The strategy that wins most often
Buy weakness, sell strength, collect the snap back. Mean reversion is intuitive, it fits how crypto feels on a quiet week, and it produces the kind of equity curve that looks like a staircase — many small wins in a row.
In our grid it did exactly that. It won 61% of its trades — one of the higher win rates across the 19 strategy grids we publish, though three sit above it (hv-squeeze 62.3%, rsi-divergence and volume-profile 61.5%). It also lost money in every single configuration.
What we tested
Our full backtest window on the top 50 coins by market cap, with fees and slippage applied to every trade (the exact window is shown on the settings page). Then the exit grid: 5 stop-loss levels x 6 take-profit levels = 30 full backtests. The live grid refreshes with our data pipeline, so the exact cells move; the shape does not.
Live grid: Mean Reversion — stop-loss & take-profit settings.
Result: 0 of 30 combinations profitable
As of 2026-07-26, the best profit factor in the surface is 0.83 — the highest of the four strategies we published verdicts for this week, and still under 1.00. A profit factor under 1.00 means the losing trades took more than the winning trades brought in. It is the number to look at first, because unlike a return figure it does not depend on how you sized positions.
The best win rate was 61.0%. Six trades in ten came back. The strategy still bled, because the four that did not come back gave up more than the six brought in.
The failure mode is inverted
Here is the part worth understanding, and it is the reason this grid is more informative than the breakout ones.
The Donchian and Heikin-Ashi surfaces are flat: averaging profit factor by stop level shows no ordering at all — no stop setting sorts either surface, unlike the mean-reversion gradient below. No stop setting rescues them, which means there is nothing to tune — the premise is what fails.
Mean reversion is different. Its grid has a real gradient — monotonic as the stop widens, and broadly degrading as the target widens (with one reversal at 8%):
- widen the stop and it improves every step: 5% -> 0.775, 7% -> 0.785, 8% -> 0.790, 10% -> 0.798, 12% -> 0.813
- widen the target and it degrades overall: 4% -> 0.804 down to 12% -> 0.760, though 8% sits slightly above 6%
That is a mechanism, not noise. When you buy weakness, price frequently keeps being weak for a while: a tight stop puts you out before the reversion you were paid to wait for, and a wide target keeps you in long enough for the bounce to turn back over. The worst single cell in the table is exactly the combination of both mistakes — a 5% stop with a 12% target.
The gradient points the right way and still never crosses 1.00. Extrapolating it is tempting and wrong — and note that the last step is the largest of the four, which is exactly when “one more step in the same direction” feels most justified and is least safe. That is how overfitting starts.
The catching-a-knife problem
Mean reversion assumes there is a mean. In a trending market — up or down — the level you are reverting to keeps moving away from you. Crypto spends a lot of time trending hard, and those are precisely the periods that produce the losses big enough to swamp a 61% win rate.
The honest takeaway
If you only look at win rate, mean reversion is among the higher of the 19 grids we publish. If you look at whether it made money, it is in the same place as the other three we graded this week: no profitable configuration in this window.
This is the clearest example we have of why we publish grids instead of headline numbers. Check it yourself: mean reversion settings or the free simulator. Our full research verdicts, failures included, are in the research log.