Trailing Stops Compared: Tight Trails Win Small and Cut the Big Winners

A trailing stop is a Stop loss that follows price as a trade moves into profit, and never moves back. It sounds like the best of both worlds: losses stay limited and winners get locked in. But every time the stop moves closer to price, it also raises the chance that a big winner gets shaken out halfway. This article compares seven methods traders actually use, with a 20,000-path price simulation and our own backtests on gold, to see what each one gains and what it gives up.

How a trailing stop works

Every method follows the same three rules.

1. Compute a new level from something: the highest high since entry, ATR, the latest swing low, or a moving average
2. Move one way only. On a long, if the new level is higher the stop moves up. If it is lower, the stop stays where it is
3. Choose when to check. Some methods move on every tick, others compute at the bar close and apply the new stop to the next bar

Rule 3 matters more than it looks. Per-tick trailing reacts fast but is sensitive to candle wicks. Bar-close trailing is slower but backtests far more reliably (see the OHLC section below).

The built-in MT5 trailing stop (right-click the position and choose Trailing Stop) is set in points. Once profit exceeds that distance, the Stop loss starts following price at the same distance. The common mistake is forgetting that it runs in your terminal, not on the server. Close the terminal or lose the connection and it stops moving, leaving only the last Stop loss already sent. More on server-side orders in Order types explained, and on running a VPS in Running an EA on a VPS.

Seven methods traders use

All examples are for a long position. Shorts are the mirror image.

1. Fixed distance

Stop = highest price since entry − a fixed distance, for example 20 pips or 1 × ATR at entry. This is what the MT5 trailing stop does. It is the simplest, but the distance does not adapt to the market: too tight on volatile days, too loose on quiet ones.

2. ATR and the Chandelier Exit

Uses ATR (the average bar range) as the unit of distance. The best-known version is Chuck LeBeau’s Chandelier Exit.

• Stop = highest high of the last 22 bars − 3 × ATR(22)

The distance widens by itself when the market gets volatile. The ATR formula is covered in ATR and Stop loss placement.

3. Swing low (price structure)

Move the stop to just below the latest swing low price has made. In this article’s simulation a bar counts as a swing low when its Low is below the two bars on each side. Note that such a point is only confirmed two bars later. If a tool draws it straight away, it is peeking at the future. More in Repainting indicators and look-ahead.

The upside is that the stop sits somewhere meaningful: if price breaks the last swing low, the uptrend’s structure really has broken. The downside is uneven distance. Sometimes the stop jumps several ATR in one go.

4. Moving-average trail

Exit when price closes below a line such as EMA 20, or use the line’s value directly as the stop. In a strong trend price can stay above a short average for a long time, but during a pause it closes below easily. Basics of moving averages are in Moving averages.

5. Parabolic SAR

Created by J. Welles Wilder (the same Wilder behind RSI and ATR) specifically as an exit. SAR stands for Stop and Reverse. For a long:

• Next bar’s SAR = current SAR + AF × (EP − current SAR)
• EP (extreme point) = highest high since the uptrend began
• AF (acceleration factor) starts at 0.02 and rises by 0.02 each time a new high is made, up to 0.20 (MT5’s defaults are Step 0.02 and Maximum 0.2)
• SAR may not be above the Low of the last two bars

What sets SAR apart is that it accelerates toward price over time. Even if price goes nowhere, the accumulated AF keeps pushing SAR up, so a trade that stalls gets closed on its own.

6. Step trail

Move the stop in steps measured in R (the initial stop distance). For example, at +1R move the stop to entry, at +2R move it to +1R, and so on. It is like moving to break-even and then continuing step by step. Between steps the stop stays put, which leaves room for pullbacks.

7. Time stop

This one doesn’t trail the stop. It requires the trade to “work” within a set number of bars, for example: if the stop hasn’t been hit after 20 bars, close at the bar close. The reasoning is that an entry with an edge usually shows it within some time. A trade that sits still is capital doing nothing.

Worked example: the first three bars of Parabolic SAR

The numbers come from the simulated price in the next figure (price units, not a real market).

• Long at a close of 84.33, ATR(14) = 1.311
• Initial stop 2 ATR away = 2 × 1.311 = 2.62, which is 1R, so it sits at 84.33 − 2.62 = 81.71
• SAR starts at the initial stop, EP starts at the entry bar’s High = 85.21, and AF = 0.02

Bar	High	EP	AF	Next bar’s SAR
1	84.46	85.21 (no new high)	0.02	81.71 + 0.02 × (85.21 − 81.71) = 81.78
2	85.83	85.83 (new high)	0.04	81.78 + 0.04 × (85.83 − 81.78) = 81.94
3	85.88	85.88 (new high)	0.06	81.94 + 0.06 × (85.88 − 81.94) = 82.18

Check: all three values are below the Lows of the last two bars (the Lows of bars 1–3 are 83.58, 83.98 and 85.26), so none needs to be pulled down. Notice that after only three bars AF has gone from 0.02 to 0.06, so SAR is already moving three times as fast.

