Digital upgrading fuels corporate green patenting — but only once local finance goes digital; Chinese listed manufacturers' digitalization increases green innovation sharply after regional digital-financial inclusion crosses a threshold, with largest gains for invention patents and firms facing financial constraints.

Main Finding

Digital technology application (firm-level adoption of AI, big data, cloud, platforms) increases corporate green innovation among Chinese listed manufacturers, but the effect exhibits a clear threshold: only after regional digital financial inclusion crosses a critical level does digitalization become strongly complementary to green innovation. Below that threshold digitalization’s positive effect is weak (and may reflect resource substitution/crowding-out), while above it the effect increases substantially. The threshold is especially important for invention-type green patents (respond only at higher inclusion levels), while utility-model green patents benefit at lower inclusion levels. The threshold pattern is stronger for non-state firms, pollution-intensive firms, firms with tight internal cash flow, and firms in less competitive markets.

Key Points

• Hypotheses:
  • H1: Digital technology application positively impacts corporate green innovation.
  • H2: Digital financial inclusion is an institutional turning point that alters the digitalization–green innovation relationship (shifts logic from substitution to strategic complementarity when mature).
• Empirical pattern: positive overall association, but with a discontinuity (threshold) in effect size as digital financial inclusion increases.
• Heterogeneous responses by innovation type:
  • Green invention patents (higher risk, longer horizon) respond only when digital financial inclusion is high.
  • Green utility-model patents (more incremental) respond even when inclusion is lower.
• Stronger threshold effects for:
  • Non-state-owned enterprises (non-SOEs)
  • Pollution-intensive industries
  • Firms with limited internal cash flow
  • Firms in less competitive regional markets
• Proposed mechanisms: digital financial inclusion operates via (1) expanded financial access (relieves resource competition), (2) improved information/credit evaluation (reduces asymmetries), and (3) better risk-sharing (raises tolerance for long-horizon, uncertain green R&D).

Data & Methods

• Sample: Chinese A-share listed manufacturing firms, 2014–2023; final sample ~6,016 firm-year observations (unbalanced panel).
• Dependent variables:
  • Corporate green innovation measured by authorized green patents (WIPO environment-related classification) — transformed as ln(1 + patent count).
  • Decomposed into green invention patents and green utility-model patents (same ln(1 + count) transform).
• Key explanatory variables:
  • Digit: firm-level digital technology application index constructed by text-mining MD&A sections of annual reports (keyword dictionary covering AI, big data, cloud, digital platforms); normalized intensity = ln(1 + keyword frequency) / ln(1 + MD&A word count). Alternative weightings checked for robustness.
  • Digital financial inclusion: regional Digital Inclusive Finance Index (Peking University’s index), matched by firm location and year, rescaled (divided by 100).
• Controls: standard firm-level controls from CSMAR; continuous variables winsorized at 1%/99%.
• Estimation strategy:
  • Fixed-effects regressions to estimate baseline associations.
  • Panel threshold regression with digital financial inclusion as the threshold variable to detect regime shifts (institutional turning point) in the Digit → green innovation relationship.
• Robustness: decomposition by patent type, subsample heterogeneity analyses (ownership, pollution intensity, cash-flow constraints, market competition). (Paper reports additional robustness checks and alternative constructions; full robustness table in supplement.)

Implications for AI Economics

• For theory and models:
  • Digital (AI) adoption effects on innovation are endogenous to institutional financial maturity — standard linear/moderation models may mischaracterize effects. Modeling should allow for nonlinearity and regime thresholds in financing institutions.
  • Resource-allocation logic can switch from substitution to complementarity as financial frictions decline; models of firm investment under uncertainty should incorporate financing-channel thresholds that change comparative statics for AI vs. R&D investments.
  • Heterogeneity matters: firm ownership, industry pollution intensity, internal cash constraints, and market competition systematically shape where thresholds matter most.
• For empirical work:
  • Useful measures: text-mined firm-level digitalization indices and regional digital finance indices capture complementary technology–finance interactions; disaggregate innovation outcomes by risk/time horizon (e.g., invention vs. utility patents).
  • Future causal work should exploit exogenous shocks to digital finance (policy pilots, rollout timing) to strengthen identification.
• For policy and practice:
  • Digital financial infrastructure (digital credit assessment, platform finance, inclusive fintech) can be a lever that unlocks complementarities between AI/digital investments and sustainability R&D—policy that supports digital financial inclusion may amplify green innovation returns to digitalization.
  • Targeting digital finance development may be especially valuable in regions with many non-SOEs, pollution-intensive firms, or financially constrained firms to avoid digital investments crowding out green R&D.
• For firms and investors:
  • Firms in low-digital-finance regions should be cautious that heavy spending on digital transformation could crowd out long-horizon green innovation unless financing access improves.
  • Investors and managers should evaluate regional digital finance maturity when assessing the long-term innovation payoff of AI/digital investments, particularly for exploratory green R&D.

Limitations noted by the paper (and relevant for further AI-economics research): potential endogeneity (reverse causality or omitted regional factors), patent counts as imperfect proxies for innovation output/quality, and the sample is limited to listed Chinese manufacturing firms—external validity beyond this context requires testing.