Do Stop Losses Work? 183 Backtest Pairs With and Without

Stop loss vs no stop loss, measured: 18 of our published backtests ran the same entries with and without a stop, and some with and without a profit target. We compared 183 such pairs, version against version, on the result after costs, the worst single loss and the drawdown.

ES and NQ futures, gold, forex, Bitcoin · 18 tests, 183 pairs of versions · no new backtest: every number comes from the published trade logs

Guide · 183 pairs compared

The stop did not make money. It made the worst day smaller.

Written22 Oct 2026

Pairs and rules frozen22 Oct 2026, before any pair was compared

Tests: 18 · Pairs: 183 · Stop cut the worst loss: 94% of pairs · Stop raised the result: 34% · New backtests: none

Short answer: they limit losses, not results. In 125 pairs of our backtested strategies with and without a stop, the stop cut the worst single loss in 94% of pairs, but raised the result after costs in only 34%; where the only change was the stop, exactly half.

00The short answer

A stop on its own did not improve the result. Where the only difference was a stop, the version with the stop did better after costs in exactly 50% of the pairs. It did cut the worst single loss in 94% of them, to a median 60% of the loss without a stop.
Stops with targets did worse. Across all 125 stop pairs the version with the stop did better only 34% of the time, and where a stop and a target were added together, 24%.
The target was the damage. Adding a profit target to a version that already had the same stop raised the result in only 9% of 58 pairs, which fits targets closing the large winners that trend rules depend on.
The drawdown fell less reliably than the worst loss: it was smaller with a stop in 58% of pairs, not much more than half; on mean-reversion rules it was usually larger, likely because stops turn trades that would have bounced into strings of small losses.

01How we compared stop loss vs no stop loss

A pair is two published versions of one test with the same entry rule, one without a stop and one with it, fixed in a list before any pair was compared (the pairs, the specification). We checked that both versions of every pair start with the same trade.

A. Stop only added: the golden cross on 1-hour bars (ES, gold, EUR/USD), the Coppock curve, the EMA 9/20 on gold, and two gap fills (ES and Bitcoin’s CME gap), where the target is the same with or without the stop.
B. Stop and target added, the signal exit kept: the same crossover rules with 1R stops and 1R to 3R targets, the EMA 9/21 on ES, NQ and gold, the NQ EMA ribbon, and Supertrend on ES and four FX pairs.
C. A fixed stop, target and time limit replacing the signal exit: Bollinger, Donchian, RSI divergence and MACD on ES and four FX pairs.
T. A target added to the same stop: the crossover rules, the gold opening-range breakout, the initial balance breakout (ES, NQ) and the London breakout.

Both versions of a pair are compared in the same money unit from their trade logs (dollars per contract or ounce, pips, percent). Each rule set weighs the same (one test on one market, e.g. Bollinger on EUR/USD), so the FX study’s many pairs do not outvote a single gap fill. Tests: Bitcoin CME gap fill (A) · Bollinger Band Reversal (C) · Coppock Curve (1H) (A, B, T) · Donchian Channel Breakout (C) · EMA 9/20 gold (M15, H1) (A, B, T) · EMA 9/21 (1H) (B) · EMA ribbon 5/8/13 (NQ) (B) · Opening gap fill (ES) (A) · Six Strategies on Four Currency Pairs (B, C) · Gold Opening-Range Breakout (T) · Golden Cross (1H) (A, B, T) · Golden Cross (EUR/USD 1H) (A, B, T) · Initial balance breakout (ES) (T) · Initial balance breakout (NQ) (T) · London breakout (GBPUSD) (T) · MACD Crossover (C) · RSI Divergence (C) · Supertrend (B).

02What the stop changed

Kind of pair	Pairs	Result after costs higher with the stop	Worst single loss smaller	Worst loss with stop, as a share of without (median)	Drawdown smaller	Drawdown with stop, as a share of without (median)	Win rate lower
A. Stop only added	15	50%	94%	60%	69%	92%	100%
B. Stop and target added	70	24%	94%	55%	65%	80%	69%
C. Stop, target and time limit replace the signal exit	40	40%	93%	52%	50%	98%	78%
All stop pairs (A + B + C)	125	34%	94%	55%	58%	88%	77%

How sure are these shares? Flipping each rule set’s result at random 10,000 times, a share as far from 50% as the 34% of all stop pairs appeared 2% of the time, and as far as the 94% for the worst loss essentially never. The 50% of the stop-only pairs is exactly a coin flip. The drawdown shares are within chance (27%).

Stops also cost wins: the win rate fell in 77% of the stop pairs, as trades that ended as winners without the stop were closed at the stop first.

03The cleanest test: a stop and nothing else

Every pair where only a stop was added, with how the stop changed each measure:

