Mean-reversion trading is the family of strategies that bet a price stretched far from a "normal" level (a moving average, a band, a recent return) will snap back toward it. It is the structural opposite of trend/momentum trading: the trend trader buys strength expecting continuation; the mean-reversion trader buys weakness expecting a bounce, and sells strength expecting a fade. Its core tension is that the same extreme reading is both the entry signal and the warning sign — a stock 3 ATRs below its average is "oversold" and a snap-back candidate, but it is also where genuine breakdowns and crashes begin. Mean reversion therefore lives and dies on not being in a stock that keeps going (the falling knife), which makes regime selection and exit discipline more important than the entry trigger itself.

How it's formed

The strategy needs three pieces: a reference level (what price reverts to), a stretch measure (how far it currently is), and a reversion trigger (evidence the snap-back has begun). Common operationalizations:

• Moving-average distance — buy when price is N% or N ATRs below an SMA/EMA (commonly the 10- or 20-day), expecting a return to the average.
• Bollinger Bands (Bollinger) — bands at the 20-day SMA ± 2 standard deviations; a touch/close outside a band flags a statistical extreme, and re-entry inside the band is the reversion cue.
• Short-period oscillators — the canonical retail tool is Connors' 2-period RSI. Per StockCharts ChartSchool, the rules are: trade only with the major trend (price above its 200-day SMA for longs), buy when RSI(2) drops below 5 (more aggressive) or 10, and exit on a move above the 5-day SMA. Connors found returns were higher buying dips below 5 than below 10.
• Statistical / pairs spreads — a cointegrated pair or a residual versus a factor model; the spread (often expressed as a z-score) is the thing that reverts, not the raw price. Pairs trading is the market-neutral institutional cousin of single-name mean reversion.

How it's used in practice

The decision-useful template most short-term mean-reversion traders converge on:

1. Filter the regime first. Buy dips only in uptrends (price above the 200-day SMA, or the broad index above its own); short rips only in downtrends. Mean reversion against a strong trend is the most expensive mistake in the book. Index/ETF mean reversion is far more reliable than single-stock because indices lack idiosyncratic blow-up risk. 2. Demand a real stretch. A Bollinger lower-band tag plus RSI(2) < 5, or a close ~2–3 ATRs below the 10-day SMA. The deeper the stretch (within a healthy trend), the better the documented edge — Connors' own testing showed lower RSI entries outperformed shallower ones. 3. Wait for a reversion trigger, not just an extreme. The single biggest improvement over naïve "buy oversold" is requiring confirmation that the bounce has started: a higher low, a close back inside the Bollinger band, RSI(2) crossing back above its centerline, or a strong up-bar/reversal candle. This filters falling knives. 4. Exit fast, into strength. Mean-reversion edges are short-lived (often 1–5 trading days). Targets are the mean itself (the 10/20-day SMA or the middle Bollinger band), not a new trend. Many systems exit on the first close above a short MA rather than waiting for a price target. 5. Hard stop on thesis failure. A break and hold below the entry low, or a close beyond ~1.5–2 ATRs against the position, says the "extreme" was the start of a new trend. (Note Connors' published RSI(2) variant used no stop, which StockCharts explicitly flags as a drawdown risk — most practitioners add one.)

Time-horizon distinction: short-term/swing mean reversion (days) exploits overreaction and liquidity demand; longer "valuation" mean reversion (months–years, e.g. reversion of P/E or CAPE) is a different, slower phenomenon and should not be conflated with the swing setup.

