Mean Reversion Strategies

Mean reversion strategies bet that prices, spreads, or ratios that have deviated from their historical equilibrium will return to that level — profiting from the 'snap back.' From pairs trading that exploits temporary divergences between correlated securities to statistical arbitrage across entire factor portfolios, mean reversion strategies generate consistent, low-correlation returns in range-bound environments but face sharp losses when trends persist longer than expected.

Level: AdvancedPart VII - Algorithmic & Quantitative InvestingPublished Deep Guide

The Statistical Foundation of Mean Reversion

Mean reversion requires that the target variable is stationary — it has a stable long-run mean around which it fluctuates, rather than trending indefinitely. Individual stock prices are not stationary (they trend over time). But price ratios between cointegrated pairs, deviations from fair value in ETF arbitrage, or spreads between near-identical securities (like on-the-run vs. off-the-run Treasury bonds) can be stationary. The Augmented Dickey-Fuller (ADF) test and the Hurst exponent are statistical tests for determining whether a series is stationary and mean-reverting vs. trending.

Cointegration (Engle-Granger or Johansen tests) identifies pairs of non-stationary series that move together in the long run — their spread is stationary even though each individual series is not. Two competing airlines' stocks might be cointegrated (driven by the same fuel costs, passenger demand, and economic cycle) even though neither is individually stationary. When the spread between them widens beyond historical norms, the mean reversion bet is to buy the underperformer and short the outperformer.

Pairs Trading and Statistical Arbitrage

Pairs trading enters a long/short position when the spread between two cointegrated securities exceeds a threshold (typically 2 standard deviations of the spread's historical distribution) and exits when the spread returns to the mean. The entry z-score (spread expressed in standard deviations from mean) triggers entry; exit at mean or at a stop-loss when the spread continues to diverge beyond the entry level. The strategy is market-neutral (simultaneous long and short), generating returns from relative performance rather than market direction.

Statistical arbitrage (stat arb) extends pairs trading to portfolios of cointegrated securities — often hundreds or thousands of pairs simultaneously. Quantitative hedge funds (Renaissance Technologies, Two Sigma, D.E. Shaw) run stat arb at massive scale, extracting tiny alpha from thousands of simultaneous mean-reversion bets whose uncorrelated errors diversify toward a smooth return stream. The edge decays as more capital chases the same signals — the research challenge is continuously finding new cointegrated relationships before they become crowded.

Risk Management in Mean Reversion Strategies

The defining risk in mean reversion is regime change: the relationship that was mean-reverting becomes permanently broken. An airline pair that was cointegrated might diverge permanently if one airline goes bankrupt (fundamentally different from temporarily underperforming). Mean reversion stop-losses must be set at levels that distinguish 'extended but recoverable' from 'fundamental relationship change' — typically 3-4 standard deviations or a time-based exit (close after N days if the spread hasn't converged).

Mean reversion strategies have characteristic return profiles: many small gains (when spreads revert normally) punctuated by occasional large losses (when relationships break). The Sharpe Ratio looks attractive in calm periods; the maximum drawdown and tail risk emerge during crises when correlations that were stable for years break down simultaneously across many pairs. The 2008 financial crisis caused catastrophic losses in many stat arb strategies as multiple historically stable pairs diverged simultaneously during the liquidity crisis.

Key Takeaways

- Mean reversion requires stationarity — prices themselves are non-stationary, but cointegrated pairs (their spread) can be stationary and reliably mean-reverting.
- Pairs trading: buy the underperformer, short the outperformer when their spread exceeds 2 standard deviations — exit at the mean; stop-loss at 3-4 standard deviations.
- Statistical arbitrage scales pairs trading to hundreds of simultaneous pairs, diversifying idiosyncratic errors into a smooth return stream.
- Primary risk: regime change breaks the cointegration relationship permanently — the strategy must distinguish 'extended but recoverable' from 'fundamentally broken.'
- Crisis correlation: mean reversion strategies suffer large simultaneous losses when multiple historically stable relationships break down in liquidity crises (2008 being the canonical example).

→ See this concept in live AIQ stock signals

Concept FAQs

Is mean reversion or trend following better?

Neither is universally better — they perform best in different market regimes. Mean reversion dominates in low-volatility, range-bound markets where prices oscillate around equilibrium. Trend following dominates in high-volatility, directionally moving markets. Professional systematic traders often combine both: using mean reversion for short-term positioning and trend following for longer-term exposure, providing more consistent performance across different market environments than either strategy alone.

Does mean reversion apply to individual stock selection?

At the stock level, short-term mean reversion (1-5 days) is well-documented: stocks that decline sharply tend to bounce slightly over the following days as liquidity providers absorb the selling pressure. This short-term reversal effect is the basis for some high-frequency strategies. For fundamental mean reversion (valuation ratios returning to historical averages), the time horizon is months to years rather than days — the mechanism behind the value premium (cheap stocks eventually re-rate to fair value).

In AIQ

See RSI, MACD, and trend structure live The concepts covered in this guide are the exact factors AIQ surfaces for every stock — apply them with live data rather than in isolation.

NVDA Technicals →