Does martingale work in forex? We doubled down on EUR/USD since 1999

How the martingale strategy works

The martingale is the oldest betting system there is. You bet one unit. If you lose, you bet two; lose again, four; then eight. When a bet finally wins, it pays back every loss in the streak plus one unit, and you start again at one. In forex it is sold as martingale robots, or EAs, that double the size of the next trade after every losing one.

The catch is how fast the doubling grows. We gave each account $10,000 and a first trade of 20,000 euros, sized so that a 50-pip move made or lost $100, which is 1% of the account. Six losses in a row cost 63% of the start, and the seventh trade has to be 1,280,000 euros.

How we tested it

We used the euro's daily reference rate against the US dollar, published by the European Central Bank, from 4 January 1999 to 16 October 2026: 7,116 days. Each trade opened at a day's rate and closed on the first day the rate stood at least 50 pips away, up or down. The next trade opened at once, at the same rate. Our trader looks once a day, so a trade could close well beyond 50 pips; we booked the real move.

We ran two versions: one that always buys EUR/USD and one that always sells it. Neither is a forecast; the only question is what doubling does. Every trade paid 1 pip in costs, a typical retail spread, and we charged no overnight financing, which flatters the martingale because its biggest trades would pay the most. Against it stood the simple alternative: the same trades at a fixed 20,000 euros.

Then we opened a fresh $10,000 account on the first trading day of every month from January 1999 to October 2025, 322 starts, and ran each for one year. An account counted as wiped out when it could no longer pay for the next doubled trade, that is, when a 50-pip loss on it would be more than the money left. We wrote down these rules, and the bar for a pass, before computing any result.

About half the accounts were wiped out

Of the 322 one-year accounts that always bought, 176 were wiped out: 54.7%. Of those that always sold, 154 were: 47.8%. Not one account trading a fixed size went bust within a year. The end usually came early. Half of the ruined accounts were gone within 4.6 months when buying and 4.3 months when selling.

The averages hide what happened to each account. On average the martingale ended close to the fixed trade size, and a little above it: $10,039 against $9,849 when buying, $9,793 against $9,721 when selling. The spread was far wider: the best martingale accounts ended the year with $35,532 and $34,926 and the worst with nothing, while the fixed-size accounts ranged from $3,768 to $15,908. That is what the arithmetic of a fair game predicts. Doubling does not change the odds of a trade; it trades many small wins for a large chance of losing the account.

Many martingale accounts did make money for a year: 45.0% of the buying ones and 51.9% of the selling ones ended with a profit, against 49.7% and 45.3% with a fixed size. Checking once a day made some endings worse. In 41 of the buying accounts and 61 of the selling ones, a single losing trade took the balance below zero, because the rate had run far past 50 pips before the trader looked. We counted those accounts as ending at $0, as a broker's negative-balance protection would.

The longer you run it, the worse it gets

Over five years (274 starts, January 1999 to October 2021), 82.8% of the buying accounts and 74.5% of the selling ones were wiped out. A single account run from 4 January 1999 shows how it happens. Always buying, it lasted until 29 October 1999: after 87 trades, six losses in a row left $4,272, too little for a seventh trade of 1,280,000 euros. Always selling, it lasted until 29 December 2003, 579 trades, and ended on a losing trade of 2,560,000 euros that took the balance below zero.

Losing streaks are not rare

It is easy to think that six or seven losses in a row hardly ever happen. On EUR/USD they happened again and again. Over the whole period our 50-pip trades came out 1,466 up and 1,481 down, 49.7% against 50.3%: about as close to a coin toss as a market gets. Always buying, there were 25 losing streaks of six or more and 11 of seven or more; the longest was 13. Always selling, there were 21 and 10, and the longest was 10.

A fair coin does the same. We tossed 2,947 virtual coins 10,000 times over: the longest losing streak was 11 in the middle case, and between 8 and 14 nine times in ten. The market was not unusually cruel. Streaks like these are simply what that many tosses produce, and the martingale needs them never to happen.

Was it the time period?

We split the starts in two. When buying, 56.5% of the one-year accounts started from 1999 to 2012 were wiped out, and 52.6% of those started from 2013. When selling, it was 73.8% and 19.5%. Selling did much better in the second half, but even then about one account in five was lost. The bar we set before the test was fewer than 5%, in both halves and both directions.

The size of the step changes the timing, not the ending. With 25-pip trades and a first trade of 40,000 euros, still 1% of the account per step, 69.9% of the buying accounts and 61.8% of the selling ones were wiped out within a year. With 100-pip trades and 10,000 euros it was 29.2% and 36.6% within a year, and 81.0% and 74.1% within five years.

Could you make it work?

Not by doubling. Every version missed the bar we set before the test, in both halves and both directions, with costs and without them. A bigger account only buys time: each doubling of the account pays for one more doubling of the trade, and the streaks above show that longer ones keep coming. Brokers also cap the size. In the EU and the UK, retail traders may trade at most 30 times their money on major currency pairs, and a 1,280,000-euro trade on what is left of a $10,000 account is far beyond that.

A fixed trade size did not make money either, which is the honest point of comparison. Over the whole period, the fixed-size account that always bought ended with $3,032 after costs ($8,928 before them), and the one that always sold was wiped out on 17 March 2008, after years of a rising euro. There was no edge in either direction. Doubling only changed how the losses arrived: rarely, but all at once.

Questions people ask