Moving the stop loss? Let's find out.

Every trader has had the thought after a stop-out, usually within about four seconds of the fill. If the stop had been a little wider, that would have been a winner. And often it is true — you can go back to the chart and watch price clip your level by a few points and then run the whole way to target without you. It is one of the most reliable sources of regret in the business.

So we tested it properly instead of arguing about it. Every trade the system has taken since January, replayed bar by bar against real one-minute price data, at twelve different stop distances from the model's own placement out to three times as wide. Same entries, same targets, same everything — only the stop moves.

The wider stops won far more often. They also made dramatically less money. That contradiction is the whole article, and the arithmetic behind it turns out to be worth understanding.

One rule that makes the comparison fair

There is a trap in this kind of test, and most versions of it published online fall straight into the hole. If you widen the stop and keep the position size the same, you have not tested a stop distance — you have tested a bigger bet. Of course it makes more when it works. It also loses more when it does not, and comparing the two tells you nothing.

So every figure in this article is risk-normalised. When the stop widens to 1.5×, the position shrinks to two-thirds the size, so that a stop-out still costs exactly one unit of risk — one R — no matter which configuration you are looking at. Every column below risks the identical amount per trade. The only thing that changes is where the stop sits and how big the position is.

With that rule in place, here is the result that started the investigation. This is the short book, which we have written about separately, because it is where our edge actually lives.

Read that table twice. Going from the model's stop to a triple-width stop converts twenty-two losing trades into winners. The hit rate goes from a respectable 60% to a frankly excellent 79%. And the account ends up with less than a third of the money.

Where the money goes

A wider stop is not protection you are given. It is protection you buy, and the currency you pay in is position size. The question is never "does a wider stop save trades" — it always does — but whether the trades it saves are worth more than the size it costs. That exchange rate is knowable, and it moves against you fast.

That is the entire mechanism, and it is not subtle once the columns sit next to each other. Widening the stop improves how often you win by about a third. It cuts what you win by roughly two-thirds. There is a narrow region near the top where the first effect outruns the second — and then it stops, permanently.

The practical lesson is worth stating in plain language, because it inverts advice most retail traders have been given for years. A higher win rate is not evidence of a better system. You can manufacture any win rate you like by moving your levels; the arithmetic simply takes the money out somewhere else. Expectancy per trade is the number that pays you, and it peaks at 1.2× — where the hit rate is a fairly ordinary 67.6%.

The full sweep

Five rows compress a lot. Here is every width we tested on the short book, with the risk measures that decide whether a configuration is actually livable.

A caution we would rather raise ourselves than have raised for us: the peak at 1.2× is about three trades wide. Move a handful of borderline outcomes and it slides to 1.1× or 1.3×. Nothing in this data supports treating 1.2× as a precise optimum, and any system that only works at exactly 1.2× is fitted to noise. The honest reading is that 1.1× through 1.5× is one broad plateau, and everything past 1.8× is decisively worse.

Within that plateau there is a real choice, and it is not a choice about returns. Compare the two highlighted rows. At 1.2× you get the most money. At 1.5× you give up about 15% of it and receive in exchange the lowest drawdown in the entire study (6.25R), a 72.2% hit rate, zero losing months in eight, and a longest losing streak of three instead of seven.

Seven consecutive losses versus three is not a statistical detail. It is the difference between a subscriber who holds through a rough patch and one who switches the system off two trades before it recovers. Those two rows are optimising different things, and which one is correct depends entirely on who is sitting behind the account.

The wider stop cannot save a bad book

The most common use of a wider stop is as a rescue: a book is losing, so give it room. We tested that directly on the long book, which over this period ran below a coin flip.

The long book loses money at every stop width we tested. At 3.0× it wins three trades out of four and finishes at −0.28R. A 76% win rate that returns nothing is the cleanest possible demonstration that hit rate and profitability are different quantities.

The highlighted row carries the more dangerous finding, and it is the one we would most want a trader to take away from this article. Widening the stop made the drawdown worse before it made it better — from 17.54R at the model's stop to 19.72R at 1.2×. The intuition that a wider stop is inherently safer is simply false. In the region where you have not yet converted enough losers into winners, all you have done is make the remaining losers more expensive while shrinking the wins that offset them. You take the same beating with less to show for it.

