Strategy style
Mean-reversion futures strategies: what the evidence says
A mean-reversion strategy bets that a move which has gone too far will partly reverse: it sells strength and buys weakness, usually winning often and losing large when it loses. The idea is simple to state and easy to fit to history, which makes the tests that separate a real reversion from a lucky one unusually important.
What the research record says about fading
The textbook reads are all on the screener's menu, faded as well as straight, because the question keeps being asked: RSI overbought, a close near the outer Bollinger band, price stretched from VWAP, the gap fill, yesterday's move reversed. Across 270 variants of familiar chart studies, none carried an edge on its own on the index contracts.
Levels drawn from the same prices they are meant to predict fared no better. Fading the edge of the prior session's volume-profile value area showed no edge at any zone definition tested, from the value-area high and low to the point of control.
Averaging into a losing position, the reversion trader's instinct, lost to a constant position at matched risk in every configuration tested. It also looks profitable on shuffled data, because deferring a loss turns it into an open position a finite sample may never close, so the only fair test of a grid is against its own shuffled-bar placebo.
Where reversion does appear
Markets differ in character. Natural gas tends to mean-revert where the index contracts trend, which is why a fade that looks promising on NG can look nothing like its twin on ES.
Reversion rules are also the style most often conditioned on a regime, such as calm markets or thin participation, and the record's warning there is specific. The state has to be measured from information available before the trade, from prior sessions or another market. A state computed from the same bars the strategy trades makes any range-sensitive rule look better in volatile buckets by construction, and the effect survives on shuffled data.
Testing a mean-reversion idea
The screener tests a reversion idea as you state it, from an oscillator such as RSI, Stochastic, Williams %R or CCI, Bollinger or Keltner bands and z-scores and the distance from VWAP, each against a level you set, on any of the 46 markets and on bars from one minute to a week, with a target at a band or the mean, a stop and a time limit.
From the research record
- SettledFading the edge of the volume-profile value areaThe reversion premise did not hold at any zone definition, and the break needs an independent tape to confirm on.
- SettledAveraging into a position on a price gridA constant position at matched risk beat every scale-in scheme tested, and the shuffled-bar placebo shows why.
- SettledFiltering a signal by a state variable from the same barsRegime conditioning works when the state is exogenous and lagged; measured on the same bars it becomes arithmetic.
- Works with conditionsAdding a filter to a strategy that already worksA filter that keeps half the trades needs a 41% lift in average R to draw level, which is the bar to design against.
Mean reversion questions
- Does mean reversion work in futures?
- Sometimes, on some markets, at some times of day; not as a general rule. None of 270 textbook oscillator and band readings screened carried an edge on its own on the index contracts, while natural gas tends to revert where the indices trend. The specific rule has to be tested, after costs, against a random-entry control.
- Why do mean-reversion strategies have high win rates?
- Because they take profit on a small snap-back and give the occasional loser a wider stop. A high win rate with a larger average loss is the style's shape, and it is why one missed stop can erase many small wins.
- Is RSI mean reversion profitable on ES or NQ?
- Not on its own, on the research record. RSI 14 and RSI 2 were among the 270 chart-study readings screened on the index tapes, faded and straight, and none carried an edge by itself. Combined with a signal from outside the chart, such as market breadth, it is a different question and one the screener can test.
- Should I average down on a mean-reversion trade?
- The record says no. Every averaging and pyramiding configuration tested lost to a constant position at matched risk, and the ones that seemed to win were carrying more risk. A grid also profits on shuffled data, so beating zero proves nothing; it has to beat its own placebo.
Screen a mean-reversion idea
Describe it in a sentence. The desk turns it into an exact specification and screens it through seven gates written down before the run, from real per-fill costs to a drift null and an era split. The verdict usually comes back in 15 to 25 seconds.