What it is
Mean reversion is the tendency of a variable to drift back toward its historical average over time, after deviating from it. A variable is mean-reverting if high observations are followed by lower ones and low observations are followed by higher ones, on average, with the magnitude of reversion proportional to the deviation.
The concept was first formalized by Francis Galton in the late 19th century, who observed that the sons of unusually tall men were taller than average but shorter than their fathers — a phenomenon he called "regression to mediocrity." The statistical pattern he identified is mathematically equivalent to what we now call mean reversion.
Where mean reversion is real — and why
Real mean reversion requires a mechanism. Without a forcing process that pushes a variable back toward its mean, there is no reason to expect reversion.
Corporate profit margins: Companies in competitive industries see their above-average margins eroded over time by new entrants attracted by those margins. Companies with below-average margins are pressured to cut costs or exit. Over decades, margins across industries tend to converge toward the cost of capital. The mechanism is competition.
Interest rates: Central banks target inflation, and interest rates are the primary instrument. When rates are very low, monetary policy eventually normalizes; when they are very high, growth slows and rates are cut. Real interest rates have shown mean-reverting behavior over long cycles. The mechanism is policy response.
Valuation multiples: When markets trade at extreme multiples — very high P/E during bubbles, very low during crises — fundamentals eventually reassert themselves. The mechanism is the relationship between price and future earnings growth: extremely high valuations imply earnings growth that is difficult to sustain indefinitely.
Where mean reversion is not real
Secular trends do not mean-revert. Industries structurally disrupted by technology do not see their business models "mean-revert" to prior profitability. Kodak's margins did not mean-revert after digital photography arrived. Retail store foot traffic has not mean-reverted after e-commerce. These are directional changes, not cyclical deviations.
Stock prices do not reliably mean-revert over short or medium horizons. If they did — if you could predict that a stock at $50 will return to its "mean" of $70 — you would be predicting future prices, which efficient markets should prevent. The random walk hypothesis specifically says there is no mean to revert to for individual stock prices. Studies have found weak evidence of long-horizon reversion (5-10 year horizons) but it is difficult to trade profitably.
The gambler who says "I've had a streak of losses, so a win is due" is applying mean reversion where there is no mechanism. If each spin of a roulette wheel is independent, past outcomes provide no information about future outcomes — the mechanism for mean reversion does not exist.
Galton's regression and the regression fallacy
Galton's original insight points to a subtle statistical trap: regression to the mean is a mathematical property of any imperfect correlation, not necessarily a causal mechanism.
When two variables are imperfectly correlated, the extreme observations of one variable will, on average, be followed by less extreme observations of the other. This is pure statistics — it would happen even if there were no causal forcing process at all. Galton's tall fathers have tall sons, but the sons are less extreme than the fathers, purely because height is not perfectly inherited.
In markets, this means some apparent "mean reversion" in prices is a statistical artifact of imperfect autocorrelation, not a causal forcing mechanism. Distinguishing genuine economic mean reversion (competitive pressure, policy response) from statistical regression-to-the-mean is essential before building a trading or investment strategy on it.
One thing most people get wrong
Mean reversion is commonly assumed to be a universal property of financial variables — that after a big move, prices will "bounce back." This is usually wrong.
The question to ask is always: what is the mechanism? What force is pushing this variable back toward its mean? For corporate margins in competitive industries, the mechanism is clear and well-documented. For the S&P 500 one week after a decline, the mechanism is unclear, the evidence is thin, and the pattern could easily be explained by statistical noise or bid-ask bounce.
Assuming mean reversion without a mechanism is a form of anchoring: you have a number in mind (the prior level), and you expect prices to return to it because you remember it. The market has no such memory. Only mechanisms create mean reversion; history alone does not.