Past return signs gave no clear gain in weekly Nasdaq Composite direction forecasts in a 2018–2025 archive test

We tested whether signed past returns improve the chance forecast for a positive weekly Nasdaq Composite price return. We used dated ALFRED archives, a fixed weekly schedule, past-only fits, and penalty selection with outcomes known before the test. The primary 2018–2025 sample has 401 weeks with fresh inputs. Adding signed returns to a past-size logistic model increased mean Brier loss by 0.23%. The 95% paired block interval for loss reduction was −1.00% to +0.51%. The past positive share had lower observed Brier and log loss than both fitted models. Both models called every return positive at a 0.5 threshold. Their 61.60% sign accuracy matched an always-positive rule. Planned period, archive-age and label checks gave no clear added gain. Independent source and scalar checks passed. The package preserves code, exact source requests and hashes, forecasts, fitted models, scores, uncertainty draws, figures, the local plan and repeat instructions. Source index levels and measured returns remain external under provider rights. This is one historical probability test. It does not establish reliable trading returns.

return direction · probability forecasts · Nasdaq Composite · point-in-time data · chronological test · negative result · Brier score