Home / Leaderboard
Leaderboard

The leaderboard that won't rank a fluke

Most trading leaderboards sort by win rate or return and stop there. That single choice inverts the table: with no minimum-sample rule, a setup that won its only three trades outranks one that has been right 55% of the time over four hundred. Below are the same 903 assets ranked both ways, side by side, with the trade count printed next to every row so you can see the naive method break.

903
assets ranked
30 / 10
minimum trades · overall / out-of-sample
10
setups that survive the bar
Ranked the usual way

Sorted by out-of-sample Sharpe, no sample rule

This is our real data, sorted the way the rest of the category sorts it. Thin trade counts are marked in red. The right-hand column is our own verdict on the very same row.

#AssetSetupOOS SharpeTradesWinOur verdict
1Sandisk (SNDK)Ehlers SuperSmoother · Daily4.441163.6%nothing beat B&H
2USX (USX)Fibonacci Pivots · Daily4.221225.0%nothing beat B&H
3Humanity (H)Detrended Price Osc. · Daily3.541163.6%nothing beat B&H
4Siren (SIREN)Volume Zone Oscillator · Daily3.231258.3%unproven
5Canton (CC)Connors RSI-2 · Daily3.141376.9%unproven
6Alphabet Inc. (Class A) (GOOGL)Morning Star · 4-Hour3.012166.7%nothing beat B&H
7Russell 2000 (IWM)Order-Flow Reversion · 1-Hour2.992982.8%nothing beat B&H
8Arm HoldingsEhlers SuperSmoother · Daily2.812544.0%unproven
9JPMorgan Chase (JPM)DeMarker · 1-Hour2.82580.0%nothing beat B&H
10United Stables (U)Fibonacci Pivots · Daily2.793426.5%unproven
What just happened

Every one of those ten failed validation — the thinnest has 11 out-of-sample trades, and the second row posts a Sharpe above 4 on a 25.0% win rate, meaning one or two outsized winners carrying a mostly-losing record. A Sharpe above 4 on a dozen trades is not an edge; it is the best of hundreds of thousands of combinations we tried, which is exactly what you would expect to surface by chance in a search that size. Nothing here is fabricated — those are genuine backtest results, and that is the point. Good numbers are easy. Good numbers that survive a sample rule are not.

Ranked honestly

The same data, sample-gated and multiple-testing corrected

Minimum 30 trades and 10 out-of-sample trades to be ranked at all, must beat buy-and-hold on both the full sample and the held-out sample, then clear a Sharpe hurdle scaled to the size of the search. Note the Sharpe numbers are lower than the table above — and the median row here has 79 out-of-sample trades rather than 11. That trade is the entire product.

#AssetSetupOOS SharpeTradesWinOur verdict
1Rain (RAIN)MA Envelope · Daily2.764151.2%validated
2GBP/JPYFisher Transform · Daily2.517535.4%validated
3NZD/JPYFisher Transform · Daily2.2717534.3%validated
4USD/CNYDMI Direction · Daily1.474231.0%validated
5Nasdaq Futures (NQ)Net Volume · Daily1.457955.7%validated
6Volatility Index (VIX)MA Envelope · Daily1.4518966.1%validated
7Texas Pacific Land Corporation (TPL)KAMA 10/30 Cross · Daily1.414663.0%validated
8Ameren (AEE)Detrended Price Osc. · Weekly1.285574.5%validated
9Micron Technology (MU)SMC: Order Block · Weekly1.211154.5%validated
10Eversource Energy (ES)Detrended Price Osc. · Weekly1.09463.8%validated

Every row links to its full report with the loss column, the drawdown, and public forward tracking from the day it was published. Educational information only — not investment advice. Hypothetical backtested results; past performance does not guarantee future results. Trading involves risk of loss.

The number under all of this

Across 660,005 out-of-sample backtests, a technical signal beat buy-and-hold 20.1% of the time. When roughly four in five attempts fail, a table of the top ten attempts is guaranteed to look spectacular whether or not anything real is happening. The sample rule is what separates the two, and it is why 703 of these 903 assets carry a verdict of "nothing beat buy-and-hold".

Questions people ask

Why isn't this leaderboard sorted by return?

Because sorting by return or win rate with no minimum-sample rule puts the least reliable rows at the top. On our own data, ranking purely by out-of-sample Sharpe fills the top ten with setups of 11–15 trades — and 10 of those ten failed our validation. The naive table on this page shows exactly that.

What counts as enough evidence here?

A setup needs at least 30 trades overall and 10 out-of-sample trades before it can be ranked at all, then it has to beat buy-and-hold both full-sample and out-of-sample, then clear a Sharpe hurdle raised in proportion to how many combinations were searched for that asset. Ten setups out of 903 assets survive.

Can a high win rate be misleading?

Very. One row in the naive top ten posts a strong Sharpe on a 25.0% win rate — a couple of outsized winners carrying a mostly-losing record. Win rate on its own tells you almost nothing without the trade count, the payoff ratio and the drawdown beside it, which is why all four are printed here.

Go deeper

Free · no spam

Get the weekly edge report

The best-performing indicator per asset, what changed this week, and the honest caveats — straight to your inbox.