Digital intelligence in e‑commerce cuts CO2 emissions, with bigger gains where post‑2020 dual‑carbon policies are stronger; however, effects differ markedly between China, the US and Germany and rest on data from market‑leading firms and imputed observations.
Digital intelligence for reducing carbon emissions and improving sustainability in E-commerce
Using PVAR and a DiD design on a 2015–2024 city/firm panel, the paper finds that higher digital intelligence adoption in e‑commerce is associated with significant CO2e reductions, with stronger effects under stricter dual‑carbon policies and substantial variation across China, the US and Germany.
E-commerce has become a powerful disruptive force in the international economy; however, it has significant environmental impacts due to its large carbon footprint. This research paper examined how digital intelligence can help achieve sustainability through the lens of the dual carbon strategy. The research studied three countries (China, the United States and Germany), using robust methodology tools (panel vector autoregressive and difference-in-differences) to assess how technology and public policy interventions affect emissions reductions. The study found that digital intelligence significantly reduces carbon dioxide emissions; however, it framed the effects of technology and policy, which vary by country due to differences in energy policy, energy market structure, regulatory frameworks, and implementation challenges. At the same time, digital tools and legal and economic legislation tended to act against each other, with the potential to facilitate and achieve sustainability-related goals and possibilities for both industry and regulators. The study presented a complementary linking theory that connected practice in sustainability and sound reasoning and consideration for future discourse on sustainable e-commerce growth strategy in the dual carbon phase.
Summary
Main Finding
Digital intelligence (AI-enabled digital tools in e-commerce and related digitalization) significantly reduces CO2 emissions, but the magnitude and direction of effects vary across countries. Technology-driven emissions gains interact with legal and economic policy in ways that can be complementary or antagonistic; country-specific energy markets, regulatory frameworks, and implementation capacity shape outcomes.
Key Points
- Scope: Comparative study of China, the United States, and Germany under a "dual carbon" strategy (policies targeting peak-carbon and carbon neutrality).
- Core result: Adoption of digital intelligence correlates with meaningful reductions in carbon dioxide emissions in e-commerce and related sectors.
- Heterogeneity: Effects differ by country due to differences in energy policy, energy-market structure (e.g., grid mix, market liberalization), regulatory design, and practical implementation barriers.
- Policy–technology interaction: Digital tools and legal/economic legislation often interact nonlinearly — sometimes reinforcing each other to accelerate decarbonization, sometimes working at cross-purposes (e.g., regulatory frictions, misaligned incentives).
- Theoretical contribution: The paper proposes a complementary linking theory that connects practical sustainability measures in e-commerce with normative and analytical reasoning to guide future dual-carbon strategies.
Data & Methods
- Geography/sample: Three national cases — China, United States, Germany.
- Empirical methods:
- Panel Vector Autoregressive (panel VAR): used to capture dynamic interdependencies and feedbacks between digitalization indicators and emissions across panels, allowing for endogenous interactions over time.
- Difference-in-Differences (DiD): applied to quasi-experimental policy or adoption events to estimate causal impacts of digital intelligence or regulatory interventions on emissions, controlling for confounders and time trends.
- Robustness: Combination of panel VAR and DiD provides complementary causal and dynamic evidence — VAR for system dynamics and DiD for identification of policy/adoption effects.
- Limitations noted (implicit in methods): cross-country heterogeneity requires careful interpretation; identification depends on plausible exogeneity of treatment timing and comparable counterfactuals.
Implications for AI Economics
- Evidence that AI/digitalization can be a net decarbonization technology strengthens the economic case for supporting targeted AI deployment in logistics, inventory management, routing, energy optimization, and demand forecasting.
- Policy design matters: Economists should emphasize complementary policies (carbon pricing, clean-energy investment, standards) that align incentives so digital adoption translates into real emissions reductions rather than rebound effects.
- Cross-country tailoring: Economic models and policy prescriptions must account for country-specific energy mixes, market structures, and regulatory capacity — one-size-fits-all incentives risk under- or over-estimating impacts.
- Regulatory interactions: Study highlights the need to analyze AI–policy interactions as potential substitutes or complements; economic evaluations should model regulatory frictions and legal constraints explicitly.
- Research directions for AI economics:
- Micro-to-macro linkage: quantify firm-level AI adoption effects and aggregate them accounting for general-equilibrium feedbacks.
- Rebound and leakage: estimate demand-side rebounds and cross-border emissions displacement in platform-mediated trade.
- Welfare and distributional analysis: evaluate who captures the gains from digital decarbonization (consumers, platforms, workers) and how policy can address distributional concerns.
