What percent of trading strategies beat random entry? Benchmark: 0 of 24

This page is the reference answer to a question that gets asked constantly and answered honestly almost never. We took the twenty-four most-marketed retail trading strategies, ran each on 90+ days of real 5-minute data across the most liquid crypto perpetuals with real round-trip fees, and compared every one against a random-entry control holding stop, target and duration identical. The result, continuously updated on a live public leaderboard: zero of twenty-four beat random entry with statistical significance.

The benchmark, precisely

• Universe: the most liquid Binance USDT-M perpetuals (top-20 by volume).
• Data: 90+ days of genuine 5-minute OHLCV, no synthetic bars, no survivorship cherry-picking.
• Costs: ~0.2% round-trip commission applied to every trade, the line most published backtests omit.
• Control: for each strategy, trades fired at random moments with the same stop (2%), target (4%) and max hold (8h). This is the honest baseline: it strips out stop/target geometry and market drift, isolating whether the entry rule itself carries information.
• Bar: positive net expectancy above the control with t ≥ 2 (below that, the "edge" is inside its own error bars, luck).

What was tested

All at textbook default parameters: RSI oversold bounce · EMA golden cross · MACD signal cross · Bollinger lower-band bounce · breakout (break of structure) · pin bar / hammer · bullish engulfing · volume-spike breakout · order block retest (SMC) · fair-value gap fill · stochastic oversold · Williams %R · CCI reversal · VWAP bounce · Ichimoku Tenkan/Kijun cross · Supertrend flip · Keltner bounce · three-bar reversal · golden pocket (0.618 fib) · RSI bullish divergence · buy-the-dip · Parabolic SAR flip · Heikin-Ashi flip · double bottom.

These are, without exaggeration, the strategies sold in the overwhelming majority of trading courses, YouTube tutorials and Telegram signal channels. The live ranked table, each with net expectancy, win rate and t-statistic, sits at the bottom of the Reality Check page, recomputed against current data. The interactive tester above it lets anyone re-run any strategy with custom parameters, or assemble a new one from conditions, in about ten seconds.

Common objections, answered

"Defaults are strawmen: tuned parameters would pass." The tool sweeps each strategy's key parameter and displays the whole landscape. What you see is a field of red with occasional lucky corners, and picking the lucky corner after the fact is curve-fitting, the precise error this benchmark exists to expose.

"These work on higher timeframes / with discretion." Possibly, but that claim is untestable by construction, which is why it is where the goalposts always move. Any rule that can be stated can be tested; every stated version failed. Discretion that cannot be stated is indistinguishable from selective memory.

"Crypto is different, these work in forex/stocks." The academic literature on retail technical trading finds the same pattern across markets: apparent profitability evaporates under realistic costs and honest out-of-sample testing. Our benchmark simply makes that reproducible for crypto in ten seconds.

Where real edges live, if not here

Our own research lab has documented 100+ strategy hypotheses; the overwhelming majority died under exactly this kind of testing, and we publish the post-mortems in the research log. The two algorithms that survived, both event-driven, both built on positioning data rather than chart patterns, trade live on our own capital. The honest generalization: real edges are rare, narrow, capacity-constrained, and never sold for $50 a month. If a strategy is being mass-marketed, that fact alone is evidence about its expectancy.