A 50% mix with mean industry risk gave no clear forecast gain in 66 US months from dated 2015–2025 archives

We tested whether a fixed 50% mix with mean industry risk improved forecasts for thirty US industry returns. Seven dated archives supplied inputs for 66 distinct monthly outcomes from December 2015 to August 2025, with gaps. Each rule used the last 120 source months and stayed fixed at its origin. The main score tested squared return errors about the same fitted means. The mix increased mean loss by 0.04988. The descriptive 95% interval for loss reduction was -0.09798 to +0.01573; a six-group sign-flip test gave p = 0.25. A full common risk estimate also had higher loss. A fixed mix of recent and long-run risk had lower loss, but its group test gave p = 0.09375. Only six capture-year groups were available. The package preserves all forecasts, source hashes, code, uncertainty draws, checks and unscored future forecasts. The test establishes no trading profit or reliable price direction skill.

industry returns · risk forecasts · dated data · shrinkage · chronological test · negative result