China’s eco-industrial parks cut manufacturing carbon emissions, firm-level study shows

When China approved its first national eco-industrial demonstration park in the sugar-producing city of Guigang, Guangxi, in 2001, the idea sounded almost paradoxical: cluster factories together, on purpose, and somehow the planet will thank you. Two decades and 93 approved parks later, a new study published in Frontiers in Environmental Science puts that paradox to a rigorous firm-level test. Using carbon emission data from China’s A-share listed manufacturing companies spanning 2007 to 2023, economists Kunyu Yang, Pei Xu and Liangyu Ye of Hunan University of Technology and Business find that the construction of these parks, known as NEIDPs, triggers a statistically significant decline in corporate carbon emissions, and they trace exactly how the reduction happens.

The backdrop is stark. The International Energy Agency’s CO2 Emissions in 2023 report recorded global energy-related emissions rising 1.1 percent, or 410 million tons, to a historic peak of 37.4 billion tons. China has ranked first in total emissions for 18 consecutive years since 2006, and its industrial sector accounts for roughly 70 percent of the national total, with manufacturing firms as the primary emitters within it. Against this backdrop, the country set its dual carbon objectives in 2020, targeting carbon peak by 2030 and carbon neutrality by 2060. The NEIDP program, refined through three distinct policy phases from pilot exploration to standardized development and finally to dual carbon guidance, has become an important policy laboratory for reconciling economic growth with decarbonization.

What makes eco-industrial parks theoretically distinct from other climate instruments is their reliance on physical space as a governance carrier. Market-based tools such as carbon emissions trading schemes work through price signals, while spatial planning policies like low-carbon city programs optimize regional layout at the macro level. Green place-based policies, by contrast, draw on industrial ecology, agglomeration economics and circular economy principles to concentrate specific industries within park boundaries, where firms can share centralized energy supply and wastewater treatment, exchange by-products and waste through industrial symbiosis networks, diffuse green knowledge and technologies, and lower the costs of regulatory enforcement. International precedents, including the Kalundborg Eco-Industrial Park in Denmark and the Burnside Industrial Park in Canada, inspired design principles that the OECD and UNIDO have since refined into international standards.

To move beyond case studies, the researchers built an endogenous growth model featuring a final goods sector that aggregates intermediate products through a constant elasticity of substitution function, and an intermediate goods sector in which representative firms produce with labor, capital, composite energy and technology. Composite energy combines fossil and clean sources through a CES function, and the policy enters the model by lowering firms’ innovation costs and raising the probability of successful innovation. The mathematics yields a clear prediction: higher policy intensity increases the number of technology types within an industry, meaning the policy promotes technological diversification, and the greater the relevance between a firm’s new technology and its existing knowledge base, the stronger this effect becomes.

The model also identifies three energy-related pathways to emission reduction. By promoting clean production technologies, building clean energy infrastructure and providing clean energy subsidies, the policy raises the substitution elasticity between fossil and clean energy, allowing firms to swap high-carbon inputs for low-carbon ones. Technological progress lowers the carbon intensity of each unit of output, and improved energy utilization efficiency means firms consume less energy while maintaining the same production level. The total effect on carbon emissions decomposes into these three channels, providing the testable hypotheses that structure the empirical analysis.

For identification, the team faced a formidable data challenge. Firm-level carbon emissions were obtained from the Listed Firm Carbon Emission Database of the China Research Data Services Platform, which collects disclosure from annual social responsibility, sustainability and environmental reports. Total emissions follow the Greenhouse Gas Protocol, with Scope 1 direct emissions and Scope 2 indirect emissions from purchased electricity and heat as the core indicator, cross-verified against the NDRC’s industry-specific accounting guidelines. Treatment status was determined with unusual geographic precision: the researchers extracted registered and office addresses from the CSMAR database, converted them to longitude and latitude coordinates using the Amap Geocoding API, and overlaid the points in ArcGIS with park boundary vector layers. A firm counted as treated only if both addresses fell within a park’s legal boundaries, ensuring genuine exposure to park policies.

