A first-principles look at why martingale and grid Expert Advisors that double lot size after each loss produce beautiful equity curves for months and then delete an account in an afternoon. We derive the exponential exposure growth, the probability of a losing streak long enough to exhaust any finite account, and the gambler's-ruin arithmetic behind it. This is educational material only, not financial advice, and not a claim that any strategy is or isn't profitable — the point is that a smooth curve and a survivable strategy are not the same thing.
A martingale Expert Advisor doubles (or otherwise increases) its position size after every loss, betting that a win will eventually recover all prior losses plus a small profit. It eventually blows the account because the sequence of losses required to bankrupt it is not rare over enough trades — it is inevitable — and when it arrives, the position size has grown exponentially large relative to what the account can support. The strategy trades a very high probability of small, steady gains for a small-but-certain-over-time probability of a total loss. The math is not controversial and it does not depend on the entry signal being good or bad.
The classical martingale comes from 18th-century gambling: bet 1 unit on a coin flip; if you lose, bet 2; then 4, 8, 16, and so on, so that the first win returns you to +1 unit net. On paper, as long as a win eventually comes, you always end ahead. The hidden assumptions are the problem: it requires an infinite bankroll, no bet-size limit, and unlimited time. Real accounts violate all three. A trading account has a finite balance, brokers impose a maximum lot size and a margin requirement, and a stop-out closes you before you reach the win that would have saved you.
The deceptive part is the shape of the results. For most of its life a martingale EA wins constantly, because most of the time a short adverse run is followed by a recovery before the doubling gets dangerous. That produces the near-straight, low-drawdown equity curve that martingale EAs are sold on. The curve is not evidence of edge — it is evidence that the fat left tail simply hasn't printed yet. This is educational only and not financial advice; trading carries a real risk of loss.
The core issue is that risk grows geometrically while the account grows arithmetically. Suppose the base position is L and you double after each loss. After n consecutive losses your next position is L · 2ⁿ, and the cumulative amount already lost on that streak (before the next bet) is L · (2ⁿ − 1) times the per-trade loss in account terms. A short table makes the ladder concrete for a strategy starting at 0.01 lots:
Losses in a row Next lot Cumulative lots committed 0 0.01 0.01 3 0.08 0.15 6 0.64 1.27 8 2.56 5.11 10 10.24 20.47 12 40.96 81.91 13 81.92 163.83
By the eighth consecutive loss the position is 256× the starting size; by the thirteenth it is over 8,000×. Two things break here. First, the position size hits the broker's maximum lot or exhausts free margin, and the EA physically cannot place the recovery trade — the martingale's one escape hatch is nailed shut. Second, even if it could, the loss on that final oversized position dwarfs every small profit the EA banked over the preceding months.
The asymmetry is the whole story. A martingale books hundreds of small wins and one catastrophic loss. Because the wins are frequent and the loss is rare, the average trade and the win rate look spectacular, while the distribution of outcomes has a tail heavy enough to be fatal. Any metric that summarises the centre of the distribution — win rate, profit factor over a short window, average monthly return — is precisely the metric that hides the tail. That is why martingale results are dangerous to evaluate with the usual dashboard numbers.
People intuitively feel a long losing streak is rare. Over enough trades it is not. If each trade loses with probability p, the chance of a specific run of k consecutive losses starting at a given trade is pᵏ. But you are not taking one shot at it — you are exposed on every trade over the EA's life, so the relevant question is the probability that at least one streak of length k appears somewhere in N trades. That probability rises quickly with N.
Take a coin-flip-like p = 0.5 and ask for a run of 10 losses — roughly where the lot ladder above becomes account-threatening. The probability of no 10-loss run in a long sequence decays geometrically with the number of trades; across a few thousand trades, seeing at least one such run goes from unlikely to near-certain. Martingale EAs are often high-frequency, taking thousands of trades a year, which means the account is effectively guaranteed to eventually meet the streak that exceeds its lot ladder. It is not a question of if but when, and the EA's smooth history tells you nothing about how close 'when' is.
Realistic trading makes it worse than the coin flip. Losses in markets are not independent — volatility clusters, trends persist, and a grid or martingale positioned against a strong directional move takes correlated losses in a burst rather than scattered randomly. Correlated adverse ticks make long adverse runs more likely than the independent-flip model predicts, not less. So the pᵏ calculation, already unforgiving, is an optimistic lower bound on how fast the fatal streak arrives in a real, trending, volatility-clustered market.
The rigorous underpinning is the gambler's-ruin problem, a classic result in probability. Model your account as a random walk that goes up by a betting increment on a win and down on a loss, with an absorbing barrier at zero (ruin) and, for a fair or unfavourable game, no protective drift pulling you away from it. The classical theorem states that for a fair game against an opponent with effectively unlimited resources — which the market, with its practically unbounded liquidity relative to a retail account, approximates — the probability of eventual ruin is 1. You will hit zero with certainty if you play long enough.
When the game is unfavourable — and after spread, commission, and swap every retail strategy without a genuine edge is unfavourable — the drift points toward ruin and the certainty only strengthens. The martingale sizing does not change this conclusion; it only changes the path. It concentrates the walk into long stretches of tiny upward steps punctuated by rare, enormous downward jumps, converting what would be a gradual bleed into a sudden collapse. The expected value is not improved by the sizing scheme; the variance is dramatically inflated and pushed into the tail.
This is why 'martingale works until it doesn't' is mathematically precise rather than a cliché. The strategy is a mechanism for delaying and concentrating the loss that the underlying negative-or-zero expectation makes inevitable, in exchange for a comfortable-looking interim. No lot cap, take-profit tuning, or 'smart' recovery multiplier removes the absorbing barrier at zero; they only move where on the ladder you meet it. None of this is financial advice — the goal here is to show that the equity curve's smoothness and the strategy's survivability are unrelated properties.
A martingale backtest is the textbook case of a result that is real but not representative. If the tested window happened not to contain a losing streak longer than the account's lot ladder, the equity curve is flawless and the drawdown column reads reassuringly small — for that window. Extend the window, change the start date, or run it across a regime it wasn't fit to, and the ruin event appears. The tester isn't lying; it is faithfully reporting one path out of a distribution whose worst outcomes it happened to miss.
The honest tools for exposing this are exactly the ones in the research literature on backtest fragility. Monte Carlo resampling — reshuffling or resampling the trade sequence thousands of times — reveals the distribution of drawdowns the single historical path hid; for a martingale, a large fraction of the resampled paths hit ruin even when the original didn't. Out-of-sample and walk-forward testing check whether the lot ladder that survived the in-sample period survives untouched data. And a simple risk-of-ruin calculation, using the strategy's own loss probability and its maximum survivable streak, turns 'it's been fine for six months' into an explicit probability of eventual account loss.
The practical honesty test is blunt: ask what happens on the k-th consecutive loss, where k is set by your account balance and the broker's max lot — not what happened last quarter. If the answer is 'margin call' or 'the EA can't place the next trade', you have found the outcome; the only open variable is its date. A single reassuring backtest is not evidence against it.
Key honesty note: this article is educational only and not financial advice. Nothing here claims any strategy is profitable or unprofitable; it argues that martingale sizing hides risk in the tail of the distribution, and that a smooth equity curve is not proof of survivability. Trading carries a real risk of loss, and a backtest is a description of the past, never a forecast.
Educational only — not financial advice. Trading involves substantial risk of loss.