Does Martingale Work in Trading? Why a 96.9% Win Rate Nets $0

No: martingale (doubling your position size after every loss so that one win recovers everything) changes the shape of your returns, not their expected value, which stays exactly $0 per sequence before costs when you have no edge. An edge does not rescue it: at an illustrative 55% win rate, a five-rung ladder on a $2,000 drawdown buffer has more likely than not breached within 38 sequences, against 1.81% lifetime ruin for fixed $100 risk.

Key takeaway

A martingale ladder does not create profit; it swaps many small wins for one rare loss of the same total size. Computed for an illustrative $100 base and $2,000 drawdown buffer, the ladder survives four straight losses, wins 96.875% of sequences at a 50% win rate and still averages $0 before costs. Optional stopping theory, as Columbia University lecture notes state it, says no stopping rule raises a fair game's expectation, so only a real edge at fixed risk makes money.

What is a martingale strategy in trading?

A martingale strategy increases position size after each loss, classically doubling it, so that the first winning trade recovers every earlier loss plus one base unit of profit. It is a negative progression: size goes up as results go down. A run of trades from the base size to the first win is one sequence, and each size step is a rung.

Three relatives get confused with it:

Grid trading places orders at preset levels around price. FundedNext Futures' help center describes it as "placing multiple long and short orders at predefined price levels to capitalize on market fluctuations". A grid becomes a martingale when the lots grow as price moves against it, which is how some grid EAs (expert advisors, automated MetaTrader strategies) are built.
Averaging down adds to a losing position. With equal adds it is a scaling decision; with growing adds it is the same negative progression.
Anti-martingale does the reverse: size rises after wins and falls after losses, so the large positions are financed by profit rather than by the drawdown buffer.

How many losses can a martingale survive on a $2,000 drawdown?

Four. With an illustrative $100 base risk and 1:1 trades, rungs of $100, $200, $400 and $800 lose $1,500 between them, and the fifth rung needs $1,600 of risk when only $500 of buffer is left.

Rung 1 risks $100 and loses. Down $100, $1,900 of buffer left.
Rung 2 risks $200 and loses. Down $300, $1,700 left.
Rung 3 risks $400 and loses. Down $700, $1,300 left.
Rung 4 risks $800 and loses. Down $1,500, $500 left.
Rung 5 risks $1,600 with $500 of room. A full loss would put the account $3,100 down, $1,100 past the limit.

The cumulative loss after k rungs is the base times (2^k - 1), so every extra rung of room needs a buffer roughly twice as large: four rungs total $1,500 and fit inside $2,000, and the fifth, at 16 times the base, takes the total to $3,100.

Each rung doubles, so the buffer that holds four losses cannot hold a fifth: the next rung alone is more than three times the room left.

The same formula settles a figure repeated across trading sites, that a $100 doubling martingale needs $12,800 after seven losses. That is the eighth stake, but by then you have already lost $12,700, 12.7% of a $100,000 account, so the eighth rung puts $25,500 of total capital at risk. The quoted number understates the requirement by half.

Does doubling after a loss change the expected value?

No. At a 50% win rate with 1:1 trades, a five-rung ladder wins $100 in 96.875% of sequences and loses $3,100 in 3.125% (1 in 32), and 0.96875 x $100 minus 0.03125 x $3,100 is exactly $0. We checked it a second way by averaging all 32 equally likely five-trade paths, and the result is again $0.

This is a theorem, not a quirk of the example. Columbia University lecture notes on the optional stopping theorem (a result about when you choose to stop a fair game) put it plainly: "Your expected fortune when stopping is the same as when you started; the stopping strategy does not help to increase your expected fortune."

The result holds whenever the stopping time is bounded, and a capped ladder plus a drawdown limit bound it. The textbook escape needs no cap at all. Imperial College London lecture notes (2018) say of the uncapped bet-until-ahead strategy in a fair coin game that "as a gambling strategy, this is hopelessly impractical", because it needs unlimited time and unlimited capital, and add that doubling after a loss has similar properties.

