also: losing streak, max consecutive losses, worst run
The longest consecutive series of losing trades or the largest cumulative loss within a contiguous run of losing trades in a backtest or live record.
The longest run of losing trades the strategy has already been through, and how much that run cost. It is the number that predicts whether a trader will still be running the EA at the end of a bad stretch.
Most accounts do not fail because of a single catastrophic trade; they fail because someone switched off a working strategy during its worst run. Maximum drawdown says how deep the hole was, and worst streak says how many times in a row the strategy was wrong on the way down — which is the part people actually cannot sit through.
Worst streak (count) = the longest run of consecutive losing trades · Worst streak (value) = the cumulative loss across that run The two definitions do not have to point at the same run. The longest run by count can be a sequence of small losses, while the costliest run by value can be shorter and heavier. Publish both, and say which one a quoted figure refers to.
No single number describes a strategy. These metrics change how this one should be read.
The chance of a run of N losses in a row follows directly from the loss rate, so the expected worst streak can be read off the win rate before any money is deployed.
The streak that feels like the EA has broken is usually the streak the win rate predicted before deployment.
Calculation 0.6⁶ = 0.0467 → a six-loss run appears roughly once every 21 trades at this win rate
Result A ten-loss run is an ordinary feature of a 40% win rate, not evidence of failure
A published streak figure is a floor, not a ceiling — it is the worst the sample happened to contain. Read it against the sample length and the win rate that produced it.
| Range | What it means |
|---|---|
| Quoted without the trade count | Uninterpretable. A short sample cannot contain a long run, so a flattering streak figure may only mean the test was brief. |
| Count only, with no cost attached | Half the picture. Eleven small losses and five heavy ones are different problems for the same account. |
| Both count and cumulative cost, over several hundred trades | The figure worth planning around. It can be turned directly into a rule for when to intervene. |
| Live streak longer than the tested worst | A decision point rather than noise — the strategy is operating outside the behaviour its sample described. |
AIStrategyMiner EA pages publish the full closed-trade ledger beside the drawdown and trade count, so the streak can be counted from the record rather than quoted. The two describe different things a deployer has to survive.
At a 40% win rate, a run of ten losses is an ordinary product of a few hundred trades. Deciding a strategy is broken because it produced the run its own win rate predicts is the most common way a working EA gets switched off.
It is the longest run inside the sample. A longer sample almost always finds a longer run, so the figure is a floor for future expectations rather than a limit.
Streak length and streak cost are separate figures, and a strategy that varies its risk per trade can have its costliest run be one of its shorter ones.
Maximum drawdown is measured from an equity peak and can be produced by a single trade. A losing streak is a run of consecutive closes and can end above the previous peak if the run started from a high. They answer different questions.
Each EA page ships a bt-trades.csv. Replaying it in close order gives the
longest run, the costliest run, and the run the win rate
predicts: ln(trades × win rate) ÷ ln(1 ÷ loss rate).
| 33 backtest ledgers | Figure |
|---|---|
| Longest run: shortest / median / longest | 2 / 8 / 38 trades |
| Records by longest run: ≤5 · 6–9 · 10–14 · ≥15 | 9 · 10 · 9 · 5 |
| Costliest run shorter than the longest run | 12 of 33 |
| Costliest run deeper than the listing’s maximum drawdown | 0 of 33 (median share 63%) |
| Longest run beats the prediction by more than 4 trades | 6 of 33, all multi-leg |
| Held-out windows (7–330 trades) beating the backtest run | 0 of 32 |
Count and cost often diverge: Iridescence’s longest run is eight trades, but its costliest is two, at 5.27%. A stop rule in trades and one in money fire on different days.
| Ledger | Win rate · trades | Predicted run | Longest run |
|---|---|---|---|
| Chrysalis | 70.7% · 693 | about 5 | 35 |
| Nautical | 30.22% · 756 | about 15 | 38 |
| Murmuration | 45.36% · 1,709 | about 11 | 26 |
| Cairn | 63.99% · 4,583 | about 8 | 14 |
| Longbow | 24.34% · 467 | about 17 | 13 |
The tester counts a portfolio’s closes in one sequence, so a week that stops every leg becomes one long run. That is how Chrysalis reaches 35 while winning seven in ten, at a cost of 12.87%. Longbow is the opposite: its 13-loss run is shorter than the 17 its own win rate allows.
The Strategy Tester report carries both definitions on two rows. “Maximum consecutive losses” gives the count with the money in brackets; “Maximal consecutive loss” gives the money with the count in brackets. On 12 of the 33 ledgers the two counts differ.
Nine portfolio builds expose it as the input
PortfolioMaxConsecutiveLosses, which halts every leg for the rest of the day
after that many losing closes and ships at 0 (off). Set it from the ledger: 3
fires inside the first year on 29 of the 33 records, and nine never exceed five
in a row. The ea-vs-manual replay prices that
early exit.
The streak is the count and the maximum drawdown is the hole; on every ledger here the hole is at least as deep as the costliest run. Predict the count from the win rate, price it with the expectancy, and let the risk of ruin calculator turn the three into odds. Then sort the EA catalogue by drawdown.