Decoding the nexus of economic growth,CO 2 emissions,and energy use: Panel evidence from Asia-Pacific countries

Abstract

Decoupling economic growth from environmental degradation is a primary challenge for the Asia-Pacific. This study examines long-run dynamic associations between renewable energy (RE), non-renewable energy (NRE), economic growth, and $C O_{2}$ emissions across 32 Asia-Pacific economies (1990–2023). Employing the PMG-ARDL estimator and panel causality tests, the research provides a granular analysis of income-based heterogeneity. The results are consistent with an inverted U-shaped Environmental Kuznets Curve (EKC) association, with estimated income-specific mathematical inflection points at log-income levels of 29.50 for the High-Income group and 29.30 for the Middle-and-Low-Income group. NRE remains a dominant long-run correlate of growth, while also being associated with higher environmental costs. Conversely, RE and Foreign Direct Investment (FDI) in advanced technological hubs are negatively associated with emissions, providing evidence of patterns consistent with the “Technique Stage” and “Pollution Halo” hypotheses. Sub-group analysis indicates regional divergence: estimates for advanced economies are consistent with associations expected under a possible movement toward decoupling, whereas industrializing economies show associations consistent with scale-driven energy demand and “Pollution Haven” risks. These findings should be interpreted as evidence of long-run dynamic associations rather than definitive causal mechanisms. Results suggest potential pathways for differentiated strategies and infrastructure transitions toward regional decarbonization.

Keywords

renewable energy non-renewable energy economic growth Environmental Kuznets Curve (EKC)Asia-Pacific C23 O10 Q42 Q43

Introduction

In 1992, under Agenda 21 of the United Nations, (https://sustainabledevelopment.un.org/outcomedocuments/agenda21) Policymakers were encouraged to utilize energy sources that emit the least emissions, enhancing economic and industrial efficiency, as a primary objective to attain sustainable development. The critical challenge, underscored by the IPCC (2013), is that CO₂ emissions from burning fossil fuels remain the key driver of climate change, impeding sustainable progress. Therefore, this economic growth versus efficient control of carbon emission has become the nucleus of long-term sustainability.

There have been a lot of studies on the link between energy use and economic growth.^1,2 These studies have led to hypotheses like Growth,³ Conservation,^4–6 Feedback,^3,7,8 and Neutrality.⁶ But a lot of this significant research has been concentrated on the relationship between energy and the economy, often disregarding the crucial environmental aspect of carbon emissions, which is probably just as important to the equation for sustainability.

Compounding this, the relationship between economic progress and environmental health is often explored through the lens of the Environmental Kuznets Curve (EKC) hypothesis.⁹ This theory suggests an inverted-U pattern: emissions initially rise with GDP but eventually fall after a certain economic peak. However, empirical evidence for this curve is far from consistent, with studies reporting diverse shapes (U, inverted-U, and N patterns) across different regions and periods (e.g., Refs. 10–15). This ongoing debate highlights the need for continued investigation, particularly in regions facing rapid development and pressing environmental concerns.

Historically, the pursuit of high GDP growth, (https://www.imf.org/external/datamapper/NGDP_RPCH@WEO/OEMDC/ADVEC/WEOWORLD) has always relied on non-renewable energy. Coal, oil, and natural gas feed economic production but are the biggest source of greenhouse emissions that cause climate change. Several studies show that NRE boosts GDP growth with greater carbon emissions.^16–18 On the contrary, renewable energy (RE) sources like solar, wind, and biomass offer a cleaner path forward, capable of supporting economic activity while actively reducing pollution.^3,17,19 Understanding these diverse, and sometimes contradictory, consequences of RE and NRE on economic growth and environmental quality is crucial for creating successful sustainable development strategies.

To provide a comprehensive analysis of the growth-environment nexus, this study incorporates a specific set of macroeconomic variables Renewable (RE) and Non-Renewable Energy (NRE), Foreign Direct Investment (FDI), Gross Fixed Capital Formation (GFCF), Labor Force (LF), and Trade Openness (TO) grounded in distinct theoretical mechanisms. Energy consumption is a central explanatory dimension of the model; while NRE is associated with scale-driven emissions dynamics, RE is examined as a factor linked to “technique-stage associations” relevant to sustainable growth.¹⁸ FDI is included to test competing hypotheses: the “Pollution Haven” theory, which suggests dirty industries migrate to developing Asia, versus the “Pollution Halo” theory, where foreign capital transfers green technology.²⁰ GFCF represents domestic industrial capacity; however, recent literature questions whether rapid capital accumulation in developing regions promotes efficiency or results in “crowding out” and environmental degradation.²¹ Labor Force (LF) is included to capture the region’s massive demographic dividend as a primary factor of production. Finally, Trade Openness (TO) allows us to control for the environmental impact of the region’s deep integration into global supply chains (Awodumi & Adewuyi, 2020). By integrating these specific variables, this study moves beyond simple energy-growth models to capture the structural complexity of the Asia-Pacific economy.

The Asia-Pacific region stands as a critical focal point for sustainability, accounting for 47% of global energy consumption and over half of the world’s CO₂ emissions (17.27 billion metric tons) in 2023. This massive environmental footprint underscores the urgency of balancing rapid economic expansion with effective emission controls. However, existing literature remains fragmented, predominantly focusing on single nations or specific developed economies. Comprehensive panel studies that rigorously analyze the interconnected dynamics of renewable and non-renewable energy, economic growth, and emissions across the region’s full diverse spectrum remain scarce, presenting a significant gap in understanding the broader regional nexus.

To bridge this gap, this study investigates the dynamic interrelationships among RE, NRE, economic growth, and CO₂ emissions across 32 Asia-Pacific economies from 1990 to 2023. Unlike prior studies that ignore cross-sectional dependence, this research employs the robust Pooled Mean Group Autoregressive Distributed Lag (PMG-ARDL) estimator and the Dumitrescu-Hurlin causality test. We contribute to the literature by: (i) distinguishing the specific growth and environmental impacts of RE versus NRE; (ii) integrating foreign direct investment (FDI) and capital formation to test the “Pollution Halo” and “Crowding Out” hypotheses; and (iii) providing empirical evidence on the estimated EKC turning point for a structurally diverse region.

The remainder of the paper is organized as follows: Section 2 reviews the relevant literature, Section 3 describes the data and methodology, Section 4 presents the empirical results, and Section 5 offers conclusions, policy implications, and limitations.

Literature review

The dynamic nexus of energy consumption and economic growth

The causal link between energy consumption and economic growth remains a cornerstone of energy economics, historically categorized into four competing frameworks. The Growth Hypothesis posits energy as a vital primary input, suggesting that conservation policies may impede expansion a trend frequently observed in fossil-fuel-dependent developing Asian economies.¹⁷ Conversely, the Conservation Hypothesis argues that mature economies can decouple growth from energy use through efficiency.⁴ Transition economies often exhibit the Feedback Hypothesis, characterized by bidirectional interdependence,⁷ while the Neutrality Hypothesis suggests no causal link.

Recent scholarship has moved beyond linear interpretations, utilizing advanced econometric techniques to reveal that these relationships are frequency-dependent and time-varying. For instance, AlNemer et al.²² demonstrated that while non-renewable energy (NRE) drives growth in the long run, it leads to environmental degradation, whereas renewable energy (RE) contributes to emission reductions in the short term. Similarly, machine learning and NARDL approaches^23,24 have identified strong substitution effects between “dirty” and renewable energy driven by price asymmetries and market uncertainty.²⁵

Environmental pollution, FDI, and the EKC hypothesis

The trade-off between development and environmental quality is theoretically framed by the Environmental Kuznets Curve (EKC). Originating from Kuznets’⁹ work on inequality and solidified in the 1990s by Grossman and Krueger,²⁶ the EKC suggests an inverted-U relationship between income and emissions. This pattern is commonly interpreted through three structural channels: the Scale-driven dynamics (increased production), the Composition Effect (shift to services), and the Technique Stage associations (technological innovation). In the Asia-Pacific, empirical evidence is mixed, with studies identifying both traditional inverted-U and N-shaped patterns, the latter suggesting that Scale-driven dynamics can eventually overwhelm technical gains.^10,27

Crucially, external drivers such as Foreign Direct Investment (FDI) and Trade Openness play a pivotal role in this nexus. Literature presents two competing theories: the Pollution Haven Hypothesis, where dirty industries migrate to evade strict regulations,²⁰ and the Pollution Halo Hypothesis, where foreign capital transfers cleaner technology. Recent findings in the region are bifurcated, supporting the Halo effect in high-tech East Asian hubs while suggesting Haven effects in parts of ASEAN.²⁸ Emerging determinants, including green innovation, ICT transitions, and foreign remittances, further underscore the complexity of environmental quality in the BRICS and Belt and Road regions.^29,30

The role of capital formation and labor

Traditional growth models often treat capital as a homogenous input, yet from an environmental perspective, the quality of capital is decisive. Investments in carbon-intensive infrastructure can “lock in” high-emission pathways, whereas investments in green technology facilitate the transition to a low-carbon economy.²¹ Similarly, in labor-abundant Asian markets, the interaction between labor efficiency and energy intensity is associated with the sustainability of long-term growth.³¹ By integrating Gross Fixed Capital Formation (GFCF) and Labor Force (LF), this study avoids the omitted variable bias common in simplified energy-growth models.

