Mean Reversion

Mean reversion is the family of trading strategies built on a single assumption: when price moves unusually far from a reference value, it tends to travel back toward it. The reference — the "mean" — can be a moving average, a rolling median, a VWAP, the previous close, a statistical fair value derived from a related instrument, or simply the middle of a recent range. Whatever the anchor, the trade has the same shape: buy weakness, sell strength, and exit when price has come back, rather than when a move has extended.

That makes it the structural opposite of trend following. A trend system needs the current move to continue; a mean reversion system needs it to stall and reverse. Both can work on the same instrument because they harvest different behaviours: trend systems tend to take a small number of large wins from persistent moves, while mean reversion systems tend to take many smaller wins from short-lived overshoots — and carry the risk of the overshoot that does not come back.

## How a mean reversion trade is built

Almost every version answers three questions. **What is the mean?** An anchor and a lookback: a short moving average, a long-term average used as a fair-value line, yesterday's close, a rolling weekly bar, or the spread against a correlated market. **How far is far enough?** A stretch measure: distance in percent, ATR multiples, standard deviations (Bollinger bands, z-score), a short-lookback oscillator reading such as RSI-2 or Williams %R, a count of consecutive lower closes, or a pure price condition like a close below the prior N-day low. **When is the trade over?** An exit: a touch of the anchor, a fixed number of bars, an opposite oscillator reading, the first profitable close, or a trailing stop.

Change any one of those three and you get a materially different system, even when two videos carry the same title.

## Main variants

**Oscillator-based** versions use a short-period indicator to define "stretched" and often exit on the mirror condition. **Statistical / band-based** versions frame the same idea in units of volatility — z-scores or standard-deviation bands — which adapts the entry threshold as volatility changes. **Price-action versions with no indicators** rely on closes, ranges and consecutive-bar counts, which removes indicator parameters but not the underlying choices.

Beyond the entry signal, versions split by scope. **Single-instrument** systems trade one index, ETF or future. **Portfolio or cross-sectional** systems rank a universe and buy the most oversold names, so the "mean" is relative to peers rather than to the instrument's own history. **Spread and pairs** versions revert a relationship between two instruments instead of a single price. **Commodity** versions often lean on inventory, seasonality or term structure as the economic reason a price should return. There are also **exit-only modules**: a mean reversion rule used to close positions opened by a different system, and **timed-exit** variants that replace the reversion target with a fixed holding period.

## What typically differentiates implementations

The exit rule is usually the biggest differentiator, and it is the part most often glossed over in a video. After that: the presence or absence of a regime filter (for example, only taking longs while price is above a long-term average); whether the system is long-only or symmetric, since the short side of an equity index rarely behaves like a mirror image of the long side; the stop-loss policy; whether entries scale in on further weakness; and the maximum holding period. Instrument choice matters as much as logic — the same rules applied to an equity index, a single commodity and an FX pair are effectively three different strategies.

## Common mistakes

Running without a regime filter, so the system keeps buying into structural declines. Using tight stops, which cut off exactly the adverse excursion the strategy is designed to absorb and can invert the edge. Grid-searching entry thresholds until one number looks exceptional. Testing only on a period dominated by a single regime. Ignoring transaction costs and slippage, which matter disproportionately when holding periods are short and per-trade edges are small — and assuming fills at extreme prices during fast moves. Reading a smooth equity curve as low risk: many mean reversion systems have negative skew, with frequent small gains and rare large losses. On stock universes, ignoring survivorship and delisting bias.

## How to evaluate and backtest a version

Define the anchor, threshold and exit before testing, and test exit variants separately so you know which component carries the result. Model costs and slippage explicitly. Look at distributions rather than averages: holding period, largest loss, adverse excursion, skew. Check parameter sensitivity — a broad plateau is more credible than an isolated spike. Segment results by regime (bull/bear, high/low volatility) and check out-of-sample and on related markets: a real edge usually degrades gracefully rather than disappearing. Two baselines are useful: buy-and-hold on the same instrument, and a fixed-bar exit replacing the strategy's exit logic. If the elaborate exit does not beat the timed exit, it is not adding what it claims to.

The versions decoded on this page cover that whole span — index ETFs and futures, individual commodities, portfolios of stocks, indicator-driven and price-only entries, and exit-focused modules. Each strategy page lists the specific rules, so you can compare the choices side by side.

Strategies in this concept (48)

Frequently asked questions

How is mean reversion different from "buying the dip"?

Buying the dip is a discretionary impulse; a mean reversion strategy is the same idea made testable. It specifies the reference value, a quantified threshold for how far price must deviate before entering, an exit condition, and a risk rule. Without those four elements you cannot backtest the idea, measure its cost sensitivity, or know whether a losing streak is normal behaviour or a broken assumption.

Which markets tend to mean revert?

Behaviour differs by asset class rather than being universal. Broad equity indices and index ETFs have historically shown short-term reversion tendencies, partly attributed to liquidity and rebalancing flows. Individual commodities can revert around supply, demand and inventory dynamics, on their own timescales. Single stocks revert but carry idiosyncratic risk, which is why portfolio versions spread that risk across a universe. The same rules ported to a new market should be re-tested, not assumed.

Should a mean reversion strategy use a stop loss?

It is a genuine trade-off, not a settled question. The strategy's premise is that price will move further against you before reverting, so a stop placed inside that expected excursion will systematically exit before the edge materialises. Removing the stop, however, leaves the position exposed to a move that never reverts. Common compromises are a wide volatility-based stop, a maximum holding period, or a position size small enough that the worst historical excursion is survivable. Test each variant separately.

What timeframe works best for mean reversion?

There is no single answer, but the timeframe determines which frictions dominate. Intraday versions generate many trades, so commissions, spread and slippage can consume the per-trade edge. Daily and swing versions hold through overnight gaps and news. Weekly or rolling-weekly versions produce fewer trades, which makes statistical validation harder because the sample is smaller. Choose the timeframe first, then judge the results against the cost and sample-size constraints it imposes.

Why do mean reversion backtests often look better than live results?

Several reasons compound. Entries occur during fast adverse moves, where realistic fills are worse than the historical price used in the test. Short holding periods amplify the impact of costs. Threshold parameters are easy to over-optimise because small changes shift results noticeably. And many test periods are dominated by one market regime, which flatters strategies with a long bias. Modelling costs conservatively and testing across regimes closes most of the gap.

Can mean reversion be combined with trend following?

Yes, and the two most common forms are different. One is a filter: use a long-term trend condition to decide whether mean reversion entries are allowed at all, which reduces trades taken against a structural decline. The other is portfolio combination: run both systems on separate capital, since their return streams often behave differently in the same conditions. Evaluate the combination as a portfolio — correlation of returns and combined drawdown — not by averaging individual results.

All strategy concepts · Explore all strategies