Transitioning to low-carbon agriculture: the non-linear role of digital inclusive finance in China's agricultural carbon emissions

The paper addresses how digital inclusive finance affects agricultural carbon emissions in China, a critical issue given agriculture’s sizable contribution to national and global greenhouse gas emissions. While China targets peaking carbon emissions before 2030 and achieving carbon neutrality by 2060, agriculture accounts for roughly 16–17% of national emissions and could reach 30% by 2050. Finance influences agricultural development and emissions, yet traditional finance often leads to uneven credit allocation and limited support for low-carbon innovation. Digital inclusive finance, leveraging digital technologies to broaden access and reduce costs, may reshape resource allocation, technology adoption, and production practices in agriculture. The research questions are whether digital inclusive finance reduces agricultural carbon emissions, whether its impact is nonlinear, and how rural organizational forms (agricultural green cooperatives) and human capital influence this relationship. The study’s importance lies in informing policies to harness digital finance for low-carbon agricultural transformation.

Prior studies show mixed effects of financial development on emissions: some find finance raises emissions via scale effects and energy use; others find finance can reduce emissions through technological progress and improved resource allocation. A growing literature on digital finance indicates it can curb emissions by spurring innovation, optimizing industrial structure, and improving factor allocation, with spatial spillovers and regional heterogeneity reported. In agriculture, digital finance may facilitate green technology adoption by easing credit constraints, improving information access, and strengthening social trust. Evidence on nonlinearity suggests an inverted U-shaped relation where early-stage financial development increases emissions before later reductions as systems mature. However, most work emphasizes individual farmers and overlooks rural business entities (e.g., green cooperatives). The paper posits four hypotheses: H1 digital inclusive finance suppresses agricultural carbon emissions; H2 the impact is nonlinear; H3 digital finance fosters green cooperatives that in turn suppress emissions; H4 a threshold effect exists linked to human capital accumulation.

Study area: 30 provincial-level regions in Mainland China (excluding Tibet; Hong Kong, Macau, and Taiwan excluded due to data) from 2011–2021, capturing diverse resource endowments and agricultural structures. Agricultural carbon emissions accounting: Emission coefficient method per IPCC 2006 guidelines, focusing on crop production (narrow agriculture). Emission sources include: fertilizers, pesticides, agricultural films, diesel use by machinery, electricity for irrigation, and soil tillage-induced organic carbon loss. Total emissions C = sum over i of Qi × ei, where Qi is input quantity and ei the emission coefficient. Emission coefficients (examples): diesel 0.59 kg/kg; fertilizer 0.89 kg/kg; pesticides 4.93 kg/kg; films 5.18 kg/kg; irrigation 266.48 kg/hm²; tillage 312.60 k/km². Data for inputs from China Rural Statistical Yearbook, China Statistical Yearbook, and National Bureau of Statistics (2011–2022 editions). Relevant variables are log-transformed to mitigate heteroscedasticity. Econometric models: - Baseline two-way fixed effects (province and year) model to estimate effects of digital inclusive finance (dif) on log agricultural carbon emissions (lnC), with and without the squared term to test nonlinearity: lnC_it = α0 + α1 dif_it + α2 Controls_it + μ_i + γ_t + ε_it lnC_it = β0 + β1 dif_it + β2 dif_it^2 + β3 Controls_it + μ_i + γ_t + ε_it Hausman tests favor fixed effects. - Moderated effects model: tests whether agricultural green cooperatives (gc) moderate the impact of dif on emissions by including interaction terms (dif×gc and, in the nonlinear model, dif^2×gc): lnC = B + B1 dif + B2 gc + B3 (dif×gc) + Controls + FE + ε and with squared term and interactions. - Threshold regression model: panel threshold model using rural human capital accumulation (hc) as threshold variable to detect regime-dependent effects of dif on lnC. The number of thresholds is determined via bootstrap (self-sampling) tests, then coefficients are estimated within regimes defined by hc thresholds. Variables: - Explained: agricultural carbon emissions (ace; log of computed C). - Core explanatory: digital inclusive finance index (dif) from Peking University Digital Finance Research Center (built from Ant Financial/Alipay data), and its three dimensions: coverage breadth (cb), usage depth (ud), digitization level (dl). - Moderator: number of actively operating agricultural green cooperatives per province-year (from CCAD, Zhejiang University). - Controls: urbanization rate (ur), degree of openness (od), transportation infrastructure (ti; log highway mileage), disaster-affected crop area proportion (ac). Some missing values interpolated. Robustness and endogeneity: - Dimension-specific estimations using cb, ud, dl to test heterogeneity. - 1% two-sided trimming to address outliers. - Difference GMM to address potential endogeneity from dynamics and omitted variables; AR tests and Hansen test reported as passed. Descriptive statistics presented for all variables; sample size N=330 (30 provinces × 11 years).

