Stop loss in a futures strategy: fixed ticks, ATR, time or trailing?

Contents

1. Where should the stop loss go in a systematic futures strategy?
2. What are the four stop types and how do they differ?
3. Why does a tighter stop often make a strategy worse?
4. How does a trailing stop formula work?
5. What happens to a stop in a fast market?
6. How do you test stop variants on the same strategy?
7. How do you read the result: expectancy, drawdown and trade count?
8. Frequently asked questions
9. References

Where should the stop loss go in a systematic futures strategy?

Put the stop where a normal bar-to-bar wiggle will not reach it, then let the data choose the type. Start with a volatility-scaled stop at a few ATR, test fixed-tick, time and trailing variants on identical rules, costs and walk-forward windows, and keep the variant that improves expectancy and drawdown after costs, not the one that feels safest.

A stop loss is a resting order that closes the position when price moves against it by a set distance. In a systematic strategy the stop is a parameter like a moving-average length: it has a value, that value changes the result, and you find the right one by testing. Average True Range (ATR) is the average size of a bar's full move, including gaps, over a chosen number of bars, so a stop measured in ATR adjusts itself when the market gets calmer or wilder. Expectancy is the average result per trade after costs; you can work it out from win rate and average win and loss with the win rate and expectancy calculator.

A practical order of work:

1. Build and test the entry and exit rules with no stop at all, so you know what the raw idea does.
2. Add an ATR stop wide enough to sit outside normal noise. Two to three ATR of the trading time frame is a common starting range.
3. Test fixed-tick, time-based and trailing variants on exactly the same rules.
4. Keep the variant that improves expectancy and maximum drawdown together, after commission and slippage.

What are the four stop types and how do they differ?

The four stop types answer four different questions: how far, how far relative to volatility, how long, and how much open profit to give back.

• A fixed-tick stop is a stop placed a set number of ticks (or points) from the entry price, the same distance in every trade.
• An ATR stop is a stop placed k times the current ATR from the entry price, so the distance grows in volatile markets and shrinks in quiet ones.
• A time stop is an exit that closes the trade after a set number of bars if the trade has not done what the rules expected, regardless of price.
• A trailing stop is a stop that follows the best price reached since entry at a fixed distance and only moves in the trade's favour.

	Fixed ticks	ATR-based	Time-based	Trailing
What sets the distance	A number of ticks from entry	k × ATR from entry	A number of bars since entry	Distance from the best price since entry
What it protects against	One large loss	A change in volatility	Dead trades that tie up margin	Giving back open profit
Typical failure	Too tight when volatility rises	Needs an ATR length that also has to be tested	Exits trades that are still valid	Exits trends on every pullback when the distance is small
Order type in live trading	Stop-market	Stop-market	Market order at the bar	Stop-market, moved each bar

The types can be combined. A common pair is an ATR stop for the first bars plus a time stop that removes trades which have gone nowhere. Every combination is one more variant to test, so add them one at a time.

Why does a tighter stop often make a strategy worse?

A stop placed inside the market's normal noise is hit by random wiggles before the expected move can happen, so the strategy loses more often and pays more in costs, even though each single loss is smaller.

Think of a smoke detector set too sensitive. It goes off every time you make toast. You do not get more safety; you get more false alarms, and eventually you take the battery out. A tight stop works the same way. Price moves up and down inside a range on every time frame, and a stop inside that range is triggered by movement that means nothing.

Three things happen at once when the stop moves inside the noise:

1. The win rate falls, because trades that would have reached the target are closed early.
2. The trade count rises, because the rules often re-enter after a stop, so one idea becomes two or three round trips.
3. Costs rise with the trade count, and every stopped-out trade also pays slippage on the exit.

There is a second effect that is easy to miss. The maximum drawdown of a strategy is usually a run of losses, not one large loss. A tight stop makes each loss smaller but the run longer, so the drawdown often does not improve at all. This is why the section on reading the result looks at expectancy, drawdown and trade count together.

How does a trailing stop formula work?

A trailing stop follows the best price since entry at a fixed distance and only moves in the trade's favour, so it locks in part of an open profit but also ends a trend on the first pullback deeper than its distance.

For a long trade the rule is:

Stop = highest high since entry − k × ATR

Here highest high since entry is the highest price the market has reached since the trade was opened, ATR is the Average True Range at the time of the calculation, and k is a multiplier you choose, for example 2 or 3. For a short trade the mirror applies: the lowest low since entry plus k × ATR. The stop is recalculated at the end of each bar. If the new value is higher than the old one, the stop moves up. If it is lower, the stop stays where it was. It never moves down.

Two details matter when you turn this into a testable rule:

• Calculate the stop from completed bars and apply it to the next bar. If the stop for the current bar uses that bar's own high, the backtest knows the high before the bar is finished, which is a form of look-ahead.
• Decide whether the trailing stop replaces the initial stop at once or only after the trade is in profit by some amount. Both are valid; they are different variants.

The value of k decides the character of the exit. A small k gives back little open profit but exits on every pullback, which hurts trend-following rules. A large k lets trends run but gives back more at the end. There is no universal k; it is found the same way as the other stop parameters, in the test described under “How do you test stop variants on the same strategy?” Setting a range for k in the exit conditions of the strategy generation of AlphaPrime (Valynox) gives you a set of trailing stop variants to compare under one entry rule.

