Signed return inputs gave no clear daily Nasdaq mean forecast gain over scaled drift in a 2018–2025 archive test
We tested whether signed past returns improve next-session Nasdaq Composite mean forecasts. A fixed ridge model used five past-return windows and a volatility scale. Its reference used the same scale and an intercept. Each input came from an ALFRED record available before the entry date. Monthly fits used fixed historical inputs and labels already released. The main 2018–2025 test had 1,933 origins with input age at most four calendar days. Added inputs gave 0.049% more mean squared error. The paired 95% interval for error reduction was -0.200% to +0.079%, from 5,000 draws of 21-origin blocks. Scaled ridge sign accuracy was 54.99%, compared with 55.41% for an always-positive rule. Planned period, block, source-age and label-version checks gave no clear primary gain. This is a negative result for fixed linear models, one index and a delayed daily target. The package preserves code, source provenance, 2,514 dated forecast rows, 120 fit states, all scores, 45,000 resampling draws, figures and repeat instructions. Provider observations remain external under their original terms. No trading return is evaluated.
return prediction · Nasdaq Composite · point-in-time data · ridge regression · negative result · chronological test
References
7- usesRecent losses gave no clear added daily forecast gain in fixed Nasdaq Composite models in a 2018–2025 ALFRED test sci:sha256:a86f9653e622fd72f8ef41036928f807d0acf6a393a31e7b890283013b1b7584
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.
- usesVolatility scaling cut lower 5% return forecast loss by 8.8% in a 2015–2025 Nasdaq Composite vintage test sci:sha256:a8d414223facf768878a01e0307db214555fb5c69e6f24f643b076010c981d6d
We reuse its retained ALFRED source and MIT source-download code. We test mean forecasts rather than lower-tail forecasts.
- Past returns reduced errors in daily return size but not direction for five US indexes in CNNpred data, 2015–2017 sci:sha256:698cddb4b89bfaf01abd536481cf17382e394cc6fa9d9934c8c957f6f0cb85de
- 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
- doi:10.1093/rfs/hhm014
- doi:10.1016/j.jfineco.2011.11.003
- doi:10.1111/jofi.12513
Cited by
6 publicationsPublic catalog entries on Sciscape.
- usesYearly turn-of-month fits gave no clear Nasdaq daily return forecast gain in a 2018–2025 archive test
We reuse its underlying retained ALFRED source and documented interval handling design for dated Nasdaq fits.
- usesCalendar inputs gave no clear daily Nasdaq return forecast gain with monthly fits in a 2018–2025 archive test
We reuse its retained dated source, interval parsing method, and MIT downloader; the shared history is not independent market evidence.
- usesTurn-of-month means gave no clear daily Nasdaq forecast gain in a 2015–2025 archive test
We build on its dated mean-forecast design and reuse the same retained raw source. We do not reuse its target labels or forecasts.
- Fitted Student-t models cut daily Nasdaq Composite log loss in a 2018–2025 dated-input test
This test finds no clear daily Nasdaq mean-return gain from signed inputs in overlapping years; density skill must be kept separate from mean skill.
- usesPast yen gains gave no clear added gain in daily Nasdaq risk forecasts in a 2018–2025 archive test
We adapt its MIT interval parsing and snapshot approach and follow its separate evaluation of mean returns.
- usesMonthly updates gave no clear gain for daily Nasdaq turn-of-month return forecasts in a 2015–2025 archive test
We use its retained ALFRED source, MIT downloader, and source interval rules. Our calendar forecasts use different inputs and monthly group models.