Parametric insurance is only as reliable as the underwriting process that designs it. A poorly calibrated trigger — one where index movement and actual loss have weak historical correlation — produces a product that disappoints clients regardless of how fast it pays. The five analytical stages below describe how Riskwright approaches every new client engagement, from initial geographic scoping through final backtesting disclosure.
Stage 1: Exposure Mapping
Before selecting an index, we map the client's exposure geographically and operationally. For agricultural operations, this means identifying county-level location, crop type, planting calendar, typical harvest date, and the specific weather phenomenon that historically drives loss in that crop-location combination. For Southeast cotton, drought during the squaring and boll development period is the primary driver. For Southeast corn, drought during pollination and grain fill dominates. These are different risk windows and require different coverage period calibration even for operations in adjacent counties.
For infrastructure, exposure mapping involves locating the asset relative to NOAA ASOS and Co-op station coverage, identifying the prevailing hazard types by latitude band and coastline proximity, and reviewing available engineering design specifications to understand the relevant damage thresholds. A solar farm's racking design specification is as important an input to our underwriting process as its GPS coordinates.
Exposure mapping also identifies what we call the negative space of the risk: loss causes that the client faces but that no index can capture. Pest pressure, equipment failure, flooding from localized events outside any station's catchment, and management-related yield variation are examples. Where these non-indexable causes represent a large share of the operation's historical loss experience, we may recommend against parametric coverage or advise sizing the limit accordingly to reflect only the indexable portion of risk.
Stage 2: Index Selection
With the exposure mapped, we identify the one or two candidate indices that best capture the relevant weather phenomenon. For drought exposure, the primary candidates are SPI-3 (precipitation deficit standardized against historical distribution), PDSI (water balance incorporating temperature-driven evapotranspiration), and PRISM-derived precipitation deficit percentage (actual versus 30-year normal). For wind exposure, the candidates are maximum sustained wind speed, peak gust speed, or a composite metric incorporating both.
Index selection is a correlation exercise, not a theoretical preference exercise. Where the client has historical production records — yield history, revenue history, documented loss events — we run correlation analysis between historical index values and the client's actual loss data. Where historical production data is unavailable, we use county-level USDA yield statistics or industry loss proxies as the correlation target.
The selection output is presented to the client with correlation statistics for each candidate index. We don't simply announce which index we've chosen; we show the correlation analysis and explain what each metric captures and where it falls short. An informed client is a client who understands the product they're buying.
Stage 3: Station Selection and Quality Screening
Index accuracy depends on the quality of the underlying station data. Selecting the index is necessary but not sufficient; the specific station feeding the index calculation is equally important. Our station screening protocol evaluates four criteria: spatial proximity to the insured asset, record length, data completeness, and station stability.
Spatial proximity threshold varies by index type. For SPI-3 drought triggers, precipitation spatial autocorrelation at 20–25 miles in the Southeast is sufficient for agricultural coverage — a station 22 miles from the operation typically shows similar drought severity patterns. For ASOS wind triggers, the acceptable proximity threshold is tighter: 15 miles for single-station design, 30 miles with multi-station weighting. Wind speed variability at mesoscale is much higher than precipitation variability.
Record length screening requires a minimum of 30 years for the SPI-3 gamma distribution fit to be statistically stable. Preferably 45+ years. Stations with records shorter than 25 years are excluded from the reference network regardless of proximity. Completeness screening flags stations with more than 5% missing daily observations in the candidate period — gaps introduce uncertainty into the historical distribution fit and reduce confidence in percentile calculations.
Stage 4: Threshold Calibration
With index and station selected, threshold calibration begins with a percentile analysis of the historical index record. We compute the index value at each percentile from the 1st through 20th, using the station's full quality-controlled historical record. This produces a calibration table: SPI-3 of -1.0 corresponds to the 16th percentile at this station, SPI-3 of -1.5 corresponds to the 7th percentile, and so on.
The threshold recommendation is driven by the client's stated risk management objective. An operation seeking coverage that responds to any meaningfully dry growing season may choose a threshold near the 15th percentile. An operation seeking coverage only against severe drought years — accepting moderate dry years as within normal operating capacity — may choose the 7th or 5th percentile. We present options across the range with premium estimates for each, allowing the client to position their coverage relative to their own risk tolerance and budget.
We also evaluate trigger frequency stability. A threshold that fires reliably at the 7th percentile over the 30-year historical record is more predictable than one that fired 6 times in the first decade and 1 time in the last two decades — a pattern suggesting possible secular change in the station's climatological record that warrants investigation before using as a policy basis.
Stage 5: Backtesting Disclosure
The backtesting report is the final analytical output before policy terms negotiation. It covers the complete available historical record at the reference station and shows: the year-by-year trigger determination for the proposed coverage window, the index value at window close for each year, the distribution of index values across trigger years and non-trigger years, and the historical trigger frequency by decade.
The report also quantifies basis risk in both directions. Upside basis risk events — years where the trigger fired but actual conditions were mild — are documented with available proxy data showing the actual conditions during those years. Downside basis risk events — years where the trigger did not fire but actual drought loss was documented in county-level agricultural statistics — are identified and explained. We require client acknowledgment of having reviewed the backtesting disclosure before proceeding to policy terms.
This disclosure practice is not legally required for parametric insurance under most US regulatory frameworks. We do it because the product's value depends on client trust, and trust requires informed purchase decisions. A client who buys a parametric policy without understanding the historical downside basis risk rate will be disappointed — and rightly so — in a year where the trigger misses a real loss event.