Margin Calls Don't Care That Your Edge Is Real

There’s a specific kind of confusion that happens when a trader gets stopped out at the broker level — not by their own sl, but by a margin call or stop-out — while running a system that backtested cleanly, validated out of sample, and had a healthy win rate in the 52-62% range. The instinct is to treat this as a contradiction. The strategy was supposed to work. How did the account blow up anyway?

It’s not a contradiction, because a margin call isn’t a verdict on your edge. It’s a mechanical event triggered by account-level math that has almost nothing to do with whether your strategy’s statistical premise is correct, and almost everything to do with how much of your equity is exposed at any given moment and in what sequence the losing trades happen to arrive. Understanding the actual mechanism here matters, because it’s the difference between fixing a real problem and second-guessing a strategy that never needed to change.

Equity, balance, and margin level are three different numbers

Most retail traders think in terms of balance — the number that goes up when you win and down when you lose. But the number a broker actually watches to decide whether to liquidate you is margin level, which is equity divided by used margin, expressed as a percentage. Equity is your balance adjusted for the floating profit or loss on open positions, marked to market in real time. Used margin is however much of your account the broker has set aside as collateral for your open lot size.

A stop-out level of 50%, which is fairly standard among retail forex brokers, means the broker starts force-closing your positions the moment your equity drops to half of your used margin — not half of your starting balance, not half of your account’s original size. This is a threshold that can be crossed purely through unrealized drawdown on positions that are still technically “on track” according to your strategy’s normal price action, if enough of them are open simultaneously and moving against you at the same time.

This is the part that a backtest genuinely cannot show you, because most backtests evaluate trades independently against a return series, not against a shared, shrinking pool of account equity in real time. A strategy can have a perfectly valid 55% win rate and still produce an equity curve, at the account level, that dips through a stop-out threshold during a losing streak that’s completely within its expected statistical range.

Position sizing is where the strategy stops being just a signal

The JSON config for a strategy — start_hour, end_hour, lot_size, sl, tp — treats lot_size as just another parameter, sitting next to the entry window and the stop distance like it’s equally interchangeable. It isn’t. lot_size is the one field that determines how much margin gets consumed per trade, and therefore how many simultaneous losing positions it takes to reach a stop-out. Widening your sl to reduce the odds of getting stopped out on noise increases the margin required per lot at most brokers, which quietly increases your liquidation risk in the opposite direction — fewer concurrent trades needed to hit stop-out, because each one is now collateralized more heavily and moving further against you before it closes.

This is why two traders can run the identical entry logic, identical win rate, identical validated edge, and one of them gets margin-called during a rough week while the other doesn’t. The difference isn’t the edge. It’s how much of the account each trade is allowed to consume, and how many of those trades were permitted to be open at once.

Sequence risk: the losing streak your backtest already told you about

Every validated system has an expected distribution of consecutive losses baked into its win rate. A system with a real 55% win rate isn’t going to produce a smooth, alternating sequence of wins and losses. It’s going to produce streaks — some short, some long — and the length of the worst realistic streak can be estimated statistically before you ever risk live capital. The problem is that most traders size their positions based on the return the strategy averages out to over hundreds of trades, not based on the worst plausible streak it can produce over the next twenty.

This is sequence-of-returns risk, and it’s brutal specifically because it’s front-loaded. A losing streak that happens in month one of live trading, before the account has built any cushion of accumulated profit, hits differently than the same streak happening after six months of gains. The strategy’s long-run edge is identical either way. The account’s ability to survive long enough to realize that edge is not. A margin call doesn’t wait for the long run. It only cares about the equity curve between now and the next few trades.

The math here isn’t pessimistic guesswork, it’s a straightforward binomial calculation, and it’s worth actually looking at rather than assuming a validated win rate makes long losing streaks unlikely:

Consecutive losses	P at 55% win rate	P at 58% win rate	P at 62% win rate
4 in a row	4.1%	3.1%	2.1%
6 in a row	0.83%	0.55%	0.33%
8 in a row	0.17%	0.10%	0.05%

Those look like small numbers until you remember they’re the probability of that streak occurring starting at any specific point, not the probability across an entire trading history. Run a strategy for a few hundred trades and the probability of encountering at least one 6-in-a-row stretch somewhere in that sample climbs into double digits even at a healthy 58% win rate. A losing streak like that isn’t a sign the edge broke. It’s an expected, periodic feature of a system that wins 55-62% of the time — and it’s exactly the scenario your lot_size and concurrent-position limits need to be sized to survive, not the scenario your average expectancy assumes will happen.

Why session overlaps make this worse, not better

The London/New York overlap, 13:00-16:00 UTC, is attractive to a lot of strategies precisely because it’s the highest-liquidity, highest-volatility window of the trading day. That’s also exactly why it’s dangerous from a margin perspective. Spread widening during volatility spikes — around news releases, or simply during the initial burst of overlap-session activity — means your floating loss on an open position can move further, faster, than it would during a quieter session, and it can do so simultaneously across multiple correlated pairs if your strategy is running more than one instrument with similar entry logic.

Correlation is the quiet killer here. If your strategy trades EUR/USD and GBP/USD with the same session-window logic, those two positions are not statistically independent bets from a margin standpoint, even if your backtest evaluated them as separate return streams. A macro move that pushes USD strength across the board will move both positions against you at the same time, consuming margin from two directions simultaneously rather than one. The win rate calculated per-trade doesn’t capture this, because per-trade statistics don’t know about the other open position sitting in the same account.

What this actually implies for how you size

None of this is an argument that the strategy is broken, and it’s not an argument for reducing position size out of fear every time a losing trade closes. It’s an argument for sizing based on the account’s ability to survive the worst statistically plausible sequence of the strategy’s own known loss distribution, not the strategy’s average expectancy. If your backtest can tell you the longest realistic losing streak at your validated win rate, that number — not your average trade — is what should determine how much margin any single lot_size is allowed to consume, and how many concurrent positions across correlated instruments you allow the system to hold at once.

The psychological trap runs in a specific direction here. After a strategy validates well, the temptation is to size up in proportion to your confidence in the edge, because the edge feels proven. But confidence in the edge and survivability of the account are two separate variables, and a margin call only tracks the second one. You can be completely right about the strategy and still lose the account, if the account was never built to survive the ordinary variance a real edge is statistically guaranteed to produce eventually. The market doesn’t owe your equity curve a smooth path to your expectancy. It only owes you the expectancy, over a long enough run that your account has to actually still be open to collect it.