The two sides of that sum are equal, $96.875 each, so they cancel. With an average loss 31 times the average win, the breakeven win rate is 3,100 / (3,100 + 100) = 96.875%, exactly the win rate the ladder shows.

The wide band of small wins and the thin sliver of one large loss hold exactly the same money, which is why a 96.875% win rate is only the breakeven for a 31-to-1 loss.

Costs then push the result below zero, because they scale with exposure while the visible win stays $100. At 50%, a sequence trades 1 + 1 + 1 + 1 + 1 = 5 base units on average (each rung is twice the size and half as likely to be reached), so an illustrative $5 of commission and slippage per $100 of risk costs $25 per sequence, a quarter of every visible win, the same house-edge logic as in our day trading versus gambling math.

How long does a martingale account survive?

At a 50% win rate, half of five-rung ladders on a $2,000 buffer have breached within 22 sequences and only 4.18% are still alive after 100. A 55% or 60% win rate stretches the life without changing its end. The table shows survival for each win rate, plus a more realistic model in which the firm closes the account mid-rung.

Model and win rate	Breach per sequence	Alive after 20	After 50	After 100	After 200
Textbook ladder, 50%	3.125%	52.99%	20.44%	4.18%	0.17%
Textbook ladder, 55%	1.8453%	68.90%	39.41%	15.53%	2.41%
Textbook ladder, 60%	1.0240%	81.40%	59.77%	35.73%	12.76%
Mid-rung liquidation, 50%	4.76% (1 in 21)	37.69%	8.72%	0.76%	0.01%

Computed by Phoenix Technologies; method: breach per sequence q = (1 - p)⁵, survival after n sequences = (1 - q)ⁿ, illustrative $100 base, $2,000 buffer, 1:1 trades, independent sequences; mid-rung row explained below. The 50% row is the honest baseline for any system that has not proven an edge, and on it a survivor at sequence 100 is about a 1-in-24 outlier, not a strategy.

Half of the 50% ladders have breached by about sequence 22, half of the 55% ladders by about 38 and half of the 60% ladders by about 68.

A better win rate moves the crossing to the right but does not flatten any curve: every ladder is heading for the same floor, only at different speeds.

Why the account breaches earlier than the textbook ladder says

The textbook row counts a breach only when the fifth rung's full $1,600 stop is hit, but the firm's floor sits $500 away and closes the account there. If price moves without drift between stop and target (a model assumption), the chance of falling $500 before rising $1,600 is 1,600 / (1,600 + 500) = 16/21, using the hitting formula in the Columbia notes. Breach per sequence becomes 1/16 x 16/21 = 1 in 21, about 4.8%, and half of accounts are gone within 15 sequences instead of 22.

The expected value is still exactly $0: 95.24% of sequences at +$100 against 4.76% at -$2,000. On a trailing floor every completed sequence ends at a new high and the floor follows it up, so the ladder never builds a cushion (see trailing versus static drawdown); on a static floor banked profit adds room, so the table is conservative there.

Does martingale work if you have an edge?

An edge makes the ladder profitable per sequence while it lives, but it still multiplies ruin compared with fixing your risk. At 55%, the ladder has an 84.47% chance of having breached within 100 sequences; fixed $100 risk with 20 losses of room has a 1.81% chance of ever being ruined.

Win rate	Ladder EV per sequence	Ladder breached within 100	Within 200	Fixed $100 EV per trade	Fixed $100 lifetime ruin
55%	$40.95	84.47%	97.59%	$10.00	1.81%
60%	$67.23	64.27%	87.24%	$20.00	0.03%

Computed by Phoenix Technologies; method: ladder as in the survival table, fixed risk by gambler's ruin ((1 - p) / p)²⁰ with no profit target. The horizons differ (lifetime against a count of sequences, and a sequence can hold up to five trades), so read it as orders of magnitude. At 53%, the fixed-risk version carries 9% lifetime ruin, as worked through in our risk of ruin on a funded account article.

The ladder earns more per sequence because it bets bigger after losses, and that extra size is drawn from the same $2,000 that keeps the account alive.

Why do martingale EAs and signals show 90%+ win rates?

