Martingale

Mechanism

After each losing trade the EA multiplies the next position size by a fixed factor, most commonly 2.0, and re-enters. A cycle ends when a trade closes at a profit, which resets the size to base. Because the cycle is defined to end on a winner, the reported win rate is a property of the definition rather than a measurement of the rule. What varies is the depth of the ladder before the winner arrives: replayed over the closed-trade lists published on this site, the longest run is fourteen losses, which puts the recovery entry at 16,384 times base size.

Suitability

There is no market state in which the ladder is safe, only states in which it has not yet been tested. The common advice to prefer range-bound or mean-reverting symbols is not supported by the records here: the mean-reversion EAs published on this site carry the two longest losing runs in the catalogue — 11 and 10 against a trend-following maximum of 9 — and under an uncapped ladder two of the six mean-reversion records lose the account while none of the five trend-following records do. A step cap bounds the loss without removing it, and raising the cap from four to five turned two surviving records into failed ones. Appropriate only for a trader who has computed the worst-case loss and can lose the account in full without consequence.

Notes

Martingale is the only strategy on this list that names its position-sizing rule instead of its entry. After a losing trade the EA doubles the next position, so that one winner repays every loss in the run plus the original target. The signal that opens the trade is close to incidental — a moving-average cross, a grid level, a coin toss — because the ladder is the product.

That makes the claim unusually easy to audit. A doubling rule can be laid over any list of closed trades, including trades that were never sized that way. So this page does not argue with martingale in the abstract. It replays the ladder over 9,339 closed trades from every expert advisor published on this site, with exits dated 2 January 2019 to 31 March 2026, and reports what the ladder did to records that are all profitable at a flat position size. The runs and testing conditions behind those trades are published under the site’s methodology.

How it works: the ladder is a definition, not an edge

The rule is one line. Start at base size. After a loss, multiply the next position by a fixed factor, normally 2.0. On a win, reset to base. Everything between two resets is one cycle.

The arithmetic that follows is fixed. After k consecutive losses the ladder has staked 1 + 2 + 4 + … + 2^(k−1) = 2^k − 1 units of average loss, and the next entry is 2^k units. Nine losses in a row is a 512-unit entry against 511 units already lost. Fourteen — the deepest run in the records below — is a 16,384-unit entry.

Two consequences are worth separating, because the sales material blends them.

The first is that the win rate stops being a measurement. A cycle is defined to close on a winner, so every completed cycle is a winning cycle. The only cycle that is not a winner is the one during which the account, or the step cap, runs out. Quoting a martingale’s win rate is quoting the definition back.

The second is that the smoothness is real, not an artefact. Small, frequent recoveries accumulate while the ladder stays shallow, so almost all of the distribution sits in one place and the rest in a tail that arrives once. This is the same shape as a grid EA without a basket stop, and for the same reason: both answer an adverse move by adding exposure rather than realising a loss.

What separates them is where the risk is set. Grid spacing decides how often the ladder deepens; the multiplier decides how much each step costs. An EA that does both is not a grid with a feature — it is a martingale with a schedule.

What the ladder does to the published records

Every figure below comes from our own data — the closed-trade lists on each EA page — replayed in exit order with a 2.0 multiplier from the base size each record was tested at, 0.1 lots except Orrery at 3 on JP225. Scaling a position by X scales that trade’s result by X, ignoring slippage, swap and margin, all of which make the real outcome worse.