- National emissions trend: Agricultural carbon emissions exhibit an inverted U-shape from 2011 to 2021, peaking in 2015 and then declining, coinciding with Chinese policy initiatives targeting low-carbon agriculture. - Spatial pattern: Higher emissions in eastern and northern regions, lower in the west and south; over time, east–west gaps narrowed while north–south gaps widened. Provinces like Shandong and Henan remain high emitters; Xinjiang, Inner Mongolia, and Jilin increased; Guangxi decreased. - Baseline regression (two-way FE): Without nonlinearity, dif significantly increases ace. With squared term, dif coefficient positive and dif^2 negative (both significant at 5%), confirming an inverted U-shaped relationship. Reported goodness of fit R^2 ≈ 0.9965–0.9966. Turning point estimated at dif = 3.7596. Below 3.7596, digital finance development raises emissions; above it, it suppresses emissions. - Controls: Urbanization (ur) positively associated with ace; degree of openness (od) weakly positive; transport infrastructure (ti) insignificant; disaster variable (ac) positive and significant. - Robustness: Dimension checks show usage depth (ud) and digitization level (dl) each display inverted U-shaped impacts on ace (linear term positive, squared term negative), while coverage breadth (cb) does not show a carbon-reduction pattern comparable to ud and dl. 1% trimming preserves inverted U results. Difference GMM confirms inverted U with significant positive dif and negative dif^2; AR and Hansen tests acceptable. - Moderation by green cooperatives: Interactions indicate that agricultural green cooperatives weaken the main inverted U-shaped effect of dif on ace. Specifically, dif×gc is significantly negative in linear interaction, while dif^2×gc is significantly positive in the nonlinear model, implying the curve flattens around the inflection, reducing digital finance’s emission-reduction potency as cooperatives strengthen. This rejects the initial hypothesis that cooperatives would enhance the emission-reduction effect of dif. - Threshold effects of human capital: Bootstrap tests support double thresholds for hc at approximately 2.313 and 2.443. Estimated dif effects by regime: hc ≤ 2.313, coef 0.0164 (not significant); 2.313 < hc ≤ 2.443, coef −0.0471 (significant); hc > 2.443, coef −0.1493 (significant). Thus, as human capital accumulates, the negative effect of dif on ace strengthens. - Mechanism interpretation: Early-stage dif expansion amplifies scale effects and energy use, raising emissions; as dif deepens and digitization advances, technology adoption, precision agriculture, and optimized resource allocation dominate, reducing emissions. - Quantitative highlights: - Turning point dif ≈ 3.7596. - Thresholds for hc: 2.313 and 2.443. - Selected R^2 values: 0.9965–0.9968 in baseline; threshold model R^2 ≈ 0.5110 (parsimonious form).

The findings show that digital inclusive finance has a stage-dependent impact on agricultural emissions: initial development increases emissions via scale and input intensification, but further maturation and digitization enable technological progress, precision input use, and efficient resource allocation, ultimately reducing emissions. This resolves the study’s central question by confirming a nonlinear inverted U-shaped relationship and quantifying the turning point. The heterogeneity across digital finance dimensions underscores that usage depth and digitization are critical for achieving emission reductions, while mere expansion of coverage breadth is insufficient. The moderating analysis indicates a current decoupling between green cooperatives and digital finance: instead of synergizing, stronger cooperatives appear to flatten digital finance’s impact curve, possibly due to path dependence and suboptimal allocation of new digital funds. Policy must therefore target better alignment between cooperative practices and digital financial products to unlock synergies. The threshold effects of human capital highlight the pivotal role of education in translating digital financial access into effective low-carbon practices; higher human capital magnifies the emission-reduction benefits, suggesting complementary investments in education and training are essential. Overall, the results are relevant for designing integrated financial, educational, and organizational policies to accelerate low-carbon agricultural transitions.

This study measures agricultural carbon emissions from crop production inputs and empirically examines the impact of digital inclusive finance across 30 Chinese provinces (2011–2021). Main contributions: (1) It identifies a significant inverted U-shaped relationship between digital inclusive finance and agricultural carbon emissions and quantifies the turning point. (2) It shows that usage depth and digitization are the principal channels through which digital finance reduces emissions, whereas coverage breadth alone is insufficient. (3) It uncovers a weakening moderating effect from agricultural green cooperatives, indicating a decoupling that calls for better coordination. (4) It demonstrates pronounced threshold effects of human capital, with higher education levels strengthening the emission-reduction effect of digital finance. Policy recommendations include: targeting emission reductions in high-emission provinces while safeguarding food security; improving rural digital financial infrastructure and penetration; tailoring financial products to the operational realities of green cooperatives and establishing feedback mechanisms; and investing in farmer education and training to enhance technology adoption and leverage human capital spillovers. Future research could extend the time horizon, conduct regional sub-sample analyses (east, central, west), compare cross-country settings, and further probe mechanisms linking digital finance and cooperative development to achieve coordinated, low-carbon agricultural growth.

- Temporal and spatial scope: The sample spans 11 years and covers 30 provinces, limiting long-run inference and regional heterogeneity analysis; future work should extend the period and analyze sub-regions. - External validity: Results are derived from China; cross-country comparisons are needed to assess generalizability. - Mechanism depth: The interaction between digital inclusive finance and agricultural green cooperatives requires deeper investigation to understand coordination pathways and performance heterogeneity. - Measurement scope: Emissions are calculated for crop production inputs (narrow agriculture); broader agricultural activities and carbon sinks could further refine estimates.