Published
When my big backtest finished, only four coins survived to the "deep test" pile. Two of their win rates sat at the far ends of the spectrum, and the contrast is the cleanest lesson I can give a new trader.
One setup on LIGHT won just 33% of its trades. One in three. By every Twitter standard that's a losing strategy. Its profit factor was 1.87 — meaning for every dollar it lost, it made $1.87 back. On paper, it made money.
Another setup, on BILL, won 96.7% of its trades — almost flawless. Its profit factor was 2.12. Higher, but not the night-and-day gap the win rates would suggest. And the 96% came with a catch the 33% didn't have.
Why a 33% win rate can win
It's just arithmetic. If your winners are bigger than your losers, you can be wrong most of the time and still come out ahead. The LIGHT setup lost small, often, and occasionally caught a move that paid for all of it. That's the entire business model of trend-following: a low hit rate, tiny controlled losses, and a few fat winners. Painful to sit through, but durable.
Why a 96% win rate can be a trap
The BILL setup won almost every trade — but 72% of those wins needed an averaging-down rescue first, and the LIGHT setup's deepest drawdown was a manageable −5.3% while the high-win-rate coins regularly sat through −10% to −14%. A 96% win rate built on "add more every time it goes against me, then exit on the bounce" is a margin call with good PR. It looks perfect right up until the one trade that doesn't bounce.
So the same search handed me a 33% strategy I'd respect and a 96% strategy I'd avoid. If I'd sorted my results by win rate — the way most people do — I'd have thrown away the better one and traded the dangerous one.
What to read instead
Three numbers beat win rate every time:
Profit factor — gross wins ÷ gross losses. Above 1.5 after fees is worth a look; below 1.2 is usually noise.
Max drawdown / adverse excursion — how deep underwater you go to earn the result. −5% is survivable; −14% on leverage is a funeral.
DCA dependence — what share of "wins" only closed green because you added to a loser. High dependence means the equity curve is borrowing from a future blow-up.
None of these four backtests is something I'd put real money on tomorrow — they're still backtests, and backtests flatter. But the win-rate spread alone, 33% to 97% in the same run, should permanently cure you of trusting that one number. Model your own trades with the PnL calculator and size them with the position size calculator — a strategy that wins 40% of the time with disciplined sizing will bury one that wins 95% and martingales into the floor.