Turn-of-month mean forecasts gave no clear error reduction for US market returns in 2016–2024

We tested a fixed turn-of-month rule for daily US market excess returns. The window has the last trading day of each month and the first three trading days of each month. We fit two means on 1990–2014 returns in a file archived in November 2015. We kept these means fixed for 2,264 test sessions in 2016–2024. The calendar forecast gave no clear error reduction against a fixed overall mean. Out-of-sample R² was −0.078%, with a paired three-month block bootstrap 95% interval from −0.315% to +0.081%. The test calendar premium was −1.12 basis points per day, with an interval from −11.38 to +8.69. A long/cash paper strategy at 5 basis points per traded dollar had total return CAGR of 2.43%, against 4.49% for a fixed 20% market allocation. Its annual arithmetic mean excess-return difference was −1.75 percentage points, with an interval from −5.77 to +2.08. Later revised outcome data gave similar forecast scores. Independent calculations and a clean rerun matched the saved results. This is one aggregate market series and a market-proxy cost simulation. It does not establish equal performance, the absence of a calendar effect, or executable profit. The package preserves code, provenance, forecasts, cost cases, bootstrap draws, figures, and negative findings.

return prediction · turn of the month · calendar effect · archived data · out-of-sample test · transaction costs · negative result