And the one long book that works wants the opposite

One long book on our roster genuinely earns its place: NAS100, which ran 71% with a +14.3 point edge over its random-walk baseline. Since widening helps the short book, the obvious move is to widen NAS100 longs too. We checked. It is the wrong move.

Look at the first two rows. Going from 1.0× to 1.2× on NAS100 longs saves zero trades — twenty-four wins and ten losses either way — and shrinks every position by 17%. That is the exchange rate at its most brutally clear: you paid for insurance that covered nothing.

This is where the study stops being about stop losses and starts being about product design. Our short book wants a wider stop. Our one profitable long book wants a tight one. There is no single number that serves both, which means any system applying one stop rule to every position is leaving money on the table by construction.

Putting it together

So we ran the combinations. Same trades, same entries, same targets — varying only which books are enabled and what stop width each one uses.

The top row is what we actually did. The highlighted row is what the same seven months would have produced with two changes: only take the books that clear their baseline, and let each book carry the stop it wants. +37.90R against +20.62R, on 5.91R of drawdown against 13.14R, with zero losing months instead of two.

The second row is worth a glance too, because it is the version that requires no selection decisions at all. Keep trading everything exactly as before and widen only the shorts, and the book goes from +20.62R to +23.53R while drawdown drops from 13.14R to 9.73R. A smaller gain than the full configuration, but it is available without switching off a single trader.

Month by month, so nothing hides

A total is easy to arrive at by luck. Here is the short book at the three widths that matter, plus the best configuration, across every month of the study.

March is the row that keeps us honest. At the model's stop the short book scraped +0.78R; at 1.5× the same trades returned +6.86R. If the study had covered March alone we would be publishing a confident article about 1.5× being correct. May and June point the other way just as firmly. The plateau is real, the peak inside it is not.

Why you should believe the replay

Every number above comes from a simulation, and simulations are easy to make say whatever you want. Before running a single sweep we checked the engine against results we had already published, month by month, with no ability to adjust anything after the fact.

Four trades were also re-verified by hand against candles pulled fresh from the database. Exit timestamps matched the recorded fills to the minute in three cases and to a single aggregation bar in the fourth. Across the whole set, the replay agrees with the recorded win-or-loss outcome on 154 of 156 trades.

Two corrections came out of that audit and both are in the numbers above. One trade's one-minute bar showed a high of 1.36303 against a target of 1.36305 — two ten-thousandths short, a level the tick-level monitor caught and the aggregated candle lost. We deferred to the monitor. Separately, our first pass sequenced drawdown by entry time, which is not a real equity path, since a trade opened first can close last. Re-ordering by exit reduced every drawdown figure in the study by 20–25%.

What we are building because of this

The clearest conclusion in this study is not a number. It is that there is no correct stop width — only a dial with return at one end and smoothness at the other, and different traders who should reasonably sit at different points along it.

So we are adding it as a setting. The SkyAnalyst Automated Trader is getting a stop-loss multiplier — 1.0×, 1.2×, 1.5× and points between — applied to the model's own placement at execution time. The model keeps deciding where the structural stop belongs, which is what it is good at; the multiplier decides how much room you personally want to give it, which is a question about you rather than about the chart.

Two details matter for it to be honest rather than decorative. Position size adjusts automatically, exactly as it does throughout this article — choose 1.5× and the size drops to two-thirds so your risk per trade does not move. A multiplier that quietly increases your risk is not a stop setting, it is a leverage setting wearing a disguise. And the multiplier is per trader, not global, because Table 5 is unambiguous that our short books and our one good long book want opposite settings. Applying one number to everything is the mistake this study exists to document.

Our own defaults, for what they are worth: 1.2× on the short books, and 1.0× on NAS100 longs. That is the highlighted row of Table 6.

What this study cannot tell you

The limits are real and we would rather list them than have them found.

The sample is thin. 237 trades over 7.7 months, of which 108 are shorts. Confidence intervals on almost every figure here span zero. This is a real edge measured over a short window, not a law.

The peak is not a location. We have said it twice already and it bears a third mention: 1.2× is roughly three trades wide. Treat 1.1× to 1.5× as one region. Anything that only works at exactly 1.2× is noise wearing a decimal point.

