Turn-of-month means gave no clear daily Nasdaq forecast gain in a 2015–2025 archive test

We tested whether a fixed turn-of-month rule improved daily Nasdaq Composite price-return forecasts. The rule used the last session of a month and the first three sessions of the next month. Each monthly fit used an ALFRED record from before the first target and the last 2,520 available daily returns. Separate calendar-group means were compared with the overall mean of the same history. The test covered all 2,766 sessions in 2015–2025. The calendar rule raised mean squared error by 0.0242%. The planned six-month paired block interval for error reduction was -0.0557% to +0.0023%. The observed turn-minus-other mean was -1.28 basis points, with interval -10.52 to +8.54. Sign probabilities gave no clear Brier loss gain. The ten-year sign forecasts had the same accuracy as always predicting a positive return. A shorter history and a fixed shrink rule did not improve the result. Independent source and scalar checks passed. All 17 analysis files matched in a second run. The package preserves code, source requests and hashes, all forecasts and fits, 80,000 uncertainty draws, figures, and repeat instructions. Raw provider prices remain external. This historical test reuses a market sample from related SCI work and assumes the session schedule was known at each forecast date. It does not establish reliable return-direction skill or a trading profit.

Nasdaq Composite · turn of the month · return forecasts · archived data · calendar effects · negative result · chronological evaluation