Adding a VIX term ratio raised weekly Nasdaq return forecast error by 2.4% in a 2020–2025 archive test

We tested whether the log ratio of three-month to one-month implied volatility helps to predict next-week Nasdaq Composite returns. The reference model used past returns, past price variation, and VIX. Each Monday forecast used ALFRED records valid by Friday. Each fit used 260 completed cases. All training features in the 2020–2025 main test came from their own earlier forecast dates. The test had 296 forecasts; a source outage removed 16 planned origins. Adding the ratio raised mean squared signed-return error by 2.42%. The paired 95% eight-week block interval for the increase was 0.38% to 4.19%. Error was higher in five of six years. Fixed penalty, source-age, later-label, and removal-of-2020 checks retained the increase. Longer block checks added after score inspection also favored the reference. Both main return models had more error than the rolling past mean. Negative-return probability models gave no clear added Brier or log-loss gain. Separate source parsing, alternate fits, timing checks, and a thirteen-file exact replay passed. We preserve code, source requests and hashes, all 557 forecasts and fit states, aggregate scores, figures, and negative findings. Complete source-derived ledgers and uncertainty draws remain retained under provider terms. This study reuses a market history from an earlier SCI variance test. It does not establish trading profit or rule out value from other term measures, assets, horizons, or methods.

return prediction · VIX term structure · Nasdaq Composite · point-in-time data · chronological test · negative result · forecast evaluation