Regression to the Mean: Why Stars Disappoint and Rookies Surprise

The Sports Illustrated cover has a reputation as a curse. Athletes appear on the cover at their peak — after an extraordinary season, a championship run, an unprecedented streak — and then seem to decline. This happens consistently enough that it's become a cultural superstition.

It's not a curse. It's regression to the mean, one of the most practically important statistical concepts in Fooled by Randomness.

The Mechanism

Any observed performance combines genuine skill with random variance. When a performance is extreme — far above or far below the average — the extreme is disproportionately likely to reflect a favorable (or unfavorable) draw from the variance component, on top of the skill component.

If skill is roughly constant across periods, and variance is random, then: the next observation will contain the same skill component and an independent new variance draw. Since the new variance draw is not conditioned on the previous one being extreme, it will, on average, be less extreme.

The extreme past performance was: skill + very favorable luck draw. The next performance will be: skill + average luck draw. The result looks like decline, even though skill didn't change.

This is not a force pulling outcomes toward the center. It's arithmetic. Extreme outcomes oversample from the lucky end of the variance distribution. The next observation doesn't continue sampling from the lucky end — it samples from the full distribution. So it tends to be less extreme.

The Sports Illustrated Illustration

The athlete on the cover is there because their performance over the recent period was exceptional. That exceptional performance combined their true skill with favorable variance: unusually good health, favorable scheduling, opponents in below-peak form, a set of conditions that maximized the skill while adding a favorable luck component.

The next season, the same skill level operates with a new draw from the variance distribution. The favorable conditions that contributed to the peak performance don't necessarily repeat. The result looks like "coming back to earth" or "struggling to meet expectations" — but what actually happened is that an extreme observation was followed by a less extreme one, exactly as the arithmetic would predict.

There's no curse. There's only the mathematical fact that peaks are peaks: by definition, the period immediately following a peak is more likely to be below the peak than above it.

Why Hiring for Past Performance Systematically Fails

The practical consequence of regression to the mean is that hiring and promotion decisions based on extreme recent performance are systematically selecting for the lucky component of a lucky period.

The star analyst who produced the best returns last year is, in expectation, going to produce returns indistinguishable from the base rate in the next year — because some fraction of what produced last year's performance was favorable variance that won't repeat.

The talent acquisition process that recruits based on top-performer lists, ranking tables, or last-year leaderboards is systematically selecting people at the peak of lucky sequences. When those people produce average performance after acquisition, it's attributed to culture fit, transition stress, or inconsistent process. The statistical mechanism was always going to produce this outcome.

The Correction

The corrective isn't to stop selecting for skill. It's to ask better questions about how much of an extreme performance is skill and how much is favorable variance.

Number of independent observations. A five-year track record spanning multiple conditions and regime changes is harder to produce through luck than a two-year track record in favorable conditions. More trials, more independence, more environments — each reduces the probability that the performance is entirely lucky variance.

Performance relative to appropriate peers. Outperformance in a bull market that lifted all boats is less informative than outperformance in a mixed environment where the peer group underperformed. The variance component of a bull-market run is shared by everyone — the genuine skill is visible in the relative performance, not the absolute.

Consistency of method, not consistency of outcome. An operator who explains clearly why their approach should work across multiple environments, and who has made specific forward predictions that came true, is a better candidate for genuine skill than one who simply shows a clean chart with no underlying explanation.

Two Heights, One Regression

The classic illustration from genetics: two unusually tall parents tend to have children who are taller than average, but shorter than themselves. Two short parents tend to have children who are shorter than average, but taller than themselves.

This is regression to the mean operating on heritable traits. The extreme parental heights combine genetic predisposition with favorable random developmental factors. The children inherit the genetic predisposition — the skill component — but draw independently from the developmental factors — the variance component. The result is movement toward the population mean relative to the extreme parental observation.

No one finds this surprising in genetics. The same mechanism is genuinely surprising to most people when applied to business performance, investment returns, or athletic achievement — even though the structure is identical.

Why This Is Humbling

The uncomfortable application: my own best periods are probably a combination of skill and favorable variance. The periods when everything I touch seems to work contain a component that is not reproducible because it was random.

Taleb's self-described discipline: after a good period, be suspicious of the methodology that produced it. Ask honestly: how much of this was the approach, and how much was a favorable environment that made the approach look better than it is? Press what was genuinely working; identify and discount what was favorable variance.

The same discipline applies in reverse: after a bad period, don't entirely abandon an approach that is structurally sound. Some of the underperformance may be unfavorable variance drawing from the same distribution that produced good results before. The next period's draw is independent.

Regression to the mean is fundamentally a lesson in humility about extreme outcomes: they are more likely to revert than to continue, and the next observation will tend toward the center of the distribution, not toward a new extreme.

For the full framework, read Fooled by Randomness: How Luck Masquerades as Skill.