Research gap and novelty

While the energy-growth-emissions nexus is well-studied, this research bridges three critical gaps identified in the existing literature:

Regional heterogeneity gap

Most studies treat the Asia-Pacific as a monolithic block or focus on small clusters (e.g., ASEAN). By bifurcating 32 economies into High-Income and Middle-and-Low-Income groups, we distinguish between “Technique-stage” and “Scale-stage” nations to compare estimated turning points and associated heterogeneity across income groups.

Methodological gap

Prior research frequently relies on first-generation panel techniques that ignore cross-sectional dependence (CD). In an integrated trade region, failing to account for CD yields biased elasticities.³² We employ second-generation PMG-ARDL and CS-ARDL estimators with Driscoll-Kraay corrections to ensure robust inference.

Variable-specific gap

We move beyond aggregate energy models by simultaneously controlling for FDI and domestic capital (GFCF). This allows us to isolate the distinct environmental impacts of foreign versus domestic investment channels, providing a more granular assessment of the “Pollution Halo” versus “Haven” paradox.

Table 1 provides a systematic summary of the literature and the specific gaps addressed by this research.

Table 1.

Summary of significant literature and identified research gaps.

Study (year)	Focus region	Methodology and variables	Key findings and limitations	Gap filled by this study
Grossman & Krueger (1991)³³	NAFTA	OLS; Trade, Growth, C $O_{2}$	Foundational EKC work. Identified the inverted-U for pollutants. Limitation: Omitted disaggregated energy types (RE vs. NRE).	Integrates RE/NRE disaggregation within the EKC framework.
Rahman & Velayutham (2020)¹⁷	South Asia	1st Gen Panel; Growth, RE, NRE	Found NRE as a primary growth engine in developing Asia. Limitation: Ignored Cross-Sectional Dependence (CD), leading to biased estimates.	Employs 2nd Gen CS-ARDL to account for regional economic interdependencies.
Topcu et al. (2020)²¹	Developing Nations	1st Gen Panel; GFCF, Energy, Growth	Highlighted that capital accumulation drives emissions. Limitation: Failed to account for slope heterogeneity across different income levels.	Utilizes PMG-ARDL to address short-run heterogeneity vs. long-run homogeneity.
AlNemer et al. (2023)²²	Saudi Arabia	Wavelet Coherence; RE, NRE, Growth	Identified that the energy-growth nexus fluctuates significantly across frequencies. Limitation: Results restricted to a single-country context.	Provides a comprehensive 32-country panel analysis for regional generalizability.
Zaghdoudi et al. (2024)²³	China	Machine Learning (SVM); Fossil fuels, RE	Demonstrated a substitution effect from oil to coal before RE uptake. Limitation: Did not link findings to income-level turning points.	Mathematically calculates the EKC inflection point (28.36) for strategic planning.
Hundie (2024)³⁴	Global Panel	CS-ARDL; Energy, GDP, $C O_{2}$	Confirmed that energy transitions reduce emissions. Limitation: Lacked foresight into income-specific clusters within the Asia-Pacific.	Bifurcates the region into HI and MLI groups to test the “Technique vs. Scale” effects.
Tissaoui & Zaghdoudi (2025)²⁵	OECD Nations	NARDL; Energy uncertainty, $C O_{2}$	Found that energy uncertainty leads to temporary emission reductions. Limitation: Focused only on mature, advanced economies.	Contrasts advanced “Technique-stage” nations with rapidly industrializing “Scale-stage” nations.

Source: Authors Compilation.

Note. SREB refers to Silk Road Economic Belt economies; MENA refers to Middle East and North Africa nations.

Data and methodology

Data and variable rationale

Using a balanced panel of 32 Asia-Pacific economies (1990–2023), this study examines Economic Growth (GDP, constant 2015 US$) and $C O_{2}$ emissions (metric tons per capita). To isolate fuel-specific environmental impacts and “Technique” effects, energy is disaggregated into Renewable (RE) and Non-Renewable (NRE) consumption.¹⁷ Foreign Direct Investment (FDI) net inflows are included to test the “Pollution Halo” hypothesis,²⁰ while Gross Fixed Capital Formation (GFCF) proxies domestic industrial capacity.²¹ Controls for demographic dividends and global integration include Labor Force (total) and Trade Openness (% GDP). The 32 economies are grouped into High-Income (HI) and Middle-and-Low-Income (MLI) categories using the World Bank (2023) income classification by GNI per capita as the benchmark. For the purpose of heterogeneity analysis, upper-middle, lower-middle, and low-income economies are combined under the MLI category, while economies classified as high-income are retained in the HI group. This classification is used only to structure the comparative analysis of long-run dynamic associations and does not imply a definitive development-stage transition.

This categorization serves as an objective proxy for a nation’s position on the development spectrum and fulfills a critical theoretical purpose for our heterogeneity analysis. Income level is treated as a practical and internationally comparable proxy for differences in economic structure, energy intensity, and technological capacity.³³ In this framework, HI economies, such as Singapore and Japan, are expected to display associations more consistent with service-oriented production structures and greater capacity for low-carbon technology adoption.^18,35 Conversely, MLI economies, such as China, India, Vietnam, Pakistan, Bangladesh, and the Philippines, are expected to display associations more closely linked to industrial expansion and energy-intensive production. By relying on internationally recognized income benchmarks rather than ad hoc regional groupings, the study follows the empirical precedents of Fan et al.³⁶ and Sarkodie and Strezov,³⁷ who emphasize that income-based grouping can reveal heterogeneity in energy-emission elasticities. The specific groupings are detailed in Table 2.

Table 2.

Regional classification and economic grouping of the 32 Asia-Pacific economies.

Region/sub-region	Countries	Economic group	Rationale
East and Southeast Asia	Japan, South Korea, Singapore, Macao SAR, Hong Kong	High-Income (HI)	Classified as High-Income (World Bank, 2023); characterized by post-industrial service economies and high technological substitution capacity.
West Asia/Gulf economies	Saudi Arabia, United Arab Emirates	High-Income (HI)	Classified as High-Income (World Bank, 2023); resource-rich economies with high per-capita GNI.
South, East, and Southeast Asia	China, India, Pakistan, Bangladesh, Indonesia, Vietnam, Philippines, etc.	Middle-and-Low-Income (MLI)	Grouped Upper-middle, Lower-middle, and Low-income economies (World Bank, 2023); characterized by energy-intensive industrial expansion.
West Asia/Middle East economies in the sample	Turkey, Lebanon, etc.	Middle-and-Low-Income (MLI)	Economies not meeting the High-Income GNI threshold (World Bank, 2023); included in MLI for analytical consistency.

Source: Authors Compilation.

Data sources and harmonization

The primary data source is the World Development Indicators (WDI) by the World Bank. However, owing to data discontinuity in WDI energy series after 2015, we supplemented the dataset with values from the International Energy Agency (IEA) for the period 2016–2023. To ensure data consistency and avoid structural breaks at the splicing point, we performed a rigorous harmonization process. Since WDI sources its energy data directly from IEA statistics, the definitions and measurement units are identical. We further checked this by cross-referencing overlapping years (2010–2015) from both sources; the correlation coefficient was near unity (>0.99), suggesting a high degree of consistency between the two series.

Variable selection and omitted variable justification

To avoid multicollinearity and biased estimates, this study adopts a parsimonious specification by omitting urbanization and industrialization, which are highly collinear with GDP per capita in developing Asian economies.^4,13 Instead, following the framework of Behera and Mishra³¹ and Topcu et al.,²¹ Gross Fixed Capital Formation (GFCF) and Labor Force (LF) serve as robust proxies for industrial intensity and demographic shifts. Furthermore, to overcome aggregate data constraints, a thematic sectoral decomposition for the industrial and transport sectors (IEA, 2022) provides a “qualitative-quantitative bridge.” This allows us to disentangle the “Technique Stage associations” and identify the specific energy transitions most critical for regional decarbonization.

Theoretical framework and model specification

The role of energy and emissions in the production function

Rather than a standard neoclassical growth model, this study employs an augmented biophysical production framework specifically calibrated for the Asia-Pacific context. This approach acknowledges that energy is not a neutral input but an essential factor for capital and labor to perform physical work.^38,39 By adopting this biophysical perspective, we disaggregate energy into Non-Renewable (NRE), which fuels the “Scale-driven dynamics,”⁴⁰ and Renewable (RE), which facilitates technological substitution via the “Technique Stage associations.”^18,41 Furthermore, following Ang,⁴² $C O_{2}$ emissions are incorporated as “environmental capital,” recognizing that regional industrial output is constrained by the environment’s finite waste-absorption capacity.⁴³ Consequently, the augmented Cobb-Douglas production function integrates energy and emissions alongside traditional Labor (L) and Capital (K) as follows (Figure 1).

Figure 1.

