The Arithmetic of Losing Streaks: What Your Backtest Did Not Tell You

Every EA loses several trades in a row. What separates buyers who survive that from buyers who do not is rarely the EA. It is whether they worked out, before the first trade, how long the streak could be and what it would cost. The numbers are simple, and almost nobody does them.

Assume trades are independent and the win rate is p. The probability of k losses in a row starting from any given trade is (1−p)^k. Over N trades you have roughly N chances for such a run to begin, so streaks of length k where (1−p)^k is about 1/N will appear at least once. Solve for k and you get the longest streak you should expect as a matter of routine, not misfortune.

Over 500 trades:

· Win rate 35%: expect a streak of about 14 losses.

· Win rate 50%: about 9.

· Win rate 65%: about 6.

· Win rate 90%: about 2 to 3.

The streak grows with trade count, but slowly, and it falls with win rate, but less than intuition says: a 65% system, which feels almost safe, still runs six in a row routinely.

The strategy report line "maximal consecutive losses" is the longest streak that happened in one particular run of history. It is one draw from a distribution with a long tail. The next 500 trades are a fresh draw, and there is a fair chance they produce a longer run than the one you saw. A practical rule: take the backtest figure, multiply it by 1.5, and size your risk so that a streak of that length is survivable. Survivable financially, meaning the drawdown does not take you near a margin call or a stop-out level. And survivable psychologically, meaning you would still let the EA take the next trade at full size, because the trade after the streak is statistically no different from any other.

The formula assumes trades are independent. They are not: markets stay in one regime for weeks, and losses cluster. So treat the figures above as the optimistic floor: the streak a 50% system will actually deliver at some point is more likely 11 or 12 than 9. If a seller's backtest shows a maximum streak shorter than the formula predicts for its win rate and trade count, the more likely explanation is a short or friendly test period, not an unusually stable system.

Take a 15-loss streak. At 1% risk per trade, each loss removes 1% of the current balance: 0.99^15 = 0.86, a drawdown of roughly 14%. At 3% risk per trade, the same 15 losses give 0.97^15 = 0.63, a drawdown of roughly 37%. That needs a 59% gain to get back to even, and most people switch the EA off well before it gets there, locking in the loss at the worst point. Same EA, same trades, same streak. The only variable was a number you typed into the inputs.

A run of losses within the range the arithmetic predicts is the system working as designed. Turning it off, halving the lot, or closing trades early during that run is the same mistake as any other manual intervention: you replace a tested rule with an untested reaction. The question to ask is not "how many in a row" but "is this streak longer than 1.5 times the backtest figure, and is the EA still behaving as described". If both answers are no, nothing has happened that you were not warned about, here, in advance.

Our three best-selling EAs: GOLD NEURON (AI) https://www.mql5.com/en/market/product/187329 · ATLAS PORTFOLIO https://www.mql5.com/en/market/product/182751