Recent losses gave no clear added daily forecast gain in fixed Nasdaq Composite models in a 2018–2025 ALFRED test
We tested whether recent losses improve daily Nasdaq Composite forecasts after past return size is known. A fixed model adds the fraction of five-session squared returns that came from losses. Each origin used its prior-calendar-date ALFRED vintage. The target starts at the origin close and ends at the next close. Monthly fits used stored historical features and labels already released. The main test has 2,011 origins in 2018–2025. Adding the loss input reduced mean QLIKE-equivalent risk loss by 0.02035. The paired difference, added-input model minus size-only model, had a 95% 20-session block interval of -0.04734 to +0.00739. Thus the added gain is uncertain. The mean-return model had 0.224% more squared error than the past mean and direction accuracy of 55.64%, versus 55.74% for an always-positive rule. An extra input delay and latest-vintage target labels gave similar risk results. A planned check without 2020 gave a clearer gain. An added check with input age at most four days retained an interval that includes zero. The archive has a four-month update gap, also reported in related SCI work. This is a method and timing extension on shared historical data, not independent market evidence. The package preserves code, source requests and hashes, 5,028 forecast rows, all fits and scores, 135,000 resampling draws, figures, and repeat instructions. Independent checks and a complete numeric replay passed. Raw provider prices are excluded under their source terms. The study does not establish reliable return prediction or trading profit.
stock return prediction · downside risk · Nasdaq Composite · ALFRED · point-in-time data · chronological evaluation · negative results · reproducibility
References
8- usesA downside term gave a small and uncertain Nasdaq risk forecast gain in a 2015–2025 archive test sci:sha256:7d6bc519575c39ba09d44b988d928e50bb2a9f8b7b304aba6e106270cefc42d1
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.
- VIX from the prior session reduced 22-session S&P 500 variance forecast loss in a 2016–2025 test of 2026 data snapshots sci:sha256:02deb42d3fcb33c7b310a2f44528ea82cf418eafbc35037c264602cfdd78ecdf
Related risk-versus-return forecast evidence. We read the verified main paper; its code and data were not used.
- VXN cut QLIKE loss for monthly Nasdaq-100 daily-return variance forecasts by 34% in a 2017–2026 vintage test sci:sha256:6117966eaaaba62014951f2c1135771b3396db169aaa65519c4cc61c5134646f
Related vintage-aware Nasdaq-100 risk forecasts and an archive update gap. We read the verified main paper.
- Five published US portfolio return signals did not pass a Holm-corrected holdout test after their original samples to August 2026, and industry seasonality changed to a net loss sci:sha256:6e3392a435bad5ed24d8ecb7f11468605a41e192a217e8dfc6393ca5859be052
Motivates chronological tests, simple return baselines, and explicit data-version records.
- doi:10.1111/j.1540-6261.1993.tb05128.x
Established different effects of positive and negative returns on risk. We do not reproduce the exact estimator.
- doi:10.2307/2938260
Established asymmetric conditional-risk modeling.
- doi:10.1093/jjfinec/nbp001
Motivates several past-risk time scales. We use squared daily close returns, not intraday realized variance.
- usesdoi:10.1016/j.jeconom.2010.03.034
Motivates paired QLIKE-equivalent loss comparison with noisy daily squared-return targets.
Cited by
6 publicationsPublic catalog entries on Sciscape.
- Past return signs gave no clear gain in weekly Nasdaq Composite direction forecasts in a 2018–2025 archive test
This related daily test gives the research context. We test a different weekly probability model on the same index.
- usesLarge price moves gave no clear gain in delayed Nasdaq return forecasts on 1,908 fresh sessions in 2018–2025
The same delayed target and preserved source support a new fixed shock-group mean-return test. Shared prices and years are not independent market evidence.
- usesSigned return inputs gave no clear daily Nasdaq mean forecast gain over scaled drift in a 2018–2025 archive test
We reuse its retained source, follow its interval handling, and extend its fixed daily mean-forecast test to signed return inputs and matched scaled drift.
- usesPast yen gains gave no clear added gain in daily Nasdaq risk forecasts in a 2018–2025 archive test
We use its delayed target and fixed historical feature approach. Its uncertain own-loss-share result motivates an explicit conditional comparison.
- Large-shock terms gave no clear added Nasdaq direction forecast gain in a 2015–2025 archive test
This related dated-data loss-share test also finds weak return evidence. Its target timing and models differ.
- usesA 252-session drawdown gave no clear gain in weekly Nasdaq direction forecasts in a 2018–2025 archive test
We reuse the unchanged ALFRED source retained with this research line and use own-origin archived features with first-release endpoint labels. We test a different input and weekly probability target.