, where early structural change reduces emissions, mid-stage service expansion temporarily intensifies them, and advanced restructuring leads to renewed environmental gains. Urbanization, by contrast, follows a clear inverted-U relationship, confirming that mature, high-density cities achieve lower emission intensities through advanced infrastructure, agglomeration economies and technological efficiency.

For the Asian economies, the results reveal no consistent evidence of an N-shaped EKC. Instead, the income–environment relationships predominantly follow the conventional inverted-U pattern. For CO₂ and total GHG emissions, pollution rises sharply at lower income levels but declines once economies surpass the upper threshold, indicating that several Asian countries are entering the stage where economic growth begins to reduce emissions through cleaner technologies, improved efficiency and stricter regulation. For PM₂.₅, the relationship is largely inverted-U, confirming that local air pollution responds more quickly to policy intervention and technological improvements.

Only limited evidence of N-type behavior appears in specific structural dimensions, particularly in industrial transformation, where mid-stage restructuring temporarily raises emissions before further diversification and technological upgrading restore environmental gains. Financial development and urbanization both exhibit nonlinear and transitional effects. Financial development initially supports emission reductions through improved access to capital and early green investment, but excessive credit expansion may later amplify emissions before declining again at higher levels of financial maturity. Urbanization shows a similar multistage dynamic, where early urban concentration increases emissions through scale effects, followed by mitigation as cities adopt cleaner technologies and more efficient infrastructure systems.

In sum, the heterogeneity analysis confirms that these dynamics differ systematically across regions. In European economies, where regulatory frameworks and technological capabilities are more advanced, nonlinear patterns tend to be more complex and exhibit clearer turning points. In contrast, Asian economies are more likely to display conventional inverted-U trajectories, reflecting ongoing industrialization and later-stage environmental adjustment. This distinction indicates that the observed nonlinearities are driven by underlying structural differences across development stages, rather than being artifacts of pooled panel estimation

This study applies a dynamic panel kink regression approach to explore the nonlinear relationship between economic development and environmental degradation across 39 ASEM countries from 2008 to 2023, identifying threshold-dependent marginal effects that reveal whether the EKC follows the traditional inverted-U pattern or a more intricate N-shaped form.

The empirical results reveal no consistent evidence of an N-shaped EKC. Instead, the relationships between economic development and environmental degradation are predominantly single-peaked or monotonic. For carbon-based pollutants (CO₂ and total GHG emissions), pollution continues to increase with GDP per capita, although the marginal effect weakens at higher income levels, signifying weak decoupling rather than genuine reversal. In contrast, PM_2.5 concentrations tend to decline as income rises, reflecting the emergence of the classical inverted-U pattern once composition and technique effects begin to offset scale effects. The influence of structural variables reinforces these findings as industrial transformation and urbanization are consistently linked to long-term emission reductions, whereas financial development exhibits transitional or rebound effects depending on credit allocation efficiency. Collectively, these heterogeneous dynamics indicate that while economic expansion remains a primary driver of aggregate emissions, structural upgrading, technological innovation and urban efficiency play increasingly critical roles in mitigating environmental pressures across the ASEM economies.

Our findings call into question the universal applicability of the EKC hypothesis within the ASEM context. Economic growth alone does not guarantee environmental improvement, and for many economies, the estimated turning points remain at relatively high levels of development.

While the presence of multiple kink points suggests nonlinear relationships between development and environmental outcomes, these thresholds should be interpreted cautiously. The results may reflect not only technological progress but also structural changes in the composition of economic activity, particularly the expansion of financial and service sectors. As economies develop, these sectors tend to increase as a share of GDP, while more pollution-intensive activities may decline domestically or shift across borders. In this context, the observed patterns do not necessarily imply that economic development alone leads to sustained environmental improvements. Instead, the relationship may partly reflect structural transformation rather than purely efficiency gains from cleaner technologies. This interpretation also raises important concerns regarding the generalizability of the development pathway. Countries that achieved environmental improvements alongside growth may have benefited from early structural transformation and global production reallocation, a pathway that may not be fully available to late-developing economies.

From a policy perspective, these findings suggest that relying on economic growth or sectoral shifts alone is unlikely to ensure long-term environmental sustainability. While structural transformation may contribute to changes in emission patterns, it does not guarantee sustained reductions. Therefore, policy efforts should place greater emphasis on accelerating the adoption of clean technologies, improving energy efficiency, and strengthening environmental regulation across all stages of development. The role of financial development should also be interpreted within this broader context. Although financial expansion may support investment and economic activity, its environmental impact depends critically on how financial resources are allocated. Without appropriate regulatory frameworks, financial deepening may reinforce carbon-intensive production patterns. As such, policies aimed at directing financial flows toward sustainable investments, including green finance instruments and climate-related risk management, are essential to ensure that financial development contributes to environmental objectives.

While this study reveals several new findings, it remains subject to several limitations. First, the analysis is constrained by the availability and comparability of long-term pollution data across ASEM economies. In addition, the external validity of the findings may be limited beyond the ASEM context. The ASEM group comprises a diverse but specific set of economies with particular institutional frameworks, development trajectories and environmental policy regimes. As such, the identified nonlinear relationships and regime-dependent dynamics may not fully generalize to regions with substantially different economic structures, regulatory capacities or stages of development. Future research could address these limitations by extending the analysis to a broader set of countries, exploring sector-specific emissions, incorporating spatial spillover effects, and further refining the identification of dynamic thresholds under alternative institutional and policy environments.

This paper is partially supported by the Center of Excellence in Econometrics, Faculty of Economics, Chiang Mai University, Thailand.

To assess the stability of the findings, three alternative estimators are employed: Fixed Effects (FE), First-Difference GMM, and System GMM. The FE estimator provides a static benchmark by controlling for time-invariant heterogeneity, although it does not account for dynamic adjustment. The First-Difference GMM estimator addresses endogeneity by eliminating individual effects through differencing, while the System GMM estimator improves efficiency by combining equations in levels and differences (Arellano and Bover, 1995; Blundell and Bond, 1998). Together, these estimators allow a comprehensive evaluation of the robustness of the kink-based results.

The results reported in  appendixTables A1–A3 confirm the stability of the main findings across estimation methods. In particular, the signs and statistical significance of the key coefficients remain largely unchanged. Although the FE estimates are less efficient in dynamic settings, their consistency with the GMM results reinforces the robustness of the core relationships. Moreover, the lagged dependent variable remains positive and statistically significant across GMM specifications, indicating strong persistence in environmental indicators, consistent with theoretical expectations. Additional robustness checks are conducted by modifying the model specification. First, excluding control variables yields qualitatively similar results, suggesting that the main findings are not driven by omitted-variable bias. Second, estimating each economic indicator separately produces consistent and statistically robust results, further supporting the stability of the identified nonlinear relationships.

This article does not contain any studies with human participants performed by any of the authors.