Volatility scaling gave no clear 21-session Nasdaq loss forecast gain in 91 dated tests in 2018–2025

We tested lower 5% thresholds for Nasdaq Composite price changes over 21 market sessions. Each forecast used an ALFRED data version available two calendar days before entry. Target periods did not overlap. Six fixed models used historical returns, volatility scaling, or a square-root horizon rule. The main 2018–2025 sample had 91 periods with recent inputs. Volatility-scaled historical residual blocks gave 2.67% higher mean quantile loss than raw historical return blocks. The 95% paired six-period block interval for loss reduction was -30.61% to +19.96%. Positive reduction means a better forecast. Fixed source-delay, label-version and block-length checks also gave no clear gain. The 5% breach rate was 7.69% for scaled blocks and 9.89% for raw blocks. Few breaches limit coverage tests. A one-year historical window gave higher loss than five years in a secondary test. The package preserves all 2,358 forecasts, 660,000 resampling rows, code, figures, source hashes and repeat instructions. Separate checks passed. A fresh run matched ten analysis files and two restricted ledgers exactly. This is a reused historical sample with one index and one origin grid. It does not establish return direction skill or trading profit. Source prices and individual observed returns remain under the original provider terms in retained research records.

Nasdaq Composite · risk forecasts · historical simulation · volatility scaling · 21-session returns · point-in-time data · negative result