EA	Symbol / TF	Trades	Win rate	Payoff	Worst losing run	Lot the ladder reaches
Beacon	AUD/CAD M15	384	79.17%	0.32	3	0.8
Peregrine	BTC/USD H4	163	78.53%	0.33	3	0.8
Zerqon	US30 H4	232	78.45%	0.42	2	0.4
Cairn	EUR/USD H1	4,583	64.02%	0.83	14	1,638.4
Lanternfish	USD/JPY M5	407	62.65%	0.75	3	0.8
Thunderhead	EUR/GBP H1	392	61.48%	0.96	9	51.2
Windrose	AUD/CAD H1	101	59.41%	0.90	4	1.6
Orrery	JP225 H1	195	55.90%	1.01	4	48.0
Iridescence	EUR/USD H1	872	51.83%	1.30	8	25.6
Tessera	USD/JPY H1	624	50.64%	1.25	8	25.6
Lattice Weave	GBP/AUD H1	492	46.34%	1.74	11	204.8
Tidewell Slack	GBP/JPY M15	303	40.26%	2.00	10	102.4
Kestrel Hover	USD/JPY M15	284	30.99%	2.95	9	51.2

The win rate does not tell you how deep the ladder goes. Lanternfish wins 62.65% of its trades and never loses more than three in a row, so its ladder tops out at 0.8 lots. Cairn wins more often — 64.02% — and has a fourteen-loss run, which puts its recovery entry at 1,638.4 lots. The two records are 1.37 percentage points apart on the column buyers sort by, and 2,048 times apart on the position the ladder demands. The worst losing run is the number that decides a martingale, and it is not predictable from the win rate.

Four of the fourteen accounts do not survive. Running the uncapped ladder on the 10,000 deposit each record was tested with, the deposit is fully consumed on Lattice Weave at trade 33 of 492, on Tidewell Slack at trade 22 of 303, on Cairn at trade 171 of 4,583, and on Thunderhead at trade 150 of 392. That is 6.11, 6.78, 2.29 and 14.36 months into records that run for years. A margin call would arrive earlier still, since the position is closed when margin fails, not when equity reaches zero.

Two of the four dead accounts finish showing a profit. The replay is deliberately generous: it keeps trading after the deposit is gone. Lattice Weave ends at 795,625.94 and Tidewell Slack at 207,902.78 — both from accounts that were already finished. This is not a quirk of the method. It is the single most important thing to know about martingale backtests, because a tester that never models the stop-out reports the performance of an account that no longer exists.

The multiplier the ladder would actually need

There is a second failure that arrives before ruin, and it is rarely stated: a doubling ladder does not always repay the run it was built for.

The condition is arithmetic. After k losses the ladder has staked 2^k − 1 units of average loss and the recovery entry is 2^k units, which returns 2^k × payoff. So the run is repaid only when the payoff ratio — average win divided by average loss — clears 0.5 at the first step, 0.75 at the second, 0.875 at the third, and approaches 1.0 as the run lengthens. Doubling is the right multiplier only for a rule whose average win already matches its average loss.

Eight of the fourteen published records have a payoff ratio below 1.0, and none of them can repay its own worst run by doubling. Beacon, Peregrine, Ballast and Zerqon fail at the very first step: their payoff ratios of 0.32, 0.33, 0.37 and 0.42 mean a doubled entry returns less than the single loss it is covering. Lanternfish and Cairn stop repaying at the third consecutive loss, Windrose at the fourth and Thunderhead at the fifth.

The pattern underneath is the uncomfortable one. Read the fourteen by win rate and the multiplier the ladder would need falls almost in order — from 4.08 on the 79.17% record to 1.34 on the 30.99% one, with three inversions, none wider than 0.13. High win rates come from targets placed close to the stop, which is exactly the arrangement that leaves the recovery trade too small. Martingale is drawn to high-win-rate rules, and those are the rules where the doubling is undersized.

Market conditions: what the ladder needs and never gets

The common advice is to run martingale on range-bound or mean-reverting symbols, since price returns and the recovery arrives. These records do not support it. The mean-reversion EAs published here carry the two longest losing runs in the catalogue — 11 and 10 against a trend-following maximum of 9 — and under an uncapped ladder two of the six mean-reversion records lose the account while none of the five trend-following records do. A symbol that reverts on average still trends occasionally, and the ladder only needs the exception.

