The 97% win rate that loses money
Win rate is the most seductive number in trading, and the least meaningful. Here is the strongest win rate our system ever produced, and why we killed the strategy that produced it.
The strategy
Sell same-day-expiry (0-DTE) put options on index ETFs and mega-cap names. Collect the premium. Most days, the market doesn't fall far enough to matter, the puts expire worthless, and you keep the money.
We ran it in shadow mode for 28 months. The results looked like this:
- 244 trades
- 97.5% win rate (238 wins, 6 losses)
- +$1,412 gross (per-contract units)
If you saw those numbers in a screenshot, you'd subscribe. A 97.5% win rate feels like a machine that prints money.
Where the number comes from
The win rate is real. It's also mechanical. Selling out-of-the-money puts that expire in hours means you win every day the market doesn't drop sharply — which is most days. Any strategy in this family produces a win rate in the 90s by construction. The win rate tells you what the strategy is, not whether it has edge.
Look at the same 244 trades differently:
- Mean profit per trade: +$5.79
- Worst single loss: −$48.16 (one assignment day)
You collect pennies on most days and hand back a stack of them on the bad day. Whether that nets out to real edge depends entirely on whether the pennies are priced generously enough — which is a statistical question, not a vibes question.
The only question that matters
Our standard for every strategy is a permutation test: how often would random entries, on the same instruments over the same period, produce results this good? We shuffle historical returns 10,000 times and count.
For this strategy, the answer was 99.56% of the time (p = 0.9956). Random does this. The win rate that looks like skill is just the shape of the trade.
We went further, because the aggregate can hide a real edge in one corner. We permutation-tested each of the 11 tickers separately, with a multiple-comparisons correction (Benjamini–Hochberg, since testing 11 things at once gives chance 11 opportunities to fool you). Result: 0 of 11 tickers pass. The strongest individual result was QQQ at p = 0.098 — not significant, and not even in the live whitelist.
One more detail worth being honest about: early in this strategy's life, an internal note cited "AAPL PF 9.21" as evidence the per-ticker edge was real. When we re-ran the full cohort, AAPL's actual profit factor was 1.15. The exciting number came from a favorable slice. This is the exact failure mode we built the audit process to catch — including when we're the ones fooled.
The verdict
Killed, April 2026. The strategy stopped producing picks, and its full history — including the 97.5% win rate — stays in our record with this verdict attached.
What to take from it
When a service leads with win rate, ask three questions:
- What's the average win versus the average loss? A 97% win rate with +$6 wins and −$48 losses is a coin-toss economics problem, not a money machine.
- How often would random do this? If they can't answer, they haven't tested it.
- Where are the losers? If the record only shows expired-worthless puts, you're looking at marketing, not measurement.
We run this same gauntlet on every strategy of our own. Most don't survive it — 24 killed so far, all published. That's not a confession. It's the method.
We run this same gauntlet on every strategy we build — most ideas die in research, 24 full plays have been killed and published, 6 are live. How the gates work: methodology. What a play is: the basics. How to judge anyone's claims, including ours: the difference.
Research publication — not investment advice — disclaimer.