Mean Reversion

Mechanism

The EA measures how far price has deviated from a moving average or Bollinger centre, usually in standard deviations or ATR units, and confirms the stretch with an oscillator in extreme territory. Entry is against the move; the target is the mean, and the stop sits beyond the level that would invalidate the reversion thesis. The distance between those two exits fixes the record. A target close to the entry produces a high win rate and a payoff ratio well below 1; a target at the full mean produces the reverse. The six mean-reversion EAs published here take 1.5 to 5.45 trades a month and hold their median winner for 48 to 187 hours.

Suitability

Best where the pair spends most of its time inside a range and returns to a definable centre — historically the CHF crosses. The mean-reversion EAs published on this site run on AUD/CAD, USD/JPY, GBP/AUD, GBP/JPY and EUR/GBP instead. Structurally unsuited to momentum currencies during trend phases, when the stretch keeps stretching and a counter-trend position is held against a move that does not return. A trend filter such as ADX below 25 is close to mandatory. Suitable for a trader who can accept a long losing run — the worst in these records is eleven consecutive losses — and who reads the stop distance before the win rate.

Notes

Mean reversion bets on elastic rather than momentum: that a price stretched far from its average is more likely to snap back than to keep stretching. The EA waits for a statistically significant deviation — typically 1.5 to 2.5 standard deviations from a moving average or Bollinger centre. It confirms the stretch with an oscillator in extreme territory, then enters against the move and targets the mean itself.

That describes the entry, and the entry is the part buyers understand. The description usually bolted onto it — mean-reversion systems win most of their trades and lose big occasionally — is a claim about numbers, not about mechanism. That means it can be checked instead of repeated. Checked against our own data, the six mean-reversion EAs published on this site, it holds for two of them. This page covers how the machine is built inside MetaTrader 5, what those records show about the shape of its results, and how to stress-test one before any capital is exposed.

How it works: the entry is a distance, the exit is the record

The entry side is a measurement, and the three common constructions are close to interchangeable:

• Deviation from a moving average. Price a set number of ATR units or standard deviations from a mean, treated as a statement that the move has over-extended. Cheap, and the whole strategy in one line.
• Bollinger touch or close beyond the band. The same idea with the deviation calculated for you. A close beyond the band is stricter than a touch, and it changes trade count more than win rate.
• Oscillator extreme with a regime filter. RSI or Stochastic past a threshold, gated by an ADX reading low enough to say the market is not trending. The filter is the difference between a mean-reversion EA and one that fades trends until it is stopped out of all of them.

The exit is where the records are actually made, and there are two of them: a target at or near the mean, and a stop beyond the level that invalidates the thesis. Expectancy per trade is win rate × average win − (1 − win rate) × average loss. That is arithmetic, not a finding — but the useful consequence is that the terms stop being independent once the two exit distances are fixed. Pull the target close to the entry and most trades reach it before noise takes them out: the win rate rises and the average win shrinks below the average loss. Push it out to the full mean and fewer trades survive the trip: the win rate falls and the payoff ratio climbs above 1.

Both are mean reversion; both enter against a stretch. They produce records that look like different strategies, and the numbers below are what that looks like in practice.

What the published mean-reversion records show

Every figure in the table is computed from the closed-trade lists published on each EA page — 1,994 trades in total, from records that begin in 2019 and run to March 2026. The full runs, parameters and testing conditions are published under the site’s methodology.

EA	Symbol / TF	Trades	Win rate	Payoff (avg win ÷ avg loss)	Profit factor	Median hold, winners ÷ losers	Worst losing run
Beacon	AUD/CAD M15	384	79.17%	0.32	1.25	0.41×	3
Ballast	EUR/GBP H1	307	78.50%	0.37	1.40	0.49×	3
Lanternfish	USD/JPY M5	407	62.65%	0.75	1.29	2.91×	3
Windrose	AUD/CAD H1	101	59.41%	0.90	1.31	0.56×	4
Lattice Weave	GBP/AUD H1	492	46.34%	1.74	1.50	2.66×	11
Tidewell Slack	GBP/JPY M15	303	40.26%	2.00	1.35	1.62×	10

Four things follow from it, and the folklore predicts none of them.

The high win rate is not a property of the strategy. The six span 40.26% to 79.17%, and not one sits inside the 65-75% band the strategy is usually sold with. Their median, 61.03%, sits above the median of the trend-following EAs published here (55.90%). Whatever separates a mean-reversion EA from a trend EA, it does not show up in the column most buyers sort by.

Win rate and payoff ratio are the same dial read from two ends. Order the six by win rate and the payoff ratio orders itself perfectly in reverse. A 79.17% win rate pairs with a payoff of 0.32, then 78.50% with 0.37, 62.65% with 0.75, 59.41% with 0.90, 46.34% with 1.74, and 40.26% with 2.00. Six out of six, no exceptions. This is not a discovery about markets. It is what happens when the target moves relative to the stop — and the reason a win rate quoted without its payoff ratio carries almost no information.

What the dial does not move is the profit factor. The payoff ratios differ by a factor of six across the group. The profit factors are 1.25, 1.29, 1.31, 1.35, 1.40 and 1.50 — the whole group inside a spread of 0.25. And the highest win rate of the six still produces the lowest profit factor of the six, while the rest of the ordering has no pattern at all. Placing the target closer buys a smoother-looking record and pays for it in size; placing it further does the opposite; neither converts into a materially better rule.