The replay is mildly generous to wide stops. Prices are mid, with no spread or slippage, and a one-minute aggregate truncates the fastest wicks — so a few stop-outs that happened in reality will not appear in the simulation. That bias helps wide stops, which means the real case against them is slightly stronger than the tables show, not weaker.

The lineup was not constant. Instruments were added and retired across the window, so this pools several slightly different versions of the same product. August is partial, through the 21st, and carried known infrastructure problems on our side that we have not finished diagnosing.

What we will stand behind is narrower than the tables might suggest, and more useful. Over the seven months we have actually traded, widening the stop bought a substantially higher win rate and a substantially lower total, the effect reversed direction depending on which book it was applied to, and the best configuration we could find returned 84% more than what we ran on roughly half the drawdown. That is enough to justify shipping the dial. It is not enough to promise where you should set it — which is rather the point of making it a setting.

The specific danger of moving a stop

Of everything in this article, the finding most likely to be misused is the one that says a wider stop returned more. Read carelessly, Table 3 says "give trades more room and make more money." It says nothing of the kind, and the distinction matters enough to spell out.

Stop placement is very close to a zero-sum exchange. Every point of room you add on one side is paid for on the other. Widen the stop and you buy a higher hit rate — genuinely, reliably, every time — but you fund it by shrinking the position, which shrinks every win. Tighten it and you buy larger wins funded by more frequent losses. The trade is real in both directions and there is no setting that quietly gives you both. Table 2 is the whole argument in six rows: between 1.0× and 3.0× the win rate improved by a factor of 1.31 while the average win fell by a factor of 3.1, and the expectancy — the only number that actually pays — went down by two-thirds while every surface-level metric was improving.

The trap is that the losing configuration feels better while you are in it. More green days, fewer stop-outs, shorter losing streaks. A trader who widens their stop and watches their win rate climb from 60% to 79% has every psychological signal telling them they improved, and an account balance that disagrees. If you take one thing from this article, take that: a rising win rate is not evidence of anything on its own, and any change that improves it should be checked against expectancy before it is trusted.

Two further cautions specific to moving stops. First, a wider stop is not a safer stop — our long book's drawdown got worse before it got better, rising from 17.54R to 19.72R on the way out, because a stop that is wider but still gets hit simply costs more. Second, widening a stop cannot repair a book with no edge. Our long book lost money at every width from 1.0× to 2.5× and reached breakeven at 3.0× only by winning three trades in four for nothing. If the underlying direction call is wrong, stop placement redistributes the damage; it does not remove it.

Please read this part before acting on any of it

This article is published for educational purposes only. It describes what our own system did over one specific stretch of market history and argues about why. It is not financial advice, not a recommendation to widen or tighten your stops, and not a suggestion that any multiplier in these tables is right for your account. Nothing here accounts for your circumstances, your risk tolerance, or your obligations.

The dataset is young. Seven and a half months and 237 trades sounds substantial in a blog post and is thin as evidence — the short book is 108 trades. A single unusual quarter could move any figure here materially, and we cannot statistically rule out that some of what we measured is chance. Past performance, replayed or live, does not correlate reliably with future performance, and the smaller the sample, the weaker that already-weak link becomes.

The overfitting risk is acute in a study shaped like this one, and we are the ones most exposed to it. We tested twelve stop widths across several book combinations against a single history and are now reporting which performed best. That is precisely the procedure that produces findings which evaporate on live data. It is also why we keep saying the peak is a region and not a point: a configuration that only works at exactly 1.2× is not a discovery, it is noise that happened to land on a decimal. Treat the flat plateau as the finding and treat any precise optimum — ours included — as an artifact until several more months of trades say otherwise.

And keep watching the side you turned down. The right stop width is a property of a market regime, not a constant. Volatility expands and contracts; the intraday liquidity structure that currently makes our shorts travel and our longs grind will not hold forever. A width that is wrong today can be right in a year, and a book we have widened, tightened or paused may need the opposite treatment when conditions turn. That is the reasoning behind shipping this as an adjustable per-trader setting rather than baking a number into the model: settings can be revisited as the evidence changes, and we intend to revisit ours. Anyone applying this to their own trading should hold it the same way — as a live hypothesis under continuous review, not a conclusion.