The impacts of high-speed railway on environmental sustainability: quasi-experimental evidence from China

China’s rapid economic growth has relied heavily on fossil fuels (especially coal and oil), creating severe resource depletion and carbon emissions challenges. The transport sector is a major emitter, and China’s expanding high-speed railway (HSR) network—now the world’s largest—offers a potentially lower-emission alternative to traditional transport modes. The study asks whether HSR operation significantly contributes to energy savings and emission reductions nationwide and through what mechanisms it promotes ecological sustainability. The paper aims to provide causal evidence on HSR’s environmental impacts using macro city-level and micro enterprise-level data, given mixed prior findings and limited micro evidence on mechanisms and spatial spillovers.

Prior work documents HSR’s transformative role in transport and regional development, with mixed evidence on environmental impacts: some studies find HSR reduces pollution via modal shifts and innovation, while others highlight construction-related emissions and agglomeration-driven increases in carbon output. Gaps include: (1) ambiguous net environmental effects, (2) limited micro-level analysis of enterprises—the main fossil fuel users and emitters, and (3) insufficient assessment of spatial spillovers beyond immediate HSR corridors. The paper develops and tests four mechanisms: H1 labor productivity channel (HSR improves firm efficiency, enabling energy savings); H2 industrial structure channel (HSR facilitates upgrading and reduces reliance on energy-intensive secondary industries); H3 elements flow channel (HSR accelerates flows of capital, labor, and information, optimizing resource allocation and reducing emissions); H4 technological innovation channel (HSR raises accessibility and collaboration, boosting especially green innovation to cut emissions).

Design: Treat HSR openings as a staggered quasi-natural experiment. Use multi-period difference-in-differences (DID) to estimate direct effects, difference-in-difference-in-differences (DDD) for heterogeneous industry effects, and spatial DID (SDID) to capture policy spillovers/conductions via spatial weight matrices decomposed into intra-group and inter-group components. Data: Panel of 285 prefecture-level cities (2003–2020) with HSR opening years collected from China National Railway Group, National Railway Administration, and 12306; city socio-economic data from China Statistical Yearbook and related yearbooks. Micro data: China Industrial Enterprise Database and Pollution Database (2003–2012). Missing values were rechecked/interpolated; some variables log-transformed. Outcome variables: Micro fossil fuel use of industrial enterprises (coal, oil). Macro carbon emissions: total carbon emissions (TCE) and per capita carbon emissions (PCE), computed from city-level natural gas, LPG, and electricity (with coal-fired generation share), then logged. Treatment: HSRit is 1 in and after opening year in treated cities; 0 otherwise. DDD includes interactions with industry types: high-carbon (HCI), technology-intensive, capital-intensive, human-intensive; firm attributes (ownership/property rights), innovation (new product), and city pivot status. Controls: Macro-level—industrial structure (secondary share), infrastructure (highway density), high-end talent, fiscal autonomy, population density (log), FDI. Micro-level—workforce, enterprise age, financing capacity, firm size (log assets), net asset turnover, and net profit margin; fixed effects for enterprise, industry, province, year, and property attributes as applicable. Spatial modeling: SDID incorporates a geographic contiguity matrix, decomposed to identify intra-group and inter-group HSR spillovers. Robustness: parallel trends and dynamic effects, PSM-DID (1:2 matching on covariates), placebo tests (500 random assignments), selection/balance tests (no correlation between initial emissions and opening years; balanced pre-trends), alternative measures/samples, policy-interference controls, and an instrumental variables approach using relief degree of terrain (RDT) interacted with annual hackney carriage counts. First-stage F≈73.6 indicates strong instrument. Heterogeneity: By administrative tier (first/second/third), resource endowment (resource-based vs non-resource; growing/mature/declining/regenerating), and city size (megacity vs non-megacity). Mechanism tests: labor productivity (per capita industrial sales, per capita gross industrial output), industrial transfer/structure (Inov, HCI_inov, HCI_emp; rationalization and advancement indices), elements flow (domestic fixed investment growth, FDI, high-end talent, human agglomeration, digital economy index), and innovation (total, green, and non-green patents).

