MACD tutorials are a genre. They all show the same thing: an indicator pane, three arrows on a chart that already went up, and the word “momentum” used eleven times. What almost none of them show is what happens when you take the crossover rule literally and make it trade every signal for years, through the boring parts, with no curation.
That’s what this backtest is.
Which MACD rule, exactly
“MACD crossover” hides two different strategies, and writeups constantly blur them:
- The zero-line cross: the MACD line (the 12-day EMA minus the 26-day EMA) crosses above zero — which is identical to the 12-day EMA crossing above the 26-day. Slower, fewer signals.
- The signal-line cross: the MACD line crosses its own 9-day EMA. Faster, far more signals, far more noise.
We test the zero-line variant: buy the day the 12-day EMA crosses above the 26-day, exit when it closes back below, on daily bars of SPY, QQQ, AAPL, MSFT and NVDA — 20% of equity per position, up to five at once, no stop, no target. The pure rule, unedited. (If you want the signal-line version, it’s a one-sentence chat edit in the lab — “use the MACD signal-line cross instead” — and comparing the two is genuinely instructive.)
Reading the record honestly
A 12/26 EMA cross is a much faster trigger than the golden cross’s 50/200, and it shows in the trade count: you get many times more signals, which makes the statistics more meaningful — and the experience more annoying. Faster crosses mean earlier entries into real trends and far more false starts. Expect the win rate to look unimpressive; trend-following systems routinely lose most of their trades and make it back on a few large winners. The number that decides whether you could actually live with this strategy is the max drawdown, not the win rate.
Compare it mentally with the golden cross on the same universe: same idea (own the uptrend), different speed setting. Faster isn’t better or worse — it’s a different bill. You pay either in lag (slow crosses give back months of gains at turns) or in whipsaw (fast crosses bleed small losses in chop). There is no setting that avoids both; anyone selling you one is selling you the arrows, not the record.
Edits worth trying
- “Only take crosses when price is above the 200-day SMA” — the classic regime filter; usually cuts the chop dramatically.
- “Add a 10% stop” — bounds the tail on individual names.
- “Switch to the signal-line cross” — see the noise trade-off yourself instead of taking anyone’s word.
- “Test it on ETFs only” — single-stock gaps are where crossover systems get hurt worst.
Then, if an edit seems to hold up: deploy it to the forward ledger and let it earn a record on data that didn’t exist when you designed it. Backtests are the audition; the ledger is the job.