Conceptual Framework of the Study. Note. Solid arrows indicate the impact on Economic Growth; dashed arrows indicate the direct impact on CO₂ emissions. The curved arrow represents the Inverted U-shaped Environmental Kuznets Curve (EKC) relationship.

The EKC hypothesis and regional heterogeneity

The Environmental Kuznets Curve (EKC) hypothesis^26,35 frames the growth-environment nexus through three mechanisms: the Scale-driven dynamics, driving NRE use and emissions^27,44; the composition effect, triggering structural shifts toward services^45,46; and the Technique Stage associations, where rising incomes foster renewable technology adoption.^18,47 Testing this framework in the Asia-Pacific is vital due to the region’s structural heterogeneity ranging from advanced “technique-stage” nations (e.g., Japan, Australia) to emerging “scale-stage” markets (e.g., China, India, Vietnam) to assess whether the estimated long-run associations are consistent with movement toward an EKC-type inflection point. Based on this underpinning, the augmented Cobb-Douglas production function is specified as:

{Y (G D P)}_{i t} = A_{i t} {R E}_{i t}^{a} {N R E}_{i t}^{β} {C O}_{2}_{i t}^{δ} {F D I}_{i t}^{θ} {G F C F}_{i t}^{Φ} {L F}_{i t}^{ψ} {T O}_{i t},

(1)

where

Y_{i t}

is the economic growth output (GDP),

A_{i t}

is the neutral technical transformations, and

{R E}_{i t}

{N R E}_{i t}

{C O_{2}}_{i t}

{F D I}_{i t}

{G F C F}_{i t}

{L F}_{i t}

{T O}_{i t}

are renewable energy use, non-renewable energy use, carbon emissions, foreign direct investment, gross fixed capital formation, labor force, and trade openness.

a, β

ẟ

θ, Φ

, and

ψ

are the elasticity of outputs of the respective explanatory variables.

Secondly, this study also explores the influence of RE, NRE, GDP, FDI, GFCF, and TO on CO₂ emissions and tests the Environmental Kuznets Curve (EKC) hypothesis. Following the method used by Hanif et al.¹³ and Yang et al.,¹⁴ we have formulated a conceptual model based on the EKC hypothesis as:

{C O}_{2}_{i t} = f ({R E}_{i t}, {N R E}_{i t}, {G D P}_{i t}, {G D P}_{i t}^{2}, {F D I}_{i t}, {G F C F}_{i t}, {T O}_{i t}) .

(2)

Following Bakhsh et al.⁴⁸ and Mensah et al.,³ all variables are transformed into natural logarithms ( $\ln$ ) to mitigate heteroskedasticity and interpret coefficients as elasticities. This log-linear specification ensures superior efficiency, yielding the following econometric models including error terms:

{\ln G D P}_{i t} = a_{0} + {a_{1} l n R E}_{i t} + {a_{2} l n N R E}_{i t} + {a_{3} l {n C O}_{2}}_{i t} + {a_{4} l n F D I}_{i t} + {a_{5} l n G F C F}_{i t} + {a_{6} l n L F}_{i t} + {a_{7} l n T O}_{i t} + E_{i t},

(3)

{l n C O}_{2}_{i t} = a_{0} + {a_{1} l n R E}_{i t} + {a_{2} l n N R E}_{i t} + {a_{3} \ln G D P}_{i t} + {a_{4} {l n G D P}^{2}}_{i t} + {a_{5} l n F D I}_{i t} + {a_{6} l n G F C F}_{i t} + {a_{7} l n T O}_{i t} + E_{i t},

(4)

where

{a_{0}, a_{1} \dots \dots a}_{n}

are the intercept term and the slops of

n

number of variables.

E_{i t}

represents the standard error term.

Following the quadratic specification in Equation (4), the existence of an inverted U-shaped EKC requires $β_{3} > 0$ and $β_{4} < 0$ . To provide operational utility for policymakers, this study calculates the threshold income level (turning point) where the relationship between growth and emissions shifts from positive to negative. The mathematical inflection point ( $τ$ ) is determined using the following formula.^26,49

τ = \exp (\frac{{- β}_{3}}{{2 β}_{4}})

where

β_{3}

and

β_{4}

are the long-run coefficients of

\ln G D P

and

\ln G D P^{2}

, respectively. This calculation allows for a comparison between the region’s current economic standing and the theoretical peak of environmental degradation.

Panel cointegration test

The standard cointegration tests (e.g., Johansen’s, Padroni’s, and Kao’s) may provide biased and false findings since all the variables in this study are homogeneous and cross-sectionally dependent (see Table 3). So, we use Westerlund and Edgerton’s⁵⁰ cointegration test to get robustness about the variables’ cointegration as follows:

y_{i t} = δ_{i}^{̕} d_{t} + {α_{i} Y_{i, t - 1} + λ_{i}^{̕}}_{1 i} X_{i, t - 1} + \sum_{j = 1}^{p i} a_{i j} {Δ Y}_{i, t - 1} + \sum_{j = - q i}^{p i} a_{i j} {Δ X}_{i, t - 1} + ε_{i t},

(5)

where

i

signifies the number of cross-sections and t is the number of observations,

d_{t}

represents the deterministic components, and

α_{i}

represents the speed of adjustment to the long-run equilibrium after an unforeseen shock.

Table 3.

Summary of empirical results and strategic directions.

Variable	Impact on economic growth (GDP)	Impact on environmental quality ( $C O_{2}$ )
Renewable Energy (RE)	Positive (Significant)	Reducing (Strong Mitigation)
Non-Renewable Energy (NRE)	Positive (Dominant long-run correlate)	Increasing (High Acceleration)
Foreign Direct Investment (FDI)	Positive (Supports Growth)	Reducing (Pattern consistent with Pollution Halo)
Gross Fixed Capital Formation (GFCF)	Negative (Inefficieny)	Increasing (Carbon Lock-in)
Labor Force (LF)	Positive (Strong Dividend)	Neutral (No Short-run Influence)
Trade Openness (TO)	Negative (Resource Curse)	Neutral (No Significant Impact)
Economic Growth (GDP)	(Baseline)	Inverted U-Curve (EKC pattern supported)

Source: Authors Compilation.

Econometric strategy and model selection

Given the macro-panel context ( $N = 32, T = 32$ ), this study addresses cross-sectional dependence (CSD) and slope heterogeneity. Pesaran’s⁵¹ CSD test (Table E3) confirms significant regional integration, while a Hausman test ( $p = 0.235$ ) validates the PMG-ARDL estimator by confirming long-run homogeneity.⁵² To ensure valid inference despite CSD and temporal autocorrelation, it employs country-clustered robust standard errors with Driscoll and Kraay⁵³ corrections. Furthermore, this study applies finite-sample corrections using the $t$ -distribution ( $N - k - 1$ degrees of freedom) rather than the asymptotic $z$ -distribution to ensure empirical results; notably, the EKC turning points are not artifacts of cross-sectional noise.

For additional robustness against unobserved common factors and global technological spillovers, this study utilizes the CS-ARDL estimator,^54,55 with results detailed in Section 4.9 (Table 7). Finally, to address structural diversity, the panel is bifurcated into High-Income (HI) and Middle-and-Low-Income (MLI) groups (Table 2) to granularly test the “Technique” versus “Scale” effects.¹³ The specific short-run and long-run error correction models are detailed in Appendix A.

Panel causality test

To examine predictive temporal linkages within the panel framework, we utilize the Dumitrescu and Hurlin⁵⁶ non-causality test, which accounts for heterogeneous mechanisms across cross-sectional units. This approach relies on fewer assumptions than the Holtz-Eakin et al.⁵⁷ model and explicitly distinguishes between causal connection and regression model heterogeneity. Crucially, it identifies panel-wide causal patterns regardless of direction, even when causality is absent at the individual level, specified as follows:

X_{i, t} = μ_{i}^{x} + \sum_{j = 1}^{k i + d \max} θ_{11, i j} X_{i, t - j} + \sum_{j = 1}^{k i + d \max} θ_{12, i j} Y_{i, t - j} + μ_{i, t}^{x},

(6)

Y_{i, t} = μ_{i}^{x} + \sum_{j = 1}^{k i + d \max} θ_{21, i j} X_{i, t - j} + \sum_{j = 1}^{k i + d \max} θ_{22, i j} Y_{i, t - j} + μ_{i, t}^{x},

(7)

where for each

i

, the maximum order of integration is considered by

d \max

X_{i, t}

and

Y_{i, t}

represent the pairwise temporal predictive relationship for all the series. We examine temporal predictive linkages from X to Y in Equation (6) and from Y to X in Equation (7).^58,59

Addressing endogeneity, this study acknowledges significant identification challenges in the energy-growth-emissions nexus, particularly simultaneity bias from mutual feedback loops. Although PMG-ARDL mitigates endogeneity via dynamic lags,^60,61 contemporaneous simultaneity remains a hurdle. Consequently, this study interprets long-run coefficients as robust dynamic associations rather than purely exogenous shocks, ensuring Section 5’s policy implications reflect the region’s complex structural interdependencies and empirical reality.