Simulating 20,000 paths: who gets what

One picture proves nothing, because every path gives a different result. So we simulated a large set of paths and ran every method on the same prices.

• Price is a random walk with a small upward drift of about 0.04 ATR per bar. Each bar has 8 sub-steps, used to check exactly where a stop is hit
• Every path is a long trade with an initial stop of 2 × ATR(14) = 1R. Every method moves its stop one way only and never below the initial stop. Maximum hold is 250 bars
• When price jumps through a stop, the fill is at the worse price, which is why the untrailed version’s average loss is −1.08R rather than exactly −1R
• No spread is deducted yet

Method	Win rate	Avg win	Avg loss	Expectancy per trade	Trades reaching 3R+	Avg bars held
Fixed 3R target (no trail)	36.7%	3.00R	−1.08R	0.415R	36.7%	22.3
1R step trail	34.3%	2.42R	−0.74R	0.348R	7.4%	17.6
Time stop 20 bars	46.4%	1.79R	−1.00R	0.297R	7.4%	14.2
Swing low	36.2%	1.80R	−0.62R	0.255R	6.9%	13.3
Parabolic SAR	42.0%	1.56R	−0.74R	0.223R	5.6%	11.2
Chandelier 3 ATR	41.0%	1.23R	−0.54R	0.183R	4.9%	9.6
Fixed 2 ATR	39.4%	1.34R	−0.60R	0.166R	4.0%	8.7
Close below EMA 20	41.5%	1.02R	−0.47R	0.153R	3.9%	7.4
Fixed 1 ATR	40.0%	0.70R	−0.36R	0.060R	0.6%	3.4

Expectancy = win rate × average win − loss rate × average loss. Check the first row: 0.367 × 3.00 − 0.633 × 1.08 = 1.101 − 0.684 = 0.417R, matching the measured 0.415R apart from rounding. The last row: 0.40 × 0.70 − 0.60 × 0.36 = 0.280 − 0.216 = 0.064R, close to 0.060R.

What the table says

1. Win rates barely differ, but the size of winners differs enormously. The 1 ATR trail wins 40.0% and the 3R target wins 36.7%, only about 3 points apart. Yet the average winner is four times larger (3.00R against 0.70R), and trades reaching 3R or more drop from 36.7% to 0.6%. The tight trail didn’t buy much extra win rate. It sold almost all the big winners.

2. If price has no direction, every method scores zero. We ran the same set with the drift removed. Every method’s expectancy landed between −0.021R and +0.007R, which is zero within the sampling error. This is a well-known mathematical result (optional stopping): on a truly random price, no exit rule can create profit. A trailing stop only changes the shape of the results, such as winning often but small, or winning rarely but big.

3. With drift, expectancy follows time in the market. Sort the table by “Avg bars held” and you get almost the same order as expectancy. Per bar held, every method earns about the same, roughly 1.8–2.1R per 100 bars. In a world like this (the math is known as Wald’s identity) average profit equals drift × time held. A tight trail isn’t “capturing profit better”. It is just leaving earlier.

A trailing stop doesn’t create profit. It only decides how long you stay in the market. If what pays is a long trend, the method that stays in longer collects more of it.

4. Per-trade costs make tight trails worse still. Deduct a cost of 0.05R per trade (such as spread) and count per 100 bars in the market: the 1 ATR trail falls to 0.31R, because it holds each trade for only 3.4 bars and so pays costs about 6.6 times as often as the 3R target over the same time. The other methods stay around 1.3–1.7R per 100 bars. More on costs and short distances in Spread cost on small timeframes.

The simulation’s limit: real prices are not a random walk with constant drift. Trends and ranges alternate, and in a range a tighter trail may lose less. This simulation explains the mechanism. It does not say which method will be best in a real market.

From our backtests

Results on gold (XAUUSD) in the MT5 Strategy Tester point the same way as the simulation: every version that pulled the stop in early made less.

A step trail on a short-target leg. A pullback system on gold, on its short-target leg (target at about 35% of the frame), real ticks 2026, 3,000 balance, 0.01 lot. We tried four step trails in the form “when price reaches X% of the frame, move the stop to Y%”.

Version	Net profit	Winning trades (of 219)	Equity drawdown
No trail	+2,532	110	265
At 25, move to 15	+2,319	109	331
At 20, move to 10	+2,150	105	309
At 20, move to entry	+2,114	94	298
At 15 to entry, then at 25 to 12	+2,096	94	292

The earlier the move, the less it made, in order, and drawdown went up despite the locked profits, because trades that used to reach the target became scratches. Pulling back to around the entry before reaching the target is normal for short trades, so a step trail has no room to work.