The honest framing is that no market state makes the ladder safe. There are states in which it has not yet been tested, and a test period that ends before the worst run arrives will report exactly that.

Parameters and settings in MT5

Input	Where the risk sits	What it controls
LotMultiplier	Any value above 1.0	Cost of each step; 2.0 is standard and is too small for a payoff ratio under 1.0
MaxSteps	Must be finite	Caps the ladder depth, and therefore the size of the loss it hands you
StepDistance	From the symbol’s own volatility	How quickly a move deepens the ladder
BasketTakeProfit	Net across open legs	Closes the cycle; a small target is what manufactures the win rate
EquityStopPercent	The one real circuit breaker	Closes everything at a defined loss instead of at the stop-out
MaxSpreadPoints	Just above normal spread	Blocks the scaled entries that arrive in the worst conditions

Three MT5 realities decide whether the ladder behaves as designed:

• The recovery order can simply be refused. Position size is bounded by SYMBOL_VOLUME_MAX and quantised by SYMBOL_VOLUME_STEP. A ladder that computes a 1,638.4-lot entry will be rejected or silently clamped, and a clamped recovery cannot repay the run it was sized for. Read those two properties and check the ladder top against them before anything else.
• Stop-out happens on margin level, not on equity. ACCOUNT_MARGIN_LEVEL falling through the broker’s stop-out threshold closes positions from the largest first, which is the recovery trade. The cycle is terminated at its most expensive point, and negative balance protection determines whether the damage stops at zero.
• Backtest fidelity matters more here than for any other strategy. The whole record depends on a handful of deep-ladder trades. Run real tick data, read the modelling quality, and treat a result built on 1-minute OHLC as untested.

Failure modes: how a martingale EA ends

• The run outlasts the account. The dominant path, and the one in the records above. Losses arrive in a block because the same conditions that beat one entry beat the next.
• The step cap converts ruin into a large realised loss. This is an improvement, not a solution. With a four-step cap the deepest drawdown still exceeds flat-lot trading on eleven of the fourteen records — by a median of 2.66 times, and by 33.65 times on Cairn.
• Raising the cap makes it worse. Going from four steps to five produced fewer failed cycles but roughly twice the loss on each. Cairn’s deficit grew from 61,978.44 to 122,460.62, and Kestrel Hover and Thunderhead went from surviving the record to having the deposit consumed.
• The recovery is too small to recover. A payoff ratio below 1.0 means the ladder is undersized at its own arithmetic, before any market view enters.
• Costs are charged on the scaled position. Slippage, spread and swap all scale with the ladder, so the deep entries — the ones the whole strategy depends on — are the most expensive trades in the record.
• The backtest reports a dead account’s profits. Without a modelled stop-out, the equity curve continues past the point where trading would have ended.

How to test a martingale EA in mt5depot before it costs anything

The useful test is not whether a martingale EA makes money — over most samples it will. It is whether the ladder it needs fits inside the account you would fund. That takes two numbers: the entry signal’s worst losing run, and its payoff ratio.

Martingale versus grid

	Martingale	Grid
What sets position size	The previous trade’s outcome	Fixed at every level
What it needs to work	A winner before the ladder outruns the account	Oscillation through the same levels
Exposure growth	Exponential in the losing run	Linear in the level count
Reported win rate	Guaranteed by the cycle definition	High, from small per-basket targets
Worst case	Full loss of the account	Full loss of the account, reached more slowly
The input to read first	LotMultiplier	MaxLevels and the basket stop

The two are sold as different products and are frequently the same one. A grid with a lot multiplier above 1.0 is a martingale whose step distance is set by the spacing, and it inherits the exponential exposure without the grid’s linear worst case. The trend-following contrast is starker: a trend EA’s bad period is a long series of ordinary losses that leaves the account intact. A martingale’s bad period is one run, and patience does not survive it. Read the multiplier before the equity curve, on every EA that has one.