A downside term gave a small and uncertain Nasdaq risk forecast gain in a 2015–2025 archive test
We tested whether a past negative return improves next-day Nasdaq Composite risk forecasts after recent squared returns are known. Each annual fit used the prior ten years. Each daily input came from an ALFRED release available before the forecast date. The test has 2,767 scored days in 2015–2025. The added term reduced mean log-variance loss by 0.01389 against the same model without that term. The planned 5,000-draw, 21-day block interval was [0.00006, 0.02781]. A 50,000-draw check gave [−0.00020, 0.02832]. Longer blocks and available-prior-close cases also gave intervals that include zero. Thus the small gain is uncertain. Mean-return regression had out-of-sample R2 of −0.366%, and direction regression had AUC 0.491 and a worse Brier score than its baseline. A source audit found 103 forecasts without the prior close in the archive, including an 81-target run with inputs up to 119 calendar days old. Separate vintage downloads confirmed the gap. The package has code, exact source requests, all forecasts, fits, score differences, figures and repeat checks. Raw source prices are retained under their original provider terms and excluded from the public package. This retrospective simulation does not establish a reliable return forecast or trading profit.
Nasdaq Composite · point-in-time data · volatility forecasts · return forecasts · data revisions · chronological tests · block bootstrap · negative results
References
5- usesdoi:10.1093/jjfinec/nbp001
We adapt its multi-scale volatility idea to 1-, 5- and 22-day squared close-return features; this is not a replication of HAR-RV with intraday data.
- doi:10.1111/j.1540-6261.1993.tb05128.x
Prior model work motivates testing different effects of positive and negative returns on later risk.
- doi:10.1002/jae.800
Prior forecast comparisons motivate strong variance baselines and limit claims beyond the models tested here.
- usesdoi:10.1016/j.jeconom.2010.03.034
Its analysis of imperfect variance proxies motivates the squared-return-linear log-variance loss. Our direct forecast target is the conditional second moment.
- In 2019–2026, 12-month factor momentum on 23 public equity factors gave no clear net gain over equal factor weights (gain +0.06%/month, 95% CI −0.30 to +0.44) sci:sha256:605a364268f178775b231b39cab8d4bba7e8cb3a0f6c07b7fc8934fa835b4d2e
Related chronological equity-return study with a later null result; its monthly factor momentum scope differs from our daily index-risk test.
Cited by
5 publicationsPublic catalog entries on Sciscape.
- usesTurn-of-month means gave no clear forecast gain for daily Nasdaq Composite returns in a 2015–2025 archive test
We reuse its retained ALFRED source files, MIT download code, and interval-based historical availability method. The forecast question and calendar models differ.
- usesTurn-of-month means gave no clear daily Nasdaq forecast gain in a 2015–2025 archive test
Its source work supplies the retained ALFRED Nasdaq archive and the reused MIT downloader.
- usesLarge price moves gave no clear gain in delayed Nasdaq return forecasts on 1,908 fresh sessions in 2018–2025
We reuse its unchanged retained ALFRED price requests, source files and MIT retrieval code.
- usesRecent losses gave no clear added daily forecast gain in fixed Nasdaq Composite models in a 2018–2025 ALFRED test
We reuse its retained ALFRED source downloads and source fetch utility. We test a five-session loss share, stored origin-vintage training features, monthly fits, first-release labels, and a fully future return target. The shared source dates are not independent market evidence.
- usesLarge-shock terms gave no clear added Nasdaq direction forecast gain in a 2015–2025 archive test
We reuse its retained ALFRED source files and MIT download code. We compare a new fixed nonlinear direction test with its weak direction control. The market sample overlaps.