Key numbers
| Measure | Value |
|---|---|
| Backtests | 13,297 |
| Distinct strategies | 1,581 |
| Symbols | 401 |
| Median Sharpe | −0.27 |
| 10th to 90th percentile | −0.64 to +0.03 |
| Beat cash (Sharpe above 0) | 12.0% |
| Sharpe above 1 | 0.05% |
| Beat buy-and-hold | 7.1% |
| Median trades | 86 |
| Of those with positive returns: still positive after 2023 | 54% |
What it means
These are full-history results, the most favourable view a rule can get. If MACD crossover rules lose here, they will not do better on new data. The best single variant looks much better than the median simply because it is the best of 1,581; compare any backtest you have with the median (−0.27) and the 90th percentile (+0.03), not with the top.
With and without filters
Each variant combines the rule with a macro or market filter (or none). By filter family:
| Filter family | Backtests | Median Sharpe | Beat cash | Beat buy-and-hold |
|---|---|---|---|---|
| liquidity (money supply, balance sheets) | 1,820 | −0.24 | 13.6% | 8.8% |
| macro releases (CPI, PMI, payrolls ...) | 2,497 | −0.25 | 14.1% | 7.0% |
| market data (VIX, indices, commodities) | 3,924 | −0.25 | 12.5% | 7.8% |
| interest rates and yield curve | 2,126 | −0.27 | 11.2% | 6.3% |
| sentiment and surveys | 2,188 | −0.29 | 10.1% | 6.0% |
| crypto market data | 742 | −0.41 | 5.8% | 4.6% |
See the study Do macro filters improve technical trading rules?
By symbol
Symbols with at least 5 variants, best and worst by median Sharpe:
| Best symbols | Backtests | Median Sharpe | Beat cash |
|---|---|---|---|
| HG | 21 | +0.78 | 100% |
| COR | 19 | +0.40 | 100% |
| CEG | 26 | +0.36 | 100% |
| AXON | 24 | +0.31 | 83% |
| HUBB | 20 | +0.17 | 75% |
| Worst symbols | Backtests | Median Sharpe | Beat cash |
|---|---|---|---|
| COP | 72 | −4.80 | 0% |
| CLX | 26 | −4.64 | 0% |
| ECL | 45 | −4.58 | 0% |
| CL | 55 | −4.54 | 0% |
| ORCL | 31 | −4.37 | 0% |
Other rules on US stocks, daily bars
| Rule family | Backtests | Median Sharpe | Beat cash |
|---|---|---|---|
| RSI zone rules | 14,056 | −0.22 | 14.5% |
| Moving-average crossover rules | 9,339 | −0.24 | 16.4% |
| Donchian breakout rules | 13,377 | −0.25 | 14.1% |
| MACD crossover rules | 13,297 | −0.27 | 12.0% |
| Bollinger squeeze rules | 10,247 | −0.73 | 0.9% |
Same rule, other timeframes
| Timeframe | Backtests | Median Sharpe | Beat cash |
|---|---|---|---|
| 15-minute | 10,965 | −2.24 | 0.0% |
| daily | 13,297 | −0.27 | 12.0% |
| 1-hour | 11,723 | −1.04 | 1.4% |
| 4-hour | 11,435 | −0.67 | 2.8% |
Caveats
In-sample, not validated on new data: full-history backtests after retail costs, Sharpe minus a 4% cash rate. "Still positive after 2023" only counts variants with positive returns over the full history, which includes 2023–2026, and 2023–2026 were rising markets, so read it as an upper bound. Monthly macro filters were applied 1–3 weeks before publication in these backtests, which flatters them. Some price files had data errors.
Terms used: in-sample upper bound, trial count, dead end.
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-LIB-MACD-US-1D, as of 2026-11-01.
Related evidence
Questions
Do MACD crossover rules work on US stocks with daily bars?
MACD crossover rules on US stocks with daily bars were backtested 13,297 times (1,581 variants, 401 symbols) over the full history after retail costs. 12.0% beat cash and 7.1% beat buy-and-hold; the median Sharpe was −0.27. That makes them mostly negative even in hindsight.
How should I read my own backtest against this?
Compare it with the median (−0.27) and the 90th percentile (+0.03) of 13,297 comparable backtests, and count how many variants you tried. A top result among many tries is expected by luck.
Is this investment advice?
No. These are research statistics about a class of rules, not a recommendation.