Mean Reversion
Mean reversion is the tendency of a series that has moved far from its historical average to subsequently move back toward it. In markets it is a property some series have at some horizons, never a law: volatility, spreads and oscillators revert strongly, while price levels mostly trend. Every mean-reversion trade is a bet that the average the series is being measured against still describes it.
A very tall parent tends to have children shorter than themselves — closer to the average. That is mean reversion in its original sense: extreme readings tend to be followed by less extreme ones.
In markets, the idea becomes a trade: when something has fallen unusually hard or stretched unusually far from its typical level, bet on the snap back toward normal. Buying a stock after five straight red days is a mean-reversion bet. So is selling volatility after a panic spike.
The catch is the word "normal." Reversion only works if the old average still applies. A stock down 80% is not automatically due for a bounce — sometimes the business changed and the old average is simply gone. Telling stretched-but-normal apart from changed-forever is the entire skill, and this page shows how we test it with data instead of assuming it.
- Category
- Risk & returns
- Entity type
- Statistical property
- Also called
- reversion to the mean, mean reversion trading, mean-reverting, reversion trade
- Last reviewed
- 2026-08-14
Current observation
This dated measurement is an instance of the concept, not the concept itself. It updates when the verified source dataset changes; the as-of date below is the freshness contract.
The last close (765.72) sits 8.2% above the 200-session average of 707.55 — the 72nd percentile of all deviations since 1993, after 95 straight sessions on the high side. The distance is the mean-reversion setup people trade; whether it reverts is the evidence question the studies below answer.
- Deviation vs 200-day
- +8.2%
- Percentile since 1993
- 72nd
- Sessions above the average
- 95
- Last 5-day losing streak
- 2025-08-01
Tested: 5-Day Losing Streak (61 occurrences since 1993, last 2025-08-01)
Five consecutive sessions with close < previous close. Triggers on the fifth.| Horizon | Median after signal | % positive | All-days median | All-days % positive |
|---|---|---|---|---|
| 1 week | +0.6% | 67% | +0.3% | 58% |
| 1 month | +2.7% | 72% | +1.3% | 64% |
| 3 months | +4.1% | 70% | +3.3% | 70% |
| 12 months | +13.6% | 77% | +12.4% | 79% |
- Win rate
- 67%
- All-days med
- +0.3%
- All-days win
- 58%
- Win rate
- 72%
- All-days med
- +1.3%
- All-days win
- 64%
- Win rate
- 70%
- All-days med
- +3.3%
- All-days win
- 70%
- Win rate
- 77%
- All-days med
- +12.4%
- All-days win
- 79%
Tested: RSI Oversold Thrust (13 occurrences since 1993, last 2026-04-09)
RSI(14) < 30 within last 20 sessions. Yesterday: RSI < 60. Today: RSI >= 60.| Horizon | Median after signal | % positive | All-days median | All-days % positive |
|---|---|---|---|---|
| 1 week | +0.6% | 69% | +0.3% | 58% |
| 1 month | +3.2% | 77% | +1.3% | 64% |
| 3 months | +6.4% | 92% | +3.3% | 70% |
| 12 months | +16.8% | 92% | +12.4% | 79% |
- Win rate
- 69%
- All-days med
- +0.3%
- All-days win
- 58%
- Win rate
- 77%
- All-days med
- +1.3%
- All-days win
- 64%
- Win rate
- 92%
- All-days med
- +3.3%
- All-days win
- 70%
- Win rate
- 92%
- All-days med
- +12.4%
- All-days win
- 79%
Signal medians and win rates come from the pipeline's SPY signal studies (20-session re-trigger cooldown); the all-days columns are every overlapping window of the same length since 1993, computed from the same closing prices. Occurrences cluster in declines and long-horizon windows overlap, so rows are tendencies on a structurally rising index, not independent samples. The comparison to make is signal vs all-days, never signal vs zero.
Why it matters
Half of trading folklore is a mean-reversion claim in disguise: "buy the dip," "overbought," "oversold," "due for a bounce." Naming the assumption lets you test it — and the tests disagree by horizon, which is the single most useful fact about the subject.
The empirical record is horizon-shaped. Short horizons (days to a few weeks) show reversal in index returns; intermediate horizons (roughly 3 to 12 months) show the opposite — momentum; multi-year horizons show reversion again in the academic record. A strategy that is right about the direction but wrong about the horizon still loses.
Some series revert by construction and some only by regime. An oscillator bounded between 0 and 100 must come back; a volatility index tethered to a long-run range usually does; a price level or a nominal aggregate can trend for decades. Knowing which kind of series is in front of you decides whether "stretched" is a signal or a description.
Calculation and identification
Distance from an n-period moving average, in percent. Our live reading below uses SPY against its 200-session average — the most-watched version of this measurement.
How many standard deviations the reading sits from its mean over a stated window. A z of ±2 marks roughly a 1-in-20 extreme if the series is well-behaved; fat-tailed market series breach it far more often than that.
If a series follows x(t+1) = φ·x(t) + noise with φ between 0 and 1, the half-life is how long a deviation takes to decay halfway back. Useful as a speed estimate; the model itself is an assumption to disclose.
