Monthly updates gave no clear gain for daily Nasdaq turn-of-month return forecasts in a 2015–2025 archive test

We tested a fixed turn-of-month rule for next-session Nasdaq Composite log returns in 2015–2025. Each monthly fit used the latest 1260 past labels available by its dated cutoff. The main forecast pulled each calendar-group mean toward the past overall mean with a weight of 60 observations. In 2766 test sessions, error reduction against the past mean was -0.069%, with a paired 95% block interval from -0.163% to +0.016%. Positive values mean lower error. Positive-return probabilities also gave no clear gain. Fixed weights of 0 and 250 gave similar results. The source had one price on a closed exchange day; we removed it before making adjacent-session labels. Labels used both prices from their first common valid ALFRED record. Source delays could not enter a fit early. Independent scalar checks passed for all labels, monthly fits, and forecasts. A future-label change left earlier forecasts unchanged. A clean replay matched all 34 numerical output and check files byte for byte. The package preserves the plan, forecasts, fitted states, uncertainty draws, source hashes, figures, code, and repeat instructions. Source prices and observed daily returns remain outside the public package under source terms. This is one historical price-index test with a reconstructed exchange calendar. It does not establish equal error or a trading profit. No strategy or real order was tested.

return prediction · turn of the month · Nasdaq Composite · point-in-time data · chronological test · negative result · shrinkage · calendar effects