Higher AI intensity in the Guangdong–Hong Kong–Macao Greater Bay Area is linked to better green economic efficiency, largely because AI encourages more rational industrial structures; the positive association is stronger in economically dependent subregions.

Main Finding

AI development significantly improves green economic efficiency (GEE) in the Guangdong–Hong Kong–Macao Greater Bay Area (GBA). This positive effect is strengthened by industrial-structure rationalisation (measured by a Theil-based index) and is larger in contexts of greater regional economic dependence (i.e., stronger economic linkages and technology diffusion across cities).

Key Points

• Primary result: AI adoption has a statistically significant positive effect on GEE across 11 GBA cities (2011–2021).
• Mediation: Industrial-structure rationalisation (TI, a log Theil index) acts as a mediator that strengthens the AI → GEE relationship. The paper reports a clear mediating role for industrial-structure rationalisation; the tertiary/secondary ratio (TS) is considered as an alternative mediator, but the abstract and findings emphasize TI as the effective mediating channel.
• Moderation: Regional economic dependence moderates the AI → GEE link. Higher economic interdependence (measured using intercity GDP, population and inverse distance weights) amplifies the positive impact of AI on GEE, consistent with greater technology diffusion and spillovers in integrated regions.
• Controls: The analysis controls for city-size (population density), openness (trade-to-GDP), education (secondary student–teacher ratio), financial development (deposits & loans), fiscal expenditure (fiscal spending/GDP), and an industrial development measure.
• Policy framing: Authors argue AI can improve resource allocation and lower environmental impact, and recommend targeted industrial/AI layouts and regional cooperation in the GBA to raise GEE.

Data & Methods

• Spatial and time coverage: Panel of 11 cities in the Guangdong–Hong Kong–Macao Greater Bay Area, 2011–2021.
• Dependent variable (GEE): Computed via a super-efficiency Slack-Based Measure (SBM) Data Envelopment Analysis that includes undesirable outputs. Inputs: labour (employees), capital (fixed-asset capital stock via perpetual-inventory method), energy (electricity consumption). Desired output: regional GDP. Undesirable output: industrial SO2 emissions.
• Mediators:
  • TS (industrial structure advancement): log(tertiary industry output / secondary industry output).
  • TI (industrial structure rationalisation): log of Theil index calculated from industry outputs and employment shares.
• Moderator (economic dependence): Intercity economic dependence coefficient Rijt built from city GDPs, populations and intercity distances; aggregated to get city-level and region-level dependence measures.
• Estimation approach: Multiple regression framework with mediation and moderation analyses to test (1) AI → GEE, (2) AI → industrial-structure measures → GEE, and (3) interaction of AI with economic dependence. Several control variables included to reduce omitted-variable bias. (Paper uses panel regressions with SBM scores as dependent variable; specifics such as fixed effects, clustering, or IVs are not detailed in the provided excerpt.)
• Period and variables sourced from regional statistics for the 11 GBA cities (2011–2021).

Implications for AI Economics

• AI as a green productivity lever: The study provides empirical evidence that AI adoption is not only a productivity engine but also a mechanism for improving environmental efficiency when combined with the right industrial configuration.
• Role of industrial structure: Gains from AI for environmental outcomes operate partly through structural change — specifically better rationalisation (more efficient sectoral composition), implying that technology policy and industrial policy should be coordinated.
• Importance of regional integration and spillovers: Economic interdependence amplifies AI’s environmental benefits. AI-related investments and policies will generate larger GEE payoffs in well-connected, economically integrated regions via faster diffusion and collaborative effects.
• Policy design: To maximize AI’s green dividends, policymakers should (a) support AI deployment in ways that enable sectoral resource reallocation (promote TI), (b) strengthen intercity linkages and knowledge diffusion, and (c) combine AI investments with fiscal, financial and education policies that enable uptake.
• Research suggestions and caveats:
  • Measurement & causality: Results rely on SBM efficiency scores and observational panel regressions. Causal interpretation would be strengthened by instruments, quasi-experimental variation, or firm-level/plant-level microdata.
  • Heterogeneity: Future work should examine sectoral heterogeneity (which industries benefit most), other pollutants, longer horizons, and cross-region comparisons.
  • External validity: Findings apply to the integrated, high-tech GBA context; effects may differ in less connected or lower-income regions.

If you want, I can (a) extract or summarize the paper’s regression tables and coefficients if you provide them, (b) prepare a short policy brief for GBA planners, or (c) map potential follow-up research designs to establish causality.