There is a statistic that should stop every allocator mid-pitch: of the US large-cap funds that sat in the top quartile in 2021, exactly zero were still in the top quartile in each year through 2025. Not one. Expand the lens slightly — funds in the top half in 2021 that stayed in the top half every year after — and 4.49% managed it, worse than the 6.25% you would expect from pure chance. The coin, it turns out, is slightly biased against you. This is S&P's persistence scorecard data, the industry's own report card, and it says what the industry least wants to hear: past performance does not predict future performance, not approximately, not with adjustments. At all.
The headline scoreboard is no kinder. In 2025, 79% of active large-cap US equity funds underperformed the S&P 500 — the fourth-worst showing in the 25-year history of the SPIVA scorecards, and a sharp deterioration from 65% in 2024. Stretch the horizon and the numbers get worse in the way they always do: 89% underperformed over five years, 86% over ten, 93% over twenty. Not a single one of the 22 US equity categories saw a majority of managers beat their benchmarks over 15 years. The last year in which most large-cap managers beat the index was 2009. We are sixteen years into a streak that the industry's marketing department would prefer you not examine.
None of this means skill does not exist. It means skill is rare, the base rate is brutal, and the standard toolkit for identifying it — trailing returns, star ratings, "top quartile" screens — is mostly a machine for identifying luck and labeling it skill. The allocators who beat the base rate do it with a different toolkit.
The Persistence Mirage
The core statistical problem is sample size dressed up as evidence. A five-year track record feels substantial; statistically, it is roughly nothing. To distinguish a genuinely skilled manager (say, 1.5% of true annual alpha) from a lucky one at conventional confidence levels takes decades of data — decades no strategy, team, or market regime survives intact. This is why the persistence studies keep coming back empty: they are testing whether a noisy, short-sample ranking predicts another noisy, short-sample ranking, and the answer is that noise does not predict noise.
Worse, the rankings actively mislead through two mechanisms. First, style effects masquerade as skill: a value manager looks brilliant in a value regime and incompetent in a growth regime, and the attribution belongs to the factor, not the person. Second, and more insidiously, the mid- and small-cap data shows strong reversal — among funds ranked by 2015–2020 performance, 83% of the bottom-half mid-cap funds and 82% of the bottom-half small-cap funds moved up over 2020–2025. Buying last period's winners is, on this evidence, closer to a contrarian indicator than a strategy. Morningstar's "Mind the Gap" research adds the behavioral kicker: the average dollar invested in US funds earned 7.0% a year over the decade to 2024, about 1.2 points less than the funds' 8.2% total return — investors systematically buy the top-quartile past and sell the bottom, donating the difference to the managers they chase.
A Better Way to Test for Skill
If trailing returns cannot identify skill, what can? The serious allocators test the process, not the outcome — and they test it the way scientists test hypotheses: with predictions made in advance, evaluated against benchmarks that strip out luck's favorite disguises.
Start with holdings-based analysis. Returns can be decomposed; luck cannot hide in a portfolio the way it hides in a time series. Does the manager's outperformance come from stock selection or factor bets? From a handful of concentrated positions or broad-based skill? Tools exist to separate the two — the question is whether anyone on the allocator's side runs them. A manager whose "alpha" is a persistent overweight to a single factor is selling beta at alpha prices.
Then demand statistical honesty. Ask for the information ratio and its t-statistic over the full available history, and discount it for the number of strategies the firm runs (multiple testing inflates someone's numbers by construction). A t-stat below 2 on a ten-year record is not evidence of skill; it is evidence of a good story about randomness. Few allocators run this test. Fewer still act on it.
Evaluate the decision process directly. The best proxy for future skill is a repeatable process that generates differentiated, falsifiable views: written investment theses with price targets and kill criteria, post-mortems on both winners and losers, turnover in ideas rather than turnover in staff. You are looking for a machine that learns. Most managers can show you performance; the ones with skill can show you the machine, the maintenance logs, and the upgrades.
Finally, invert the burden of proof. The quotable version: in manager selection, the null hypothesis is luck, and the manager's job is to reject it — with data, not narrative. The industry default runs the other way: impressive returns are assumed to be skill until proven otherwise, and the proof never arrives because nobody demands it. Flip the presumption and the entire process sharpens — fewer managers hired, longer evaluation periods, harder questions, better outcomes.
Skill exists. It is just much rarer than the fees imply, much harder to identify than the pitch books suggest, and utterly invisible to anyone screening on past returns. The allocators who find it are not the ones with the best screens. They are the ones who stopped trusting screens.
Identifying real investment skill is a research discipline. managerresearch.com is available for acquisition. Inquiries: info@primedomaincapital.com.