# Potential forecasting-cost savings

## Finding as of September 28, 2026

**Neither case currently supports a defensible savings percentage.** Public evidence establishes that modeling services can require substantial spending and that some traded forecast inputs have inexpensive access. It does not price the same forecasting job on both routes, or establish which existing costs a user could eliminate. This is an evidence finding, not an estimate of zero savings.

The denominator is the three-year prospective cost of producing and maintaining one specified peril forecast used in underwriting. It excludes property assessment, policy design, claims, premiums, capital and profit. No employee wage rate or assumed staff hours enter the calculation. We do not adopt the previously proposed 15–30% range.

## Rhine: a seasonal trigger forecast

The output is the probability of each of the current ten Kaub July–November low-water duration thresholds, at matching information dates before and during the season. For the cost comparison, assume the same daily snapshot from each route; that frequency is a research design choice, not an existing service commitment.

[BfG's public service](https://6wochenvorhersage.bafg.de/) provides six-week forecasts of weekly mean water levels and discharge, including Kaub, in PDF and CSV. The [BfG Open Data catalogue](https://geoportal.bafg.de/CSWView/) lists Rhine forecast datasets under CC BY 4.0. This provides a public alternative with no dataset licence charge under that licence; integration and validation remain costs. Six weekly means cannot directly determine the probability of a count of daily low-water days over a five-month season. Public availability is not evidence that the full underwriting forecast costs nothing.

[SpotBarge](https://spotbarge.com/rhine-waterlevels/) advertises a 14-day forecast for four Rhine locations including Kaub. Its reviewed page does not publish a price. Neither source supplies a matched price for the exact seasonal duration curve. We found no public separable build-and-maintain cost for that curve, nor a live PMF venue tariff or priced adoption scope. Baseline total, replacement total, savings percentage and numerical break-even ceiling therefore remain unavailable.

The opportunity to test is avoiding a separately commissioned seasonal trigger model. A firm already using free inputs with a simple inexpensive method may have little spending to avoid. No fraction of a river-system model budget is assigned to this case.

## California: hypothetical statewide index cover

The output is the probability of each Draft 0.3 annual CAL FIRE structures-destroyed threshold, at matching dates before and during the reporting year. Use the same daily snapshot comparison. This is a hypothetical index-cover workflow; no matching purchase or insurer model retirement is established.

[CDI's fact sheet](https://www.insurance.ca.gov/01-consumers/180-climate-change/upload/FOR-PUBLICATION-WILDFIRE-MODEL-FACT-SHEET-3.pdf), page 1, describes anticipated multi-million-dollar initial funding for a center and model. Its scope includes research and education as well as modeling. The [program timeline](https://www.insurance.ca.gov/01-consumers/180-climate-change/public-wildfire-model.cfm) records an August 2026 notice of intent to award, with grant issuance anticipated in October. This is not evidence of completed spending. Neither source isolates our annual count forecast. Page 3 of the fact sheet also distinguishes hazard, property vulnerability and financial loss calculations.

A statewide count curve cannot replace location-specific wildfire analysis or individual policy pricing. No public matched build/maintenance cost or market adoption cost was identified. Both three-year totals, savings percentage and numerical break-even ceiling remain unavailable. Broad program funding is not used as a denominator.

## Public cost comparables

| Evidence | Public amount | What it establishes | Why it is excluded from peril savings |
|---|---|---|---|
| [Central Bank of Ireland, Q1 2023 awards](https://www.centralbank.ie/docs/default-source/tns/about---tns/freedom-of-information/procurement/q1-2023-contracts-awarded.pdf?sfvrsn=ac9d991d_2), page 1 | €185,450 excluding VAT; Mazars; January 13, 2023 | Award value for catastrophe-modeling services, with completion-based duration | No exact peril forecast, recurring-cost breakdown or cash-paid record; cannot annualize or allocate arbitrarily |
| [Climate Finance Lab, Oasis analysis, 2016](https://www.climatepolicyinitiative.org/wp-content/uploads/2016/12/Oasis-Platform-Instrument-Analysis.pdf), PDF pages 7 and 9 | Model development about $100,000 to $1 million or more; projected 25–50% modeling savings | Historical expert estimates concerning Oasis and broader model development | Projected rather than measured savings; different mechanism, geographies and scope; not a PMF percentage |
| [CME FedWatch API](https://www.cmegroup.com/market-data/market-data-api/fedwatch-api.html) and [June 2026 fee schedule](https://www.cmegroup.com/market-data/files/analytics-data-fee-list-june-2026.pdf), page 2 | End-of-day access starts at $25/month; internal end-of-day use has no additional licence charge in that schedule | A published price for accessing a futures-derived rate signal | Different event; excludes adoption and operation; no PMF pricing established |

