A fixed variance gap rule gave 10.7% more squared error for three-month Nasdaq-100 returns in 2017–2026

We tested whether implied variance minus recent daily variance improved three-month Nasdaq-100 return forecasts. Each forecast used an ALFRED data version dated before its target period. The fixed primary rule shrank the fitted gap slope by one half. The test had 114 monthly origins from January 2017 to June 2026, with overlapping targets ending by August 2026. Mean squared error was 10.7% higher than for the mean of past returns. Out-of-sample R-squared was -10.73%, with a descriptive 95% 12-month block interval of -28.05% to -0.15%. Sign accuracy was 76.32%, versus 77.19% for an always-positive forecast. The 38 nonoverlapping quarterly origins gave R-squared -21.12%, with interval -57.23% to +0.27%. Origins in 2020 supplied 90.4% of the extra error. A freshness check and later outcomes gave similar point results. The one-month control reproduced an earlier negative SCI result. Stronger shrinkage reduced the error disadvantage but gave no gain. At an assumed 5 bp trading cost and 20 bp annual equity fee, the primary paper rule had 0.78 percentage points less annual utility than the past-mean rule, with interval -2.14 to +0.09 points. This result concerns one price index, fixed rules, and a daily variance proxy. It does not reject all variance risk measures. The package preserves the plan, source hashes, code, forecasts, fit records, uncertainty draws, figures, checks, and repeat instructions. Raw provider data are retained separately. November 2026 paper forecasts remain unscored.

stock return prediction · Nasdaq-100 · variance risk premium · historical data versions · negative result · chronological forecast test · ridge regression · paper trading