The VIX term ratio gave no clear gain for weekly Nasdaq 3% loss probability forecasts in a 2020–2025 archive test

We tested whether the log ratio of three-month to one-month implied volatility adds information about weekly Nasdaq Composite price losses of at least 3%. Each input used an ALFRED archive valid by the preceding Friday. All main-test training inputs came from their own origin dates. Fixed logistic models used past price variation and VIX, with or without the term ratio. The 2020–2025 test had 296 available weeks and 39 events. Adding the ratio changed mean Brier loss by -0.001298, a 1.12% reduction, with a paired 95% eight-week block interval from -0.003928 to +0.000552. The year 2020 supplied 97.9% of the observed reduction. Without that year, the change was -0.000033, with interval -0.001148 to +0.000954. Binary log loss and fixed 2% and 5% event checks gave no clear added gain. We preserve all 400 forecast rows, model states, fixed comparisons, source requests and hashes, code, figures, checks, and repeat instructions. Direct source-derived records are retained separately under provider rights. The test covers one index, a small model family, and a reused historical sample with a 16-week source outage. It does not establish trading profit or a reliable added benefit from this ratio.

stock return forecasts · loss probability · VIX term structure · point-in-time data · chronological evaluation · negative result