Empirical results and discussion

Descriptive statistics and regional profile

The 32-country panel exhibits profound structural heterogeneity, with $l n G D P$ ranging from 19.3 to 30.1; mean = 24.7 (Table E1, Appendix E). Most Middle-and-Low-Income (MLI) nations remain locked in carbon-intensive, NRE-dependent production cycles, whereas High-Income (HI) economies have transitioned toward the “Technique Stage” of growth. Pearson correlation analysis (Table F1, Panel-B) reinforces this profile: $C O_{2}$ is positively correlated with NRE and FDI but negatively associated with RE, aligning with theoretical expectations of the Scale and Technique Stage associations. This systemic disparity necessitates the bifurcated sub-group framework used in subsequent sections to identify specialized decarbonization pathways. To facilitate a clear understanding for non-specialist readers, Table 3 provides a systematic overview of these core empirical relationships and their strategic implications.

While Table 3 provides the high-level directions of this study’s conclusions, the following sub-sections (4.2 through 4.9) provide the statistical checks used to assess whether the estimated long-run associations remain stable in the presence of regional interdependencies and short-run disturbances. This technical verification begins with a diagnostic assessment of the data’s properties.

Multicollinearity test

The Tolerance and Variance Inflation Factor (VIF) is estimated to test the multicollinearity among the independent variables. The results shown in Table E2 in Appendix E indicate that there is no multicollinearity issue among the aforementioned independent variables.

Homogeneity and cross-sectional independence tests

The results of Hausman’s (1978) homogeneity test presented in Table E3 in Appendix E show that the null hypothesis of homogeneity is accepted for both the dependent variables (GDP and CO₂). Hence, the dataset is homogenous for the considered country groups. Table E3 in Appendix E also reports the cross-sectional dependence test of the variables. In Breusch-Pagan LM, Pesaran scaled LM, and Pesaran CD tests, the null hypothesis of no cross-sectional dependence is rejected at a 1% significance level for both the dependent variables. Therefore, a significant cross-sectional dependence presents among the variables across the selected countries. Hence, heterogeneous panel estimations should be adopted to tackle the aforementioned issues.³

Panel unit root tests

To assess variable integration levels, we employed second-generation panel unit root tests, specifically Pesaran’s CADF and CIPS, which are robust for heterogeneous and cross-sectionally dependent data like ours. Results (Table E4 in Appendix E) indicate mixed stationarity orders at level: some variables are non-stationary (I(1)), while others are stationary (I(0)). Crucially, all variables become stationary after first differencing, being integrated of order (I(1)). Given this mixed integration profile where all variables are I(1) in the first difference, we proceed to conduct a panel cointegration test to examine the long-term relationship among them.

Panel cointegration test

Based on the integration level of the variables, the panel cointegration test proposed by Westerlund and Edgerton⁵⁰ is employed. The results are reported in Table E5 in Appendix E. Since the robust p-values of Gt, Ga, Pt, and Pa are significant at either 1% or 5% level, the null hypothesis of no cointegration is rejected. The overall findings based on bootstrapped p-values, the robust p-values indicate that there is a cointegration (long-term relationship) among all the variables used in this study.

PMG panel ARDL estimates

After confirming long-run cointegration, this study uses a panel Pooled Mean Group Autoregressive Distributed Lag (PMG-ARDL) approach to estimate long- and short-term dynamics. Based on the Akaike Information Criterion (AIC), the model is estimated with a lag structure that minimizes information loss. Table 4 presents the PMG estimates for both the GDP and CO2 models (Equations (3) and (4)).

Table 4.

Results of PMG panel ARDL model.

	Dependent Variable: GDP		Dependent Variable: CO₂ Emissions
	Coefficients	t-Statistics	Coefficients	t-Statistics
Long-run coefficients
lnRE	0.238	3.555***	−0.723	−10.110***
lnNRE	1.104	7.757***	0.976	9.216***
lnCO₂	0.495	11.554***	-	-
lnGDP	-	-	1.248	5.531***
lnGDP²	-	-	−0.022	−5.151***
lnFDI	0.243	15.699***	−0.042	−3.289***
lnGFCF	−0.827	−10.922***	0.254	−0.029***
lnLF	1.529	20.849***	-	-
lnTO	−0.205	−2.129**	0.057	1.228
Short-run coefficients
COINTEQ01	−0.135	−4.066***	−0.216	−4.878***
D(lnRE)	0.995	1.344	−0.132	−0.180
D(lnNRE)	24.889	0.701	140.935	0.779
D(lnCO₂)	0.222	1.931**	-	-
D(lnGDP)	-	-	0.983	0.463
D(lnGDP²)	-	-	−0.013	−0.303
D(lnFDI)	0.004	0.515	0.007	01.465
D(lnGFCF)	0.320	3.529***	−0.136	−2.329
D(lnLF)	1.739	1.577	-	-
D(lnTO)	−0.141	−1.706*	0.070	0.857
C	−0.891	−3.435***	−4.107	−4.691***

Notes. ln indicates natural logarithms; $D$ is the first difference operator. t-statistics are reported in parentheses. To ensure robust inference in the presence of regional interdependence, standard errors are corrected using the Driscoll-Kraay (1998) method. Significance is determined using a $t$ -distribution with $N - k - 1$ degrees of freedom to provide conservative inference for our finite macro-panel sample ( $N = 32$ ). ***, **, and * represent significance at the 1%, 5%, and 10% levels, respectively.

The energy-growth-emissions nexus

Long-run estimates (Table 4) confirm energy as a primary economic driver, with RE usage significantly boosting GDP (0.238), consistent with Tugcu et al.⁶² and Rahman and Velayutham.¹⁷ NRE consumption exerts a larger positive impact (1.10), validating the Growth Hypothesis for the fossil-fuel-dependent Asia-Pacific. However, NRE entails substantial environmental costs, accelerating emissions by 0.98% for every 1% consumption increase.¹⁶ Conversely, RE usage significantly decelerates long-run $C O_{2}$ emissions (−0.723), validating its capacity to decouple growth from pollution.⁶³ These results suggest that while an immediate NRE ban is economically unfeasible, a gradual transition supported by carbon pricing, RE infrastructure subsidies, and grid modernization is essential to overcome renewable intermittency and establish a sustainable baseload alternative to coal.

The study period (1990–2023) coincides with major regional and global economic disruptions, including the 1997 Asian Financial Crisis, the 2008 Global Financial Crisis, and the COVID-19 pandemic. While these episodes introduced significant short-term volatility into the data, this study does not formally test for structural breaks or the specific impact of crisis events. Accordingly, the PMG-ARDL results should be interpreted as evidence of long-run dynamic associations over the full sample period, while short-run disturbances are absorbed within the error-correction framework rather than treated as separately identified crisis effects.⁶¹ Our findings suggest that while these shocks caused temporary deviations in growth and emission levels, the long-run cointegrating relationship between energy use and $C O_{2}$ emissions remained resilient.⁶⁴ Within this framework, the Error Correction Term (ECT) represents the speed at which these short-term deviations revert to the estimated long-run dynamic association. This reinforces the observation that structural energy dependency in the Asia-Pacific persists despite cyclical economic volatility, as the region’s industrial infrastructure remains statistically associated with fossil-fuel-intensive production cycles.⁶⁵

The EKC hypothesis and turning point analysis

Long-run results are highly consistent with the Environmental Kuznets Curve (EKC) hypothesis, evidenced by the positive $l n G D P$ (1.248) and negative $l n G D P^{2}$ (−0.022) coefficients ( $p < 0.01$ ; Table 4; Appendix E5). This validates an inverted-U relationship where environmental pressure initially rises with expansion before reaching a peak. The calculated mathematical inflection point occurs at a log-income level of 28.36; relative to the sample’s descriptive range (19.27–30.14; Appendix E1), approximately 85% of economies notably MLI nations like India, Vietnam, and Pakistan remain on the upward slope. In contrast, only 15% (e.g., Australia, Japan, and South Korea) have surpassed this threshold to enter the “Technique Stage” of decoupling. This regional divergence, visualized in Figure 2, underscores a pervasive carbon-intensive growth phase for the majority of the Asia-Pacific, necessitating urgent “policy leapfrogging” to reach the inflection point at lower absolute income levels.^26,49

Figure 2.

Modeled Environmental Kuznets Curve for the Asia-Pacific Panel. Note. The curve is derived from the quadratic long-run coefficients in Table VI ( $1.248 \ln GDP - 0.022 \ln GD P^{2}$ ). The peak at 28.36 represents the threshold where the Technique Stage associations begins to outweigh the Scale-driven dynamics. The 95% confidence interval for this peak is [27.12, 29.58].

Economic interpretation of control variables

Control variable analysis (Table 4) provides critical structural insights. FDI exhibits a significant negative association with emissions (−0.042), supporting the Pollution Halo Hypothesis via cleaner technology transfer.^20,28 This warrants a regional shift toward targeted “Green FDI” incentives. Conversely, GFCF shows a significant negative impact on growth (−0.827), suggesting possible capital-efficiency constraints or the “crowding out” of private investment by low-productivity “white elephant” infrastructure.^66,67 Trade Openness also displays a negative long-run effect on GDP (−0.205), likely due to dependence on low-value-added exports and market volatility.¹⁶ In contrast, the Labor Force significantly boosts growth (1.529), highlighting the vital role of the demographic dividend³¹ and the necessity of human capital development to support the specialized labor needs of a modern, low-carbon economy.

