The calibration credential for catastrophe pricing.
RMS, Verisk AIR, and Munich Re NATCAT publish loss exceedance curves. None publish pre-registered probabilistic forecasts with subsequent Brier scoring on a public ledger. Forecast Registry is the neutral standards body that defines what audit-defensible calibration looks like — and makes it citable in state insurance commissioner rate filings under NAIC Model Bulletin 30.
Reference benchmarks: 0.05–0.07 is the FiveThirtyEight state-level election band; 0.10–0.15 is the Good Judgment superforecaster tier; a coin flip is 0.25. Metric definition →
Why this exists
Three vendor catastrophe-model firms drive an estimated $100B+ in annual reinsurance and cat-bond pricing. None of them publish pre-registered, time-stamped, out-of-sample probabilistic forecasts subsequently scored against realized events on a public ledger. Their out-of-sample accuracy is not auditable by the regulators, state residual-market funds, and parametric MGAs whose pricing depends on it.
Documented vendor miss patterns are catalogued in methodology §4.1 with sources (Aon Benfield 2018, PG&E 10-K FY2018, Swiss Re sigma 2023, Munich Re NATCAT 2024).
NAIC Model Bulletin 30 (adopted Dec 2023) requires governance, testing, monitoring, and validation of AI systems used by insurers. The Bulletin is silent on which specific calibration metric satisfies the validation requirement under §3(D). State adoptions — NY DFS Circular 7, CO DOI Reg 10-1-1, CT DOI MC-25-08, CA DOI Bulletin 2024-1 — leave the gap open.
Forecast Registry proposes a Brier-graded, SHA-locked, OpenTimestamps-anchored credential as one acceptable answer. The proposal is methodology §5. Public-comment text is in governance.
For your institution
Segmented entry points. The methodology is identical; the asks differ by what each audience can do with the credential today.
State insurance commissioners
What an audit-defensible AI-model disclosure looks like under NAIC Model Bulletin 30 §3(D). Free to regulators by structural commitment.
For state insurersParametric MGAs
Calibration credentials for trigger probability claims. NOAA IBTrACS / USGS / SPC / JMA resolution sources wired into the scoring engine.
For parametric MGAsCat-bond issuers
SHA-locked, OTS-anchored credential for the modeled probabilities disclosed in offering memoranda. Investor-grade audit trail.
For cat-bond issuersReinsurance brokers
Independent calibration reference citable to cedents and capital providers. API-callable. Adds defensibility without replacing vendor models.
For reinsurance brokersAudit infrastructure
Every forecast is hashed at the moment of registration — question, resolution criterion, probability, lock timestamp. The hash is published in the canonical ledger and never altered.
Before resolution, the canonical ledger JSON is submitted to OpenTimestamps. The aggregated hash is anchored in a Bitcoin transaction. .ots receipts are publicly verifiable.
Stable canonical URL. Full history preserved. Corrections issued as superseding rows that point to the prior original_sha, never as in-place edits.
The institutional API
Three endpoints. Free to regulators. Bearer-token authenticated for institutional buyers. Documented OpenAPI spec.
GET /pricing/{peril}?date=YYYY-MM-DD
GET /forecaster/{id}
GET /methodology/versionReturns calibrated probability surface, audit anchor, methodology version, and source-ledger SHA. CORS-enabled for institutional buyer integration. Rate-limited per token.
Founding contributors
Author: Addie Conner. Founder, Chorus Public Ledger (the reference forecaster). Prior: SocialCode (acq. S4 Capital), Decoded Advertising (acq. S4 Capital), Breathwrk (acq. Peloton). Avenue100 / Link-Ventures alumna. UVM Econ + Stats.
Proposed academic advisors (outreach pending; no commitments): Philip E. Tetlock (Penn / Good Judgment), Andrew Gelman (Columbia), Frauke Kreuter (LMU München / Maryland JPSM), Roger Pielke Jr. (Colorado). Standards-board seats also reserved for one current or former state insurance commissioner and one reinsurance-market practitioner. The methodology will not be ratified as v1.1 until at least three advisory seats are filled.