- Measurement and standards: develop standardized emissions accounting for digital services and e-commerce supply chains to improve empirical identification and policy targeting.
- Policy recommendation summary: align digital-technology incentives with clean-energy supply and regulatory reforms; foster industry–regulator collaboration to design interoperable standards and avoid counterproductive rules that dampen digital decarbonization potential.
Assessment
Paper Typequasi_experimental
Evidence Strengthmedium — The paper applies reasonable quasi‑experimental tools (PVAR and DiD) to a multi‑year panel and exploits a clear policy timing (post‑2020), which supports causal interpretation; however, identification depends on a coarse binary treatment, limited convincing tests of parallel trends are reported in the provided text, key variables rely on firm self‑reports and constructed city samples, and substantial imputation is used for missing data—reducing confidence in causal magnitude and external validity.
Methods Rigormedium — The authors employ advanced time‑series/panel methods (PVAR for dynamics and DiD for policy effects), include fixed effects, and describe cross‑validated imputation, but the empirical design raises concerns: treatment definition is a simple pre/post dummy rather than staggered or policy intensity measures; reliance on three market‑leader firms and self‑reported DI adoption and CO2e metrics risks measurement error/selection bias; the paper does not (in the excerpt) present thorough diagnostics for DiD assumptions or alternative identification strategies to address endogenous adoption.
SampleCity‑year panel covering ~150 cities (about 50 per country: China, USA, Germany) for 2015–2024, combining corporate sustainability reports from three large e‑commerce firms (Alibaba, Amazon, Zalando) for firm‑level DI adoption and emissions, UNCTAD E‑Commerce Outlook city data, and national statistics (energy, production); key variables include CO2e (metric tons, logged), DI adoption rate (percent), logistics efficiency (km/order), revenue and operational costs (logged); some 2025 quarter data are imputed using a random forest model with cross‑validation.
Themesadoption governance innovation
IdentificationUses a multi-country city-year panel (2015–2024) with dynamic modelling (panel vector autoregression) to account for lagged relationships and a difference‑in‑differences (DiD) design that treats the post‑2020 period (binary 'dual carbon policy' indicator) as the policy 'treatment'; includes country and year fixed effects and controls (logistics efficiency, revenue, operational costs) and conducts robustness checks and imputation (random forest with cross‑validation) for missing quarter data.
GeneralizabilityOnly three countries (China, US, Germany) — results may not extend to developing economies or other institutional contexts, Heavy reliance on three market leaders (Alibaba, Amazon, Zalando) and their self‑reported metrics limits firm‑level representativeness, City‑level sample may not capture rural or national supply‑chain heterogeneity, Short panel (2015–2024) and coarse binary policy indicator (pre/post‑2020) may miss staggered or intensity variation in policy implementation, Imputation of missing 2025 data introduces additional uncertainty, DI adoption measure aggregates heterogeneous technologies (AI, IoT, big data) which may have different environmental footprints
Claims (7)
| Claim | Direction | Confidence | Outcome | Details |
|---|---|---|---|---|
| E-commerce has significant environmental impacts due to its large carbon footprint. Other | negative | medium | environmental impact / carbon footprint (general) |
0.29
|
| This research examined three countries (China, the United States, and Germany) using panel vector autoregressive (panel VAR) and difference-in-differences (DID) methods to assess how technology and public policy interventions affect emissions reductions. Other | null_result | high | methodological scope / ability to assess emissions reductions |
n=3
0.48
|
| Digital intelligence significantly reduces carbon dioxide emissions. Other | negative | medium | carbon dioxide (CO2) emissions |
n=3
0.29
|
| The effects of technology and policy on emissions vary by country due to differences in energy policy, energy market structure, regulatory frameworks, and implementation challenges. Other | mixed | medium | carbon dioxide (CO2) emissions / emissions reductions (heterogeneous effects) |
n=3
0.29
|
| Digital tools and legal and economic legislation tended to act against each other, though both have potential to facilitate and achieve sustainability-related goals. Other | mixed | low | sustainability-related goals (primarily emissions reductions) |
0.14
|
| The study presents a complementary linking theory that connects sustainability practice and reasoning to inform future discourse on sustainable e-commerce growth strategy in the dual carbon phase. Other | positive | medium | conceptual linkage / theoretical framework for sustainable e-commerce strategy |
0.29
|
| Robust methodology (panel VAR and DID) was used to assess the impact of technology and public policy interventions on emissions reductions. Research Productivity | null_result | medium | methodological robustness in estimating effects on emissions |
0.29
|