Three-month blocks gave no clear annual 10% US market loss forecast gain in 20 tests under assumed archive dates, 2005–2025

We tested whether three-month blocks improve forecasts of a US market loss of at least 10% at one or more monthly closes in the next year. We used dated French annual archives and 20 separate test years, October 2005 through September 2025. Each input had 120 months through July. We assume archive release by September 30. Five years had the event. The primary Brier gain over independent monthly draws was -0.000810, with a descriptive 95% annual resampling interval from -0.009199 to +0.008840. Thus the test gave no clear gain. A 5% loss threshold gave a positive secondary interval, without correction for multiple tests. We preserved forecasts, source hashes, code, all planned cases, uncertainty draws, a failed numerical boundary check and corrected runs. Independent checks passed, and a full repeat matched seven result files. The small annual sample, monthly closes and assumed archive dates limit the result. No trading profit is established.

market returns · path loss · historical simulation · block bootstrap · dated archives · forecast evaluation