Because the ladder manufactures them: a five-rung doubling ladder wins 96.875% of sequences at a coin flip. The losses arrive rarely, all at once, and often after the record has been sold, copied or ranked.

The number of winning sequences before the first breach follows a geometric distribution with a mean of 31 and a median of 21. On that average path the record shows 31 rising steps of $100 to a $3,100 peak and one sequence that drops $3,100 straight back to the start: 31 wins, 1 loss, a 96.9% win rate and a net $0. On a trailing floor the account is closed $2,000 below its $3,100 peak during the fifth rung.

The average martingale record climbs smoothly for 31 sequences and then gives it all back in one, so a record without drawdowns may simply be one that has not reached its cliff.

FTMO's blog ran a martingale EA on EURUSD across 2016 and reported "7 tests out of 20 tests lost its entire capital", about 35%, after three earlier runs had each made 25% to 35%; on DAX the system wiped the account in the first or second month in all three tests.

The U.S. Commodity Futures Trading Commission warned in January 2024 that scammers claim AI-created trading algorithms can generate huge returns or 100 percent win rates; our article on whether AI trading bots work covers that evidence. A leaderboard sorted by return or win rate adds a selection effect, because the ladders that have not hit their cliff yet rise to the top, which is why the top leaderboard trader is often the riskiest.

How do you spot a martingale in a track record before copying it?

Check whether position size rises after losses and whether the average loss is many times the average win. Those two numbers identify a martingale or grid record regardless of what the provider calls it.

What to check	Martingale or grid record	Fixed-risk record
Size after a loss	Up, x2 per rung (rung 5 = 16x base)	Same
Sequence or basket win rate	96.875% at zero edge	About 50% at 1:1
Average loss / average win	31 (3,100 / 100)	About 1
Breakeven win rate	96.875%	50%
Drawdown history	None, then one cliff	Regular small drawdowns
Largest position	16x base	1x base

The fastest single test is the breakeven: divide the average loss by the sum of the average loss and the average win. If the stated win rate is barely above that figure, the win rate is a product of sizing and proves nothing about skill.

Two numbers from any track record, size after losses and average loss against average win, separate a ladder from an edge before you copy it.

The behavioural tell behind every row is a refusal to take a loss. No public data exists on what share of signal providers, copy-trading leaders or marketplace EAs use martingale or grid sizing, so you have to check each record yourself.

Is martingale allowed at prop firms?

It depends on the firm, and most rule pages never use the word: of six rule pages checked in October 2026, one says it does not restrict martingale EAs, one allows martingale in a single account only, and the other four never name it, though three forbid the sizing pattern or grids. The table summarises each firm's own page.

Firm	Names martingale?	What its own page says
FTMO	No	Forbids substantially larger positions, overexposure and EA hyperactivity
FundedNext (CFD)	Yes, allowed	Martingale EAs not restricted (April 2026); EAs only on MT4/MT5 under $50,000 (September 2026)
FundedNext Futures	No	Grid trading prohibited
The5ers	No	Forbids sizes substantially larger than typical and gambling-like positions
Topstep	No	Seven prohibited strategies plus a simulator-exploit list, none about increasing size after losses
Funded Trader Markets	Yes, one account only	Allowed within a single account, not across several (page undated)

FundedNext's CFD help center says it "does not impose any limitations or restrictions on trader's strategy, whether it involves discretionary trading or EAs that employ martingale strategies". Its EA page, updated September 2026, allows EAs only on MetaTrader 4 and 5 accounts below $50,000, so a martingale EA is limited to those. Its futures help center takes a hard line on grids: "Grid trading is prohibited because it can lead to market manipulation and create artificial trading activity."

FTMO's Forbidden Trading Practices page bans "opening substantially larger position sizes compared to your other simulated trades, on any of your accounts", and a doubling ladder opens a rung 16 times the base on its fifth step by construction. The5ers' prohibited practices page has a matching clause on positions "substantially larger or smaller than your typical trading activity".

Topstep's prohibited strategies page lists seven prohibited strategies (plus a separate list of simulator exploits), and the closest to sizing is "Trading Maximum Position Size into Major News Events". FTMO and The5ers do not publish the threshold for "substantially", and none of the six firms checked publishes its detection logic or how many accounts it has closed for martingale.

