US jobless claims gave no clear added gain for weekly Nasdaq loss forecasts at the 2% threshold in a 2018–2025 archive test

We tested whether a 13-week rise in the four-week mean of US initial jobless claims improves next-week Nasdaq Composite loss probabilities after past price returns and variation are known. Fixed logistic models used historical ALFRED inputs and only outcome labels released before each fit. The main event was a price loss of at least 2%. In 394 fresh weeks in 2018–2025, with 79 events, adding claims increased Brier loss by 0.48%. The paired 95% 13-week block interval for loss reduction was -1.64% to +0.47%. This gives no clear added gain. The past event share had lower point loss than either fitted model. Fixed penalty, period, block, label and 1% and 3% loss checks gave no clear added gain. A direction check also gave no clear gain. Four separate dated claims downloads matched the reconstructed archive. All 4,992 fit states passed timing checks. A full repeat matched 14 result files exactly. Removing future source releases preserved 155 earlier forecast rows and 1,860 fit states. The package preserves code, public claims data, source records, all forecasts, fit states, scores, 15,000 block draws, figures and negative findings. Protected source features and individual outcomes are retained separately under provider rights. This is a retrospective benchmark on one index and a known market history. It does not establish trading profit or reliable future prediction skill.

stock return forecasts · jobless claims · loss probability · point-in-time data · chronological evaluation · negative result · Brier loss · Nasdaq Composite