The EA has a 91% win rate. The equity curve rises steadily with barely a dip. Five years of backtest data. Maximum drawdown 4.2%. The vendor’s myfxbook shows six months of live results that match almost perfectly. You buy it, fund a $5,000 account, and eighteen months later you get a margin call on a Tuesday morning when the dollar gaps 300 pips on a surprise Fed announcement.
This is not an unusual story. It happens constantly, to experienced traders as well as beginners, because the strategy involved — martingale or one of its variants — is specifically designed to look safe in backtests while carrying a tail risk that only reveals itself in live trading under the right conditions.
Understanding how martingale works, why it looks so good on paper, and exactly how to detect it is one of the most practically important skills in EA evaluation. The detection method is not complicated. Most traders simply never apply it.
What Martingale Actually Is
Martingale is a position sizing system, not a trading strategy. In its pure form, the rule is simple: after every losing trade, double your position size. After every winning trade, return to the base size.
The mathematical logic behind it sounds superficially appealing. If you double after every loss, the first win recovers all previous losses plus the original profit target. As long as you eventually win, you always come out ahead. In a coin-flip game with infinite capital and no table limits, martingale guarantees profit in the long run.
The problem is that neither condition holds in trading. Capital is finite. And forex markets can trend against a position for far longer than any account can sustain exponentially increasing position sizes. A sequence of eight consecutive losses — not unusual in volatile markets — requires a position size on the ninth trade that is 256 times the original. On a $10,000 account trading 0.01 lots to start, that is 2.56 lots on trade nine. Most retail accounts are wiped long before that point.
What makes martingale genuinely dangerous is not that traders don’t understand this intellectually — most do, at some level. What makes it dangerous is that backtests hide the tail risk almost completely.
Why Martingale Backtests Look So Good
In a backtest run over historical data, the martingale sequence almost always completes. The market eventually reverses, the accumulated position closes in profit, and the equity curve steps upward. The rare extended losing sequence that would blow a live account either didn’t occur in the test period, or the backtest was designed around it — tested on a period of low volatility, a ranging instrument, or conditions that suited the strategy’s recovery mechanism.
The result is a backtest with three characteristics that look like signals of quality but are actually signals of danger:
Very high win rate. Because most martingale sequences complete with a winning trade that recovers all losses, the percentage of individual trades that close in profit is extremely high — typically 85-95%. A legitimate trend-following EA might win 45% of trades. A martingale wins 91% because it waits and adds to positions until the market gives it a win.
Low reported maximum drawdown. In the backtest, the worst losing sequence happened to resolve before breaching the account. The maximum drawdown reflects the deepest trough that was recovered, not the potential depth of a sequence that was never interrupted. The real risk — the risk of a losing sequence that doesn’t resolve in time — is invisible in the equity curve.
Smooth, consistent equity growth. Because the wins are large (recovering multiple accumulated losses) and the losses are small (cut before the sequence completes), the equity curve has a characteristic pattern: small regular dips followed by sharp recoveries. It looks like a high-quality, low-variance strategy. It is not.
These three characteristics together produce backtests that routinely outperform legitimate strategies by every conventional metric. Higher win rate. Lower drawdown. Better Sharpe ratio. Better profit factor. This is why martingale EAs dominate leaderboards on myfxbook and MQL5 signals — the metrics used for ranking reward exactly the properties martingale produces.
The Variants You Need to Know
Pure martingale — strict doubling after every loss — is relatively easy to spot because the position size escalation is obvious in the trade history. But most commercial martingale EAs use one of several variants that are harder to detect:
Soft martingale / multiplier scaling. Instead of doubling, the EA increases position size by a multiplier of 1.3x, 1.5x, or 1.7x after each loss. The escalation is slower and less obvious in the trade data, but the tail risk is identical in structure — it simply takes a longer losing sequence to trigger it.
Grid trading. The EA opens multiple positions at fixed intervals as price moves against the initial trade, then closes all of them together when the combined position reaches a profit target. This is martingale in spatial rather than temporal form. The total exposure grows as price moves against you, the equity curve looks smooth and consistent, and the tail risk is a directional move that exceeds the grid spacing before any recovery occurs.
Recovery mode / averaging down. The EA trades normally but has a “recovery mode” that activates when a trade is underwater beyond a threshold. In recovery mode, it opens additional positions in the same direction at lower prices, effectively averaging down. The trade eventually closes at a weighted average entry with a smaller loss or small profit. Until the market doesn’t recover.
Hidden lot size escalation. The EA appears to trade fixed lot sizes but has a subtle logic that increases size after a defined drawdown level, news event, or time condition. This is the hardest variant to detect from the equity curve alone, because the escalation trigger is infrequent and the position size change may be modest.
What all of these share is the core characteristic that defines martingale behavior: winning trades following a loss are larger than typical winning trades. This is the signature that statistical detection looks for.
How to Detect It: The MG Indicator
The statistical detection of martingale behavior is based on one comparison: the average size of winning trades that followed a losing trade versus the average size of winning trades that followed a winning trade.
In a clean, fixed-lot-size strategy, these two numbers should be approximately equal. A win is a win regardless of what happened on the previous trade. The size of the position should not depend on prior outcomes.
In a martingale strategy, winning trades that follow a loss are systematically larger — because the position was scaled up during the losing sequence and the win closes that larger position. This produces a measurable ratio:
MG Indicator = Average win size after a loss ÷ Average win size after a win
A ratio of 1.0 means identical behavior — no evidence of martingale. A ratio of 1.5 means post-loss wins are 50% larger on average than post-win wins — minor recovery bias. A ratio of 2.5 or above means post-loss wins are more than twice the size of typical wins — strong evidence of martingale or recovery-based sizing.