What happens to a stop in a fast market?

A stop is usually a stop-market order, so in a fast move or across a session gap it fills at the next available price, which is beyond the stop level, and a backtest that fills every stop exactly at the stop price is optimistic.

Slippage is the distance between the price in the test and the price of your real fill. For a stop order the assumption is the stop level. In a normal market the fill is usually one tick or so worse. In a fast market, around a news release or at a session open, the first trade after the stop is triggered can be many ticks away. Across a session break there may be no trade at all between the last close and a much lower open, and the stop fills at that open.

What to do in the test:

1. Give stop exits their own slippage setting, larger than the slippage on entries and limit-style exits. One tick per side is a common minimum for liquid futures with market or stop orders; stops deserve more.
2. Fill a stop at the bar's open when the bar opens beyond the stop level, not at the stop level.
3. Rerun the test with the stop slippage doubled. A strategy whose result depends on exact stop fills has a cost problem, not an edge.

The same logic as in trading costs in backtests applies here: tighter stops are hit more often, so they pay this extra slippage more often. In backtesting settings, fees, slippage and next-bar execution should be fixed before any stop variant is compared.

How do you test stop variants on the same strategy?

Keep the entry rules, data, session, costs and next-bar execution identical, change only the stop, run every variant over the same period and through the same walk-forward windows, and compare the joined out-of-sample results rather than one in-sample number.

The test is only fair if one thing changes at a time. If variant A was run on day-session bars and variant B on full-session bars, or with different slippage, the difference in results says nothing about the stop.

1. Lock the base. Entry and exit rules, the data series, the trading session, commission, slippage for entries and for stops, and next-bar execution. Write them down.
2. Define the variants. For example: no stop; fixed 20 ticks; 1, 2 and 3 ATR; a 10-bar time stop; a 2-ATR trailing stop. Keep the list short enough that you can look at every result.
3. Run all variants on the same full period. This gives the in-sample view.
4. Run the same walk-forward windows on each variant. Walk-forward analysis tunes on one window and tests on the next, so the joined out-of-sample curve shows how the stop behaved on data the tuning did not see.
5. Compare out-of-sample expectancy, maximum drawdown and trade count across variants, as in the next section.

Every variant you compare is one more draw from the same data, and the best of many draws is partly luck. Bailey and López de Prado (2014) describe how the best result among many trials is inflated by this selection, and why the number of trials must be taken into account when reading it. Keeping the variant list short and reading the out-of-sample curve, not the in-sample peak, is the practical defence.

How do you read the result: expectancy, drawdown and trade count?

Judge a stop by expectancy per trade after costs, maximum drawdown and the number of trades together; a stop that lowers drawdown a little but cuts expectancy and multiplies stopped-out trades has made the strategy worse.

Read the three numbers as a set:

• Expectancy per trade after costs. This is the number the stop must not destroy. If it falls toward zero as the stop tightens, the stop is inside the noise.
• Maximum drawdown, which you can work out from an equity series with the drawdown calculator. This is what the stop is supposed to improve. If it does not fall when the stop tightens, the stop is only changing the shape of the losses, not their total.
• Trade count. A rising count as the stop tightens means re-entries after stops. Each one pays a round trip of costs.

Check the same table on the out-of-sample curve from the walk-forward run. The in-sample table tells you which stops are worth looking at; the out-of-sample table tells you which ones held up.

What this means if you use AlphaPrime: the stop types and their distances are parameters that AlphaPrime evaluates with the same fees, slippage and next-bar execution on every variant, so a stop comparison is one more run on the same rules, and strategy validation applies the same walk-forward windows to every variant.

Frequently asked questions

What does stop loss mean in futures trading?

A stop loss is a resting order that closes an open futures position when price moves against it by a set distance. It is usually a stop-market order: once the stop price trades, the order becomes a market order and fills at the next available price. In a systematic strategy the stop distance is a tested parameter, not a fixed habit.

How do you set an ATR stop loss?

Measure the Average True Range over a chosen number of bars on your trading time frame, multiply it by a factor k, and place the stop that far from the entry price. Two to three ATR is a common starting range for the first test. Then test neighbouring values, because the right k depends on the entry rules and the market.

What is the trailing stop formula?

For a long trade the stop equals the highest high since entry minus k times ATR; for a short trade it is the lowest low since entry plus k times ATR. The stop is recalculated at the end of each bar and only moves in the trade's favour, never back. Calculate it from completed bars and apply it to the next bar to avoid look-ahead.

Is a strategy without a stop loss acceptable in a backtest?

Yes, as a baseline. Testing the raw rules with no stop shows what the idea does on its own, and every stop variant is then compared with that baseline. Many rule sets already exit on a signal or at the session close, so the stop's job is to limit the worst run of losses, which the maximum drawdown measures.

References

• Bailey, D. H., & López de Prado, M. (2014). The Deflated Sharpe Ratio: Correcting for Selection Bias, Backtest Overfitting and Non-Normality. The Journal of Portfolio Management, 40(5), 94–107.