Gold Martingale Strategy: Why Doubling After Losses Blows Accounts, Simulated on Real XAUUSD Data

The short answer

We ran a classic doubling martingale on real gold prices: start at 1 ounce, double after every loss, return to 1 ounce after a win, with a $10,000 account and 1:100 leverage.

• Two years of hourly data: the longest losing streak was 6, so the martingale never broke. It turned a base strategy that lost $906 per ounce into an account that grew to $18,185. This is exactly why martingale is so persuasive.
• Daily data since 2015: the account crawled up to $20,635 by 13 February 2025. Then came nine losses in a row. Position size went 1, 2, 4 … 256 ounces, equity fell to $10,319 by 27 February, and the next trade required 512 ounces, more margin than the account had. Ten years of gains disappeared in two weeks.
• Bigger starting size: starting at 5 ounces instead of 1, 61% of six-month windows on hourly data ended in ruin. At 10 ounces, 84% did.

Martingale does not create an edge. It trades many small wins for one rare, enormous loss, and the only question is when the loss arrives.

How martingale works, and why it feels safe

Suppose each trade risks $44 per ounce (our base strategy’s average). The sequence after a run of losses looks like this:

One win at any point recovers the whole sequence plus one unit. Most of the time, the win comes within two or three trades, so the equity curve rises steadily. The danger is the right-hand column: after nine losses, the next trade needs margin worth twice the starting account, at a moment when the account has already lost 511 units.

The base strategy we used

Martingale needs something to trade. We used a deliberately ordinary rule so that the results show the sizing, not a clever entry. The rules:

Both base strategies lose money, which is typical of a simple rule after costs. That is the point: martingale promises to fix a losing strategy.

Simulation 1: two years of hourly data (it survived)

The longest losing run in these two years was 6 trades (it happened twice). At 1 ounce, a 6-loss run needs a 64-ounce seventh trade, about $2,600 of margin at $4,000 gold: affordable for a $10,000 account. So the martingale won.

Rolling six-month windows (75 start dates, every week)

Nothing about the market changed between those rows, only the starting size. At 5 ounces a 6-loss run means a 320-ounce trade, which a $10,000 account cannot margin.

Simulation 2: daily data since 2015 (ten years, then two weeks)

Daily data covers 2,630 trades, enough for a rarer streak to appear. The base strategy’s losing streaks:

The equity path of the 1-ounce martingale, sampled through the years:

And then the final two weeks:

The account had to abandon the sequence with a $10,316 loss from its peak: ten years of gains, gone in nine trading days. Over 114 rolling twelve-month windows on daily data, 11 ended in ruin and 54 finished below the starting balance.

What the simulations say, in plain English

• Martingale makes a losing strategy look like a winning one, for a while. The base rule lost money at fixed size in both datasets. Martingale showed profits for two years on H1 and for ten years on D1.
• Survival depends on the longest streak you happen to meet. Two years of hourly trades contained no streak longer than 6; eleven years of daily trades contained two of 9. Short backtests systematically hide the streak that kills you.
• Size is the hidden variable. The same rules and the same market gave 0% ruin at 1 ounce and 84% at 10 ounces.
• Caps do not remove the risk; they realise it in instalments. The six-doubling cap took two losses of about $1,000 each on hourly data and left 16 of 75 windows below the starting balance.

Red flags in martingale robots

Many gold expert advisors sold with “95% win rates” use martingale or its cousins (grid and recovery systems). Warning signs:

• An equity curve with almost no losing months, and no stop loss shown.
• Lot sizes in the trade history that double or multiply after losses.
• Backtests of only a year or two, or only on one market phase.
• Marketing that talks about “recovery” or “never closing at a loss”.

Our 12-point EA checklist covers what to ask before trusting any robot, and our page on grid and hedging systems simulates the related approaches on the same data.

What to do instead

Fixed fractional sizing: risk the same small share of the account on every trade (many traders use 0.5% to 1%), sized from the stop distance with the position size calculator. After losses, your position gets smaller, not larger. Our risk management lesson explains the method, and how to handle a losing streak shows the maths of streaks and drawdowns. If you want to understand why a strategy needs a positive expectancy before any sizing method can help, read trading math and expectancy.

Check it yourself

Take any strategy you trade, count its longest losing streak over the longest history you have, and add two. Then calculate the position size martingale would need at that streak length and compare it with your account. That one calculation is usually enough.

Learn it with other traders

Bad sizing is easier to spot from the outside. In the XAUUSD Global community, traders review each other’s position sizing and robot results before money is at risk. The first month is free, then $19.99 a month.

Educational content only, not financial advice. Trading gold with leverage carries a high risk of loss.

Frequently asked questions