Using insurance data to quantify the multidimensional impacts of warming temperatures on yield risk

The study investigates how warming temperatures affect not only average crop yields but also the risk of yields and revenues falling below expected levels for U.S. corn and soybeans. While prior work shows strong sensitivity of crop yields to climate variables—particularly temperature and precipitation—disentangling direct heat stress from indirect moisture-related effects at large scales remains difficult. Understanding these mechanisms is critical for guiding technological innovation, on-farm decision-making, and climate adaptation strategies. Existing literature largely emphasizes changes in mean yields, but shortfalls below expectations can trigger price spikes, social unrest, and threaten farm viability, with implications for the expanding crop insurance sector. This paper leverages insurance-based measures to quantify how warming changes downside risk, attribute losses to specific causes (heat vs. drought), and assess implications for crop insurance pricing.

Prior research documents nonlinear temperature effects and significant damages to crop yields with warming and highlights roles of precipitation, CO2, and extreme heat. Studies have begun to examine variability and interannual yield risk, noting that production shocks can be large and spatially correlated. However, most observational analyses focus on mean yields and struggle to separate direct heat stress from moisture deficit mechanisms. Work on crop insurance has underscored the potential for warming to increase premium rates and government outlays. This paper adds by using insurance Cause of Loss data to attribute losses to specific perils (e.g., heat, drought, excess moisture, cold) and to quantify how warming affects the distribution of outcomes rather than only the mean.

Data: The authors compile county-year crop insurance data for U.S. corn and soybeans from 1989–2014 from the Risk Management Agency’s Summary of Business (SOB) and Cause of Loss (COL) databases. SOB provides county-level losses and liabilities; COL provides county-level losses by 44 causes. Weather data (daily temperature and precipitation) are from PRISM. Degree days above each 1°C threshold are computed using the sinusoidal method and aggregated over the April–September growing season. The analysis focuses on predominantly dryland counties east of the 100th meridian. Outcome measure: Loss-cost ratio (LCR) = total indemnity payments divided by insured liabilities, computed at county-year level. LCR is also computed for specific causes: heat, cold (sum of cold wet weather, freeze, frost), drought, and excess moisture. Empirical model: Panel regressions relate county-year LCRs to weather covariates with controls for county fixed effects, year fixed effects, and state-specific time trends. Temperature enters via a piecewise linear function of degree-day exposure with three intervals. The first cutoff is fixed at 10°C; two additional cutoffs are selected by grid search to maximize fit (overall LCR cutoffs ~27°C for corn and 28°C for soybeans; higher cutoffs for heat/drought, lower for cold-related). Precipitation enters quadratically. Standard errors are clustered by year. Estimation is conducted in Stata/SE 15. Cause-specific models: Separate regressions are estimated for heat, cold, drought, and excess moisture LCRs using the same weather specification, enabling attribution of warming impacts to direct heat stress versus moisture-related mechanisms. Warming simulations: A uniform +1°C scenario is simulated by increasing daily minimum and maximum temperatures by 1°C and recomputing degree-day exposure. Predicted changes in total and cause-specific LCRs are obtained using estimated coefficients, holding precipitation constant. Robustness and adaptation: The authors test alternative temperature quantifications and insurance product subsets; add controls (e.g., insurance participation patterns, vapor pressure deficit); and estimate models that allow for adaptation over longer horizons: long-differences, cross-section, and 5- and 10-year moving averages. Results are compared on warming impact magnitudes and significance. Spatial heterogeneity: County-level impacts are mapped for total and cause-specific LCRs to assess geographic patterns.

