Key numbers
| Measure | Value |
|---|---|
| Backtests | 38,761 |
| Beat cash after costs | 9.3% |
| Median Sharpe | −0.84 |
Why this is an upper bound
These backtests ran over the full history, including the period the rules were tuned on. That is the most favourable view a rule can get. If it loses here, it will not do better on new data.
All rule families
| Rule family | Backtests | Beat cash | Median Sharpe |
|---|---|---|---|
| RSI zone rules | 84,481 | 3.8% | −1.09 |
| MACD crossover rules | 81,574 | 3.8% | −1.12 |
| Donchian breakout rules | 75,865 | 4.6% | −0.90 |
| Bollinger squeeze rules | 46,928 | 0.7% | −1.61 |
| Moving-average crossover rules | 38,761 | 9.3% | −0.84 |
| Turtle soup (false breakout) rules | 8,836 | 9.8% | −0.92 |
Caveats
Sharpe here subtracts a 4% cash rate. Rules were combined with macro filters; costs follow a retail broker model. Some symbols had data errors that we did not correct for this layer.
Let your AI check how ideas like yours did before it runs a single test. In our pilot, agents that checked first spent 1% of their backtests on known dead ends instead of 15%, and every backtest not wasted is time and tokens you keep. Cite this entry as ER-IS-SMA_, as of 2026-11-01.
Related evidence
Questions
Do moving-average crossover rules work?
Moving-average crossover rules were backtested 38,761 times on up to 557 symbols over their full history after retail costs. Only 9.3% beat cash (Sharpe above zero); the median Sharpe was −0.84. Even looking back, this is a dead end.