Lab story: the week fees dismantled our strategy

The week the win rate lied

One of our systems traded at a 78% win rate — and still lost money. No bug, no bad luck: geometry.

The average take-profit gain (after fees) was +0.68, the average stop-loss −2.98. Plug that into the break-even formula and this system needs an 81.4% win rate just to reach zero. Reality delivered 77.8%. The rest is arithmetic.

Where fees come in

On the most expensive venue in our setup, buying plus selling costs about 0.4% round trip. The tighter the profit target, the larger the share the fee alone devours from the gross win — on our second system, the average winning trade netted essentially zero after costs.

The uncomfortable truth: the venue was not the problem — our geometry was. On daily timeframes (larger moves per trade), the same fee rate was nearly irrelevant in our tests.

Three rules we distilled

1. Compute the break-even rate before anything goes live — and require a safety margin of at least five percentage points across at least 100 out-of-sample trades.
2. Profit targets must amortize the fees. Tight targets + high fees = a structural loss machine, however good the signal.
3. In-sample winners are not winners. Every strategy that shines in the rear-view mirror must prove itself forward — our prettiest backtest curves died in droves in the holdout.

Why this is published

This magazine documents building our trading OS in the workshop, not the showroom. Nothing is sold before a public track record exists. Until then: the numbers, even when they sting.