Mean Reversion
Mean reversion is the tendency of an asset's price or other financial variable to drift back toward its long-run average after deviating from it, forming the basis of contrarian trading strategies.
Mean reversion is the statistical tendency of a series, such as a stock price, a spread between two assets, or a valuation ratio, to return toward its historical average over time. When the variable rises far above its mean, a mean-reverting view expects it to fall; when it drops far below, the view expects a rebound. This contrasts with trend-following: instead of betting that recent moves continue, a mean-reversion strategy bets that extreme moves reverse. Common implementations include pairs trading (going long the cheaper of two correlated assets and short the richer one), buying oversold stocks flagged by oscillators like RSI, or fading prices that pierce the outer Bollinger Bands.
Whether a series actually mean-reverts is an empirical question, not a guarantee. Some series are mean-reverting (many spreads, short-term interest rates, realized volatility), while others, like broad equity index levels, trend upward over the long run and revert only over shorter horizons or after dislocations. Practitioners test for reversion with tools such as the Augmented Dickey-Fuller test or by estimating the half-life of a deviation. The central risk is that an apparent deviation is not noise but a regime change: a cheap stock can keep getting cheaper because its fundamentals deteriorated, turning a reversion bet into a losing position that never comes back.
For investors, mean reversion matters because it shapes both entry timing and risk management. It explains why mechanically buying dips sometimes works and sometimes is ruinous, and why position sizing and stop-loss discipline are essential when the assumed average shifts. It also underlies a large family of quantitative strategies and is one of the two classic return drivers studied alongside momentum.
On hedgewing.ai, mean-reversion structure is captured implicitly rather than hard-coded. The platform's 45 engineered features include oscillator and band-based signals (such as RSI distance and Bollinger position) that encode how stretched a price is relative to recent norms, and the four-model ensemble of LSTM, GRU, TCN, and Transformer networks learns from data when those stretches tend to revert versus persist. Because regimes shift, every model is validated with nightly walk-forward backtesting so a reversion edge is measured strictly out-of-sample rather than assumed.
Related terms
Momentum (Investing) · Bollinger Bands · RSI (Relative Strength Index) · Feature Engineering
Back to the hedgewing.ai glossary · See AI stock forecasts