Martingale doubles the bet after every loss so the next win recovers everything, and against a finite account that guarantee inverts: one ordinary losing streak escalates the exposure past what the account can hold. The curve looks perfect until it ends. Here is the math, and how to spot martingale hiding inside a smooth track record.
It converts a small frequent win into a rare total loss. Double after every loss and the arithmetic runs away immediately: a streak of seven losses starting from 1R of risk puts 128R on the table for the eighth trade, with 127R already gone. Streaks of seven are not rare events, they are scheduled ones, a normal feature of any win rate over a few hundred trades.
The strategy does not fail because the streak was unlucky. The streak was ordinary. The sizing made it fatal.
Martingale does not remove risk. It schedules it.
Here is the part that trips people up... the curve is genuinely smooth, because every dip gets bought back within a few trades and the win rate prints in the high 90s. What the curve hides is the position size column, where the real story lives. Equity charts plot outcomes, and martingale's risk lives entirely in exposure, so the chart looks calm right up to the bar where it stops existing.
This is why a martingale record passes the eyeball test that a real edge sometimes fails. The eyeball reads smoothness as skill. The math reads it as a fuse.
Feller's classic result: a gambler with finite capital playing against a table limit reaches ruin with probability approaching one as play continues, even in a fair game, and faster in an unfavorable one. Doubling schemes do not escape this, they accelerate it, because each escalation moves the player toward the limit in bigger steps. The only martingale that works requires infinite capital and no limit, and no broker offers either.
Trading dresses the same structure in different clothes. The table limit is your margin, your account, or your prop firm's daily loss rule, and risk of ruin is the formal name for the probability of meeting it.
Four tells, readable from any honest trade log.
| Tell | What it looks like |
|---|---|
| Size grows after losses | Position column climbs exactly when the streak does |
| Win rate too clean | High 90s with no visible losing stretches |
| Dents vanish instantly | Every drawdown recovered within a handful of trades |
| Truncated loss streaks | Loss streak distribution shows almost nothing past two |
Quantprove surfaces the raw material for all four: the loss streak distribution, the drawdown profile, and the per trade R values that expose a size column doing the heavy lifting. A record where losses simply never accumulate was either traded by the luckiest person alive or sized by escalation.
Same family, slower fuse. Adding to a loser lowers the average entry and raises the exposure, so the position needs a smaller bounce to break even and does more damage if the bounce never comes. Run repeatedly, it produces the same signature: high win rate, smooth curve, one catastrophic column.
One distinction worth keeping, because it saves legitimate strategies from the same bucket: scaling into a planned position with the total risk capped before entry is not martingale. The difference is one question. Was the full size decided before the first fill, or did the losses decide it?
Backwards from martingale in every direction: size grows after evidence accumulates, not after losses do. The edge gets validated out of sample, the worst case gets resampled at the new size, and the increase happens in steps with real trade blocks between them. When should you increase your position size walks the gates. Martingale raises size to escape the math. Scaling raises it because the math agreed.