- Warming increases downside yield risk: A uniform +1°C warming increases the aggregate LCR by about 0.038 for corn (≈30% increase over historical average LCR of 0.12; p=0.023) and by 0.011 for soybeans (≈10% over 0.11; statistically insignificant at conventional levels). In the abstract, the increases are summarized as ~32% for corn and ~11% for soybeans. - Heterogeneity: County-level impacts vary widely (corn: −0.044 to +0.28; soybeans: −0.039 to +0.24). Southern counties tend to experience higher losses; northern counties often benefit from warming. - Cause attribution under warming: On a percent basis, +1°C increases drought and heat losses substantially and reduces cold and excess moisture losses: • Corn: drought +92%, heat +105% (p<0.01); excess moisture −43%, cold −256% (p<0.05). • Soybeans: drought +59%, heat +105% (p<0.01); excess moisture −59%, cold −234% (p<0.05). In absolute terms, combined increases from drought and heat exceed combined decreases from excess moisture and cold, yielding a net increase in total losses. - Temperature response: Losses decrease with additional exposure to “beneficial heat” below the first threshold, but increase sharply starting around 29–30°C, intensifying toward 38–40°C (similar thresholds for both crops). - Precipitation effects: A 10% increase in precipitation reduces drought LCR by ~0.008 (corn) and ~0.012 (soybeans), and increases excess moisture LCR by ~0.008 (corn) and ~0.006 (soybeans). Precipitation has no statistically significant effect on heat losses; effects on cold/excess moisture under warmer seasons are less precisely estimated. - Correlations: Degree days >29°C are positively correlated with total, heat, and drought LCRs; precipitation exhibits weak simple correlations except for excess moisture, consistent with nonlinear effects captured in regressions. - Robustness and adaptation: Across alternative specifications and models incorporating longer-run variation (allowing for adaptation), warming-induced risk increases remain: corn +17% to +32%; soybeans +11% to +37%. Additional controls do not improve out-of-sample performance and do not alter conclusions. - Insurance implication: Increased production risk implies higher premium rates and increased costs for crop insurance programs and producers.

Findings indicate that modest warming increases production risk by elevating the frequency and magnitude of losses, especially from drought and heat, while reducing cold and excess moisture losses. Elevated risk affects farm decision-making—potentially reducing input use or shifting toward risk-mitigating inputs—and can synchronize losses across regions due to widespread heat and drought, thereby amplifying aggregate supply shocks. Consequently, warming is expected to lower average food supply and increase intra-seasonal variability. The cause-specific analysis clarifies that drought-related losses contribute more to the increase than heat alone, informing targeted adaptation strategies (e.g., drought tolerance, soil moisture management, irrigation where feasible). Implications for the Federal Crop Insurance Program include higher actuarially fair premium rates and government outlays; cause-specific sensitivities can guide design of new insurance products (e.g., peril-specific or index products) to manage climate-sensitive risks more efficiently. Spatial heterogeneity suggests adaptation should be localized, even within major production regions like the U.S. Corn Belt.

Using large-scale, publicly available insurance data, the study quantifies how warming temperatures increase downside yield risk for U.S. corn and soybeans and disentangles mechanisms via cause-specific losses. A +1°C warming raises overall risk (notably for corn), primarily through increased drought and heat losses, with reductions in cold and excess moisture insufficient to offset the increases. Results are robust across specifications and models allowing for adaptation, implying higher crop insurance premiums and program costs. Policy and technology responses should focus on localized adaptation emphasizing drought and heat resilience, while leveraging insurance design to better target climate-sensitive perils. Future research should exploit monthly loss data to analyze intra-seasonal timing, allow weather effects to vary across growth stages, seasons, space, and time, endogenize the growing season and producer/insurer adaptations within a structural framework, and jointly consider changes in precipitation and atmospheric CO2 alongside warming.

- Cause assignment in insurance claims may not perfectly reflect true causal mechanisms; while this constitutes measurement error in the dependent variable (not biasing coefficients), misperceptions can influence producer behavior and adaptation. - The analysis aggregates weather over a fixed April–September growing season and does not model phenological shifts or stage-specific sensitivities; monthly matching is possible but not implemented here. - Weather effects are assumed homogeneous within the specified piecewise and quadratic forms; spatial/temporal heterogeneity beyond state-specific time trends is not fully modeled. - Warming simulations hold precipitation and CO2 constant; joint changes could alter impacts. - Insurance liabilities and program changes are controlled via fixed effects and trends but not endogenized; structural responses of the insurance program and farmer choices are not explicitly modeled.