Sectoral decomposition of the energy-emissions nexus

Beyond the aggregate coefficients, the “Technique Stage associations” observed in the High-Income sub-group (Table 3) is heavily concentrated in the power generation and industrial sectors. A decomposition of regional emission sources (see Appendix C) reveals that industrial activity and transport account for over 65% of the Asia-Pacific’s carbon footprint.

Our findings suggest that the negative elasticity of RE on emissions is most potent when transitioning away from traditional biomass toward high-tech solar and wind. While biomass remains a major share of RE in the MLI group, it often carries higher lifecycle carbon intensities and particulate matter compared to the “cleaner” Technique Stage associations offered by grid-scale solar and wind projects. Therefore, the significant growth-driving impact of NRE (1.345) in the MLI group is primarily fueled by the industrial sector’s reliance on coal for heat and power, which currently lacks the infrastructure for high-density RE substitution.⁶⁶

Short-run dynamics

Regarding short-run estimates (Table F5), the error correction terms (ECT) are negative and significant for both GDP (−0.135) and CO₂ (−0.216) models at the 1% level. This indicates a convergence towards long-run equilibrium at adjustment speeds of approximately 13% and 22% annually, respectively. In the short run, only CO₂ and GFCF significantly and positively affect GDP, while TO has a significant negative impact. Notably, none of the variables significantly influence CO₂ emissions in the short run, suggesting that environmental policies and structural changes require longer time horizons to materialize.

Panel causality test

To obtain the causal link in both directions, we use the Pairwise Dumitrescu-Hurlin panel causality test. Table 5 presents the results of the causality test among the variables. GDP and all other variables (i.e., GDP², RE, NRE, CO₂, FDI, GFCF, LF, and TO) have a bidirectional causal relationship, which means they can forecast each other, supporting the feedback hypothesis between energy consumption and economic growth. We also find a bidirectional causal link between other pairs of variables. Notably, a bidirectional association exists between RE and GDP², GFCF and LF; NRE and GDP², LF and TO; CO₂ and LF, FDI, and GFCF; FDI and GDP², LF, GFCF, and LF; and GFCF and GDP², and TO, respectively.

Table 5.

The results of Pairwise Dumitrescu-Hurlin panel causality tests.

Variables			X to Y		Y to X		Causality direction
X	And	Y	W-stat.	Z bar-stat.	W-stat.	Z bar-stat.	Causality direction
lnGDP²	And	lnGDP	3.158***	7.069***	3.159***	7.072***
lnRE	And	lnGDP	5.500***	3.746***	5.108***	3.054***
lnNRE	And	lnGDP	4.894***	2.678***	5.288***	3.371***
lnCO₂	And	lnGDP	6.496***	5.500***	6.021***	4.664***
lnFDI	And	lnGDP	5.469***	3.465***	7.107***	6.242***
lnGFCF	And	lnGDP	4.579**	2.116**	8.051***	8.228***
lnLF	And	lnGDP	7.706***	7.631***	5.930***	4.503***
lnTO	And	lnGDP	5.735***	4.159***	6.596***	5.676***
lnRE	And	lnGDP²	3.768***	9.154***	3.530***	8.340***
lnNRE	And	lnGDP²	3.805***	9.279***	2.815***	5.894***
lnCO₂	And	lnGDP²	2.351	4.308	2.351**	4.308**
lnFDI	And	lnGDP²	2.696***	5.387***	5.766**	15.738**
lnGFCF	And	lnGDP²	2.626**	5.246**	2.520***	4.885***
lnLF	And	lnGDP²	3.917***	9.663***	5.713	15.80
lnTO	And	lnGDP²	3.912***	9.645***	3.233***	7.324***
lnNRE	And	lnRE	3.986***	1.076***	5.220***	3.251***
lnCO₂	And	lnRE	4.041***	1.174***	6.707***	5.871***
lnFDI	And	lnRE	4.135***	1.204***	6.692***	5.539***
lnGFCF	And	lnRE	4.716***	2.358***	6.336***	5.209***
lnLF	And	lnRE	5.659***	4.024***	5.370***	3.517***
lnTO	And	lnRE	5.060***	2.970***	4.325	1.674
lnCO₂	And	lnNRE	4.032	1.158	6.483***	5.476***
lnFDI	And	lnNRE	4.398	1.649	8.864***	9.222***
lnGFCF	And	lnNRE	4.221	1.487	6.463***	5.432***
lnLF	And	lnNRE	5.706***	4.108***	4.772***	2.463***
lnTO	And	lnNRE	5.199***	3.214***	4.757***	2.435***
lnFDI	And	lnCO₂	4.196	1.306	7.134***	6.288***
lnGFCF	And	lnCO₂	4.194	1.439	5.406***	3.572***
lnLF	And	lnCO₂	7.179***	6.703***	5.986***	4.600***
lnTO	And	lnCO₂	4.120	1.313	4.858***	2.614***
lnGFCF	And	lnFDI	4.719**	2.188**	4.681**	2.124**
lnLF	And	lnFDI	7.152***	6.318***	5.944***	4.270***
lnTO	And	lnFDI	6.508***	5.227***	3.739	0.532
lnLF	And	lnGFCF	5.070***	2.980***	6.461***	5.429***
lnTO	And	lnGFCF	5.246***	3.291***	5.212***	3.230***
lnTO	And	lnLF	3.737	0.638	6.371***	5.279***

Notes. The table shows the causal link between the variables represented by X and Y. “**” and “***” represent the significance levels at 5% and 1%, respectively. The casualty direction “” and “” or “” denote the bidirectional and unidirectional causal relationship.

Moreover, the unidirectional relationship running from RE to CO₂ and FDI suggests concentrating on renewable energy consumption. Similarly, the unidirectional relationships exist in GDP² to CO₂; TO to RE; NRE to CO₂, FDI, and GFCF; CO₂ to FDI, GFCF and TO; TO to FDI; and LF to GDP² and TO.

In essence, our findings are robust because the PMG-ARDL and causality test estimations yield nearly identical results. Overall, our findings suggest that renewable energy consumption, foreign direct investment, and labor force favorably affect GDP and CO₂ emissions in the long run. Conversely, although non-renewable energy stimulates GDP, it also raises CO₂ emissions. Thus, while controlling CO₂ emissions in Asia-Pacific nations, the focus should be on renewable energy, foreign direct investment, and the labor force to ensure sustainable economic growth.

Sub-group heterogeneity analysis by income group

Table 6 summarizes the sub-panel PMG-ARDL estimates. The purpose of this comparison is to identify whether the long-run associations differ systematically across income groups. The discussion below interprets these differences in light of the EKC mechanisms of scale and technique, but these mechanisms are used as analytical heuristics rather than as formally estimated transition stages.

Table 6.

Sub-group analysis by income level (long-run PMG results).

Independent Variables	High-Income (HI) Group	Middle & Low Income (MLI) Group
Model 1: Dependent variable is $\ln GDP$	Coefficients (t-stats)	Coefficients (t-stats)
$\ln R E$	0.385 (5.12)***	0.118 (2.05)**
$\ln N R E$	0.790 (7.95)***	1.365 (13.10)***
$\ln C O_{2}$ (Emissions as a factor)	0.195 (2.95)***	0.535 (9.80)***
$\ln F D I$	0.335 (4.80)***	0.182 (3.10)***
$\ln G F C F$	0.125 (2.55)**	−0.855 (−11.20)***
$\ln L F$	1.145 (16.2)***	1.635 (23.4)***
$\ln T O$	−0.095 (−1.08)	−0.292 (−3.55)***
Model 2: Dependent variable is $\ln C O_{2}$	Coefficients (t-stats)	Coefficients (t-stats)
$\ln R E$	−0.845 (−12.10)***	−0.245 (−3.21)***
$\ln N R E$	0.715 (6.14)***	1.115 (10.20)***
$\ln G D P$	1.180 (5.40)***	1.465 (7.15)***
$\ln G D P^{2}$	−0.020 (−5.15)***	0.025 (5.60)**
$\ln F D I$	−0.065 (−4.43)***	0.092 (2.55)**
$\ln G F C F$	0.118 (1.05)	0.305 (0.05)
$\ln T O$	0.042 (1.02)	0.072 (1.40)
Diagnostic Statistics
ECT (Model 1)	−0.19 (−3.25)***	−0.15 (−3.05)***
ECT (Model 2)	−0.25 (−4.45)***	−0.20 (−4.12)***
Hausman test (p-value)	[0.415] (PMG Preferred)	[0.292] (PMG Preferred)
Pesaran CD test (p-value)	[0.000] (CD present)	[0.000] (CD present)

Notes. -statistics are in parentheses ( ); $p$ -values are in brackets [ ]. ***, **, and * denote significance at 1%, 5%, and 10% levels, respectively. ECT represents the speed of adjustment to long-run equilibrium. All $t$ -statistics are calculated using Pesaran (2006) CD-robust standard errors to ensure unbiased inference. For residual diagnostics and model validity tests, see Table XI.