Reverting is what the winners do slowly. The strategy is described as taking quick profits and holding losers, which would put the median winning hold below the median losing hold. It is true of three EAs out of six. Beacon holds winners 0.41× as long as losers, Ballast 0.49× and Windrose 0.56×, as advertised. The other three hold winners 1.62× to 2.91× longer, and the group median is 1.09× against the trend group’s 1.10× — with six records the two groups are indistinguishable on this axis, and the half that holds winners longer is exactly the half whose target sits at the full mean.

Market conditions: when mean reversion pays and when it does not

The bad environment is not a losing market — it is a directional one, the exact mirror of how trend-following fails in a range. That symmetry is the honest argument for looking at the two together rather than as substitutes: they lose money in different weather. It is also why the losing runs cluster. Eleven consecutive losses is not eleven unrelated mistakes; it is one trending stretch met by the same rule eleven times.

What caused the deviation matters as much as its size. A price stretched by thin liquidity in a quiet session is a candidate for reversion. A price stretched by a rate decision has been repriced, and there is no old mean for it to return to. No indicator separates the two, which is why a news filter matters more here than for a strategy entering with the move.

The same elastic assumption drives statistical arbitrage, which trades the spread between two correlated instruments rather than one instrument’s deviation from its own mean. The thesis is identical — applied to a relationship instead of a price — and so is the failure mode: it breaks precisely when the relationship stops holding. On a single pair such as EUR/GBP, the relationship assumed is the pair’s own history with its average.

Parameters and settings in MT5

Input	Typical starting point	What it controls
MA_Period / BB_Period	Long enough that the mean is a level, not a lagging copy of price	What the EA treats as the value price should return to
Deviation (σ or ATR multiple)	1.5–2.5σ, or a volatility-scaled equivalent	How far price must stretch before the entry is valid — the trade-count dial
RSI_Period / thresholds	Confirmation only, not the trigger on its own	Filters stretches that are still accelerating
ADX_Max	Around 25, tested rather than assumed	Suppresses entries in trends — the single most important input on this page
TakeProfit (distance to mean)	The mean itself, or a fraction of the distance to it	Together with the stop, this sets win rate and payoff ratio
StopLoss	Beyond the level that would invalidate the reversion	The line between mean reversion and an accidental martingale
MaxSpreadPoints	Just above the symbol’s normal spread	Extremes and wide spreads arrive together; this blocks the worst fills
OncePerBar	true for bar-close logic	Stops one extended bar from firing the same signal repeatedly

Three MT5-specific realities decide whether a promising backtest survives contact:

• The stop has a minimum distance, and this strategy wants a wide one anyway. Read SymbolInfoInteger(symbol, SYMBOL_TRADE_STOPS_LEVEL) and clamp to it; treat SYMBOL_TRADE_FREEZE_LEVEL the same way for modifications near price. A stop the broker silently rejected turns a defined-risk trade into an open-ended one.
• Spread widens exactly where this strategy enters. The extremes it fades often occur in thin conditions. A test run on average spread overstates the edge of an EA whose entries cluster where spread is worst. Test with realistic or real-tick spread, and read the modelling quality before the profit.
• Holds are longer than the strategy’s reputation implies. The median winner in these records is held 48 to 187 hours. That makes SYMBOL_SWAP_LONG and SYMBOL_SWAP_SHORT a real line item rather than a rounding error, and triple-swap day differs by broker.

Failure modes: how a mean-reversion EA loses money

• The stretch keeps stretching. The normal way this strategy loses. Every entry is against a move that does not return, and because the rule keeps re-qualifying, the losses arrive in a block. The eleven-loss run in the table is this failure mode, not a broken EA.
• The mean moves. A pair that has repriced has a new average, and the old one is no longer a level. The EA keeps measuring distance from a number that has stopped meaning anything.
• The stop is removed to avoid the loss. Widening a stop, or replacing it with an average-down, converts a defined-risk strategy into a grid with a good story. The equity curve improves immediately and the tail risk becomes unbounded.
• The win rate is treated as a margin of safety. At a 79% win rate with an average loss three times the average win, the strategy is solvent only while the hit rate holds.
• Cost is measured on the wrong bar. Entries cluster at extremes where spread is widest, and multi-day holds pay swap. Both are invisible in a test configured with fixed average spread and no financing.
• Parameters fitted to a range-bound sample. A deviation threshold optimised over a period without trends will be the wrong one in the next. Neighbouring values producing very different results is the tell that the number was fitted rather than chosen.

How to build and stress-test a mean-reversion EA in mt5depot

You do not need to write MQL5 to assemble the machine described above. A deviation measure, an oscillator confirmation, an ADX gate, a target at the mean and a stop beyond the extreme are all blocks in the no-code builder. The reason to build it yourself: the test that matters is not “does it make money”. It is the comparison between two target placements on one entry rule, and no vendor page runs that comparison for you.

Mean reversion versus trend

	Mean reversion	Trend-following
What it assumes	A stretched price returns to its average	An established direction continues
Entry relative to the move	Against it, at an extreme	With it, after confirmation
Loses when	The market trends and the stretch keeps stretching	The market ranges and moves fold back
Typical exit	Fixed target at or near the mean	Trailing stop or signal reversal
Worst case	One position, or a run of them, held against a move that does not return	A long series of ordinary losses
Published win-rate median	61.03% (six EAs)	55.90% (five EAs)
Published worst-run median	3.5, longest 11	4, longest 9
Complexity	Intermediate — the two exit distances carry the risk	Beginner — few inputs, hard to sit through

Neither is safer in general, and their failure modes are close to complementary: one needs stillness, the other needs movement. That is a real argument for running both, with one caveat the table cannot show. A mean-reversion EA and a trend EA with a fixed take-profit are not as different as their labels suggest, because both then depend on price stopping where the rule says it should. Check what each exit actually does before assuming that holding two EAs has diversified anything.