Preparing FCIP transactions for yield calibration

Introduction

Because producer-level FCIP records are exempt from public disclosure, empirical work with the program’s loss data has historically been confined to what indemnities reveal. Indemnities are censored, observed only when yield declines exceed the elected deductible, which frequently exceeds 25 percent (Tsiboe, Turner, & Yu, 2025). Analyses built on loss-cost ratios therefore identify behavior over a specific range of downside outcomes while the rest of the yield distribution remains dark. The FCIP’s rating system, however, embeds more information than its loss records release: premium rates are deterministic functions of reported production history, so the rates themselves carry the yield signal.

This article documents the first stage of the FCIP data system: the transformation of raw administrative experience records into standardized, yield-basis transaction records augmented with the rating primitives needed for yield calibration (Tsiboe, Turner, & Yu, 2025). Two operations define the stage. Revenue-based policies are first converted to yield-equivalent terms, placing all continuously rated plans in a common metric; the RMA rating function is then inverted to recover each transaction’s rate yield, the production-history summary that generated its observed premium rate.

Unit of observation

Let i index an aggregated transaction: the collection of insured units that share an insurance pool j(i) (county, commodity, type, and practice), a contract design (insurance plan, coverage level \theta_i \in [0.50, 0.85], and unit structure), and a crop year t. Because rating in the FCIP is a deterministic function of pool and contract characteristics, all producers within a transaction faced identical premium schedules and policy menus; the transaction is therefore interpreted as a representative producer, and the collection supports inference at that level rather than at the level of individual farms. Retained records are buy-up business (coverage type “A”) in the continuously rated plans: Yield Protection (YP), Revenue Protection (RP), Revenue Protection with Harvest Price Exclusion (RP-HPE), and Actual Production History (APH).

For each transaction the administrative source reports insured acreage a_i, liability L_i, total premium P_i, premium subsidy S_i, and indemnity I_i, along with the projected price p^{p}_{jt} and, where applicable, the harvest price p^{h}_{jt} established for the pool under RMA price-discovery rules.

Yield equivalence

Revenue plans insure the product of yield and price, so their dollar magnitudes are not directly comparable to those of yield plans. Comparability is restored by deflating each revenue policy by its guarantee-relevant price. Define

\tilde{p}_{i} = \begin{cases} \max\left(p^{p}_{jt},\, p^{h}_{jt}\right) & \text{RP, whose guarantee is re-established at the higher of the two prices},\\[4pt] p^{p}_{jt} & \text{RP-HPE, YP, and APH.} \end{cases}

The yield-basis liability and the indemnified yield (the per-acre yield shortfall implied by the observed indemnity) are then

L^{y}_{i} = \frac{L_{i}}{\tilde{p}_{i}}, \qquad \tilde{y}_{i} = \frac{I_{i}}{\tilde{p}_{i}\, a_{i}},

with indemnities recomputed under a zero floor so that negative implied shortfalls (possible when harvest prices exceed projected prices under RP) do not enter. Missing harvest prices for yield plans are backfilled with the projected price, which is innocuous because p^{h} does not enter the guarantee for those plans. The approved yield implied by the coverage election follows from the identity defining the yield guarantee:

y^{a}_{i} = \frac{L^{y}_{i}}{\theta_{i}\, a_{i}},

where \theta_i is the coverage level, so that \theta_i y^{a}_i is the per-acre guaranteed yield.

Recovering the rate yield

FCIP premium rates for the continuously rated plans are generated by a published rating function f(\cdot) mapping a contract’s rate yield y^{r}_{i} (a summary of its production history) and a vector of pool-level rating parameters into the base premium rate. Writing the observed base rate as r^{obs}_{i} and the pool parameters (reference rates, exponents, and fixed loadings drawn from the actuarial records) as (\boldsymbol{\alpha}_{j},\boldsymbol{\delta}_{j}), the rate yield is recovered as the solution to

\hat{y}^{r}_{i} = \arg\min_{y^{r}} \left[\, f\!\left(y^{r};\, \boldsymbol{\alpha}_{j}, \boldsymbol{\delta}_{j}\right) - r^{obs}_{i} \,\right]^{2}.

Because f is piecewise-smooth but not globally monotone in all parameter regimes, the program solves the inversion by differential evolution, a global stochastic optimizer, rather than by local root-finding. Transactions for which no rate yield reproduces the observed rate within tolerance are flagged rather than forced, so downstream users can condition on inversion quality. The recovered \hat{y}^{r}_{i}, the implied approved yield y^{a}_{i}, the adjusted base rate, and the components of (\boldsymbol{\alpha}_{j},\boldsymbol{\delta}_{j}) constitute the rating primitives carried into the calibration stage.

Economic interpretation

The inversion embodies a revealed-information argument: under a known pricing rule, prices reveal the private information on which they were conditioned. Its validity rests on the deterministic character of FCIP rating (there is no underwriting discretion in the continuously rated plans) and on the completeness of the published actuarial records. The argument parallels the recovery of marginal cost from prices under a known markup rule in empirical industrial organization, and it is what distinguishes this data system from approaches that impute farm yields from area aggregates, which cannot separate within-pool heterogeneity from measurement error.

Limitations

Aggregation is the binding constraint: a transaction pools producers with identical contracts, so within-transaction dispersion in yields is not identified, and moments computed from these records are moments of pool-by-contract means. Administrative revisions to the experience records propagate to this collection when it is refreshed. Finally, the zero floor in the yield-equivalence step censors the small set of revenue-plan records whose implied shortfalls are negative, a conservative treatment that slightly understates revenue-plan indemnification in high harvest-price years.

Data availability

The collection comprises one file per crop year from 2011 onward (calibration_data_<year>.rds), distributed as release assets:

piggyback::pb_download(
  file = "calibration_data_2022.rds", dest = tempdir(),
  repo = "ftsiboe/USFarmSafetyNetLab", tag = "calibration_data")

Recommended citation

Tsiboe, F. (2026). Preparing FCIP transactions for yield calibration. In FCIP calibrated and synthetic data catalogue. https://ftsiboe.github.io/rfcipCalibrate/articles/calibration-data.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

Coble, K. H., Knight, T. O., Goodwin, B. K., Miller, M. F., & Rejesus, R. M. (2010). A comprehensive review of the RMA APH and COMBO rating methodology: Final report. Prepared for the USDA Risk Management Agency.

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

U.S. Department of Agriculture, Risk Management Agency (USDA-RMA). (n.d.). Summary of Business, Actuarial Data Master, and program information. https://www.rma.usda.gov