Extra weight on past falls cut 5% Nasdaq Composite tail forecast loss by 1.1% in 2015–2025 with a two-day input cutoff

We tested whether extra weight on past price falls improves next-day Nasdaq Composite loss forecasts. Each forecast used an ALFRED data version dated two calendar days before its target. We fixed the rules before score calculation and tested 2,767 days in 2015–2025. The main filter gave squared falls weight 1.5 and squared rises weight 0.5. Its reference used equal weights. Both used exponential variance updates and 1,260 past standardized returns. Extra weight on falls cut mean lower 5% quantile loss by 1.11%. The paired 95% interval from 21-day blocks was 0.31% to 1.91%. The absolute gain was 0.170 basis-point score units, with interval 0.047 to 0.302. The result stayed positive with 5-day and 63-day blocks. There was no clear gain at the lower 1% quantile. The main rule had higher 5% loss in 2024. A source audit found 78 targets with inputs more than seven days old, due to an archive gap. An added check without those dates gave a 1.18% gain. The primary test keeps all dates. Independent scalar checks passed, and a second run matched all eight core numerical files. Code, source requests and hashes, all 38,738 forecasts, model states, 95,000 resampling rows, figures and negative findings are preserved. Raw source prices remain separately retained under provider terms. This is a small risk-forecast gain in one index and period. It does not establish return-direction skill or trading profit.

Nasdaq Composite · quantile forecasts · historical data versions · filtered historical simulation · asymmetric volatility · chronological evaluation · point-in-time data · negative findings