Where should your stop-loss go? We tested nine answers

We ran one strategy over six years of Bitcoin and changed nothing between runs except the stop-loss. Same entries, same exits on the opposite signal, same costs, same 2% risk per trade. Nine different answers to "where should the stop go."

The best answer returned +278.97%. The worst returned -0.42%. Between those two runs, every single entry signal was identical.

Most traders treat stop placement as an afterthought, a number you pick once and stop thinking about. The spread above says it deserves roughly as much attention as the entry.

The protocol

• Strategy: EMA 20/50 crossover on BTCUSDT, 4-hour candles. Long on the cross up, short on the cross down, and an opposite cross always closes the trade if the stop hasn't already.
• Window: April 2020 to August 2026, which contains two bull runs, the 2022 bear, and a lot of chop.
• Sizing: 2% of equity risked per trade, sized off the stop distance. This detail turns out to run the whole experiment, more on it below.
• Costs: the full model. Intra-bar fills on 1-minute data, order-book fills where we have book coverage, dynamic slippage, funding.
• The nine stops: fixed at 1%, 2.5%, 5%, 10%, and 20% from entry; at 1.5× and 3× ATR; a 2.5% stop that trails; and a 2.5% stop that jumps to break-even once the trade is +5%.

Every run produced 258 to 260 round trips from the same signal stream (the tiny differences come from what was still open at the window's edge).

The results

Three of the most common beliefs about stops die somewhere in that table.

Tight isn't safe and wide isn't rich

The intuition says a tight stop protects you and a wide stop gives the trade room. The table says the relationship isn't even monotone.

The 1% stop got stopped out on 85% of its trades and won only 13% of them, yet still made +91%, while suffering a 70.6% drawdown that almost nobody would sit through. The 10% stop, the one "giving trades room," made approximately nothing over six years. And the 20% stop, wide enough that it never fired once, beat the 10% stop while drawing down only 12.4%.

The resolution to this weirdness is position sizing. With risk-based sizing, the stop distance is the position size: risking 2% of equity across a 1% stop means a position 10 times larger than the same 2% across a 10% stop. Tighten the stop and you get slapped out constantly, but in size; widen it and each loss still costs the same 2%, while the position becomes too small for the winners to compound. That's why 10% is the graveyard: wide enough to shrink the size, not wide enough to dodge the 2022-style swings.

So the real question was never "how much room do I give the trade." It's "what's the best exchange rate between being wrong often and being wrong big," and for this trend system the answer sat near 2.5%, not at either extreme.

The win rate will lie to you

Look at the trailing stop. It has the best win rate in the table, 37.3%, and one of the worst returns, +7.96%.

A trailing stop ratchets toward price as the trade moves your way, which sounds like pure upside. On a trend-following system it's quietly fatal: trends breathe, and a 2.5% trail gets hit by the first healthy pullback. 94% of all exits became stop-outs. The system banked hundreds of small scratches and never once held a position through the moves that pay for everything else. It won often and earned nothing.

Meanwhile the 1% stop won 13% of the time and made eleven times as much. Win rate measures how often you're right, not how much being right is worth, and an exit rule can buy a flattering win rate with money.

What the ATR stops add

A fixed percent ignores that Bitcoin's 4-hour volatility in 2020 and in 2026 are different animals. An ATR-based stop scales with current conditions: 1.5× ATR lands near 2.5% in a normal regime, tighter in quiet markets, wider in violent ones.

It shows. 1.5× ATR returned +218.81%, the best of any plain stop, with the same Sharpe as fixed 2.5% (0.51) and a deeper 52.1% drawdown, the price of running tighter, larger positions through 2021. If you can't stomach that, 3× ATR gave a smoother +95.92% at 37% drawdown. Neither is magic; both are the fixed-percent logic with the distance denominated in the market's units instead of yours.

The one modification that beat everything

The winner wasn't a placement at all but a rule: start with the plain 2.5% stop, and once the trade is up 5%, move the stop to the entry price. Nothing else changes. +278.97%, drawdown 37.9%, and the best Sharpe and profit factor in the table.

The asymmetry is the whole trick. Unlike the trailing stop, the break-even ratchet fires once, only after the trade has already proven itself, and then leaves the position alone. Winners keep their full room to run; what gets removed is only the specific disaster of a +5% trade returning all the way through entry into a full 2% loss. A quarter of its stop-outs, 42 of 172, were these scratches at the entry price: exits that cost fees and slippage instead of the full risk budget.

It's not a free lunch in general, and we only tested one trigger level here. But it's the same mechanic our prop-firm strategy uses, found independently by a different search, and it survived both experiments with the best risk-adjusted numbers on the board. That's a pattern worth noticing.

What to do with this

Treat the stop as a sizing decision first. Under risk-based sizing, stop distance sets position size. Before asking what the stop protects, ask what it makes your position size, because that number drives the equity curve more than the protection does.

Denominate the stop in volatility if you can. ATR-based distances kept pace with regime changes that fixed percentages can't see coming.

Never evaluate an exit rule by win rate. Demand the full picture: return, drawdown, and what fraction of exits the stop is taking. A stop that fires on 94% of trades isn't a safety net, it's the strategy.

Test the break-even ratchet before fancier ideas. One line of config, and in both of our studies it was the single highest-value change to an exit.

Honest caveats

One strategy family, one asset, one window, one sizing rule. These are the exit dynamics of a trend follower, and they partially invert for mean-reversion systems, where tight stops cut off the snap-back you're trading for. The break-even trigger of +5% is a single tested value, not an optimized one, and its edge over the plain 2.5% stop, while large here, deserves the same overfitting suspicion as any number this good.

And the caveat over all of it: no stop placement made a bad entry good. The same crossover signals under every rule above never turned into a strategy we'd call finished. The stop decided how much of the trend's edge survived contact with reality, between roughly nothing and roughly all of it, which is exactly why it's worth testing instead of defaulting.