Micro-level fossil fuel consumption: HSR significantly reduces enterprise coal and oil use. Example coefficients (logs): Coal around -0.51*** and Oil around -1.78*** to -2.17*** in baseline variants; effects are stronger in high-carbon industries and more pronounced for capital- and human-intensive industries. Distance attenuation: a 1% increase in distance to the nearest HSR station raises fossil fuel consumption by at least 0.001%, indicating diminishing HSR benefits with distance. Property rights and pivot-city analyses show larger reductions in state-owned enterprises and in pivot cities. Spatial spillovers in resource consumption (SDID): Both direct HSR and spatial terms are negative and significant; intra-group spillovers exceed inter-group spillovers, indicating reductions in fossil fuel use propagate along and beyond HSR corridors. Macro-level carbon emissions: DID results show HSR reduces emissions. Representative coefficients: TCE -0.020** (full sample with controls often around -0.015, sometimes not significant at 5%); PCE -0.023** to -0.034* depending on specifications and subsamples. Dynamic DID confirms no pre-trends and sustained post-opening reductions. SDID shows significant negative spillovers: spatial HSR terms for TCE/PCE around -0.009*** to -0.017***. Robustness: PSM-DID confirms negative effects on TCE (-0.020**) and PCE (-0.031***). Placebo tests (500 simulations) center near zero while true estimates (-0.015, -0.023) lie in the tails. Selection/balance tests show no relation between initial emissions and HSR opening years and balanced covariates across groups. Alternative measures/samples and policy controls preserve results. IV estimates corroborate large negative effects of HSR on TCE (-0.460***) and PCE (-0.424***), with a strong first-stage (F≈73.6). Mechanisms: Labor productivity rises with HSR (ISV +0.098***; GIV +0.634***), especially in high-carbon industries and in firms with new products and in pivot cities. Industrial transfer and structure upgrading: HSR reduces industrial shares consistent with energy-intensive activity (e.g., Inov -0.042***; HCI_inov -0.035**; HCI_emp -0.031**) and improves rationalization/advancement (RIS +0.012***; AIS +0.067***). Elements flow: HSR increases domestic fixed investment growth (+0.045**), high-end talent (+0.027***), and digital economy index (+0.014***), while slightly reducing FDI (-0.004***); human agglomeration effects are small. Technological innovation: HSR boosts total (+0.179**) and green patents (+0.318***); green innovation associates with lower emissions (e.g., TCE/PCE fall when TIP/GIP increase), whereas non-green innovation relates to higher emissions, indicating the importance of green innovation. Heterogeneity: Emission reductions are strongest in third-tier cities, non-megacities, and resource-based cities (notably mature resource-based). In megacities, HSR is associated with higher PCE, possibly reflecting capacity and agglomeration pressures.

The findings demonstrate that HSR openings causally reduce fossil fuel consumption and carbon emissions, addressing the core research questions by linking HSR expansion to measurable environmental sustainability gains. Reductions are driven by multiple channels—enhanced labor productivity, industrial upgrading and transfer away from energy-intensive sectors, accelerated flows of domestic capital, talent and digital information, and especially green technological innovation. Spatial analyses reveal that benefits extend beyond HSR cities through spillovers and conductions, though stronger within the treated group. The heterogeneity results suggest prioritizing HSR-related interventions in third-tier, non-megacities, and mature resource-based cities to maximize environmental gains, while carefully managing megacity pressures. Overall, the results highlight HSR as a strategic non-policy lever complementing formal climate and energy policies, aiding the transition toward cleaner energy use and lower urban emissions.

This study integrates macro city-level and micro enterprise-level evidence to show that HSR openings deliver significant reductions in fossil fuel use and urban carbon emissions, with immediate and strengthening effects over time. The work advances the literature by jointly analyzing energy consumption and emissions, incorporating spatial spillovers, and detailing mechanisms across productivity, industrial structure, factor flows, and innovation. Policy implications include leveraging HSR’s green attributes, targeting supportive measures that foster green innovation and factor mobility, and tailoring strategies by city type and resource endowment. Future research should refine emissions measurement (e.g., carbon emission efficiency) and integrate route- and city-specific characteristics to further unpack heterogeneity of HSR’s environmental impacts.

Two main limitations are acknowledged: (1) Emissions measurement at the city level relies on TCE and PCE, which may be too coarse; future work should incorporate efficiency-based metrics and broader components of the urban carbon inventory. (2) Greater integration of HSR route characteristics with city attributes is needed to assess heterogeneous effects along specific lines and nodes, enabling more granular policy guidance.