Standing & evidence

Short-horizon reversal is one of the better-documented effects in finance — but the gap between the academic anomaly and the tradable retail strategy is the key honest point. Lehmann (1990) and Jegadeesh (1990) showed weekly/monthly losers outperform winners the following period — Lehmann's zero-cost weekly contrarian strategy generated a commonly cited ~1.79% per week gross over 1962–1986, and Jegadeesh documented significant negative serial correlation in monthly returns (~2%/month abnormal return to a short-term reversal sort, 1934–1987). De Bondt & Thaler (1985) framed long-horizon reversal as overreaction. However, most subsequent work attributes the short-term reversal largely to liquidity provision and microstructure (bid-ask bounce), not pure overreaction. Critically, Avramov, Chordia & Goyal (2006) and others find the largest reversals concentrated in the smallest, most illiquid stocks, and that a large portion of reversal profits disappears after realistic trading costs — because the strategy demands frequent rebalancing in high-cost names. de Groot, Huij & Zhou (2011) show the cost drag comes mainly from small-caps: restricting to large-caps and lowering turnover leaves roughly 30–50 bps per week net of costs, and Blitz, Huij et al. show residual (factor-adjusted) reversal preserves more of the edge. So: the anomaly is real and robust academically, but the net-of-cost edge for a retail trader is far thinner than gross backtests (the widely circulated "75%+ win rate" RSI(2) figures are gross, single-asset, and exclude slippage). Treat published win rates as upper bounds.

Adoption: heavily used by quant/stat-arb desks (pairs, short-term reversal factors) and by a large retail community via Connors' RSI(2) and Bollinger Bands; less favored by discretionary trend followers, who view it as picking up pennies in front of the falling-knife steamroller.

Strengths & limitations

Works when: the market is range-bound or in a steady uptrend with normal volatility; on liquid indices/ETFs; when high win-rate, short-hold, frequent-trade exposure suits the trader. Reversal strategies also tend to do well in crises by acting as liquidity providers (the academic literature notes this).

Fails when: a real trend or regime change is underway — strong trends turn "oversold" into "more oversold" for days. It is structurally a negative-skew strategy: many small wins, occasional large losses (the bounce that never comes). Volatility expansion (gaps, earnings, macro shocks) destroys it, which is why earnings dates should be a hard exclusion filter.

#1 misuse: running it counter-trend with no stop on single names — buying a stock making new lows because it "must be oversold." That converts the strategy's small edge into uncapped downside. The second most common error is mistaking a gross backtest win rate for a net-tradable edge.

System relevance

This node defines the style; the swing-specific operational mechanics (exact entry/stop/target rules, hold windows) belong in the Swing Trading branch — cross-link rather than duplicate. It pairs directly with the Bollinger Bands and RSI indicator-definition nodes and is the conceptual mirror of the Momentum / Trend-Following style node. For Augustus, the load-bearing caveat is regime-conditioning: a mean-reversion setup should only be surfaced when the regime engine reads range-bound or healthy-uptrend conditions and the name is liquid and earnings-clear — never as a standalone "oversold = buy" signal during a downtrend.

Sources

• StockCharts ChartSchool — RSI(2) (Connors strategy rules, 200-day filter, exits, no-stop caveat).
• Lehmann, B. (1990), Fads, Martingales and Market Efficiency — weekly contrarian reversal (~1.79%/wk gross).
• Jegadeesh, N. (1990), Evidence of Predictable Behavior of Security Returns — short-horizon reversal.
• De Bondt, W. & Thaler, R. (1985), Does the Stock Market Overreact? — overreaction/long-horizon reversal.
• Avramov, Chordia & Goyal (2006), Liquidity and Autocorrelations in Individual Stock Returns; "Another Look at Trading Costs and Short-Term Reversal Profits" (de Groot et al.) — costs erode profits, illiquid/small-cap concentration.
• Blitz, Huij et al. — Short-Term Residual Reversal (liquidity/residual filtering preserves edge).
• QuantifiedStrategies & Quantpedia (Short Term Reversal Effect in Stocks) — landscape/backtests; note gross-of-cost win-rate figures (e.g. "75%+") are upper bounds.
• Bollinger, J., Bollinger on Bollinger Bands — band construction (20-SMA ± 2σ) and band-tag/re-entry use.