Mean reversion is the idea that a financial time series (price, returns, yields, valuation ratios, etc.) tends to move back toward its long‑term average (its “mean”) after deviating from it. Large deviations from the mean raise the probability that future observations will move back toward that mean. Traders and investors use this idea to time entries and exits: buy when a price is far below its historical mean, sell (or short) when it is far above.
Source: Investopedia — “Mean Reversion” (Sydney Saporito).
Key concepts and formulas
– Mean (sample average): Mean = (Sum of observations) / N
– Deviation for an observation: Deviation = Price − Mean
– Sample standard deviation: SD = sqrt( Sum[(Price − Mean)^2] / (N − 1) )
– Z‑score (how many SDs an observation is from the mean): Z = (Price − Mean) / SD
Interpretation: common thresholds are |Z| ≥ 1.5–2 to flag potential over/under‑valuation (overbought if Z ≳ +2, oversold if Z ≲ −2).
Why it’s useful
– Identifies potential entry/exit points based on statistically extreme moves.
– Underpins many technical indicators (moving averages, Bollinger Bands, RSI, stochastic oscillators).
– Works well in range‑bound markets and for securities with stable mean behavior (e.g., some FX pairs, interest rates, mean‑reverting commodities).
Practical step‑by‑step: how to build a basic mean‑reversion trading strategy
1. Define objective and horizon
• Decide whether you’re day trading, swing trading (days–weeks), or investing (months–years). This determines your lookback period and execution style.
2. Choose the universe and timeframe
• Pick liquid assets (high liquidity reduces execution cost).
• Common choices: major FX pairs, large-cap equities, ETFs, interest rates, volatility indices.
• Shorter horizons → shorter lookback (e.g., intraday 5–60 min); longer horizons → longer lookback (e.g., 20–200 daily bars).
3. Gather and clean data
• Get historical price series (close, midprice, intraday ticks if needed).
• Remove errors/outliers, ensure consistent timestamps and corporate actions (splits, dividends) adjusted if using equities.
4. Compute the mean and volatility
• Choose a lookback window N (e.g., 20 bars for a swing strategy).
• Compute Mean_N and SD_N using formulas above. Optionally use exponential moving average (EMA) for a “faster” mean.
5. Compute the Z‑score (or normalized deviation)
• Z_t = (Price_t − Mean_N_t) / SD_N_t.
• Use Z to rank how “extreme” current price is relative to recent history.
6. Define entry and exit rules (example)
• Entry: Buy when Z ≤ −2; Short when Z ≥ +2.
• Exit: Close when Z reverts to 0 (price near mean), or when Z crosses a smaller threshold (e.g., ±0.5).
• Alternate: Use Bollinger Bands (middle band = moving average, upper/lower = ±k·SD) — buy at lower band, sell at middle/upper band.
7. Risk and position sizing
• Use fixed fractional sizing or volatility‑adjusted size (e.g., risk 0.5–1% of account per trade).
• Set stop loss: absolute price level, volatility multiple (e.g., 3·SD), or time‑based (exit if no reversion in X bars).
• Consider maximum simultaneous exposure across correlated positions.
8. Backtest and evaluate
• Backtest over multiple market regimes (trending, range‑bound, high/low volatility).
• Evaluate metrics: CAGR, Sharpe, max drawdown, hit rate, average win/loss, and time to mean.
• Use out‑of‑sample and walk‑forward testing to avoid overfitting.
9. Implementation and monitoring
• Add transaction costs and slippage to backtest.
• Monitor strategy performance in live trading with small capital / simulated paper trading before scaling.
• Recalibrate lookbacks, thresholds, or filters if performance degrades.
Hypothetical numeric example
– Asset: daily close price. Lookback N = 20 days.
– Suppose Mean_20 = $100, SD_20 = $2.50. Today’s price = $94.
– Z = (94 − 100) / 2.5 = −2.4 → substantially below the mean.
– Rule: Buy on Z ≤ −2; exit when Z ≥ −0.2 or after 10 trading days.
– Position sizing: risk 1% of account; stop‑loss at $91 (3·SD below mean), take‑profit near mean ($100).
This trade assumes the price will revert toward $100; if a structural change altered fundamentals, reversion may not occur.
Mean reversion in different trading styles
– Day trading: use intraday means (VWAP, short EMAs), RSI and stochastic for overbought/oversold signals, and Bollinger Band squeezes for mean pullbacks. Execution speed and slippage matter; some use algorithmic/HFT approaches.
– Swing trading: use longer MAs (20–50 days), daily Z‑scores, and combine with pattern/context (support/resistance).
– Forex: many major currency pairs show mean‑reverting tendencies over certain windows; use price vs. moving average or carry/interest rate differentials as additional mean signals.
Which assets and timeframes work best?
– Best timeframes: no universal “best.” Shorter timeframes produce more signals but higher noise and transaction costs; longer timeframes reduce noise but slower signals. Choose based on your trading costs, execution, and capital.
– Best assets: liquid instruments with history of range behavior—major FX pairs, large‑cap stocks/ETFs, bond yields, some commodities. Assets in strong structural trends (fast‑growing tech stocks, new market leaders) can break mean behavior.
Mean reversion vs trend‑following
– Mean reversion assumes prices return to a long‑term average; you buy dips and sell rallies. Profits come from pullbacks.
– Trend‑following assumes price moves persist in the direction of the trend; you join and ride strong moves (buy breakouts, ride winners).
– They are complementary: mean reversion struggles in trending markets; trend followers struggle in choppy, mean‑reverting ranges.
Benefits of mean reversion
– Clear, statistically grounded signals (Z‑scores, Bollinger Bands).
– Often high hit rate when markets are range‑bound.
– Intuitive risk control (distance from mean maps to stop sizing).
Limitations and risks
– Not all deviations revert—some represent changes in fundamentals or regime shifts; prices can trend away from the historical mean for long periods.
– Whipsaw and “false mean reversion” during trending markets can cause repeated small losses.
– Model risk: choice of lookback, using sample mean vs. EMA, and non‑stationarity of data can all produce poor results.
– Transaction costs and slippage can erase small mean‑reversion profits, especially on short horizons.
Practical tips and improvements
– Combine mean‑reversion signals with filters: trend filter (avoid mean‑reversion trades when long‑term trend is strong), volume confirmation, or macro/fundamental checks.
– Use volatility‑adaptive sizing: smaller positions when volatility is higher or when Z is marginal.
– Consider pair trading / statistical arbitrage: trade the spread between two correlated assets that diverge, expecting the spread (not the absolute price) to revert.
– Regularly re‑estimate parameters and test across regimes — avoid “set it and forget it.”
Common indicators used for mean reversion
– Moving averages (simple and exponential)
– Bollinger Bands (MA ± k·SD)
– Relative Strength Index (RSI) and stochastic oscillator (overbought/oversold)
– VWAP (intraday mean)
The bottom line
Mean reversion is a widely used concept in finance and technical trading: price and return series often oscillate around a historical average. It provides actionable rules (z‑scores, Bollinger Bands, RSI) to identify potentially profitable countertrend trades. However, its success depends heavily on timeframe, choice of asset, market regime, robust risk management, and realistic accounting for costs. Always backtest thoroughly and use risk controls because what appears as an “extreme” can be the start of a new trend rather than a reversion.
Primary source for this article
– Investopedia, “Mean Reversion” (Sydney Saporito)
Editor’s note: The following topics are reserved for upcoming updates and will be expanded with detailed examples and datasets.