Digitalization protects farm productivity only when climate risk is high: Chinese listed agricultural firms see meaningful TFP gains from prior digital investment once city-level climate risk surpasses a threshold, with benefits emerging after two years and working through improved financing, risk-taking, and green innovation.

Main Finding

The productivity returns to agricultural digitalization are climate-risk contingent and nonlinear. Using Chinese A-share agricultural firms (2007–2023), the authors find that two-period-lagged firm digitalization has a negative marginal effect on total factor productivity (TFP) below a city-level climate-physical-risk threshold (~18.96) and a positive effect above it. At mean climate-risk the marginal effect of lagged digitalization on TFP is 0.244 (a 1‑SD rise in digitalization → +0.066 TFP, = 8.06% of TFP SD). At high climate-risk (mean + 1 SD) the marginal effect rises to 0.496 (a 1‑SD rise → +0.134 TFP, = 16.36% of TFP SD). Mechanisms: digitalization alleviates financing constraints, raises firms’ risk‑taking capacity, and promotes green innovation. Digitalization acts as adaptive capital whose productivity value depends on climate-risk intensity and requires ex‑ante accumulation (lagged effect).

Key Points

• Nonlinear, threshold relationship: climate-risk intensity (city CPRI) ≈ 18.96 separates regimes where digitalization’s marginal return to TFP flips from negative to positive.
• Temporal structure: the productivity-protective effect appears for digitalization lagged two periods (ex‑ante preparedness), not contemporaneously.
• Quantitative effects:
  • At mean climate-risk: marginal effect = 0.244; 1 SD increase in digitalization → +0.066 TFP (8.06% of TFP SD).
  • At mean + 1 SD climate-risk: marginal effect = 0.496; 1 SD increase → +0.134 TFP (16.36% of TFP SD).
• Mechanisms validated (mediating roles):
  • Financing constraint alleviation (digital signals/transactions improve credit access under climate stress).
  • Risk-taking capacity enhancement (better monitoring/forecasting reduces uncertainty and supports adaptive investment).
  • Green innovation promotion (digital data/platforms accelerate adoption and diffusion of resource‑saving/low‑carbon tech).
• Heterogeneity: climate-adaptive value of digitalization is higher where human capital, network infrastructure, industry exposure, market institutions, and firms’ baseline productivity are more conducive.

Data & Methods

• Sample: Unbalanced panel of Chinese A‑share listed agricultural firms, 2007–2023.
• Firm digitalization measure: firm‑level indicator constructed via text analysis of annual reports (capturing mentions / investments in digital technologies such as IoT, big data, AI, platforms).
• Climate-risk measure: city-level Climate Physical Risk Index (CPRI) capturing extreme low/high temperature, extreme rainfall, and extreme drought; each firm matched to its city-year CPRI.
• Outcome: firm-level agricultural total factor productivity (TFP).
• Empirical strategy (as reported):
  • Nonlinear/threshold analysis linking lagged digitalization to TFP across climate-risk regimes (threshold ≈ 18.96).
  • Use of two-period lags to capture ex‑ante accumulation effects.
  • Mediation analyses to test financing constraints, risk‑taking capacity, and green innovation pathways.
  • Heterogeneity checks across human capital, infrastructure, industry exposure, institutions, and initial productivity.
• Controls and standard panel specifications (fixed effects, robustness checks) are used to isolate effects (paper details full econometric implementation).

Implications for AI Economics

• Valuation of AI/digital investments must account for environmental context: returns to AI in agriculture are state‑dependent and can be substantially higher in high climate‑risk settings.
• Timing and absorptive capacity matter: AI/digital investments show lagged returns (two-period accumulation). Economic models and ROI calculations should include implementation/learning lags and complementarities (human capital, infrastructure).
• Policy targeting: Subsidies, credit support, and public digital infrastructure yield larger social returns where climate physical risk is high and where firms lack complementary assets. Blanket promotion of digitalization may produce weak or negative short‑term returns in low‑risk areas.
• Finance and risk pricing: Digitalization reduces information asymmetry and can improve access to credit under climate stress; lenders and insurers should incorporate firm digital signals into contract design and climate‑contingent pricing.
• Innovation and climate co‑benefits: AI/digital tools not only boost operational efficiency but also speed green-tech development and diffusion—important for models of the twin transition (digital + green).
• Modeling recommendations for AI economists:
  • Incorporate nonlinear threshold effects of external shocks (e.g., climate risk) when estimating marginal productivity of AI.
  • Explicitly model slow-moving digital capital and interaction terms with environmental risk.
  • Consider heterogeneous returns across regions/firms driven by complementary assets and institutions.
• Research gaps suggested: disentangle specific AI technologies (ML/vision/forecasting) from broader “digitalization”; micro‑level causal evidence for each mechanism; generalizability outside Chinese listed firms and to smallholders.