This ratio is calculable from any backtest that records individual trade sizes and outcomes. You do not need the source code. You do not need to understand the EA’s logic. You need only the trade history, and the calculation is straightforward.
What the Numbers Look Like in Practice
Here are three real-world scenarios and what the MG Indicator reveals:
Scenario A — Clean trend-following EA: 847 trades over 4 years on EURUSD. Win rate 54%. Average win after a loss: $38.20. Average win after a win: $41.10. MG Indicator: 0.93. The ratio is slightly below 1.0 — if anything, post-loss wins are marginally smaller, which is the opposite of martingale. This EA passes the test cleanly.
Scenario B — Soft martingale scalper: 1,240 trades over 3 years on GBPUSD. Win rate 88%. Average win after a loss: $127.40. Average win after a win: $44.20. MG Indicator: 2.88. Post-loss wins are nearly three times larger than post-win wins. The EA is clearly scaling up after losses and closing at a recovery profit. High win rate, smooth equity curve, catastrophic tail risk. This is the pattern that blows accounts.
Scenario C — Grid EA on XAUUSD: 634 trades over 2 years. Win rate 79%. Average win after a loss: $89.70. Average win after a win: $52.30. MG Indicator: 1.71. Moderate recovery bias — not as extreme as scenario B but clearly present. The EA is adding to losing positions. The tail risk depends on how many grid levels are open and what the maximum total exposure reaches before triggering a close.
In each case, the conventional metrics tell a very different story from the MG Indicator. Scenario B has a better Sharpe ratio, lower drawdown, and higher win rate than Scenario A by every standard measure. The MG Indicator is the only metric that identifies it as the more dangerous strategy.
The Holding Time Test
A secondary indicator of martingale behavior is the asymmetry between how long winning trades and losing trades are held open. In a clean strategy, the average holding time for winners and losers should be broadly comparable. In a martingale or averaging strategy, losing trades are held open significantly longer — because the EA is waiting for a recovery that may not come.
A ratio of losing trade holding time to winning trade holding time above 2.0 is a warning sign. Above 3.0 is a strong indicator of averaging or grid behavior. Some martingale EAs show ratios of 8x or higher — losers held eight times longer than winners on average, because the strategy is accumulating positions through an extended adverse move and waiting for a multi-pip recovery before closing everything.
This test requires access to the trade timestamps, which are included in MT4 and MT5 strategy tester HTML reports. It is
worth calculating manually on any EA with a win rate above 75% or an MG Indicator above 1.4.
The Live Account Test That Vendors Use Against You
Many martingale EA vendors maintain live myfxbook accounts that show months or years of consistent performance. This is used as the primary trust signal — “look, it works in real trading, not just backtests.”
The problem is selection bias. A vendor running ten martingale accounts across different instruments and brokers presents the one or two that haven’t blown yet. The others are quietly closed or restarted. The surviving accounts genuinely show consistent performance — because they haven’t yet encountered the market conditions that trigger the catastrophic losing sequence.
Alternatively, the live account is funded with a large enough balance that the margin call threshold is never reached during the demonstration period. An EA that blows a $1,000 account in eighteen months might run for years on a $50,000 account before the same sequence occurs — not because it is safer, but because it has more margin buffer before the escalating position sizes exceed the account’s capacity.
The way to verify a live account is not to look at the equity curve but to download the trade history and calculate the MG Indicator directly. If the ratio is above 1.5, the account is running martingale regardless of what the equity curve shows. The question is not whether it will eventually blow — it is when.
What a Safe EA Looks Like by Comparison
For contrast, here is what clean position sizing produces in the same metrics:
Win rate typically between 40% and 65%. MG Indicator between 0.85 and 1.20. Losing trade holding time within 1.5x of winning trade holding time. Drawdowns that look rough on the equity curve but are proportionate to the risk taken. Profit factor in the 1.3-2.0 range rather than 3.0-8.0.
A clean EA looks worse than a martingale EA by almost every conventional metric. It wins less often. Its equity curve has visible drawdowns. Its profit factor is modest. But when you apply Monte Carlo simulation to its trade history, the drawdown distribution is tight and predictable — the P95 drawdown is close to the historical maximum drawdown because the risk profile doesn’t depend on a favorable sequence of outcomes.
A martingale EA run through Monte Carlo produces the opposite: a wide, unpredictable distribution where the P99 drawdown is many multiples of the historical maximum — because the historical maximum was a product of the specific sequence tested, and many other sequences produce catastrophic outcomes.
Checking Any EA Before You Trade It
The full martingale detection process on any MT4, MT5, or cTrader backtest takes under a minute using the free tool on this site. Upload the strategy tester HTML report and the analysis calculates the MG Indicator automatically — showing you the raw ratio, the score on a 0-40 scale, and a plain-language interpretation of what the number means.
The MG Safety score accounts for 40 of the 100 points in the overall robustness score — more than any other single component — because martingale detection is the single most important test in EA validation. A strategy can have suboptimal drawdown positioning or modest predictability and still be deployable. A confirmed martingale is not deployable at any position size, because the tail risk is unbounded relative to the account.
Before you fund any EA on a live account, run the backtest through the tool. The calculation takes ten seconds. The alternative — finding out eighteen months later on a Tuesday morning — is considerably more expensive.