cd /news/machine-learning/we-graded-13-famous-trading-strategi… · home topics machine-learning article
[ARTICLE · art-57878] src=dev.to ↗ pub= topic=machine-learning verified=true sentiment=↓ negative

We graded 13 famous trading strategies on 6 years of data. All 13 failed.

Tessen.ai graded 13 famous trading strategies on six years of crypto data and found that all 13 failed to pass all five pass/fail gates. The best performer, the ADX trend rider, passed only four of five gates and netted just 10.7 basis points per trade after costs. The results highlight that win rate is a trap, trend-following outperformed mean-reversion, and trading costs turn marginal edges into losses.

read3 min views1 publishedJul 13, 2026

(https://tessen.ai/strategies) Every trading book, YouTube channel, and Discord server teaches the same classics: RSI mean-reversion, the golden cross, Bollinger bands, MACD momentum, Turtle trading. They're taught because they're intuitive and they backtest beautifully — if you let yourself be a little dishonest about how you backtest.

We run a strategy-grading engine, so we did the obvious thing: we took 13 of the most widely taught strategies and graded them all the same way we grade everything —

6 years of crypto data across 10+ major pairs

A strict out-of-sample split: parameters are chosen on the training window only; the grade comes exclusively from data the strategy never saw

Real costs: taker fees and slippage modeled on every fill

Five pass/fail gates: positive out-of-sample expectancy, clears the cost hurdle, robust across assets, survivable drawdown, and not overfit

Here's the honest scoreboard:

Strategy Gates passed Net/trade (after costs) OOS trades Win rate

ADX trend rider 4/5 +10.7 bp 4,235 37.8%

Bollinger breakout 4/5 +3.8 bp 6,382 36.8%

MACD momentum 3/5 +3.5 bp 8,345 37.0%

Momentum (rate of change) 2/5 +1.6 bp 5,944 36.6%

Golden cross 2/5 +0.7 bp 8,126 37.6%

Ichimoku cloud 1/5 +1.4 bp 7,134 36.8%

Connors RSI-2 1/5 −3.7 bp 12,141 47.8%

Turtle trading 1/5 −8.0 bp 7,054 35.9%

Bollinger reversion 1/5 −9.7 bp 6,946 49.0%

Stochastic reversion 1/5 −13.5 bp 9,313 44.3%

RSI mean-reversion 1/5 −19.8 bp 4,629 37.8%

Quiet dip buyer 1/5 −28.3 bp 1,439 36.3%

CMF money flow 0/5 −5.4 bp 3,615 33.3%

Zero out of thirteen passed all five gates.

Three things worth actually learning from this

Win rate is a trap. The two highest win rates on the board — Bollinger reversion at 49% and Connors RSI-2 at 47.8% — both lose money after costs. Meanwhile the closest thing to a passing strategy, the ADX trend rider, wins only 37.8% of its trades. A high win rate with small wins and large losses is how a strategy feels good while bleeding out. (This is personal: the founder ran a live bot with a 78% win rate that turned out to be flat over 6 years. The win rate was hiding a 1:0.34 reward-to-risk.)

Trend-following beat mean-reversion — everywhere. Look at the top of the table: ADX trend rider, Bollinger breakout, MACD momentum. Now look at the bottom: RSI reversion, stochastic reversion, dip buying. On this data, every strategy built on "it went down, so buy the bounce" lost money after fees, and every strategy that got close to passing was riding moves, not fading them.

"Close to passing" is still failing. Two strategies passed 4 of 5 gates. That sounds encouraging until you remember what the fifth gate was protecting you from. A strategy that isn't robust across assets, or that only worked because its parameters were tuned to the test period, will do to your real money exactly what it couldn't be caught doing in a sloppy backtest.

Why fees change everything

Most published backtests of these classics quietly assume zero or near-zero trading costs. At thousands of trades over six years, even a few basis points per trade compounds into the difference between "works" and "wipes out." Connors RSI-2 is the cleanest example: with 12,141 out-of-sample trades, it's gross-positive and net-negative. The strategy isn't wrong about the market — it's just too small an edge to pay its own way.

Check the work

Every grade above has a public verify page with the full per-asset breakdown — the methodology is the same for all 13, and the inputs can't be edited after the fact: tessen.ai/strategies

And if you have a strategy of your own — from a book, a video, or your own head — you can run it through the same five gates free, no signup: tessen.ai/grade. Fair warning, most ideas fail. Knowing that before you fund an account is the entire point.

Nothing here is financial advice. A failed grade on historical data doesn't prove a strategy can never work; a passed one wouldn't guarantee it keeps working. These are measurements, not predictions.

── more in #machine-learning 4 stories · sorted by recency
── more on @tessen.ai 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/we-graded-13-famous-…] indexed:0 read:3min 2026-07-13 ·