When to Stop Trading a Strategy

[Fairman]

When to Stop Trading a Strategy

Knowing when to quit a strategy is as important as knowing when to start. Both mistakes are expensive: abandoning a working system too early, or clinging to a broken one too long.

Define the stop conditions in advance, while you are calm. Useful triggers are numerical. A drawdown larger than the maximum seen in testing by a set margin, for example 1.5 times. A rolling expectancy over the last 100 trades that falls below zero. A loss of the market condition the strategy requires, such as the end of a trend regime. Or a structural change, like new broker costs.

When a trigger occurs, do not abandon the system automatically. Reduce risk, and investigate. Check execution, costs, and rule adherence first. Many apparent failures come from deviations rather than the strategy.

If the analysis shows a genuine edge loss, retire or rework it, and document why.

Equally important are conditions where you should not stop: a normal losing streak within the range of expectation, or a short period of underperformance.

The statistics from your testing tell you what normal looks like. Compare live results to that range, using the expected maximum losing streak and drawdown.

Write the criteria in your plan, so that the decision is rule-based.

Practical step: define two numbers for your strategy, a drawdown limit and a rolling expectancy floor, that trigger a formal review.

[Fairman]

I like writing the stop conditions on the first page of the strategy doc with a date next to them. When the drawdown hits, you're reading a decision you made calmly instead of making one in the middle of it.

[Fairman]

The investigate step usually turns up something boring. A spread change, a session I wasn't trading before, moving stops when I shouldn't. Fix that first, then judge the strategy.

[Shadow Trader]

The 1.5 times drawdown trigger makes sense, though it is worth checking how the test drawdown was measured. A maximum drawdown from one backtest is a single path, and the same trades in a different order could easily have produced a deeper one.

Shuffling the test trades a few hundred times and taking the 95th percentile drawdown gives a sturdier baseline. Say the backtest shows 8 percent, the shuffled figure might be 12, and that changes where the review trigger should sit.

A rolling expectancy over 100 trades is also slow for a low frequency system. At five trades a week that window covers almost five months, so a strategy can lose its edge for a whole quarter before the number turns negative.

A shorter check over 30 trades is too noisy to act on alone, but it works as a warning that brings the formal review forward. It does not decide anything, it just tells you to look sooner.

On the investigate step, comparing live fills with the fills the tester assumed is often the quickest answer. Has anyone set a trigger based on execution quality alone, such as average slippage doubling, separate from the results?

[FXS Prop Desk]

The rolling expectancy over the last 100 trades is a good trigger, though it reacts slowly. Combining it with a faster warning, such as a rolling 30-trade expectancy below zero, which leads to reduced size rather than stopping, gives two levels of response. The slow trigger decides whether to stop, while the fast one limits damage in the meantime.

Recording the reason for stopping helps later reviews.

[Shadow Trader]

One trigger that gets less attention than depth is time under water. A strategy can sit inside its drawdown limit and still go far longer without a new equity high than anything seen in testing.

Say the longest flat stretch in the test was 14 weeks. A live stretch past 20 weeks deserves the same formal review as a deep drawdown, even if the losses are mild.

The other half of the plan is the restart rule. Once a strategy is paused, keep recording its signals on demo, and decide in advance what would bring it back, such as a positive expectancy over the next 40 signals.

Without that, a paused strategy tends to stay paused forever, or gets restarted on a hunch right after a good week that was never traded. Both outcomes waste the work that went into testing it.

Results far better than the test deserve a look too. If live expectancy runs at three times the backtest for a few months, something has changed, and it may be the risk being taken rather than the skill.

[Shadow Trader]

Scalping stops are small, so the spread can represent a meaningful fraction of the risked distance. With a 5-pip stop and a 1-pip spread, one fifth of the stop is consumed the moment a trade opens. The real risk is not the stop distance alone but stop distance plus spread, plus any slippage.

Position sizing should therefore use the full cost-inclusive distance. When the stop is measured from the chart price rather than the actual entry price, the loss at the stop includes the spread. Sizing from the nominal 5 pips understates the true loss by around 20 percent in this example.

The ratio also shapes reward. A target that looks like 1:1 on the chart becomes unfavorable once costs are counted, and the break-even win rate rises well above 50 percent. The tighter the stop, the more severe the distortion.

Spreads widen at rollover and around news releases, so a size that works at midday can be oversized then. A common guideline is to skip setups where the spread takes more than a modest share of the planned stop, with the threshold derived from the strategy's own trade data.

[Shadow Trader]

A strategy that works on a 100,000 dollar account may not work on a 500 dollar one, and the reverse is also true.

Small accounts face minimum lot sizes, which restrict risk control. A 1% risk per trade might be impossible with a 10 pip stop. Margin limits can block multiple positions. Costs take a larger bite relative to equity when position sizes can't shrink further.

Larger accounts face the opposite issue: execution. Bigger orders can slip, partial fills appear, and some strategies depend on small size to work at all.

Before committing to a strategy, it helps to run the numbers at the intended account size. Minimum lot, stop distance, risk percent, and costs all interact, and the result can differ greatly from a backtest done in the abstract.

A mini-case to illustrate. A strategy with a 6 pip stop and a 1% risk rule needs, on a 300 dollar account, a position size of 0.05 lots. Many brokers offer 0.01 minimums, so that's fine. On a 100 dollar account, the same rule demands 0.016 lots, which can't be placed. The strategy hasn't changed, but the account can't express it properly.

What strategy changes became necessary when the account size changed, up or down?