thetrading.tools
Statistical propertylive data through 2026-08-21

Mean Reversion

Direct definition

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.

In plain English

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.

Site observation · S&P 500 (SPY) vs its 200-session average
SPY closed 8.2% above its 200-session average

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
Measurement: Daily SPY closes vs their trailing 200-session simple average, 1993+; percentile ranks the current deviation against every prior session. The forward-return studies use the pipeline's signal computations with a 20-session re-trigger cooldown. Raw JSON →
+8.2%
SPY vs its 200-session average (2026-08-21)
72nd
Percentile of that stretch, all sessions since 1993
95
Straight sessions above the average
708
The 200-session average itself (last close 765.72)
-39.5-9.520.51995200020052010201520202025+8.2%
SPY's distance from its trailing 200-session average, 1994 through 2026-08-21. The zero line is the average itself — the "mean" this page's concept reverts to. Every 15%+ stretch and every crossing is retained; quiet stretches are sampled. Note how long the series can stay on one side: the mean pulls slowly and on no schedule.

Tested: 5-Day Losing Streak (61 occurrences since 1993, last 2025-08-01)

Five consecutive sessions with close < previous close. Triggers on the fifth.
1 week
+0.6%
Win rate
67%
All-days med
+0.3%
All-days win
58%
1 month
+2.7%
Win rate
72%
All-days med
+1.3%
All-days win
64%
3 months
+4.1%
Win rate
70%
All-days med
+3.3%
All-days win
70%
12 months
+13.6%
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.
1 week
+0.6%
Win rate
69%
All-days med
+0.3%
All-days win
58%
1 month
+3.2%
Win rate
77%
All-days med
+1.3%
All-days win
64%
3 months
+6.4%
Win rate
92%
All-days med
+3.3%
All-days win
70%
12 months
+16.8%
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

Stretch vs a reference average
dev(t) = P(t) / MA_n(t) − 1

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.

Z-score of the current reading
z = (x − μ) / σ

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.

Half-life of reversion (AR(1) model)
HL = −ln(2) / ln(φ)

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.

  1. 1z = (130 − 100) / 12 = +2.5 — a two-and-a-half sigma stretch versus the past year.
  2. 2If the oscillator historically mean-reverts, the expectation for coming readings is drift back toward 100, and the trade is to fade the extreme.
  3. 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.
A +2.5σ reading, an explicit reversion hypothesis, and a test that either supports or kills it.

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

  1. 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.

  2. 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.

  3. 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.

  4. 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.

  5. 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.

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.

  • SPY signal-study methodologyThe Trading Toolsprimary

    Rules, cooldowns and forward-return computation behind the oversold studies quoted on this page.

Machine-readable object

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.