People keep sending me MACD strategies that lose money in the same way every time. Take every 12-26-9 signal-line crossover, long the bullish ones, short the bearish ones, done. The equity curve grinds sideways for a while and then bleeds. The usual conclusion is that MACD is broken, or that indicators in general are astrology for men. I think the actual problem is more boring. Most people using MACD cannot say what its three components measure, so they treat every wiggle as a signal, and most wiggles on this particular indicator are noise by construction. So before settings and strategies, it is worth being precise about what the thing is.
What the three parts actually measure
The MACD line is the gap between two exponential moving averages, the 12-period minus the 26-period by default. Nothing more exotic than that. When the line rises, the fast average is pulling away from the slow one to the upside, which is a workable definition of momentum. When it falls, momentum is fading or rolling over. Because both inputs are moving averages, everything MACD tells you is smoothed and late. That lateness is a feature in trends and a tax everywhere else.
The zero line is where the two averages are equal. MACD above zero means the 12 EMA sits above the 26 EMA, which is a slow but honest way of saying the market has been going up. Crossing zero is exactly the same event as those two moving averages crossing on your price chart, so people who treat the zero cross as a separate confirmation are counting one piece of information twice.
The signal line is a 9-period EMA of the MACD line itself, a smoothed copy of an already smoothed thing. The histogram is the distance between the MACD line and its signal line, which makes it roughly a second derivative of price. It tells you whether momentum is accelerating or decelerating, and it turns before the MACD line does, which makes it both the most useful and the most dangerous part of the display, because early and noisy travel together in indicators.
Why crossovers alone bleed in ranges
Here is the mechanical problem with trading every signal-line cross. When a market ranges, the 12 and 26 EMAs braid around each other, the MACD line hugs zero, and the signal line sits right on top of it. In that state a small wobble in price flips the crossover. You get a long signal, then a short signal two candles later, and every flip costs a spread, a fee, and usually some slippage. On crypto perps with taker fees the bleed compounds quickly.
Crossovers are a trend-following signal. In a real trend they fire rarely and each one is worth something. The catch is that markets typically spend far more time chopping than trending, so a system that takes every cross spends most of its life paying for signals generated by noise, and the occasional trend winner has to cover the whole tab. Sometimes it does. Usually it does not, and you can verify that on almost any pair and timeframe in an afternoon of backtesting.
The three signals worth acting on, and the two that are noise
After years of watching people use this thing, I would keep exactly three signals.
- A signal-line cross on the same side as the zero line, well away from zero. MACD is comfortably above zero, dips toward the signal line, then crosses back up. That is a pullback ending inside an established uptrend, which is the highest-quality thing MACD knows how to say. Mirror it below zero for downtrends.
- A zero-line cross after a long stretch on the other side. Too late for a precise entry, but useful as a regime flag. I treat it as permission rather than a trigger, long ideas only while MACD holds above zero, short ideas only while it holds below.
- Histogram divergence into a level you already cared about. Price makes a new low, the histogram makes a visibly higher low, and it happens at support you marked before the indicator said anything. Divergence alone is weak, but divergence plus a pre-existing level plus a reversal candle is a real trade.
And the two I would ignore. First, any crossover that prints while MACD is hugging the zero line. That is the chop signature from the section above, and it is where nearly all the losing trades live. If the line is oscillating in a tight band around zero, the honest reading is that MACD has nothing to say right now. Second, divergence on its own against a strong trend. Strong trends print divergence after divergence, sometimes three or four in a row, while price keeps going. Betting on the first one to nail the top has historically been one of the most reliable ways to donate money to the market.
Settings for crypto, and the filter that matters more
The default 12-26-9 comes from a different world. The usual story is that those numbers mapped to roughly two weeks, one month, and a week and a half of sessions back when stocks traded six days a week. Crypto trades every hour of every day, so the calendar logic behind the defaults means nothing here, and there is no sacred reason to keep them. On faster timeframes, the one hour chart and below, compressed settings like 8-21-5 are popular because they react sooner. The tradeoff is exactly what you would expect, earlier entries in trends and more whipsaws in chop. No setting fixes the range problem, because the range problem lives in the market rather than in the parameters.
What actually fixes it is a filter from one timeframe up. My plain version, if I am trading signal-line crosses on the one hour chart, I only take longs while the four hour MACD is above zero and only shorts while it is below. A 200 EMA on the higher timeframe gets you similar behavior if you prefer price-based filters. Either way, the filter deletes most of the range trades, because a range on your trading timeframe usually sits inside some larger drift, and the filter keeps you on the side of that drift.
Two smaller habits worth stealing. Exit on the histogram rather than the opposite crossover, since a histogram that peaks and then shrinks for two or three bars usually gets you out well before the full cross, which tends to give back a big chunk of open profit. And write the rules down before you trust them. The whole setup, filter, entry, skip zone near zero, histogram exit, fits in about five lines and backtests cleanly. I run versions of it through the backtester we built at Blockcircle whenever someone insists their custom settings are special, and twenty minutes of data usually settles what hours of chart screenshots never do.
If you take one thing away, run the raw crossover system yourself on whatever you actually trade, look at where the losses cluster, then add the higher timeframe filter and run it again. Watching the difference in your own results teaches more than any explanation of EMAs will. MACD has been around since the late 1970s and it still does its narrow job well. That job is measuring the gap between two moving averages, and nothing about it involves telling you what to do without context.