The baseline estimate, from a staggered difference-in-differences specification covering 28,690 firm-year observations, shows a coefficient of minus 0.451 on the policy dummy with firm and year fixed effects, a substantial decline in log emissions. The finding survived an unusually demanding battery of checks. An event study confirmed parallel trends, with no divergence between treated and control firms before park approval and a sharp decline afterward. A placebo test randomly reassigned the policy 500 times, and the true coefficient sat far outside the distribution of random estimates. Propensity score matching with nearest neighbor, radius and kernel methods produced coefficients between minus 0.445 and minus 0.450. Two instrumental variables, a Bartik-style interaction of national Five-Year Plan emission reduction intensity with 2005 city-level pollution abatement investment, and city river density interacted with a time trend, yielded second-stage coefficients of minus 0.448 and minus 0.436, with weak-instrument statistics far exceeding critical values.

The mechanism tests are where the study becomes genuinely revealing. Park construction raised firms’ technological diversification, measured as knowledge variety across patent technology categories, with a coefficient of 0.237, and the Sobel test confirmed that this diversification significantly suppresses emissions. The policy also increased green invention patent applications, classified using the WIPO Green Inventory, improved energy structure through reduced standardized raw coal consumption, and boosted energy efficiency measured as operating revenue per unit of total energy. Technological relevance, the degree of matching between a firm’s current and previous year’s patent classifications, positively moderated the diversification pathway, and a Bootstrap moderated mediation analysis showed the indirect emission reduction effect strengthening monotonically from the 25th through the 90th percentile of relatedness, becoming significant only once firms’ old and new technologies reached a certain threshold of matching.

Heterogeneity analysis sharpened the policy picture further. Among certified high-tech enterprises, the emission reduction coefficient was minus 0.578 and highly significant, while among non-high-tech firms it was minus 0.381 and statistically insignificant, a between-group difference confirmed by Bootstrap testing. Heavily polluting firms, identified across 16 industry categories including coal, mining, textiles, papermaking and metallurgy, showed a stronger effect at minus 0.524 than cleaner firms at minus 0.412, consistent with the theoretical prediction that high fossil energy dependence implies greater marginal emission reduction space, and that park infrastructure sharing gives high-emission firms low-cost access to clean technologies. Firms simultaneously covered by the Innovative City Pilot Policy showed an amplified effect of minus 0.493 versus minus 0.430, evidence of a genuine policy synergy rather than policy crowding.

Perhaps the most consequential finding extends beyond park walls. Using a Spatial Durbin Model with a geographic distance weight matrix treating firms within 100 kilometers as neighbors, the study finds that park construction reduces emissions not only inside parks but also among surrounding manufacturing firms. The direct effect was minus 0.380 and the indirect spillover effect minus 0.071, accounting for roughly 15.7 percent of the total effect, with the result robust across alternative thresholds of 50, 150 and 200 kilometers and an inverse distance specification. A Bacon decomposition showed that clean comparisons between treated and never-treated units carried 87.6 percent of the estimator’s weight, and a Sun-Abraham corrected estimator produced a consistent minus 0.436. For developing countries searching for place-based tools to control emissions, the message is striking: the green transition of a single industrial cluster can radiate outward through knowledge spillovers, technology diffusion and industrial linkages, turning a park into a point that drives an entire region.

Subject of Research: Can green place-based policies reduce carbon emissions in manufacturing firms? Empirical evidence from China's national eco-industrial demonstration parks

Article Title: Can green place-based policies reduce carbon emissions in manufacturing firms? Empirical evidence from China’s national eco-industrial demonstration parks

Article References: Yang, K., Xu, P., & Ye, L. (2026). Can green place-based policies reduce carbon emissions in manufacturing firms? Empirical evidence from China’s national eco-industrial demonstration parks. Frontiers in Environmental Science, 14, Article 1898261. https://doi.org/10.3389/fenvs.2026.1898261

Image Credits: AI Generated

DOI: 10.3389/fenvs.2026.1898261

Keywords: carbon emissions, eco-industrial parks, green place-based policies, manufacturing firms, China, environmental policy, green, place-based, policies, reduce, carbon, emissions