Mean Reversion

Mean Reversion

A mean-reversion strategy bets that a price stretched far from its recent average will snap back toward it -- like a rubber band that has been pulled and released. Where a trend follower buys strength and expects it to continue, a mean-reversion trader buys weakness and expects it to reverse. Both are betting on what happens next; they just disagree about which way the odds lean.

This article covers the classic mean-reversion setups, the regime filter that keeps them out of trouble, the strategy's characteristic failure mode, and what to watch when you backtest one.

The Opposite of Trend Following

Mean reversion inverts the trend-following profile. A reversion trade targets a modest, well-defined move -- the distance back to the average -- so wins are small and frequent. The losses are the problem side. A mean-reversion entry buys while price is falling, and nothing about the entry logic limits how much further it can fall. When the snap-back does not come, the loss keeps growing until something else -- a stop, an exit rule, or the trader's nerve -- forces the position closed. The result is the mirror image of a trend system: many small wins, punctuated by occasional large losses.

Classic Mean-Reversion Setups

Bollinger Band Bounces

Bollinger Bands draw a moving average with an envelope two standard deviations above and below it, so the bands themselves define what "stretched" means -- statistically far from the recent average, by the market's own recent volatility. The bounce setup: enter when price closes below (or touches) the lower band, exit when it returns to the middle band. The middle band is the mean being reverted to, which makes the exit unusually well-defined for a trading rule.

RSI Extremes

RSI compresses recent gains versus losses into a 0-100 scale, with readings below 30 conventionally labeled oversold and above 70 overbought. The reversion setup buys when RSI drops below 30 and exits when RSI recovers to its midline around 50 -- or, more aggressively, holds until it reaches 70. The same idea runs on RSI's cousins: Stochastic below 20, CCI below -100, and the other oscillators each draw their own version of the "too far, too fast" line.

One mechanism to understand before trusting any oversold signal: oversold is a statement about the recent past, not a floor under the price. RSI below 30 says recent candles were dominated by losses. It does not bound how much further price can fall, and in a strong downtrend RSI can pin below 30 for weeks while price keeps sliding. Oversold becomes more oversold.

The Short Side

Every setup above has a mirror: price above the upper band, RSI above 70, and a short position betting the stretch resolves downward. Backtesting both directions separately is worth the effort -- markets do not behave symmetrically, and a bounce setup that works long may test very differently short on the same instrument.

Buy the Dip -- But in Which Market?

The single most common upgrade to a mean-reversion system is a regime filter: only take oversold longs when the bigger picture is still healthy, for example when price remains above its 200-period moving average. The reasoning follows directly from the failure mechanism above. In an uptrend, a dip to oversold is a pullback inside a rising structure -- the snap-back has the trend's help. In a downtrend, the same signal is an invitation to catch a falling knife: the "stretch" is not an anomaly, it is the trend itself.

This is one instance of a general technique -- gating one indicator's signal on another indicator's context -- covered in depth in Combining Indicators.

Pairs Trading

The professional world's favorite mean reversion does not revert a price -- it reverts a relationship. Pairs trading finds two instruments that historically move together (two oil majors, two exchange-listed banks), waits for the gap between them to stretch unusually wide, then shorts the outperformer and buys the underperformer. The bet is that the gap closes, regardless of where the overall market goes. The "mean" is the pair's historical relationship, and the position is insulated from market-wide moves because the long and short legs cancel them out.

The Failure Mode: One Big Loss

Mean-reversion risk concentrates in the tail. The equity curve grinds upward through dozens of small wins, then a position that never reverts hands back weeks of profit at once. On the metrics page this shows up as a max drawdown that looks out of proportion to the day-to-day results, and as ugly tail-risk numbers -- CVaR exists for exactly this shape of strategy.

The obvious fix -- a tight stop-loss -- carries a built-in tension here. A stop below a mean-reversion entry triggers at exactly the point where the entry logic reads the setup as even more attractive: price is further from the mean than when you bought. The stop is not wrong -- it is the only thing that caps the falling-knife scenario -- but placed too tight, it systematically sells the bottoms of dips that were about to revert. Alternatives that limit tail risk without fighting the entry logic: a time-based exit that closes any trade still open after a set number of candles (if the snap-back has not happened by then, the thesis is stale), and smaller position sizing so that the inevitable outlier is survivable. Backtest the variants rather than debating them -- the trade-off shows up clearly in the numbers.

Costs Eat Small Edges

Mean reversion trades often and profits little per trade, which makes it far more sensitive to trading costs than slow-moving strategies. Say the average winning trade gains 0.8% and a round trip costs 0.2% in spread and commission -- that is a quarter of the average win handed over before the trade earns anything (0.2 / 0.8 = 25%). A backtest run with zero costs can show a solid edge that vanishes entirely the moment realistic costs are applied. Set commission and spread to match your broker before believing any high-frequency reversion result.

Testing in Simula

Reversion systems generate plenty of trades, so sample size is rarely the problem -- believability of the risk numbers is. Focus on max drawdown, CVaR, and the largest individual losses in the trade list, and make sure the test window includes at least one period where the market fell hard. A reversion backtest that only spans a calm uptrend has never met its own worst case -- the "buy the dip" results are real but the tail risk is invisible until the window includes a market that kept falling.

If you find yourself tweaking the RSI threshold (25? 30? 35?) or the band width to make results look better, that is parameter optimization done by hand -- the systematic version tests the grid for you, and walk-forward analysis tells you whether the winning threshold survives out-of-sample.

Common Mistakes

No regime filter. Buying every oversold signal means buying every stage of every downtrend. Gate the entries on the larger trend.

Trusting the win rate. The strategy's architecture produces high win rates whether or not the edge is real. Expectancy and profit factor are the honest numbers.

Testing without costs. Frequent small wins are exactly the profile that spread and commission erase.

Stops that fight the entry. A very tight stop under a dip-buying entry sells bottoms by design. If the backtest shows most stopped trades reverting right after the stop, widen it, size down, or switch to a time-based exit -- and let the backtest arbitrate.

Where to Go Next

Trend Following -- the opposite bet, and why the same Bollinger Band can serve both strategies.

Combining Indicators -- the regime-filter idea generalized: gating signals on context.

Bollinger Bands -- how the bands are calculated and what band width says about volatility.

RSI -- the oversold/overbought oscillator behind the second setup.

Win Rate & Expectancy -- why high-win-rate systems still need the sizes to work out.

Monte Carlo: How It Works -- stress-testing a strategy whose risk hides in rare losses.