The common claims that FTMO bans martingale by name, that The5ers or Topstep explicitly prohibit it, or that every prop firm bans it, are not supported by the firms' own pages. Apex Trader Funding's help center could not be checked for this article, so we state no rule for it. For EA and copy-trading permissions firm by firm, see prop firms that allow EAs and copy trading, and check the firm's current rules before buying.

What happens when a copier replicates a martingale across many accounts?

Every follower runs the same ladder, so the fifth loss breaches every account in the same sequence. Ten copies of one ladder are one ladder at ten times the size, which is the correlation risk of copying one master in its starkest form.

Funded Trader Markets' FAQ says "martingale and layering strategies are allowed within a single account" but "must not be used across multiple accounts". Copying an external provider's martingale into a funded account can also break own-trades rules before the maths does, as covered in why copied signals get accounts banned.

A copier cannot fix a negative-progression master. Thor replicates one master to unlimited follower accounts and has built-in drawdown kill-switches and max position limits; a limit can cap rung size and a kill-switch can stop copying at a set loss, which trades the cliff for a smaller, earlier loss but leaves the expectation where optional stopping puts it. Proportional or multiplier sizing scales every rung rather than removing any, as explained in copy trading sizing methods.

Before you copy or run a high-win-rate system on a funded account:

If size rises after losses anywhere in the record, treat it as a martingale whatever its label.
If the stated win rate is within a few points of average loss / (average loss + average win), the edge is unproven.
If base x (2^rungs - 1) exceeds your buffer, the account breaches on that rung: with a $100 base and a $2,000 buffer, that is rung 5.
If your firm's page has a "substantially larger" clause, a rung 16 times your base fits it.
If you do have an edge, keep risk fixed: the edge earns the money, and the ladder only adds the cliff.

Go deeper

Frequently asked questions

Why do people call martingale risk-free?

Because it is risk-free only in a world with no limits. FTMO's blog puts it this way: "If a player had unlimited capital and an unlimited number of rounds, then he would realize the endless risk-free profit." A funded account has a fixed drawdown buffer and a trader has finite time, so the losing streak that ends the ladder eventually arrives.

Does Topstep allow martingale?

Topstep's Prohibited Trading Strategies page, as of October 2026, does not mention martingale, grid or averaging. It lists seven prohibited strategies, including account stacking and "Trading Maximum Position Size into Major News Events". That is not the same as permission, so check the current page and the loss limits before running a ladder.

Can AI make a martingale bot safe?

No. The doubling arithmetic is the same whoever or whatever places the orders, and the expected value at zero edge stays $0. The CFTC's January 2024 advisory on trading bots states that "AI technology can't predict the future or sudden market changes."

Does a smaller multiplier fix martingale?

No. With a gentler multiplier one win no longer recovers every earlier loss, so sequences run longer and the cliff arrives later and smaller, but optional stopping still applies to any bounded rule in a fair game. The shape changes; the zero average before costs does not.

Does capping the ladder at four rungs make it safe?

It makes the loss smaller and more frequent, with the same average. Capped at four rungs with a $100 base at 50%, the ladder loses $1,500 once in 16 sequences on average and wins $100 in the rest, and the two cancel to exactly $0. You swap a rare account breach for a regular $1,500 hit.

Will a prop firm detect a martingale copied from a signal?

None of the firms checked here publishes the detection logic it uses, so nobody outside a firm can say exactly what triggers a review. What is public is the rule text: FTMO and The5ers forbid position sizes substantially larger than your usual trades, and a doubling ladder produces that pattern on every losing streak.

Is grid trading the same as martingale?

Not always. A grid places orders at preset price levels and becomes a martingale only when the lots grow as price moves against it, which is how some grid EAs are built. Firm rules can treat the two differently: FundedNext Futures prohibits grid trading, while FundedNext's CFD help center does not restrict martingale EAs, though since September 2026 it allows EAs only on MT4 and MT5 accounts under $50,000.