Digital financial inclusion and grain production stability: dimensional decomposition and heterogeneous associations in Chinese cities

Background Grain production stability is fundamental to food security and to Sustainable Development Goal 2 (SDG 2: Zero Hunger), yet existing studies have not isolated the separate relationships between different digital finance dimensions and agricultural output volatility. Aggregate digital finance indices may mask opposing patterns across the dimensions of financial inclusion. Methods Using a city-level panel of 338 Chinese cities from 2011 to 2023, we decompose the Peking University Digital Financial Inclusion Index into three dimensions (coverage breadth, usage depth, and digitization level) and estimate their separate associations with grain output volatility (3-year rolling coefficient of variation) using two-way fixed effects models with weather controls (mean temperature, temperature variability, relative humidity, and wind speed). We probe robustness with Bartik and province-level leave-one-out mean instrumental variables, wild-cluster bootstrap inference, non-overlapping volatility windows, and an acreage–yield decomposition of the dependent variable, and we examine the descriptive correlation between digital finance, agricultural insurance premiums, and grain volatility. Heterogeneity is examined across ten city-level dimensions and disciplined by formal interaction tests. Results A one-within-city-standard-deviation increase in the digitization level is associated with approximately 10% lower grain output volatility relative to the sample mean ( p < 0.001; province-level wild-cluster bootstrap p = 0.012); the overall standard deviation yields a larger benchmark (29%) that overstates the identifying variation because fixed effects absorb most cross-city dispersion. The digitization sign is consistent under instrumental variable estimation and in non-overlapping windows, though it loses significance in spatial specifications that absorb provincial grain-volatility dependence. Coverage breadth is associated with higher volatility ( p < 0.01 under city-level clustering; p = 0.040 under province-level clustering), an association concentrated in yield volatility, though structural-transformation confounds operating through yield cannot be ruled out. Usage depth is insignificant contemporaneously but positive with a one-period lag. The officially weighted aggregate index is itself insignificant ( p = 0.70), as the opposing coverage and digitization associations offset each other. Agricultural insurance premiums are positively, not negatively, correlated with grain volatility, so insurance does not account for the digitization association. Formal interaction tests indicate that the negative digitization association is concentrated in cities with strong digital infrastructure and service-oriented economies; the remaining subgroup contrasts are exploratory. Conclusion Digital financial inclusion has dimension-specific (and sometimes opposing) conditional associations with grain production stability. The evidence suggests that the digitization dimension, rather than undifferentiated geographic expansion, carries the most consistent negative association with grain volatility in our city-level data, though the mechanisms remain conjectural and the associations do not survive spatial specifications.