Moving the stop to entry at 1R. A pullback system on large swings on gold M1, real ticks 2026, 3,000 balance, 0.02 lot. Without the move it made +11,485. Moving the stop to entry once profit reached 1R left +5,636, a 51% drop, and the win rate fell from about 32% to 24%, because trades stopped out at entry are not winners. When we looked at how deep winners pulled back before running, the trades that became big winners often came back through the entry area first. A stop parked at entry removes exactly those.

Moving later, at 2R, told a different story. 2026 made +10,717, close to unchanged, and over five years (2022–2026) it made +16,590 against +16,355 without the move, with the maximum drawdown down from 23.2% to 20.0% (2022–2025 used simulated data, not real ticks). A profit difference of about 1% is within noise, so this counts as lower risk, not more profit. The full break-even discussion is in Moving the Stop loss to break-even.

Turning the tightness dial one step at a time. A small-box breakout system on gold M5, real ticks 2026, 3,000 balance, 0.01 lot. The trailing stop starts once profit reaches X% of the target distance, then follows Y% of the target distance behind.

Start / follow distance (% of target distance)	Net profit	Profit factor	Win rate	Equity drawdown
7 / 5	−392	0.82	88.3%	16.9%
15 / 10	−118	0.96	83.5%	13.8%
30 / 20	−28	0.99	74.5%	12.6%
50 / 30	−174	0.97	64.8%	12.7%

This shows the trade-off most clearly. The tightest trail won 88.3% and lost the most. Loosening it lowered the win rate at every step, down to 64.8%, while the money improved for three steps and then got worse again. So there is a sweet spot, not “wider is always better”. And no setting made that year profitable. A better exit can lose less, but it cannot create an edge the entry doesn’t have. (30/20 has not been tested on other years. Picking it from one year is fitting to the past.)

One more example covered earlier: the same entries on gold M5 in 2026 (real ticks, 0.02 lot), exiting with a profit lock at +2 dollars, won 78.5% but lost 493 dollars, while a far target won 38.9% and made 3,185 dollars. Details in A high win rate doesn’t mean profit.

All of these are backtest results. Past results do not predict future results, and they are not returns you should expect.

The OHLC trap: tight trailing stops look too good

Test modes that don’t use real ticks have to guess how price moved inside each bar. A trailing stop that moves on every tick is the system most affected by that guess, because its result depends on whether price reached the High or the Low first.

• An 8-bar box breakout on gold M1 with a tight trailing stop, 2026, 0.01 lot: generated ticks made +3,460.68, real ticks +915.30, inflated 3.78 times
• A multi-position system on gold M1, 3,000 balance, changing only the stop-lock method: in 2026 the OHLC mode showed profit rising from about +33,000 to +155,804, but on real ticks both versions lost about 2,900, nearly the whole balance (the two real-tick runs covered slightly different months, but the scale is the same)

If a system trails its stop inside the bar, judge it on real ticks only, or switch it to computing the stop at the bar close, which gives similar results in every mode. More in MT5 test modes and tick data quality.

A trailing stop that looks good on OHLC but has never run on real ticks should be treated as untested.

How to choose

• Know what makes the system money. A trend system whose profit comes from a few big trades needs a wide trail, such as Chandelier 3 ATR or a swing low, or no trail at all with a far target. A system with short targets doesn’t need a step trail. For the trend-following view, read Trend following research
• Trail wider than normal noise. Measure how deep the system’s winners pull back before running, and set the distance wider than that, not by feel
• Start trailing late. In systems whose winners pull back deeply, moving to entry at 2R or later tends to cost less than at 1R
• Want a higher win rate? Close part instead of tightening the stop. Close half and let the rest run with the original stop, so you keep part of the big winners. See Partial close
• Size from the initial stop. A trailing stop doesn’t reduce the risk at entry. The real risk is the distance to the first stop, so size the lot from it, as in Position size from Stop loss
• Always compare with no trail. Test a fixed-target version as the benchmark, then compare profit, drawdown and trade count, on real ticks

Common mistakes

• Trailing tight to “lock in profit.” You get a portfolio of small winners, and the big winners that pay for every stop are gone
• Moving the stop backward. Widening it when ATR expands or price pulls back isn’t trailing any more. It is adding risk in the middle of a trade
• Believing the MT5 trailing stop works with the computer off. It stops with the terminal
• Using an unconfirmed swing low. A swing low needs the bars to its right to close first, otherwise the backtest looks better than reality
• Judging from one painful trade. The trade that ran and then came back to the stop is easy to remember. The trade that pulled back and then ran far because it wasn’t trailed is not. Look at hundreds of trades together
• Tuning the trail distance on one year. The best value in one year is rarely the best the next. Test several periods before using it live