The reproducible access-only illustration is **$25 × 36 months = $900**, before tax, assuming the starting price stays unchanged for one end-of-day access arrangement. This is neither a future price guarantee nor an estimate of model cost or savings. Intraday non-display use has a separate $280/application monthly fee, excluding connectivity, listed as valid through March 31, 2027. Do not apply that temporary fee across three years or treat all commercial uses as free.

## Calculation and evidence gate

For each route, total cost = setup + three times annual recurring commitments. Calculate `(existing cost − market-based cost) / existing cost × 100` only when all material components on both routes have matched public cost evidence, a common price basis and currency, and a positive denominator. No inflation or currency conversion is applied to the current context-only figures; they are never combined.

The component ledger covers development, integration, licences, infrastructure, maintenance, validation, fallback and market-specific access/monitoring/support. Null means unknown; it never means zero. Zero is allowed only with supporting evidence, including for components confirmed inapplicable. Numeric inputs require sources eligible for this specific forecast. All sources currently have context-only status, so none can populate a case's cost total.

Classify prospective commitments as avoided, reduced, retained or added using a specified architecture and licence scope. Exclude sunk development expenditure. Count a bundled licence or shared resource as saved only if a documented payment can actually end. A retained fallback remains in replacement cost; do not also count it in another row. Record any setup costs for market creation within integration, and recipient-funded recurring liquidity support within market operations. Unknown integration, monitoring or fallback costs block a savings estimate.

When the existing total is known, the largest **total** replacement cost that breaks even equals that existing total. The remaining adoption budget is that ceiling minus already priced retained and replacement costs. Neither case has a documented existing total, so no numerical ceiling is published. If the old model stays fully operational and market adoption adds positive cost, savings are negative. The calculation preserves this outcome; it never floors savings at zero.

A future range must use source-supported low/high line items and coherent scenarios, including retained-model sensitivity. It is a conditional planning range, not a confidence interval or measured saving. Equivalent forecast quality, coverage, availability and decision quality must be demonstrated separately before claiming realized savings.

## Reproduce and inspect

Download `forecasting-cost-calculations.py`, `forecasting-cost-inputs.json` and `forecasting-cost-evidence.json` into an empty directory, then run:

```sh
python3 forecasting-cost-calculations.py --out results
python3 forecasting-cost-calculations.py --out results --verify
```

No network calls, paid credentials or third-party Python packages are required. Outputs include separate case results, component CSV, evidence CSV, inputs and calculation source. The public assessment freezes dated factual summaries and source identities. Provider originals live only in the repository's research archive and are excluded from deployment. This research snapshot is separate from both the current Rhine ladder and the archived settlement definitions.

## Source and reuse notices

The evidence ledger records exact URLs, locators, versions, retrieval times, checksums, source categories and reuse limits. Provider PDFs, website HTML, imagery and datasets are not included in public cost downloads. The Oasis report is CC BY-NC-SA 3.0 and requests permission for commercial reuse; we link to it and summarize a few facts, without reproducing its report, charts or tables. CME material remains copyrighted. Public access to procurement notices does not establish redistribution rights. BfG's dataset licence is documented separately from page access. PMF's specification and code licence remains pending.

## Research coverage and limits

Reviewed public supplier offerings and fees (CME, SpotBarge), BfG forecast/data catalogues, an awarded Central Bank of Ireland modeling contract, CPI's historical analysis, and CDI's program and funding factsheet. Searches included government procurement and catastrophe-model licence terms. The linked California procurement event returned HTTP 403 in this review; no unseen detailed budget is assumed. CME direct re-downloads also returned 403; previously archived September 27, 2026 provider bytes were reused and the official facts were rechecked through browser retrieval. No vendor or insurer was contacted. “Not identified” describes this review, not proof that no qualifying public contract exists.

## Validation assessment

**Share with caveats:** the published findings are suitable as a cost-evidence assessment. A numerical peril savings claim remains unsupported. Award values, budgets, expert projections and supplier prices are labeled separately. Public evidence cannot establish that a company has retired a model, that forecast quality is preserved, or that the proposed markets will be liquid.