Test and pair	Result after costs with the stop	Worst loss, with as a share of without	Drawdown, with as a share of without	Gates passed
Golden Cross (1H) ES: ES GH-L-N → ES GH-L-S	higher	46%	36%	1 → 2
Golden Cross (1H) ES: ES GH-LS-N → ES GH-LS-S	higher	41%	24%	1 → 3
Golden Cross (1H) Gold: Gold GH-L-N → Gold GH-L-S	lower	27%	96%	6 → 2
Golden Cross (1H) Gold: Gold GH-LS-N → Gold GH-LS-S	lower	49%	131%	5 → 1
Golden Cross (EUR/USD 1H) EURUSD: EURUSD GH-L-N → EURUSD GH-L-S	lower	35%	92%	1 → 1
Golden Cross (EUR/USD 1H) EURUSD: EURUSD GH-LS-N → EURUSD GH-LS-S	higher	30%	65%	1 → 1
Coppock Curve (1H) ES: ES CC-L-N → ES CC-L-S	higher	74%	101%	1 → 2
Coppock Curve (1H) ES: ES CC-LS-N → ES CC-LS-S	higher	74%	83%	1 → 1
EMA 9/20 gold (M15, H1) Gold M15: M15-L-N → M15-L-S	lower	95%	95%	2 → 3
EMA 9/20 gold (M15, H1) Gold M15: M15-LS-N → M15-LS-S	higher	95%	89%	2 → 2
EMA 9/20 gold (M15, H1) Gold H1: H1-L-N → H1-L-S	lower	98%	105%	5 → 5
EMA 9/20 gold (M15, H1) Gold H1: H1-LS-N → H1-LS-S	lower	101%	120%	4 → 4
Opening gap fill (ES) ES: GF-NS → GF-1X	higher	79%	84%	2 → 2
Bitcoin CME gap fill BTC: CG-NS → CG-S1	lower	33%	88%	1 → 1
Bitcoin CME gap fill BTC: CG-NS → CG-S2	lower	60%	101%	1 → 1

The version that came closest to an edge on this list had no stop: the long-only 1-hour golden cross on gold passed all six gates in its run without a stop (later Inconclusive after a data check) and 2 with a 1R stop. Its stop cut the worst loss to 27% of the loss without, but the result after costs was lower.

04Take profit vs no take profit

The 58 target pairs keep the stop and add a fixed target. A target raised the win rate in every pair (it is reached more often than a distant exit) and lowered the result after costs in about 91% of the pairs (rule sets weighted equally); 13 versions went from positive to negative after costs, 3 the other way. Every target pair here is a trend-following or breakout rule. Such rules typically earn much of their profit from a few long moves, which a fixed target closes early; that fits the result, though this comparison does not measure it directly. The same pattern, with target sizes of 1R, 2R and 3R, is in our risk-reward guide.

05Trend rules and mean-reversion rules

Stop pairs	Pairs	Result higher with the stop	Worst loss, with as a share of without	Drawdown, with as a share of without
Trend following (trend and breakout rules)	102	36%	61%	80%
Mean reversion	23	29%	34%	117%

Labels from our trend following vs mean reversion comparison. On mean-reversion rules the stop cut the worst loss the most (to 34%), which fits rules that buy weakness, where a falling market can keep falling; but their drawdown was usually larger with the stop (117% of without), likely because trades that would have bounced were closed at small losses first. With 12 mean-reversion rule sets, these are indications, not findings. The share of pairs where the stop raised the result was similar for both styles.

06Did the stop change the verdict?

Across the 125 stop pairs, the stop turned 20 versions from positive to negative after costs and 14 from negative to positive. The gates passed went down in 42 pairs, up in 30 and stayed the same in 53. No version with a stop passed all six gates.

07What this does not say

It does not say trade without a stop. The stop bought something real: a smaller worst loss in almost every pair. A single loss twice the size can end an account or a prop firm evaluation, and that is outside a comparison of totals.
These are the stops the specs defined (mostly 1R, 1.5 to 2 ATR, the other side of a range or the gap). Other stop distances can give other answers.
A backtest fills stops at the level or at the next price. In fast markets real stops slip further, which would make the stopped versions somewhat worse than shown.
It is our sample: 18 tests of popular rules, mostly trend following, with few mean-reversion pairs.

08How to use a stop loss, from these results

Use the stop to size the position, not to make the strategy profitable. It limits the worst trade; it did not improve the average one.
Be careful with fixed targets on trend rules. They raised the win rate and lowered the result in most of our pairs.
Test the strategy with the stop you will really use, and compare it with the same entries without the stop: if the stop changes the verdict, the edge was in the exits, not the entries.
On mean-reversion rules, look at the drawdown with the stop, not only the worst trade.
Count your costs: where a fixed stop, target and time limit replaced the signal exit (the C pairs), the number of trades rose by a median 15%, each paying the full cost; see how much trading costs ate across our backtests.

FAQCommon questions

Do stop losses work?

They do what they promise, cap the loss of a trade: in our tests the worst single loss was smaller with the stop in 94% of pairs, a median 55% of the loss without. They did not make strategies more profitable: the result after costs was higher with the stop in 34% of pairs.

Is it better to trade without a stop loss?

Not on this evidence. Where only a stop was added, the result was higher with it in exactly 50% of pairs, a coin flip, while the worst loss fell in 94%. A smaller worst loss is what keeps an account or a prop firm evaluation alive.

Does a stop loss reduce drawdown?

Less reliably than it reduces the worst trade: the largest drawdown was smaller with the stop in 58% of our pairs. On mean-reversion rules it was usually larger with the stop (a median 117% of without).

Should I use a take profit target?

On trend-following rules, our tests say be careful: adding a fixed target to the same stop raised the win rate in every pair but raised the result after costs in only 9%. The target closes the large moves these rules depend on.

Why did stops with targets do worse than stops alone?

In our pairs the stop alone was neutral (50% higher) and the target was the damage (9% higher when added to the same stop), so versions with both did worse (24%). What is a good risk-reward ratio.

09Check our arithmetic

Specification and pairsThe pair rules, questions and checks, frozen before the comparison, and the pair list written before any pair was compared.
sl_spec.md · sl_pairs.js · sl_pairs.json

Scripts and dataThe run reads both versions of every pair from the published trade logs in the same unit, checks trade counts and totals against the summaries and that both start with the same trade, then answers the questions; a second, independent script reads the logs itself and must give identical numbers.
sl_run.js · sl_ref.js · fm_logs.js · sl_rows.json · sl_summary.json · check results

The testsEach test’s frozen specification, code, trade logs and gates are on its own page, linked above.

10Corrections

None since publication.