Martingale versus a bigger fixed lot: 55 strategies, same profit

We do not use martingale in any Wonder EA. Until recently our reason for that was the ordinary one: it is dangerous. That is a weak answer, and we knew it. Plenty of things in trading are dangerous and still worth doing if they pay for the danger.

So we stopped arguing and measured it. The result was sharper than we expected, and it is not the usual moral lecture. Martingale is not simply risky. On our data it is dominated: for the same profit, a plain bigger fixed lot gave less drawdown and roughly a third to a half of the risk of ruin.

Here is how we tested it, so you can disagree with the method rather than the conclusion.

The one property that makes this testable exactly

We tested a sequential martingale, the classic kind: one position at a time, a hard stop loss always present, the lot multiplied after a losing trade, reset to base after a win, with a cap above which the size resets. That is the same logic as the standard martingale money management module in most strategy platforms.

The useful property is this: a sequential martingale does not change which trades happen. Entries and exits are identical, only the size of each trade changes. So it does not need a new backtest at all. We took the real closed trade lists of strategies we already had, and reapplied the sizing rule trade by trade. No re-simulation, no approximation, no optimizer involved.

The benchmark almost nobody uses

Most martingale discussions compare martingale against the base strategy at 0.01 lots. That comparison is meaningless: of course a scheme that trades bigger makes more money than one that trades smaller. It also makes more drawdown.

The honest benchmark is a bigger fixed lot. If a martingale earns 40 percent more than baseline, then the thing it must beat is baseline traded at 1.4 times the size. Same profit, then compare the risk. That comparison is the whole experiment.

Setup

• Two independent samples: 44 strategies that survived our out-of-sample selection process, and 11 strategies from our live book. 55 in total, no overlap.
• Trades closed from 1 January 2020 onward.
• Everything normalized in R, where R is that strategy's own average losing trade. The account starts at 100 R, so an average loss costs 1 percent of capital. Ruin is defined as equity down 50 percent.
• Risk of ruin from Monte Carlo, 300 resampled runs per strategy per scheme, using a block bootstrap with blocks of 10 trades. The blocks matter: consecutive losing streaks are exactly what kills a martingale, and an ordinary independent resample destroys those streaks and produces an answer that is far too kind to martingale.

Sample one: the 44 selected strategies

Net and drawdown are averages in R. Ruined means the account hit minus 50 percent on the real historical trade order.

Now read it in pairs, at matched profit. That is where the argument lives:

• Around 52 net: martingale 1.3 cap 8 gives drawdown 21 and 3.6 percent ruin. Fixed 1.5x gives drawdown 20 and 1.6 percent ruin.
• Around 71 net: martingale 2.0 cap 8 gives drawdown 35 and 11.0 percent ruin. Fixed 2x gives drawdown 26 and 4.0 percent ruin.
• Around 91 net: martingale 2.0 cap 16 gives drawdown 52, 18.9 percent ruin, and three accounts actually wiped out by the real sequence of trades. Fixed 2.5x gives drawdown 33, 6.7 percent ruin, and zero wiped out.

Not one matched pair goes the other way.

Why it has to come out like this

Look at the Net / DD column on the fixed lot rows. It is 2.77 all the way down, and that is not a coincidence: multiplying every trade by the same constant scales profits and drawdowns by that same constant, so the ratio cannot move. Fixed leverage is a clean dial. You buy return with risk at a fixed exchange rate.

Every martingale row sits below 2.77, and it sinks as the scheme gets more aggressive: 2.52, then 2.41, then 2.03, then 1.76. The reason is geometric. A martingale does not spread size evenly. It puts the largest positions exactly where the losses are already stacking up, because that is its definition. It is leverage, delivered at the worst possible moments.

Sample two, and the part that argues against us

The 11 live book strategies are a harder sample: baseline Net / DD is 0.59 against 2.77. Martingale degrades further there, up to 4 accounts out of 11 wiped out by the real 2020 to 2026 trade sequence at multiplier 2.0 with cap 8, a scheme that ended with an average net of just 3 R against a drawdown of 51 R.

But this sample also contains a result we are not going to hide, because it qualifies our own headline. On these 11 strategies, high fixed leverage breaks too. At fixed 2.5x and 3x, two accounts were also ruined on real history, and once ruin truncates the equity curve the tidy constant 0.59 ratio stops holding.

So the correct conclusion is narrower than "fixed lots are safe". It is this: at any given profit target, the martingale version is the worse way to get there. If a strategy is not strong enough to carry the leverage you want, martingale will not rescue it. It will get you to the same place with a deeper hole and a higher chance of not arriving.

The cosmetic appeal, and why it is a trap

Martingale is popular for a reason that has nothing to do with expected value: it produces beautiful screenshots. Recovering after losses makes the equity line look smooth and the monthly table look green, right up until the sequence that ends it.

That smoothness is worth almost nothing as evidence. In a separate study on 1,195 of our generated strategies, we split each track record in two and measured how well the straightness of the equity curve in the first period predicted straightness in the second. The correlation was 0.005. Essentially zero. A pretty curve is a description of the past, not a property of the system.

What this test does not prove

We want the limits on the record:

• It covers sequential martingale, not grid. A grid opens several positions against the move at once, which genuinely changes which trades happen and what the average entry price is. That is a different mechanism and it deserves its own test. It now has a bar to clear though: at equal net, beat the fixed lot on Net / DD and on risk of ruin. If it cannot, it is just leverage in costume.
• 300 Monte Carlo runs per strategy per scheme is enough to separate 4 percent from 11 percent. It is not enough to argue about tenths of a percent.
• All of this is simulation applied to historical trade lists. It is backtest analysis, and past performance does not guarantee future results.

What we do with it

Nothing changes in our EAs, but the reason changed, and that matters to us. Every Wonder EA opens each position with a fixed stop loss and take profit, one position at a time, no martingale, no grid, no averaging down. That is not a marketing position we adopted because it sounds responsible. It is what the matched-profit comparison says to do.

If you want to see the rule set in practice before spending anything, Wonder 8 is free on MT5 (also on MT4): USDJPY H1, fixed 0.01 lot, stop loss on every trade. Run it on a demo through a losing week and watch what it does not do.

And if you are evaluating any EA with progressive lot sizing, ours or anyone's, the useful question is now a single one: what does this scheme give me that the same strategy at a bigger fixed lot would not give me more cheaply? Ask it before you fund the account, not after.

All figures in this article are backtest and simulation results computed on historical trade data from 2020 onward. Backtests are hypothetical, past performance does not guarantee future results, and trading involves risk of loss.