Reexamining the impact of foreign direct investment on carbon emissions: does per capita GDP matter?

Global FDI rebounded strongly in 2021 and continues to shape both developed and developing economies. The literature offers conflicting views on whether FDI degrades or improves environmental quality: the pollution haven hypothesis suggests FDI inflows to countries with weak environmental regulations increase emissions, while the pollution halo hypothesis posits technology transfer and better management reduce pollution. Heterogeneity across countries—in income levels, institutions, financial development, and environmental regulation—may underlie these mixed findings. This study asks: (1) Is there a nonlinear relationship between FDI and environmental quality? (2) If nonlinear, is the effect related to income level? (3) Is there heterogeneity across income groups? Using panel data for 67 countries (1990–2019), the paper contributes by modeling nonlinearities and potential structural breaks via threshold models, explicitly incorporating income level as a threshold variable, and linking the two opposing hypotheses through income-dependent effects.

Theoretical perspectives include the pollution haven hypothesis (firms relocate pollution-intensive activities to countries with lax regulation, worsening environmental quality) and the pollution halo hypothesis (FDI brings cleaner technologies and superior management, improving environmental outcomes). Empirical evidence is mixed across regions and income groups, with studies finding both positive and negative effects of FDI on emissions. The environmental impact of FDI can be decomposed into scale, technology, and composition effects, whose relative strengths likely vary by development stage and regulatory stringency. Prior research also indicates that income levels influence environmental regulations and public demand for environmental quality, suggesting that income heterogeneity may mediate the FDI–emissions nexus. Based on this, the study proposes: Hypothesis 1: a nonlinear relationship between FDI and carbon emissions; Hypothesis 2: heterogeneous impacts of FDI when income levels differ; Hypothesis 3: heterogeneity across income-group-specific threshold regressions supporting robustness.

The study first estimates a baseline linear panel model of per capita CO2 emissions (rjCO2) on FDI and controls using FMOLS to address endogeneity and serial correlation. To investigate nonlinearity and structural breaks, it employs Hansen’s panel threshold regression (PTR), with GDP per capita (constant 2010 USD) as the threshold variable splitting the FDI coefficient into regimes. Single and multiple thresholds are tested using bootstrap procedures (300 replications). Models include country fixed effects and time effects. Data cover 67 countries from high-, upper middle-, and lower middle-income groups (World Bank classification), 1990–2019. Variables: rjCO2 (metric tons per capita), FDI (net inflows as % of GDP), rjGDP (GDP per capita), and controls: financial development (FD: domestic credit to private sector % GDP), population size (TP), industrial structure (ES: industry value added % GDP), trade openness (TS: trade % GDP). Series are log-transformed to mitigate heteroskedasticity and autocorrelation; limited missing values are imputed via simple moving average. Panel unit root tests (LLC, IPS, Fisher-ADF, Fisher-PP) indicate stationarity in first differences; Pedroni, Kao, and Johansen-Fisher tests confirm cointegration among variables for the full panel and by income group. Estimation proceeds with FMOLS for long-run elasticities and PTR for nonlinear threshold effects; group-specific threshold models are also estimated for robustness by income group.

- Linear FMOLS (global): EKC confirmed with rjGDP coefficient ≈ 1.4597 (p<0.01) and rjGDP^2 ≈ -0.1832 (p<0.01), indicating inverted-U relationship between income and emissions. FDI elasticity ≈ 0.0229 (p<0.01), implying a small but significant positive association with per capita CO2 globally. FD ≈ 0.1212 (p<0.01) and ES ≈ 0.2303 (p<0.01) positively affect emissions; TS and TP are statistically insignificant. - FMOLS by income group (Table 3): EKC holds across high-, upper middle-, and lower middle-income panels (negative squared income terms). FDI effects are heterogeneous: positive in high- and lower middle-income groups and negative in upper middle-income group; magnitudes are small but significant in several cases (e.g., lower middle FDI ≈ 0.0342, p≈0.0151; high income FDI ≈ 0.0208, p≈0.0091; upper middle FDI ≈ -0.0135, p≈0.0158). FD generally increases emissions; ES effects vary by group. - Threshold effects (global PTR): A double threshold in ln(GDP per capita) is detected (F=42.92, p≈0.063). Estimated thresholds: γ1=2.7339 and γ2=4.6676. Converted to levels, GDP per capita ≈ $541.87 and ≈ $46,515. Regime-specific FDI elasticities of rjCO2: below γ1: 0.1180 (p=0.001); between γ1 and γ2: 0.0122 (ns); above γ2: -0.0887 (p<0.001). Interpretation: FDI significantly raises emissions at very low income levels, the effect is negligible in the middle range, and becomes significantly negative at high income levels. - Spatial-temporal patterns: Most sample countries lie between the two thresholds (about 85% over time). Since 2014, countries in the highest regime include Switzerland, Iceland, Denmark, Sweden, the United States, Singapore, and Australia; in this regime FDI is associated with emission reductions. - Robustness: Income-group-specific single-threshold PTRs confirm heterogeneity: High income threshold ln=4.5027 (~$31,819.99): FDI below threshold 0.0331 (p<0.10), above threshold -0.0290 (ns). Upper middle ln=3.0601 (~$1,148.42): FDI below threshold -0.4355 (p<0.01), above 0.0113 (ns). Lower middle ln=2.7144 (~$518.08): FDI below threshold 0.1135 (p<0.01), above -0.0087 (ns). These reinforce the nonlinearity and income-dependent direction of the FDI–emissions relationship.

The findings address the research questions by demonstrating a clear nonlinear, income-dependent relationship between FDI and carbon emissions. Using GDP per capita as a threshold reconciles conflicting results in the literature: at very low income levels, weak regulations and composition effects dominate, leading to pollution haven outcomes; at high income levels, stricter regulations, higher environmental preferences, and technology spillovers yield pollution halo effects. In the broad middle-income range, the net effect of FDI on emissions is small or insignificant. Thus, the two hypotheses can coexist across development stages rather than being mutually exclusive. The robustness checks by income group substantiate these mechanisms and suggest policy relevance: as countries develop, the net environmental impact of FDI can shift from harmful to beneficial provided regulatory and technological conditions are favorable.

Using FMOLS and panel threshold regression on 67 countries (1990–2019), the study confirms an EKC and uncovers double income thresholds at which the FDI–emissions relationship changes sign. Specifically, when GDP per capita is below about $542, FDI significantly increases emissions; between about $542 and $46,515, the effect is negligible; above about $46,515, FDI significantly reduces emissions. Group-specific models corroborate the heterogeneity by income level, indicating that the pollution haven and halo hypotheses are development-stage contingent. Policy recommendations include: (i) low-income countries should foster economic growth to cross the lower threshold while strengthening environmental regulation to deter pollution-intensive FDI and to attract cleaner technologies; (ii) high-income countries and multinational firms should ensure outward FDI complies with environmental standards to avoid pollution offshoring; (iii) all countries should leverage FDI for green innovation and energy efficiency. Future research should expand samples, periods, and environmental indicators beyond CO2 to validate and generalize these findings.

Data are limited to 67 countries over 1990–2019, constraining sample size and representativeness. Environmental quality is proxied only by CO2 emissions, omitting other pollutants and composite indicators (e.g., ecological footprint). Some missing values are imputed via moving averages, and low-income group data did not pass cointegration tests for certain analyses. Future work should broaden temporal and country coverage and include a wider set of environmental indicators to enhance generalizability.