The Evidence
The one manager skill that persists
We peel each manager's realized, position-level return — in one fixed sequence — through market, sector, subsector, and a size/value style block, down to a single stock-selection residual. Then we ask, out of sample, which decomposed component actually predicts forward results. The answer is asymmetric, and the asymmetry is the point.
Selection persists. Timing doesn't.
256 diversified US equity funds, 2006–2026, 15 annual non-overlapping rebalances. Managers are ranked into quintiles on their trailing stock-selection residual; the table reports the forward top-minus-bottom-quintile spread. All three skills run through the identical point-in-time machinery — which is what rules out that the result is an artifact of the method rather than a property of the residual.
| Skill (same test) | Forward Q5−Q1 | Reliability | Verdict |
|---|---|---|---|
| Stock selection (residual)13 of 15 yrs positiveThe headline result | +2.3 pp | t ≈ 3.4 | Persists |
| Style timing4 of 15 yrs positive | −2.1 pp | t ≈ −2.1 | Weak negative |
| Sector timing8 of 15 (coin flip) | +0.0 pp | t ≈ 0.0 | No pattern |
And it's durable manager quality, not hot hands: ~94% of the predictive power is the between-manager level of skill, only ~6% is within-manager timing. The discipline that follows is identify and hold, not chase recent winners.
It survives the obvious attacks
Raw p ≈ 0.004 → p ≈ 0.013 under Bonferroni, Holm, and Benjamini–Hochberg across all three skills. Still significant at 5%.
Lag holdings by the actual ~60-day 13F delay and the forward gap moves ~0.2pp — reliability is unchanged, if anything higher (t 3.3 → 3.5).
Fixed-survivor and balanced-panel re-runs hold or strengthen (t ≈ 3.3), bounding the cohort's backfill bias.
The top-five peer-relative overweights of high-residual managers outperform the bottom quintile's by +3.2pp forward (t ≈ 4.0) — the signal lives in the specific stocks they emphasize.
Why this isn't in your current toolkit
Manager databases show you exposures. Brinson attribution is accounting. Neither tests, out of sample, which decomposed component actually persists. From the paper:
“We are not aware of a single incumbent product or published manager-selection framework that combines all of: a signal that is holdings-based, factor- and sector-neutral by construction, and validated out of sample on which decomposed component persists.”
What this does not claim
- The signal is gross and holdings-derived. Whether skill reaches the investor net of fees is a separate question.
- It is a within-mandate ranking signal that improves forward odds — not a guarantee about any single fund.
- This is not a claim that active management beats indexing net of fees. It narrows the allocator's search problem.
- The numbers are established on a US large-cap diversified cohort. The framework is portable; the specific figures are not extended to sector, international, small-cap, or non-equity mandates.
Figures are published results from Part 3 — The One Manager Skill That Persists (riskmodels.org): gross, holdings-derived, within-mandate, US large-cap diversified cohort, 2006–2026. Informational use only — not investment advice.