What Is Mean Reversion?
Michael Whitman
·
Trading since 2017, full-time since 2020
·
Published 2026-09-05
·
Updated
·
Reviewed
Mean reversion means trading against a move on the expectation that price returns toward its average. It rests on a move tending to be followed by one in the opposite direction, and on the S&P 500 since 1950 that tendency after large daily drops changed sign from one decade to the next.
The other half of the pair. Everything the trend following page assumes, this one assumes the reverse of, on the same market at the same moment. The useful question is not which idea is right in general, but which one a given market has shown, at what speed, and when.
How it works
Price strays from its average, so trade the return.
The edges are where you act and the middle is where you exit. That is the whole shape of it, and it needs a trading range or some other measured idea of “normal” to make any sense.
The premise, stated so it can be tested
Mean reversion bets that a move is more likely to be followed by a move the other way. Trend following bets the reverse. Both are claims about the same series, so the data can settle which one a market has shown, at a given speed, over a given period.
The test needs enough cases to mean anything. The share of next days that closed higher after a drop has a standard error of roughly the square root of p × (1 − p) ÷ n. With 23 drops and a rate near 54%, that is about 10 percentage points, so a result six or seven points from the base rate is well inside the noise.
So the two strategies are not opinions about markets in general. They are opinions about a specific series at a specific resolution, and the same index can give a different answer in different decades, which is what the data below shows.
Where it is applied
A range with edges you can name. Bollinger Bands or a distance from a moving average turn “stretched” into a number, which is the operational form of the idea.
And the target is the average, not the far edge. Taking the middle asks price for a smaller move, while stretching for the other side asks it to cross the whole range against the move that set up the trade.
The shape of the returns
Mean reversion produces many small gains and occasional large losses. The account rises steadily and the method feels correct for a long time.
Trend following is the mirror: many small losses and occasional large gains, which feels wrong almost all the time.
Neither shape is better. But a long run of small gains is evidence of the shape, not of an edge, and the number that matters is what the losses cost when they come.
What makes a market revert
Mean reversion’s story is about supply. A move that ran without new information has left people holding positions at prices they did not intend to pay, and some of them want out, which supplies selling into a rise that had no reason behind it, or buying into a fall.
That predicts something checkable: reversion should be stronger where the move had no news behind it, and weaker where something actually changed. A stock that gapped on an earnings release is a poor candidate; the same stock drifting on a quiet afternoon is a better one. The band or the oscillator says “stretched” without asking why, and the why is what separates a move that comes back from one that has repriced.
A worked example
The S&P 500 in April 2025, closes only. On 3 April 2025 the index closed at 5,396.52, down 4.84% from 5,670.97. A rule that buys the close after a 2% drop buys there.
The next close was 5,074.08, another 5.97% lower, the second drop in a row. Five sessions after the entry, on 10 April, it closed at 5,268.05, still 2.38% below the entry.
Two and a half weeks later the same rule turned out the other way. On 21 April the index fell 2.36% to 5,158.20; the next day it closed at 5,287.76, up 2.51%, and five sessions later it was 7.18% higher.
Same rule, same month, opposite outcomes. That is why the plan has to start from a range drawn from market structure, an entry against the move at its edge, a target at the middle, and a stop beyond the edge that bounds the case where price keeps going.
The original data
What was measured. S&P 500 index daily closes from Yahoo Finance, 3 January 1950 to 2 October 2026, downloaded 3 October 2026. A drop is a close at least 2% below the previous close. For each drop, the next session either closed higher or did not, and the comparison is every day in the same decade. Before 4 March 1957 this series is the earlier 90-stock index, which covers 17 of the 1950s drops.
Over the whole period, there was no reliable bounce. The next session closed higher after 245 of 444 drops, 55.18%, against 53.12% for all 19,308 days (p = 0.385). Drops bunch together in sell-offs, so they are not fully independent cases, and the p values here use the normal approximation to the binomial.
By decade, the answer changed sign.
| Decade | 2% drops | Next day higher | All days | p |
|---|---|---|---|---|
| 1950s | 23 | 60.87% | 54.18% | 0.520 |
| 1960s | 14 | 64.29% | 53.03% | 0.399 |
| 1970s | 30 | 26.67% | 50.55% | 0.009 |
| 1980s | 51 | 37.25% | 52.81% | 0.026 |
| 1990s | 42 | 71.43% | 53.56% | 0.020 |
| 2000s | 146 | 58.90% | 52.25% | 0.107 |
| 2010s | 71 | 59.15% | 54.85% | 0.466 |
| 2020s | 67 | 55.22% | 54.07% | 0.849 |
In the 1970s and 1980s a big down day tended to be followed by another down day, the opposite of reversion: the average next day was −0.43% and −0.41%. In the 1990s it was followed by an up day more often than usual. Since 2000, no decade separates from chance. With eight decades tested, one result under 0.05 could be luck; three, with the sign flipping, look more like a relationship that changed than like a fixed property of the market.
A week later the effect had gone. Counting only drops at least five sessions apart, so no two windows overlap, the share of closes higher five sessions later sat within chance of the everyday five-day rate in every decade (the smallest p was 0.297, in the 1980s). The decade table is published with both measures. For SPY alone, the fade trading page tests the next-day bounce on the fund itself.
In the search study behind this site, 157 of 24,971 videos have mean reversion in the title, at a median of 3,857 views; 93% never pass 50,000.
When it fails
Fading a trend
This failure is not symmetrical with trend following’s. A trend follower in a range loses a little repeatedly; a mean reverter in a trend loses a great deal once. April 2025 above is the shape: the first drop was followed by a bigger one.
The regime changes under you
A rule tested on one decade can meet the opposite in the next. A next-day bounce rule fitted to the 1990s would have been fitted to the one decade where it looked strongest, after two decades where the same signal pointed the other way.
There is no stop
The natural exit is “when price comes back”, which is not a price. A method whose losing case has no defined end is the one that ends an account, and it is an easy mistake here because the winning case is so frequent.
You judged the edge before it broke
The top of a range and the start of a breakout are the same bar. Fading is right in one case and very costly in the other, and the choice has to be made before anyone knows which it is.
Related
Trend following is the opposite bet, and neither page means much without the other.
Trading range is the condition this needs, read from price.
And Bollinger Bands is the usual way of putting a number on “stretched”.
Size a mean-reversion position for the move that does not come back, not for the many that do. A run of small gains can make the method feel safe, and then one move keeps going and takes back more than many small gains added.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.