Calibrating sub-county yields from insurance transactions

Introduction

Farm-level program evaluation requires farm-level yields, yet no public data source reports them: official statistics stop at the county, and survey panels are too sparse to support the within-county analysis that risk-management questions demand. This article documents the second stage of the FCIP data system, which recovers sub-county yield realizations for every aggregated FCIP transaction by exploiting the intertemporal structure of the program’s rating rules, following the methodology developed and validated in Tsiboe, Turner, and Yu (2025).

The economic logic is that of a state-space inversion. A transaction’s premium rate in year t{+}1 is a deterministic function of its production history through year t; the year-t yield realization is a state variable of that history. Conditional on the rating primitives recovered in the preparation stage (see the companion article Preparing FCIP transactions for yield calibration), the year-t yield is therefore identified, up to the history configuration, by the requirement that it reproduce the rate observed in year t{+}1.

Notation

Let i index aggregated transactions and j(i) their insurance pools; \theta_i is the coverage level, y^{a}_{i} the approved yield, \tilde{y}_{i} the indemnified yield (the per-acre shortfall implied by the observed indemnity), I_i the indemnity, and r^{obs}_{i,t+1} the base premium rate observed for the matching transaction in the subsequent crop year. Let f_{t+1}(\cdot) denote the rating function applicable in year t{+}1 and H\!\left(y;\, T_0, T_1\right) the production-history operator that inserts a year-t yield realization y into a history spanning T_1 years, of which the most recent T_0 are actual (as opposed to transitional) records.

The calibration problem

For each transaction matched across adjacent years, the procedure solves the constrained minimum-distance problem

\left(\hat{T}_0, \hat{T}_1, \Delta\hat{y}\right) = \arg\min_{T_0,\, T_1,\, \Delta y} \left[\, f_{t+1}\!\Big( H\!\big(y^{c}_{i}(\Delta y);\, T_0, T_1\big) \Big) - r^{obs}_{i,t+1} \,\right]^{2}

\text{subject to}\quad T_0 \in [2, 10],\quad T_1 \in [3, 10],\quad T_1 \geq T_0,\quad \Delta y \geq 0,

where the calibrated yield is defined piecewise by the loss experience:

y^{c}_{i} = \begin{cases} \theta_i\, y^{a}_{i} - \tilde{y}_{i}, & I_i > 0 \quad \text{(the guarantee less the indemnified shortfall)},\\[4pt] \theta_i\, y^{a}_{i} + \Delta\hat{y}, & I_i = 0 \quad \text{(the guarantee plus a nonnegative optimized offset)}. \end{cases}

For indemnified transactions the yield is point-identified by the indemnity itself: the payment reveals exactly how far the realization fell below the guarantee. For non-indemnified transactions the indemnity reveals only the censoring event y^{c}_{i} \geq \theta_i y^{a}_{i}, and the offset \Delta\hat{y} resolves the censoring by selecting the realization most consistent with the next year’s observed rate, in effect using the rating system’s own memory as the identifying instrument.

Benchmarking to official statistics

Because the non-indemnified branch is identified only through the rating function, its level is disciplined against official yield histories. Let \bar{y}^{off}_{g} denote the official mean yield for benchmark group g and \bar{y}^{c}_{g} the corresponding mean of calibrated yields. Candidate adjusted yields are formed by the proportional rescaling

y^{adj}_{i,g} = y^{c}_{i} \cdot \frac{\bar{y}^{off}_{g}}{\bar{y}^{c}_{g}}, \qquad y^{adj}_{i,g} \geq \theta_i\, y^{a}_{i},

where the floor preserves the censoring constraint, at three nested scopes: the insurance pool, the state–commodity–type–practice group, and the commodity–type–practice group. The final calibrated yield coalesces these candidates in order of scope specificity (pool, then state, then commodity), falling back to the unadjusted value where no benchmark exists. Transactions are then aggregated to producer-level contracts, each carrying a simulation weight (its share of insured exposure) and a revealed budget (the producer-paid premium implied by the observed election).

Validation and use

The procedure is validated in Tsiboe, Turner, and Yu (2025) against 148,243 farm-level observations from the Kansas Farm Management Association: dryland and irrigated corn, sorghum, soybean, and winter wheat records for which actual yields are known. Regressing calibrated on actual yields produces a coefficient of 0.87 with an R^2 of 0.88, indicating that the calibration closely tracks realized yield variation. When calibrated yields serve as the dependent variable in regression analysis, estimated coefficients are statistically indistinguishable from those obtained with actual yields; when used as an independent variable, attenuation consistent with classical measurement error appears, so errors-in-variables corrections are advisable in that design. The original application detects asymmetric information in unit-structure elections: disaggregated units, whose field-level deductibles raise indemnification probability, exhibit systematically lower yields than aggregated ones. It further shows that the 2008 Farm Bill’s subsidy changes for aggregated structures attenuated the effect.

Beyond the validation study, the calibrated yields underpin the risk-reduction measurement of Tsiboe et al. (2025), the program-combination analysis of Gaku and Tsiboe (2025), the prospective contract-design analysis of Tsiboe and Turner (2025), and the supplemental coverage assessment of Tsiboe, Biram, and Hagerman (2026), which applies the calibration and simulation machinery to over one million sub-county observations.

Limitations

The most recent crop year is calibrated against the latest available subsequent-year records and is therefore provisional until that year’s experience is complete. The procedure recovers first moments of pool-by-contract yields; higher moments within a transaction remain unidentified, and the proportional benchmarking preserves relative, not absolute, dispersion. Users should treat the five reported yield concepts (rate, approved, average, adjusted, and calibrated) as distinct economic objects; in the current vintage the average and adjusted series coincide with the rate and approved series respectively, but the schema keeps them separate so future refinements do not break consumers.

Data availability

One file per crop year from 2011 onward (calibrated_yield_<year>.rds):

piggyback::pb_download(
  file = "calibrated_yield_2022.rds", dest = tempdir(),
  repo = "ftsiboe/USFarmSafetyNetLab", tag = "calibrated_yield")

Recommended citation

Tsiboe, F. (2026). Calibrating sub-county yields from insurance transactions. In FCIP calibrated and synthetic data catalogue. https://ftsiboe.github.io/rfcipCalibrate/articles/calibrated-yield.html

Data users should additionally cite Tsiboe, Turner, and Yu (2025).

Disclaimer

This product uses data provided by USDA/RMA but is neither endorsed by nor affiliated with USDA or the U.S. Government.

References

Gaku, S., & Tsiboe, F. (2025). Evaluation of alternative farm safety net program combination strategies. Agricultural Finance Review, 85(2), 254–273. https://doi.org/10.1108/AFR-11-2023-0150

Tsiboe, F., & Turner, D. (2025). Incorporating buy-up price loss coverage into the United States farm safety net. Applied Economic Perspectives and Policy, 47. https://doi.org/10.1002/aepp.13536

Tsiboe, F., Turner, D., & Yu, J. (2025). Utilizing large-scale insurance data sets to calibrate sub-county level crop yields. Journal of Risk and Insurance, 92(1), 139–165. https://doi.org/10.1111/jori.12494

Tsiboe, F., Turner, D., Williams, B., Miller, M., Baldwin, K., & Dohlman, E. (2025). Risk reduction impacts of crop insurance in the United States. Applied Economic Perspectives and Policy, 47(5), 1832–1847. https://doi.org/10.1002/aepp.13513

Tsiboe, F., Biram, H., & Hagerman, A. (2026). Low participation and untapped benefits of supplemental crop insurance in the United States (working paper). Agricultural Risk Policy Center, North Dakota State University.