VXN reduced daily 5% quantile loss by 4.8% in a 2017–2026 Nasdaq-100 archive test, with 60% of the gain from March 2020

We tested whether option-implied volatility improves forecasts of large daily Nasdaq-100 losses. Each of 116 monthly origins used an ALFRED vintage two calendar days before the month. All daily thresholds stayed fixed for that month. Eight fixed models used past returns, Cboe VXN, both inputs, or direct historical quantiles. The test covers 2,429 sessions from January 2017 to August 2026. Adding VXN to a past-return variance model with an empirical return shape reduced lower 5% pinball loss by 4.80%. The planned paired month-block 95% interval was 0.01% to 10.06%. March 2020 supplied 60.3% of the gain. Without that month, the reduction was 2.07%, with interval -0.19% to 4.14%. Changing the combined model from a normal to an empirical shape gave almost no 5% score gain. At 1%, the normal model had 1.98% hits; the empirical model had 1.44%, and its score gain was uncertain. Stale-input and later-outcome checks left the main result nearly unchanged. We preserve code, provenance, all forecasts, scores, bootstrap draws and negative findings. The result applies to month-fixed daily risk forecasts for one price index. It does not establish return-direction skill or trading profit. The study directly reuses the source data and variance-model functions of an earlier SCI test and is not an independent market sample.

Nasdaq-100 · VXN · quantile forecasts · value at risk · point-in-time data · historical simulation · tail risk · out-of-sample evaluation