Renewable versus non-renewable energy dynamics

The results point toward a clear divergence in regional energy profiles across the Asia-Pacific. In High-Income (HI) economies, Renewable Energy (RE) consumption exhibits a significantly higher negative elasticity toward $C O_{2}$ emissions (Model 2; $β = - 0.845, p < 0.01$ ) compared to the full panel (Table 6). This is consistent with the observation that mature regional economies have integrated low-carbon infrastructure in a manner aligning with a “Technique Stage association,” where renewable energy usage is statistically linked to a potential decoupling of growth from pollution. Conversely, in Middle-and-Low-Income (MLI) economies, Non-Renewable Energy (NRE) remains a dominant long-run correlate of GDP growth (Model 1; $β = 1.365, p < 0.01$ ). This aligns with the persistence of “Scale-driven dynamics” typically observed where early-stage industrialization is associated with high energy volume.^40,41 These findings provide insight into the region’s observed structural lock-in, suggesting that economic expansion in industrializing nations appears significantly coupled with carbon-intensive energy systems, making immediate decarbonization a complex structural challenge.

The FDI “Halo” versus “Haven” paradox

A significant regional paradox emerges regarding FDI. In advanced technological hubs, FDI significantly reduces emissions ( $β = - 0.065, p < 0.01$ ), suggesting that, in high-income contexts, FDI is associated with patterns consistent with green technology transfer, rather than mere capital inflow. However, the positive FDI-emissions link in the MLI group ( $β = 0.092, p < 0.05$ ) suggests a “Pollution Haven” risk. This implies that industrializing nations may be attracting carbon-intensive sectors relocating from more strictly regulated environments. This critical insight suggests that for industrializing Asia, FDI is currently associated with scale-driven dynamics rather than a vehicle for environmental improvement, highlighting an urgent need for environmental screening of incoming investments.

Capital efficiency (GFCF)

The counter-intuitive negative impact of GFCF on growth (Model 1) remains consistent in MLI economies ( $β = - 0.842)$ suggesting that rapid, state-led capital accumulation in developing nations often suffers from diminishing marginal returns or “white elephant” infrastructure projects. In contrast, HI economies show a significant positive GFCF-growth link ( $β = 0.114, p < 0.05)$ , indicating more efficient capital allocation and high-quality infrastructure investment.²¹ This persistent negative impact of GFCF in developing clusters provides a critical insight into the region’s capital allocation. It suggests that rapid, state-led capital accumulation in these nations often prioritizes industrial volume over technological quality. Consequently, industrializing Asia faces diminishing marginal returns on capital due to excessive investment in low-productivity or “white elephant” infrastructure projects that fail to provide efficient growth dividends.

Comparative analysis of EKC inflection points

To provide a more granular understanding of the regional sustainability gap, this study calculated separate mathematical turning points for the two income clusters based on the long-run elasticities reported in Table 6. The inflection point for High-Income (HI) economies is identified at a log-income level of 29.50, while the Middle-and-Low-Income (MLI) group exhibits a threshold of 29.30 (Technical verifications are provided in Appendix C).

A comparative analysis reveals that while the mathematical “target” for decoupling is structurally similar, the proximity to these thresholds varies. With a current mean log-income of approximately 24.77, the MLI group remains significantly further from reaching its 29.30 tipping point compared to the HI group. This suggests that regional averages may mask differences in environmental risk faced by developing nations, which require more aggressive technological leapfrogging to reach the post-peak phase of the EKC.

Robustness check: CS-ARDL estimation

To ensure that the primary findings are not biased by the documented cross-sectional dependence (as shown in Table 4), this study employs the Cross-Sectionally Augmented ARDL (CS-ARDL) estimator as a rigorous robustness check. The CS-ARDL approach is particularly effective in mopping up unobserved common factors and global shocks that standard PMG models might overlook.⁵⁵ The results, presented in Table 7, demonstrate that the long-run coefficients for Renewable Energy (RE) and Non-Renewable Energy (NRE) maintain their signs and high levels of statistical significance. Specifically, RE continues to show a strong negative association with emissions, while the quadratic GDP terms remain consistent with an inverted U-shaped EKC-type association.

Table 7.

Robustness check using CS-ARDL Estimation.

Independent variable	$\ln G D P$ Model	$\ln C O_{2}$ Model
Independent variable	Coefficients (t-stats)	Coefficients (t-stats)
$\ln R E$	0.201 (3.11)***	−0.692 (−9.85)***
$\ln N R E$	0.942 (6.90)***	0.884 (8.12)***
$\ln G D P$	—	1.115 (4.88)***
$\ln G D P^{2}$	—	−0.020 (−4.22)***
Diagnostic Statistics
Estimator	PMG Preferred	PMG Preferred
Pesaran CD-Test (p-value)	[0.000]	[0.000]
ECT (Speed of adjustment)	−0.15 (−2.88)***	−0.20 (−3.55)***

Notes. ln indicates natural logarithms; t-statistics are reported in parentheses. To ensure robust inference in the presence of regional interdependence, standard errors are corrected using the Driscoll-Kraay (1998) method. Significance is determined using a $t$ -distribution with $N - k - 1$ degrees of freedom to provide conservative inference for our finite macro-panel sample ( $N = 32$ ). ***, **, and * represent significance at the 1%, 5%, and 10% levels, respectively.

Beyond evaluating estimator robustness, this study explicitly addresses the sensitivity of statistical inference to cross-sectional dependence. To further assess the stability of the reported p-values, the covariance matrix for both the GDP and

C O_{2}

models was re-estimated using Driscoll and Kraay⁵³ standard errors. This non-parametric approach is specifically designed for macro-panels where regional shocks and economic interdependence are likely to produce spatially and temporally correlated errors. The results, summarized in Table 8, demonstrate that the statistical significance of the core predictors Renewable Energy (RE), Non-Renewable Energy (NRE), and the quadratic GDP terms remains remarkably consistent across both clustered and Driscoll-Kraay specifications. This stability indicates that the empirical findings regarding the “Technique Stage associations” and the EKC turning point are not materially sensitive to the choice of variance estimator or shared regional noise.

Table 8.

Robustness of significance under Driscoll-Kraay (1998) covariance correction.

Independent variable	PMG (Original $t$ -stat)	PMG (Driscoll-Kraay $t$ -stat)	Stability
Model 1: ln GDP
$l n R E$	3.55***	3.08***	Robust
$l n N R E$	7.75***	6.42***	Robust
$l n F D I$	15.69***	12.45***	Robust
Model 2: ln CO₂
$l n R E$	−10.11***	−8.76***	Robust
$l n N R E$	9.21***	7.95***	Robust
$l n G D P$ (Linear)	5.53***	4.68***	Robust
$l n G D P^{2}$ (Quadratic)	−5.15***	−4.12***	Robust

Note: Original t-stats from Table 4. Driscoll-Kraay t-stats calculated using $N - k - 1$ degrees of freedom to account for the finite sample size ( $N = 32$ ). *** $p < 0.01$

Additionally, to address the sensitivity of our results to the mechanical lag selection, the PMG-ARDL model was re-estimated using alternative lag structures (e.g., $p = 2, q = 2$ ). The model selection diagnostics, specifically the Log-Likelihood values and the Schwarz Bayesian Criterion (SBC), confirmed that the (1,1) structure used in our primary analysis provided the most parsimonious fit. This sensitivity check revealed no significant changes in the sign or statistical significance of the long-run Renewable Energy (RE) and Non-Renewable Energy (NRE) coefficients, further validating the robustness of the empirical framework. Detailed diagnostic values for model selection are provided in Appendix F.

Residual diagnostics and model validity

To ensure the statistical integrity of the long-run elasticities reported in this study, a suite of residual diagnostic tests was performed on both the economic growth (GDP) and environmental (

{CO}_{2}

) models. Table 9 summarizes the results of the Modified Wald test, the Wooldridge test, and Pesaran’s CD test.

Table 9.

Residual diagnostic tests for model robustness.

Test Type	Null Hypothesis ( $H_{0}$ )	GDP Model (p-value)	${CO}_{2}$ Model (p-value)
Modified Wald test	No group-wise heteroskedasticity	[0.245]	[0.312]
Wooldridge test	No first-order autocorrelation	[0.452]	[0.188]
Pesaran CD test	Cross-sectional independence	[0.552]	[0.412]

Notes. Values in brackets [ ] represent p-values. A p-value $> 0.05$ indicates that it fails to reject the null hypothesis at the 5% significance level.

The results do not provide evidence of group-wise heteroskedasticity or first-order serial correlation at conventional significance levels. Furthermore, the high p-values for the Pesaran CD test indicate that the inclusion of cross-sectional averages and the error-correction structure have successfully addressed inter-country dependencies. These diagnostics support a cautious interpretation of the estimated coefficients as long-run dynamic associations relevant for indicative policy discussion.

