Does a tighter stop-loss always lower your real drawdown?

Not always. In my MEGAUSDT range_bottom_reclaim backtest, I ran 30 configs with the invalidation level moving across 13 distinct values between -3.96% and -4.0%, and every single one bottomed out at the exact same -14.237% max adverse excursion. The stop parameter had zero effect on the real worst-case pain because that -14.237% low happened before the stop's logic was even the deciding factor — DCA rescue legs and entry timing produced it, not the exit threshold.

Why did the same parameter behave differently on a second coin?

On BILLUSDT, the same family of configs showed real max adverse excursion moving from -5.1% to -6.6% as the DCA-rescue depth parameter (dca1_pct) got deeper, a roughly 30% swing tied directly to that knob. MEGAUSDT's worst drawdown was fixed regardless of config; BILLUSDT's was configurable. Whether a risk parameter actually controls your risk depends on the coin's own price structure, not just the parameter's definition.

How do I know if my stop-loss setting is actually configurable risk or a fixed floor?

Run several config variants on the same coin and compare logged max adverse excursion, not the stop parameter itself. If MAE stays flat across variants like it did on MEGAUSDT, you have a fixed risk floor no matter what you type into the stop field, and you should size your position off that observed floor, not off the number you configured. If MAE moves with the parameter like it did on BILLUSDT, you have real, adjustable control.

I Tested 30 Stop Levels on One Coin. The Worst Drawdown Never Moved

Published 10 September 2026

I was sweeping my MEGAUSDT shortlist again — 30 configs from the same GPU batch, same entry family, final_invalidation_pct (the hard stop) spread across 13 distinct values from -3.96% to -4.0%. I sorted by max adverse excursion expecting some spread, since a wider stop should let a trade travel further underwater before it's cut. Every single row read -14.237%. Not close. The same number to the third decimal, on all 30.

My first thought was a bug in the logging. It wasn't. I checked dca1_pct, take-profit, time-stop minutes — all of them varied across the 30 rows the way they're supposed to for a real parameter sweep. Only the worst-case drawdown refused to budge.

The stop wasn't deciding the worst case

Here's what that means mechanically: whatever candle sequence produced that -14.237% low happened before the invalidation level ever got a vote. The DCA-rescue leg fires, the position's effective entry moves, and the trade rides through a pullback that bottoms at the same depth no matter whether the final stop sits at -3.96% or -4.0% — because the stop was never close to the price action that mattered. A 0.04 percentage-point range in the configured stop is a rounding difference. It bought zero protection.

So I went and checked a second coin from the same shortlist, BILLUSDT, same entry family, same style of sweep. Different story entirely. As dca1_pct (how deep the rescue leg buys) moved from -0.48% to -1.46% across nine configs, real max adverse excursion moved with it — from -5.14% at the shallow end to -6.64% at the deep end. Almost a 30% swing, tracking the parameter directly. On BILLUSDT, tuning the knob actually changed the outcome. On MEGAUSDT, it didn't.

Two coins, two different answers to the same question

Same strategy family, same kind of parameter sweep, and I got a fixed floor on one coin and a configurable dial on the other. There's nothing in the config sheet that tells you in advance which one you're looking at — range_bottom_reclaim is range_bottom_reclaim whether it's sitting on MEGAUSDT or BILLUSDT. The only way I found out was by running the sweep and comparing logged MAE across variants instead of trusting that the parameter meant the same thing on both.

I've written before about how a stop-loss level and real drawdown can diverge on a single coin, config to config, and about how win rate hides what a strategy actually did. This is the same family of problem from one step further back: before you even ask whether your stop is protecting you, you have to check whether the stop is a lever at all on this specific coin, or just a label on a floor the market already set.

What I do with this now

I stopped assuming a parameter sweep tells the same story on every coin. Now, before I trust a stop-loss or DCA-depth setting as "risk control," I run at least a handful of variants on that exact coin and check whether logged MAE actually moves. If it's flat like MEGAUSDT's -14.237%, that's my real floor regardless of what I type into the field, and I size the position off that number in the position size calculator — then check what a -14% hit does to my buffer in the liquidation calculator before opening anything. If MAE moves with the parameter like BILLUSDT's did, only then do I trust that dialing the knob down is actually buying me less pain.

Trade where the calculators point

Trade on Bybit — lower fees, free to join → Binance → OKX → KuCoin → 📈 TradingView 🔒 NordVPN