Mean Reversion Is 140 Years Old. It Wasn't Invented for the Stock Market.
Everything about Francis Galton, De Bondt and Thaler, and the origin of pairs trading below is general public history — documented in academic papers and widely reported financial journalism — not a claim tested against our own data. The backtest numbers in the second half of this piece are ours, from our own archive, and we're keeping that line clearly drawn the same way we always do.
Every "buy the pullback" strategy, ours included, rests on an idea that's about 140 years old. It wasn't invented for the stock market. It was discovered by a Victorian scientist measuring how tall people's children turned out to be.
Francis Galton, 1886: height, not stocks
In 1886, the English polymath Francis Galton published a paper called "Regression Towards Mediocrity in Hereditary Stature." He'd been measuring the heights of parents and their adult children, expecting tall parents to produce tall children and short parents to produce short children in roughly the same proportion. That's not what he found. Very tall parents tended to have children who were tall, but on average shorter than they were. Very short parents tended to have children who were short, but on average taller than they were. Each generation drifted back toward the population's average height — not perfectly, not immediately, but reliably.
Galton called this "regression toward mediocrity." The name stuck as "regression," and it's the root of the statistical technique — linear regression — taught in every introductory stats course today. Galton wasn't thinking about markets at all. He was trying to understand heredity, and stumbled onto a pattern that shows up anywhere a measurement is part signal and part noise: extreme readings are disproportionately made of noise, and noise doesn't repeat, so the next reading tends to land closer to the average.
Wall Street catches up: De Bondt and Thaler, 1985
It took about a century for that idea to get formally tested on stock prices. In 1985, economists Werner De Bondt and Richard Thaler published "Does the Stock Market Overreact?" in the Journal of Finance — now one of the most-cited papers in behavioral finance. They sorted stocks into "winners" and "losers" based on their prior three-to-five-year returns, then tracked what happened next. The losers — the stocks that had fallen the most — went on to outperform the winners by a wide margin over the following three to five years. The market, they argued, overreacts to both good and bad news, pushing prices further than the underlying fundamentals justify, and then drifts back. Thaler would go on to win the 2017 Nobel Prize in Economic Sciences for this line of work on behavioral finance.
The trading-desk version: Morgan Stanley, mid-1980s
Around the same time academics were formalizing the idea, a quantitative trading group at Morgan Stanley was building a strategy on essentially the same premise, independently of the paper. Industry accounts credit Gerry Bamberger, working under Nunzio Tartaglia's group, with the original insight behind what became known as pairs trading: find two stocks that normally move together, and when they diverge further than usual, bet on them snapping back into line. It's mean reversion applied to a spread between two related assets instead of one stock's own price history. The desk reportedly produced enormous profits for Morgan Stanley for a period in the mid-1980s, and pairs trading — and the broader field of statistical arbitrage it helped launch — has been a fixture of quantitative trading ever since.
What this actually means for a score based on a 20-day average
Our own entry score is a modern, narrow application of the same idea Galton described: a stock's price is compared to its own 20-day moving average, and the further it's pulled back, the higher the score — up to a point. That "up to a point" is the part naive mean reversion skips, and it's worth being precise about, using our own numbers.
We backtested 1,991 historical Buy Zone entries (score ≥79), split into quartiles by how far each one had run over the prior 60 trading sessions before the entry:
| Prior 60-session return | Stopped out before target 1 |
|---|---|
| -73% to -10.6% (biggest declines) | 81.6% |
| -10.5% to +2% | 70.5% |
| +2.1% to +14.7% | 65.3% |
| +14.8% to +213.4% (biggest run-ups) | 60.8% |
That's a straight line, and it runs the opposite direction from what "buy what fell the most" would predict. The entries that had fallen hardest — the purest, most Galton-style reversion candidates — stopped out most often. The entries with the mildest pullback inside an existing uptrend held up best. De Bondt and Thaler's three-to-five-year losers-beat-winners effect is a real, well-documented phenomenon, but it's measured over years, on a portfolio of many stocks, with survivorship and rebalancing assumptions built in. It says nothing about whether a single stock down 70% in three months is a good trade this week. Those are different claims wearing the same name.
This is why our score anchors to a 20-day average instead of a multi-year low, and why the checklist looks for a stock that's merely pulled back — not one that's collapsed. Mean reversion is real. It's just a much narrower, more conditional edge than "it fell, so it should bounce" — and 140 years of the idea's own history, from Galton's height data through Q1 of our own backtest, says the same thing: the version of reversion worth trading is the mild one, not the extreme one.
Scores in this piece come from the ZonedInOut model. How the score works →
Frequently asked
What is mean reversion in investing?
Who discovered mean reversion?
Does buying stocks that have fallen the most actually work?
How is ZonedInOut's score related to mean reversion?
Get the Sunday Brief
One email a week, before the open: what the board did, which calls worked, which didn't, and what changed. Free. No spam, unsubscribe anytime.