Separate slopes for market loss months gave no clear joint forecast gain for three US factors in 2005–2025 under assumed archive dates

We tested whether a fixed change in factor slopes in market loss months improves joint monthly forecasts for US market, size and value returns. Twenty annual source versions gave 240 later monthly outcomes from October 2005 to September 2025. Each fit used ten years of data through July. Forecasts stayed fixed for twelve later months. The test assumes that each archive preserves its stated version and was available by September 30. A linear model and a model with half the fitted slope change used the same market Student t margin and conditional residual covariance. The joint forecast integrates over the unknown future market return. The half change gave a log score gain of -0.00396 nats per monthly vector against linear, with a paired 95% interval from -0.01188 to +0.00364. It gained in 8 of 20 annual groups. The full change, longer training window, separate residual scales and source-version checks gave no evidence of improvement. A joint elliptical Student t control scored better on average. The size and value mean forecasts also gave no added mean squared error gain. This is a fixed negative test on a reused sample. It does not reject all asymmetric market relations or establish a trading advantage. The package preserves code, source provenance, forecasts, fitted states, uncertainty draws, figures, independent checks, an exact replay and fixed unscored future forecasts. Provider return rows remain in a separate retained record under provider rights.

joint return forecast · market losses · size factor · value factor · conditional density · dated source versions · negative result · chronological test