Yearly turn-of-month fits gave no clear Nasdaq daily return forecast gain in a 2018–2025 archive test

We tested yearly calendar forecasts of daily Nasdaq Composite price returns. Each fit used 1,260 returns from dated ALFRED records. The calendar window has the last session of a month and the first three sessions. The main test has 2,011 returns in 2018–2025. Two calendar means had 0.0792% more squared error than an overall mean from the same data. The paired 95% interval for error reduction was -0.2149% to +0.0240%. This test gave no clear added forecast gain. Yearly calendar fits also had more error than a frozen calendar fit. That comparison includes changes in the overall mean. A paper allocation held the calendar group with the higher fitted return. At 5 basis points per traded dollar, it trailed a reference with similar expected equity exposure. Its annual arithmetic mean return difference was -4.48 percentage points. The interval was -9.98 to +0.96 points. The price-index simulation excludes dividends and cash yield. It does not establish executable profit. Independent checks matched all forecasts, labels and 25 paper accounts. The package preserves forecasts, fit states, cost cases, 70,000 bootstrap rows, code, source records, figures and negative findings. It extends a prior fixed US-market test to yearly Nasdaq fits. The shared market and years are not independent validation.

return prediction · turn of the month · point-in-time data · Nasdaq Composite · chronological evaluation · paper trading · negative result