Smoothing absolute returns gave no clear daily Nasdaq risk forecast gain in a 2018–2025 archive test

We tested whether smoothing absolute daily returns improves next-session Nasdaq Composite risk forecasts over smoothing squared returns. Both rules used the same decay rate and separate scale fits from 504 earlier forecast records. Each input used an ALFRED archive version from before the target. Original past features stayed fixed. The main test had 1,933 sessions in 2018–2025. The QLIKE gain for absolute returns was -0.00530, with a paired 95% 21-origin block interval from -0.01885 to +0.00937. Positive gain means lower loss for the absolute rule. Thus the test gave no clear gain. Slower decay, longer input delay, different label versions and removal of 2020 also gave no clear gain. Variance mean squared error and lower-5% quantile loss gave no clear gain. The main source outage removed 78 sessions. A separate check found two archive validity starts before their observation dates. The main parser already used an observation-date gate. We corrected one secondary label date per delay run and preserved the initial records. Independent checks passed. A fresh run matched all 13 analysis files. The package has code, provenance, all 4,369 forecast rows for eight rules, 212 fitted states, uncertainty draws, figures, negative results and repeat instructions. Provider inputs and measured outcomes remain separate under source terms. This is a fixed retrospective comparison on a reused index sample. It does not establish reliable return-direction forecasts or trading profit.

risk forecasts · absolute returns · exponential smoothing · point-in-time data · Nasdaq Composite · negative result · chronological test