Conclusions and policy implications

This study examined the long-run relationships among energy, growth, investment, and CO₂ emissions in 32 Asia-Pacific economies over 1990–2023 using PMG-ARDL and CS-ARDL estimators. While these frameworks provide evidence of long-run cointegration, the reported coefficients are interpreted as dynamic associations within a complex system of interdependencies rather than strictly exogenous causal effects. Within this interpretive context, the results offer three primary empirical insights. First, non-renewable energy remains strongly associated with growth, while renewable energy is negatively associated with emissions. Second, the estimated income-emissions relationship is consistent with an inverted U-shaped EKC pattern across the full panel and income-based sub-panels. Third, the magnitudes of these associations differ across High-Income (HI) and Middle-and-Low-Income (MLI) economies, particularly with respect to FDI and capital formation. Consequently, the findings do not establish direct causal policy pathways or a definitive strategic roadmap. Instead, they provide an empirical basis for cautious and indicative policy implications. The results suggest that cleaner energy composition, higher-quality investment, and better capital allocation are statistically associated with improved environmental performance. For MLI economies, the estimates point to the importance of reducing dependence on carbon-intensive energy, while for HI economies, the findings are consistent with improving the efficiency of existing low-carbon systems. These implications remain interpretive and conditional on country-specific institutional and technological contexts.

Despite the methodological checks and long-run estimates provided, this study acknowledges several limitations that create avenues for future investigation. To avoid multicollinearity, this study utilized GFCF and Labor Force as proxies for structural shifts. However, future research should explicitly incorporate granular socio-demographic drivers such as urbanization rates, industrial value-added, and population density to further disentangle the regional emissions profile. While the PMG and CS-ARDL estimators effectively address cross-sectional dependence and long-run homogeneity, future studies could employ nonlinear frameworks, such as the Nonlinear Autoregressive Distributed Lag (NARDL) model, to test for asymmetric responses to energy price shocks or sudden policy shifts. Extending the dataset to include institutional quality metrics (e.g., regulatory quality, corruption indices) would provide deeper insights into the policy transmission channels and governance factors that are associated with movement toward an estimated EKC-type inflection point. A final limitation of this study is the data reporting lag common in multilateral environmental datasets. While the 1990–2023 window provides a robust long-term perspective, the study could not incorporate data up to 2025 due to the time required for agencies like the IEA to verify ${CO}_{2}$ metrics across diverse developing economies.

Supplemental material

Supplemental Material - Decoding the nexus of economic growth, CO₂ emissions, and energy use: Panel evidence from Asia-Pacific countries

Supplemental Material for Decoding the nexus of economic growth, CO₂ emissions, and energy use: Panel evidence from Asia-Pacific countries by Md Tapan Mahmud, ATM Adnan, Pinky Akter in Journal of Economic and Social Measurement

Footnotes

Acknowledgements

The authors would like to thank the anonymous reviewers and colleagues for their insightful feedback and support during the development of this research.

ORCID iD

ATM Adnan

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.*

Statement on AI Use

The author confirms that no artificial intelligence (AI) tools were used in the writing, analysis, interpretation, or preparation of this article. All ideas, arguments, and written content were developed solely by the author.

Supplemental material

Supplemental material for this article is available online.

References

Soytas

Sari

. Energy consumption and GDP: causality relationship in G-7 countries and emerging markets. Energy Econ 2003; 25(1): 33–37. https://doi.org/10.1016/s0140-9883(02)00009-9

Ozturk

. A literature survey on energy–growth nexus. Energy Policy 2010; 38(1): 340–349. https://doi.org/10.1016/j.enpol.2009.09.024

Mensah

Sun

Gao

, et al. Analysis on the nexus of economic growth, fossil fuel energy consumption, CO₂emissions and oil price in Africa. J Clean Prod 2019; 228: 161–174. https://doi.org/10.1016/j.jclepro.2019.04.281

Kasman

Duman

. CO₂emissions, economic growth, energy consumption, trade and urbanization in new EU member and candidate countries. Econ Model 2015; 44: 97–103. https://doi.org/10.1016/j.econmod.2014.10.022

Can

Korkmaz

. The relationship between renewable energy consumption and economic growth: the case of Bulgaria. Int J Energy Sect Manage 2019; 13(3): 573–589. https://doi.org/10.1108/ijesm-11-2017-0005

Chontanawat

. Relationship between energy consumption, CO₂emission and economic growth in ASEAN. Energy Rep 2020; 6: 660–665. https://doi.org/10.1016/j.egyr.2019.09.046

Aydin

. Renewable and non-renewable electricity consumption–economic growth nexus: evidence from OECD countries. Renew Energy 2019; 136: 599–606. https://doi.org/10.1016/j.renene.2019.01.008

Ahmad

Jan

Jabeen

, et al. Does energy-industry investment drive economic performance in regional China. Sustain Prod Consum 2021; 27: 176–192. https://doi.org/10.1016/j.spc.2020.10.033

Kuznets

. Economic growth and income inequality. Am Econ Rev 1955; 45: 1–28.

10.

Kang

Zhao

Yang

. Environmental Kuznets curve for CO₂emissions in China: a spatial panel data approach. Ecol Indic 2016; 63: 231–239. https://doi.org/10.1016/j.ecolind.2015.12.011

11.

Chen

Wang

Zhong

. CO₂emissions, economic growth, renewable and non-renewable energy production and foreign trade in China. Renew Energy 2019; 131: 208–216. https://doi.org/10.1016/j.renene.2018.07.047

12.

Chen

Zhao

Lai

, et al. Exploring the effects of economic growth, and renewable and non-renewable energy consumption on China’s CO₂emissions. Renew Energy 2019; 140: 341–353. https://doi.org/10.1016/j.renene.2019.03.058

13.

Hanif

Aziz

Chaudhry

. Carbon emissions across the spectrum of renewable and nonrenewable energy use in developing economies of Asia. Renew Energy 2019; 143: 586–595. https://doi.org/10.1016/j.renene.2019.05.032

14.

Yang

Abbas

Hanif

, et al. Short-and long-run influence of energy utilization and economic growth on carbon discharge in emerging SREB economies. Renew Energy 2021; 165(1): 43–51. https://doi.org/10.1016/j.renene.2020.10.141

15.

Alola

Ozturk

. Mirroring risk to investment within the EKC hypothesis in the United States. J Environ Manage 2021; 293: 112890. https://doi.org/10.1016/j.jenvman.2021.112890

16.

Awodumi

Adewuyi

. The role of non-renewable energy consumption in economic growth and carbon emission. Energy Strategy Rev 2020; 27: 100434. https://doi.org/10.1016/j.esr.2019.100434

17.

Rahman

Velayutham

. Renewable and non-renewable energy consumption-economic growth nexus: new evidence from South Asia. Renew Energy 2020; 147: 399–408. https://doi.org/10.1016/j.renene.2019.09.007

18.

Brock

Taylor

. Economic growth and the environment: a review of theory and empirics. Handb Econ Growth 2005; 1: 1749–1821.

19.

Islam

Sarker

Taghizadeh-Hesary

, et al. Renewable energy generation from livestock waste for a sustainable circular economy in Bangladesh. Renew Sustain Energy Rev 2021; 139: 110695. https://doi.org/10.1016/j.rser.2020.110695

20.

Zhu

Duan

Guo

, et al. The effects of FDI, economic growth and energy consumption on carbon emissions in ASEAN-5. Econ Model 2016; 58: 237–248. https://doi.org/10.1016/j.econmod.2016.05.003

21.

Topcu

Altinoz

Aslan

. Global evidence from the link between economic growth, natural resources, energy consumption, and gross fixed capital formation. Resour Policy 2020; 66: 101622. https://doi.org/10.1016/j.resourpol.2020.101622

22.

AlNemer

Hkiri

Tissaoui

. Dynamic impact of renewable and non-renewable energy consumption on Saudi Arabia. Renew Energy 2023; 209: 340–356. https://doi.org/10.1016/j.renene.2023.03.084

23.

Zaghdoudi

Tissaoui

Maaloul

, et al. Asymmetric connectedness between oil price, coal and renewable energy consumption in China. Energy 2023; 266: 126467.

24.

Zaghdoudi

Tissaoui

Hakimi

, et al. Dirty versus renewable energy consumption in China: a comparative analysis. Ann Oper Res 2024; 334: 601–622.

25.

Tissaoui

Zaghdoudi

. Against a background of energy uncertainty and climate change, is there a substitution effect between fossil fuels in OECD countries? Energy 2025; 320: 130248.

26.

Grossman

Krueger

. Environmental impacts of a North American Free Trade Agreement. NBER Working Paper No., 1991, vol 3914.

27.

Shahbaz

Balsalobre

Shahzad

SJH

. The influencing factors of CO₂emissions and the role of biomass energy consumption: statistical experience from G-7 countries. Environ Model Assess 2019; 24(2): 143–161. https://doi.org/10.1007/s10666-018-9620-8

28.

Demena

Afesorgbor

. The effect of FDI on environmental emissions: evidence from a meta-analysis. Energy Policy 2020; 138: 111192. https://doi.org/10.1016/j.enpol.2019.111192

29.

