Large-shock terms gave no clear added Nasdaq direction forecast gain in a 2015–2025 archive test

We tested whether large prior-day price moves improve next-session Nasdaq Composite direction forecasts. Annual logistic fits used ten earlier years. Each test input came from an ALFRED version valid by the prior calendar day. A session calendar removed closed-day rows. The test had 2,663 eligible sessions in 2015–2025; 103 sessions lacked the prior close. The fixed nonlinear model added separate terms beyond positive and negative standardized shocks of 1.5. Mean Brier loss rose from 0.247644 for the linear model to 0.247685. The paired gain was -0.0000406, with a 95% 21-session block interval of -0.0002969 to +0.0002065. The past positive-return frequency had lower loss, 0.247138. The nonlinear gain against that reference was -0.0005464, with interval -0.0012332 to +0.0001582. Direction accuracy was 55.50%, equal to the always-positive rule. A separate nonlinear mean-return model had 0.166% more squared error than the past mean. First-receipt outcome prices changed 463 return values but no direction labels. The result does not establish reliable return prediction or trading profit. It does not rule out a small gain or other reversal methods. Code, source hashes, forecasts, fits, uncertainty draws, figures and repeat instructions are preserved. Original provider prices remain external. The source and test years overlap earlier SCI work, so this is not independent future market evidence.

return direction · Nasdaq Composite · short-term reversal · point-in-time data · out-of-sample forecast · negative result