Worked example
Reading a two-sigma stretch
Suppose an index oscillator prints 130 when its trailing one-year mean is 100 with a standard deviation of 12.
- 1z = (130 − 100) / 12 = +2.5 — a two-and-a-half sigma stretch versus the past year.
- 2If the oscillator historically mean-reverts, the expectation for coming readings is drift back toward 100, and the trade is to fade the extreme.
- 3Check the assumption before the trade: pull every prior reading above +2σ and compute what actually followed. If forward outcomes after past extremes were no better than average, the stretch is a description and carries no edge.
The number 130 alone says nothing — the same reading is a fade in a stable regime and a trend confirmation in a shifting one. The historical follow-through of comparable extremes is the evidence; the extremity itself proves nothing.
Where it can mislead
- 01
The mean can move. The most expensive mean-reversion mistake is averaging into a series whose average has changed — bank stocks in 2008 looked "cheap versus their mean" the whole way down. Reversion logic assumes a stable regime; regime breaks are exactly when it fails hardest.
- 02
Prices and returns behave differently. Bounded oscillators (RSI, percent-above-average measures) revert by construction; index price levels mostly trend and only their short-horizon returns show reversal. "The market always comes back" is a claim about a structurally rising index, not evidence that any given stretched reading must close.
- 03
Horizon decides the sign. Reversal in days-to-weeks, momentum in months — the academic record (De Bondt-Thaler on multi-year reversal, Jegadeesh-Titman on 3-12-month momentum) puts both effects in the same market at different clocks. A reversion entry held into the momentum window fights the stronger documented effect.
- 04
The payoff shape is short-vol. Fading extremes typically wins often and small, and loses rarely and large — the rare occasions when the extreme keeps extending are precisely the crises. Win rates flatter these strategies; the tail does the damage.
- 05
Our forward-return studies use overlapping history and a structurally rising index. Signal occurrences cluster in bear markets, long-horizon windows overlap, and SPY's baseline drift is positive — so "beats baseline" is the claim to check, never "was positive."
Relationships
Concept-to-concept edges are typed and reciprocal. Tools, manuals, signals and datasets are separate resource nodes that measure, explain or operationalize the concept.
A drawdown is the stretch a mean-reversion trade tries to fade: buying inside one bets the decline reverts toward the old average rather than marking a regime change — and drawdown's recovery asymmetry prices what failure costs.
The cleanest mechanical reversion setup we track: five straight red closes on SPY, with every occurrence and forward return since 1993.
The oscillator version: SPY's RSI stretched to an oversold extreme, graded against the same forward-return framework.
The deeper stretch: what followed each 10% decline from an all-time high — reversion logic applied to corrections.
A cross-sectional cousin: its beaten-down regimes historically mean-reverted to the best forward returns — the contrarian U this page's logic predicts.
The closing-price series behind the live 200-day-average reading and the studies' baselines.
Frequently asked questions
Is the S&P 500 stretched above or below its long-term average right now?
As of the 2026-08-21 close, SPY sits 8.2% above its 200-session average — the 72nd percentile of all readings since 1993, after 95 consecutive sessions on the high side. The stretch is a measurement. It is not a signal by itself; the studies on this page show what followed comparable setups.
What is mean reversion in trading?
A strategy family that bets a stretched reading — a price far below its moving average, an oscillator at an extreme, a spread far from its norm — will move back toward its historical average. Every version of it assumes the old average still describes the series; testing that assumption per series and horizon is what separates a strategy from a slogan.
Do stock prices actually mean revert?
By horizon. Index returns show short-term reversal (measured in days to a few weeks) and some multi-year reversion in the academic record, but 3-to-12-month horizons show momentum — the opposite. Individual stocks are less reliable than indexes because a single business can change permanently, taking its old average with it.
What indicators are used for mean reversion?
Distance from a moving average, RSI and similar bounded oscillators, Bollinger-band position, and z-scores of spreads or ratios. All of them measure the stretch; none of them establish that the stretch tends to close. The follow-through record of comparable past extremes is the part that carries evidence.
Is buying the dip a mean-reversion strategy?
Yes — it bets that a short-term decline reverses toward the trend rather than continuing. On the S&P 500 the computed record on this page has leaned in its favor at short horizons since 1993, with the honest caveats attached: occurrences cluster in bear markets, the index's baseline drift is positive, and the rare failures were large.
Sources, provenance and machine access
- Does the Stock Market Overreact?De Bondt & Thaler, Journal of Finance (1985)primary
The foundational evidence for multi-year return reversal — long-term losers outperforming long-term winners.
- Returns to Buying Winners and Selling LosersJegadeesh & Titman, Journal of Finance (1993)primary
The counter-evidence: 3-to-12-month momentum, the horizon where reversion logic runs backward.
- Mean Reversion in Stock Prices: Evidence and ImplicationsPoterba & Summers, Journal of Financial Economics (1988)
The classic study of transitory components in stock prices across horizons.
Rules, cooldowns and forward-return computation behind the oversold studies quoted on this page.
Stable ID: https://www.thetrading.tools/concepts/mean-reversion#term. Dated observations have their own IDs and point back to this term; they never overwrite its definition.