The "91% win rate" strategy, tested honestly
Grade C — reproducible and thoroughly ablated, but a full-sample forward-return study, not pre-registered.
Status — FAILED · score v1
The claim
It's one of the most-shared setups on trading YouTube: a five-indicator "confluence" that supposedly wins 91% of the time. Stack five popular signals — price above the 21 and 34 EMAs, a TTM Squeeze firing, Stochastic Slow turning up, and OBV confirming — and when all five line up, buy long-dated call options (LEAPS). Five green lights, near-certain win. The pitch sells itself.
We tested it before spending a single day building an options backtest, using the cheapest possible question: in the underlying stock, does this signal predict anything at all?
The verdict, in one line
No. The win rate is about 65%, not 91% — and 65% is exactly what you'd get buying any S&P 500 stock at random in the same years. The signal's forward returns are statistically indistinguishable from noise, and slightly negative relative to just being long. There is no directional edge to leverage.
How we tested it
- Universe: S&P 500 (503 names), weekly bars, 2010–2026 — about 406,000 weekly observations.
- Method: mechanize the exact five-condition checklist, then measure the average forward return 4, 8, 12, and 26 weeks after each signal, against the unconditional baseline — the return of a randomly chosen stock on a random week. Not a trade simulation; a pure "does the signal predict direction?" test. If it can't beat owning the stock, the leveraged options version is strictly worse.
What we found
| Horizon | Signals | Signal return | Baseline return | Edge | Signal win % | Baseline win % | t-stat |
|---|---|---|---|---|---|---|---|
| 4 weeks | 2,688 | 1.18% | 1.38% | −0.20% | 58.0% | 58.2% | −1.40 |
| 12 weeks | 2,683 | 3.99% | 4.08% | −0.09% | 65.1% | 62.9% | −0.33 |
| 26 weeks | 2,642 | 8.21% | 9.00% | −0.79% | 68.3% | 67.3% | −2.07 |
There's the "91%," demystified. The signal does win ~65% of the time at 12 weeks — but that's because almost everything went up 2010–2026. Buying at random won 63% of the time. The five-light confluence adds nothing; if anything it slightly subtracts.
We then took the strategy apart, indicator by indicator. Every single condition, tested alone and in combinations, was flat or negative. The trend-following pieces (the EMAs and OBV) were consistently negative with enormous statistical significance — buying after a stock is already extended reliably underperformed buying at random. The one faintly positive cell (the squeeze alone, at 12 weeks, +0.18%) was economically trivial and reversed by 26 weeks. Nothing lit up.
Companion finding: the TTM Squeeze on its own
The squeeze is the glamorous ingredient, so we gave it a separate trial: do breakouts from a "volatility squeeze" outperform? On a hand-picked set of 31 well-known names, it looked great — +1.36R vs +0.78R. Then we ran it on a broad, honest universe of 74 names that included the losers and the delisted (a company yfinance can no longer even serve — survivorship made concrete). The edge vanished and inverted in every market regime (−0.12R overall). The whole "edge" was an artifact of only testing names we already knew had won. Same lesson, different indicator.
Why the claim looks true anyway
Three ordinary illusions manufacture a "91% win rate" with no edge behind it:
- 1.A bull market makes almost any long strategy look like a genius (baseline win rate ~63%).
- 2.Survivorship — backtesting on today's index means backtesting on the survivors.
- 3.No benchmark. "91% of my trades won" means nothing without "…versus 63% for buying at random."
How the score breaks down
| Dimension | Score | Why |
|---|---|---|
| Edge | 3 / 25 | Forward returns indistinguishable from — and slightly worse than — buying at random. The claimed 91% is a bull-market win rate, not an edge. |
| Robustness | 4 / 25 | Robustly dead: every indicator, alone or combined, was flat or significantly negative across horizons. |
| Practicality | 3 / 25 | Buying LEAPS calls on a zero-to-negative signal stacks premium, theta, and leverage on top of no edge — a structural loser even before costs. |
| Evidence | 13 / 25 | Reproducible, thoroughly ablated, honest benchmark. Capped as a full-sample forward-return study reacting to a claim, not a pre-registered test. |
| Total | 23 / 100 | No demonstrated edge — the famous number is real arithmetic on a meaningless base rate. |
Evidence Grade C: we're confident there's no directional edge here; the test was a clean, cheap falsification rather than a pre-registered head-to-head.
What it means for you
When a strategy advertises a win rate with no benchmark, that number is almost always the market's base rate wearing a costume. Ask one question of any "X% win rate" claim: versus what? Here, the honest answer is "versus 63% for buying blindfolded" — and the fancy version costs you options premium for the privilege. The cheap test saved us weeks of building an options backtest for a strategy that was never going to work.
References
- The cheap-test-before-you-build principle: kill expensive research paths with a simple forward-return study first.
- The four-dimension scoring rubric behind the number above: The Tapelab Score.
Scored under score v1, the fixed public rubric summarised on the research index. Report last updated 2026-07-03; published from the research repository's canonical write-up, unedited. Nothing here is investment advice.