How to Set a Stop Loss
To set a stop loss, first find the price at which your reason for the trade stops being true, then check that it sits outside normal one-bar movement. The distance from entry to that price then determines your position size, never the other way round.
A stop is not a safety device and it is not a prediction. It is the price at which you have agreed in advance that the reason for the trade no longer holds, and everything else follows from that.
Before you start
A measure of how far the instrument normally moves in one bar. Average true range, or simply the median bar height over the last fifty bars. Without this you cannot tell noise from signal.
The price at which your reason for the trade is wrong. Below the level, beyond the structure, past the swing. This is a location on the chart, not a percentage of your account.
The amount you are willing to lose on this trade, in currency. A fixed figure decided before the order exists, because it is the numerator in the sizing arithmetic.
The steps
1. Find the price that makes the idea wrong
If you bought because a level held, the stop belongs below that level. If you cannot name the price that disproves the trade, there is no trade to size yet.
2. Measure the normal noise before placing anything
On the shared series a typical bar spans 0.493, the tenth percentile 0.17, and the ninetieth 1.101. A stop inside one ordinary bar’s movement will be hit by nothing in particular.
3. Place it beyond the structure, not at a round number
Below the swing low plus a margin for noise. Round numbers and obvious levels are where everyone else’s stops sit, which is exactly where price is drawn.
4. Let the distance set the size
Divide the currency amount you will risk by the distance from entry to stop. That is the size. A size chosen first and a stop fitted afterwards inverts the whole procedure.
5. Decide in advance whether it trails
A trailing stop converts an open profit into a decision made by price. Choose the distance before the trade, using the survival figures below rather than a feeling.
6. Add a time stop for trades that never work
If the move has not happened within the number of bars your reasoning implied, the reasoning was wrong even though the price stop was never touched.
7. Check the liquidity of the instrument
A stop is a market order once triggered. In a thin book it fills below the level you named, and the gap between the two is the real cost.
8. Never move it away from price
Moving a stop closer is a decision. Moving it further away is abandoning the one number the whole position was built around, after the position started losing.
How to tell it worked
Audit your last 30 trades against three counts.
Count the stops placed before entry rather than after. 30 out of 30 is the standard. A stop added once a position is open was placed by the position rather than by the plan.
Count the times you widened a stop after entry. The target is 0 out of 30. Being stopped and re-entering costs 2% of a median bar’s range on the shared series, and widening feels cheaper than that at the moment it happens while usually costing far more.
Then measure your average stop distance in typical bars. Under 1 bar is inside the noise. Over 5 is a stop chosen for comfort rather than for structure, and the size will be too small to matter.
What the trailing-stop data actually shows
Trailing stops were tested at four distances across 562 trials each on the shared series. The median survival was 3 bars at 1 ATR, 10 bars at 2, 22 bars at 3, and 32 bars at 4. Between 91% and 100% of positions were eventually stopped out at every distance tested.
Read the second number first. Almost everything gets stopped eventually, so the choice of trail distance is not a choice between being stopped and not being stopped. It is a choice about how long you stay in before it happens.
And the first numbers are not linear. Quadrupling the trail distance from 1 ATR to 4 multiplied
survival roughly tenfold, from 3 bars to 32 — so a wider trail buys disproportionately more time,
paid for with a larger give-back when it finally triggers. The figures are in
research/series-measurements.json.
The original data
Of the 24,971 unique videos in research/search-study-corpus.jsonl, 89 have an instruction-shaped
title mentioning stop losses, at a median of 20,904 views across 75 channels, with a maximum of
953,165. Position sizing, the arithmetic step four depends on, appears in only 34 at a median of
6,157. The counts come from site/rank_howto.py.
Eighty-nine videos on stops against 34 on position sizing, and the stop videos pull over three times the median. The two are the same decision — the stop distance is the input to the size — and the half people search for is the half that feels like protection rather than the half that is arithmetic.
The answer to the question on that chart is no, and the trailing data says why. Being stopped before a reversal is the ordinary outcome, not evidence of a badly placed stop: 91% to 100% of positions were stopped eventually across every distance tested. Widening after a stop-out is optimising on one observation — and the one observation you have chosen is the one that hurt.
When it fails
A range is where a tight stop fails repeatedly and expensively. Price oscillates without going anywhere, so a stop placed inside normal movement is touched again and again, each time costing the round trip. The instinct is to blame the entries, but the measurement in step two is what identifies the real problem: the stop was inside the noise, and in a range the noise is all there is.
The second failure is a gap through the level. A stop is a request that becomes a market order, and a price that never traded cannot fill you.
A third is sizing first and fitting the stop afterwards. It produces a stop placed where the loss is tolerable rather than where the idea is wrong.
A fourth is a stop at an obvious round number. That is where everyone else’s sits, which is precisely why price reaches for it.
A fifth is a stop in a thin instrument. The fill arrives well past the level and the planned loss was never the real one.
And a sixth is widening after a stop-out. It converts a defined loss into an undefined one at the worst possible moment.
Related
Stop loss covers what the order type actually does and why the fill is not promised. Trailing stop is the moving version and where the survival figures above come from. And position sizing is the arithmetic in step four, which is the half of this decision people skip.
The measurement that settled this for me was the trailing-stop survival test. I had assumed a wider trail was a meaningfully different strategy, and the numbers say it mostly buys time rather than outcomes — at one average true range the median position lasted three bars, at four it lasted thirty-two, and virtually everything was eventually stopped either way. That reframed the stop as a cost of participation rather than as protection.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.