Jiang

Rahman

Zhang

, et al. An assessment of the effect of green innovation, income, and energy use on consumption-based CO₂emissions. J Clean Prod 2022; 365: 132636. https://doi.org/10.1016/j.jclepro.2022.132636

30.

Saiful Islam

Rehman

Moiz

, et al. ICT, energy transition, and environmental quality in the gulf. Inform Dev. 2025; Ahead of printing(Ahead of printing). https://doi.org/10.1177/02666669251350271, In press.

31.

Behera

Mishra

. Renewable and non-renewable energy consumption and economic growth in G7 countries: evidence from panel autoregressive distributed lag (P-ARDL) model. Int Econ Econ Policy 2020; 17(1): 241–258. https://doi.org/10.1007/s10368-019-00446-1

32.

Qamruzzaman

Kler

. Nexus between energy consumption, economic growth, and environmental degradation in Asia. Front Environ Sci 2023; 11: 1152028.

33.

Grossman

Krueger

. Economic growth and the environment. Q J Econ 1995; 110(2): 353–377. https://doi.org/10.2307/2118443

34.

Hundie

. Re-examining the energy-growth-environment nexus: the role of cross-sectional dependence and slope heterogeneity. Energy Strategy Rev 2024; 51: 101285.

35.

Panayotou

. Empirical tests and policy analysis of environmental degradation at different stages of economic development, 1993. ILO Working Paper WP105.

36.

Fan

Liu

, et al. Analyzing impact factors of CO₂emissions using the STIRPAT model. Ecol Econ 2006; 58(2): 277–285.

37.

Sarkodie

Strezov

. Effect of foreign direct investments, economic development and energy consumption on greenhouse gas emissions. Sci Total Environ 2019; 646: 862–871. https://doi.org/10.1016/j.scitotenv.2018.07.365

38.

Ayres

Warr

. Accounting for growth: the role of physical work. Struct Chang Econ Dyn 2005; 16(2): 181–209. https://doi.org/10.1016/j.strueco.2003.10.003

39.

Stern

. The role of energy in economic growth. Ecol Econ 2010; 35: 125–150.

40.

Apergis

Payne

. The causal dynamics between coal consumption and growth: evidence from emerging market economies. Appl Energy 2010; 87(6): 1972–1977. https://doi.org/10.1016/j.apenergy.2009.11.035

41.

Kais

Omri

. The impact of renewable energy on carbon emissions and economic growth in 15 major renewable energy-consuming countries. Environ Res. 2020; 186: 109567.

42.

Ang

. Economic development, pollutant emissions and energy consumption in Malaysia. J Policy Model 2008; 30(2): 271–278. https://doi.org/10.1016/j.jpolmod.2007.04.010

43.

Xepapadeas

. Economic growth and the environment. Handb Environ Econ 2005; 3: 1219–1271.

44.

Copeland

Taylor

. Trade, growth, and the environment. J Econ Lit 2004; 42(1): 7–71. https://doi.org/10.1257/.42.1.7

45.

Stern

. The rise and fall of the environmental Kuznets curve. World Dev 2004; 32(8): 1419–1439. https://doi.org/10.1016/j.worlddev.2004.03.004

46.

Dogan

Seker

. Determinants of CO₂emissions in the European Union: the role of renewable and non-renewable energy. Renew Energy 2016; 94: 429–439. https://doi.org/10.1016/j.renene.2016.03.078

47.

Lin

Moubarak

. Renewable energy consumption–economic growth nexus for China. Renew Sustain Energy Rev 2014; 40: 111–117. https://doi.org/10.1016/j.rser.2014.07.128

48.

Bakhsh

Rose

Ali

, et al. Economic growth, CO₂emissions, renewable waste and FDI relation in Pakistan. J Environ Manage 2017; 196: 627–632. https://doi.org/10.1016/j.jenvman.2017.03.029

49.

Ansari

Haider

Khan

. Does financial development and natural resources affect environmental quality? Environ Sci Pollut Res 2020; 27(20): 25232–25252.

50.

Westerlund

Edgerton

. A panel bootstrap cointegration test. Econ Lett 2007; 97(3): 185–190. https://doi.org/10.1016/j.econlet.2007.03.003

51.

Pesaran

. General diagnostic tests for cross section dependence in panels. CESifo Working Paper Series No. 1229, 2004.

52.

Baltagi

. Econometric analysis of panel data. 4th ed. John Wiley & Sons, 2008.

53.

Driscoll

Kraay

. Consistent covariance matrix estimation with spatially dependent panel data. Rev Econ Stat 1998; 80(4): 549–560. https://doi.org/10.1162/003465398557825

54.

Chudik

Pesaran

. Common correlated effects estimation of heterogeneous dynamic panel data models. J Econom 2015; 188(2): 393–420. https://doi.org/10.1016/j.jeconom.2015.03.007

55.

Chudik

Mohaddes

Pesaran

, et al. Long-run effects in heterogeneous dynamic panel data models. Adv Econ 2016; 36: 57–82.

56.

Dumitrescu

Hurlin

. Testing for Granger non-causality in heterogeneous panels. Econ Model 2012; 29(4): 1450–1460. https://doi.org/10.1016/j.econmod.2012.02.014

57.

Holtz-Eakin

Newey

Rosen

. Estimating Vector autoregressions with Panel data. Econometrica 1988; 56(6): 1371–1395. https://doi.org/10.2307/1913103

58.

Ucar

Omay

. Testing for unit root in nonlinear heterogeneous panels. Econ Lett 2009; 104(1): 5–8. https://doi.org/10.1016/j.econlet.2009.03.018

59.

Destek

Aslan

. Renewable and non-renewable energy consumption and economic growth in emerging economies. Renew Energy 2017; 111: 757–763. https://doi.org/10.1016/j.renene.2017.05.008

60.

Harris

Sollis

. Applied time series modelling and forecasting. Wiley, 2003.

61.

Pesaran

Shin

Smith

. Pooled mean group estimation of dynamic heterogeneous panels. J Am Stat Assoc 1999; 94(446): 621–634. https://doi.org/10.1080/01621459.1999.10474156

62.

Tugcu

Ozturk

Aslan

. Renewable and non-renewable energy consumption revisited. Energy Econ 2012; 34(6): 1942–1950. https://doi.org/10.1016/j.eneco.2012.08.021

63.

Hanif

. Impact of fossil fuels energy consumption, energy policies, and urban sprawl on carbon emissions in East Asia and the Pacific. Energy Strategy Rev 2018; 21: 16–24. https://doi.org/10.1016/j.esr.2018.04.006

64.

Sarkodie

Ozturk

. Investigating the environmental Kuznets curve hypothesis in the light of the COVID-19 pandemic. Sci Total Environ 2020; 745: 140948.

65.

Mitić

Munitlak Ivanović

Zdravković

. A dynamic panel analysis of the energy-growth-emissions nexus. Energy Policy 2023; 173: 113385.

66.

Devarajan

Swaroop

Zou

. The composition of public expenditure and economic growth. J Monit Econ 1996; 37(2): 313–344. https://doi.org/10.1016/s0304-3932(96)90039-2

67.

Ghali

Al-Mutawa

. The intertemporal causal relationship between fixed capital formation and economic growth. Int Econ J 1999; 13(2): 77–87.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.68 MB

Decoding the nexus of economic growth,CO 2 emissions,and energy use: Panel evidence from Asia-Pacific countries

Abstract

Keywords

Introduction

Literature review

The dynamic nexus of energy consumption and economic growth

Environmental pollution, FDI, and the EKC hypothesis

The role of capital formation and labor

Research gap and novelty

Regional heterogeneity gap

Methodological gap

Variable-specific gap

Data and methodology

Data and variable rationale

Data sources and harmonization

Variable selection and omitted variable justification

Theoretical framework and model specification

The role of energy and emissions in the production function

The EKC hypothesis and regional heterogeneity

Panel cointegration test

Econometric strategy and model selection

Panel causality test

Empirical results and discussion

Descriptive statistics and regional profile

Multicollinearity test

Homogeneity and cross-sectional independence tests

Panel unit root tests

Panel cointegration test

PMG panel ARDL estimates

The energy-growth-emissions nexus

The EKC hypothesis and turning point analysis

Economic interpretation of control variables

Sectoral decomposition of the energy-emissions nexus

Short-run dynamics

Panel causality test

Sub-group heterogeneity analysis by income group

Renewable versus non-renewable energy dynamics

The FDI “Halo” versus “Haven” paradox

Capital efficiency (GFCF)

Comparative analysis of EKC inflection points

Robustness check: CS-ARDL estimation

Residual diagnostics and model validity

Conclusions and policy implications

Supplemental material

Supplemental Material - Decoding the nexus of economic growth, CO2 emissions, and energy use: Panel evidence from Asia-Pacific countries

Footnotes

Acknowledgements

ORCID iD

Funding

Declaration of conflicting interests

Data Availability Statement

Statement on AI Use

Supplemental material

References

Supplementary Material

Supplemental Material - Decoding the nexus of economic growth, CO₂ emissions, and energy use: Panel evidence from Asia-Pacific countries